Roe Et Al., 2018. Mathematics For Sustainability

Texts for Quantitative Critical Thinking
John Roe
Russ deForest
Sara Jamshidi
Mathematics
for Sustainability
Foreword by
Francis Edward Su
apprecia@ucalgary.ca
Series Editors:
Stephen Abbott
Middlebury College, Middlebury, VT, USA
Kristopher Tapp
St. Joseph’s University, Philadelphia, PA, USA
Texts for Quantitative Critical Thinking (TQCT) is a series of undergraduate textbooks, each of
which develops quantitative skills and critical thinking by exploring tools drawn from mathematics
or statistics in the context of real‐world questions. Topics are high sophistication, low prerequisite,
offering students of all disciplines the opportunity to build skills in the understanding, evaluation,
and communication of quantitative information. These are books about mathematical and statistical
thinking, not computation and procedure. Each book explores an application or idea in depth, offering
students in non‐STEM fields a focused, modern context to develop quantitative literacy, while students
with a technical background will gain a broader perspective to bridge beyond procedural proficiency.
Books in Texts for Quantitative Critical Thinking develop writing and communication skills, which
are essential to such cross‐disciplinary discourse and are highly‐transferable for students in any field.
These titles are ideal for use in General Education courses, as well as offering accessible quantitative
enrichment for an independent reader.
More information about this series at http://www.springer.com/series/15949
John Roe • Russ deForest • Sara Jamshidi
Mathematics for Sustainability
John Roe Russ deForest
Department of Mathematics Department of Mathematics
Pennsylvania State University Pennsylvania State University
University Park, PA, USA University Park, PA, USA
Sara Jamshidi
Department of Mathematics
Pennsylvania State University
University Park, PA, USA
ISSN 2523-8647 ISSN 2523-8655 (electronic)

ISBN 978-3-319-76659-1 ISBN 978-3-319-76660-7 (eBook)
https://doi.org/10.1007/978-3-319-76660-7
Library of Congress Control Number: 2018935919
Mathematics Subject Classification (2010): 00A06, 00A69, 00A71, 91A35, 91B76, 97-01, 97F70, 97M99
© Springer International Publishing AG, part of Springer Nature 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or
in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or
by similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even
in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore
free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be
true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or
implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher
remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Cover illustration: Cover image of Lake Erie courtesy of NASA Earth Observatory (Joshua Stevens, using Landsat data from
the U.S. Geological Survey)
Printed on acid-free paper
This Springer imprint is published by the registered company Springer International Publishing AG part of Springer Nature.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents
Contents i
Foreword iii
Before we begin v
0.1 To the Student . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
0.2 To the Instructor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
0.3 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
I Fundamental Concepts 1
1 Measuring 3
1.1 Units and Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Scientific Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3 Estimates, Precision, and Orders of Magnitude . . . . . . . . . . . . . . . . . . . . . 33
1.4 Communicating Quantitative Information . . . . . . . . . . . . . . . . . . . . . . . 44
1.5 Exercises for Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2 Flowing 67
2.1 Stocks, Flows, and Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.2 Energy Stocks and Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.3 Calculating Equilibrium States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
2.4 Energy Flows in the Climate System . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3 Connecting 127
3.1 Networks and Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.2 Networks and Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
3.3 Feedback and Dynamic Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
3.4 The Exponential Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4 Changing 183
4.1 Logarithms and Change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
4.2 Logistic Models and the Limits to Growth . . . . . . . . . . . . . . . . . . . . . . . 206
vi CONTENTS
4.3 Measuring Feedback Strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

4.4 Tipping Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
5 Risking 259
5.1 Understanding Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
5.2 Probabilities and Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
5.3 Expectations and Payoffs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
5.4 Assimilating New Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
6 Deciding 345
6.1 Market Perspectives and Large-Scale Change . . . . . . . . . . . . . . . . . . . . . 348
6.2 The Strange Behavior of Rational People . . . . . . . . . . . . . . . . . . . . . . . 360
6.3 The Tragedy of the Commons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
6.4 After Math: Decision-Making and Ethics . . . . . . . . . . . . . . . . . . . . . . . . 388
II Case Studies 409
7 Case Studies 411

7.1 Mathematics and Persuasive Writing . . . . . . . . . . . . . . . . . . . . . . . . . . 413
7.2 The Changing Risks of Wildfires . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
7.3 Is Recycling Worth It, Really? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
7.4 World Population Growth and the Demographic Transition . . . . . . . . . . . . . . 433
7.5 Genetic Engineering and the Future of Food . . . . . . . . . . . . . . . . . . . . . . 443
7.6 Nuclear Power Is a Commitment to the Future . . . . . . . . . . . . . . . . . . . . . 450
7.7 Using Electricity Efficiently at Home . . . . . . . . . . . . . . . . . . . . . . . . . . 458
7.8 Growth and Payback Time for Solar Energy . . . . . . . . . . . . . . . . . . . . . . 466
7.9 Energy Return on Energy Invested . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
III Resources 481
8 Resources for Student and Instructor 483

8.1 Resources for Further Reading and Writing . . . . . . . . . . . . . . . . . . . . . . 483
8.2 Useful Numbers for Sustainability Calculations . . . . . . . . . . . . . . . . . . . . 483
List of Figures 491
List of Tables 498
Bibliography 501
Index 519
Foreword
I’m excited to introduce this book to you, because it may be different from any math text you’ve read
before. It will change the way you look at the world and enlarge the way you think about mathematics.
No longer will you be just a spectator when people give you quantitative information—you will
become an active participant who can engage and contribute new insights to any discussion. Just
look at the verbs that underlie the chapter titles: measure, flow, connect, change, risk, decide!
Here’s what stands out to me when I read this book: there are many math books that will feed you
knowledge, but it is rare to see a book like this one that will help you cultivate wisdom.
There is a deep difference between knowledge and wisdom. A knowledgeable person may be armed
with facts, but a wise person considers how to act in light of those facts. A knowledgeable person may
think an answer is the end of an investigation, whereas a wise person considers the new questions that
result. And a knowledgeable person might ignore the human element of a problem that a wise person
deems essential to understand. As the authors illustrate, mathematics that pays attention to human
considerations can help you look at the world with a new lens, help you frame important questions,
and help you make wise decisions.
Sustainability asks: how can we be wise stewards of Earth’s resources? One way or another this
question will impinge on some aspect of your life, if it hasn’t already. Sustainability is an economic
concern because resources are limited. Sustainability is a moral concern, because any scarcity of
Earth’s resources will harm the weak and vulnerable first. And sustainability is a scientific concern,
because we may have the power to improve the lives of those who will be affected.
I know that each of the authors shares a deep vocational commitment in bringing this book to you,
and as evidence, I’ll speak personally of the author I have the privilege to know as a friend: John
Roe, a man of deep grace and humility who made this book his highest professional priority while
battling a difficult illness. For him, this project grew out of a conviction and prayerful reflection that
his knowledge as a mathematician and an educator could be channeled into wise action on matters
that will impact us all.
The authors have poured their hearts into this remarkably important and timely book, and I hope
you will engage it with the same fervor. Because it concerns the world you live in, how you will need
to live in it, and the problems that you—yes YOU—can solve so that all of us can live in it well.
Francis Edward Su
Benediktsson-Karwa Professor of Mathematics, Harvey Mudd College
Past President, Mathematical Association of America
vii
Before We Begin. . .
0.1 To the Student

A Letter from the Authors
Dear Student,
The world that you are inheriting is full of bright possibilities—and also of big problems. Many
of the problems center on sustainability questions like “can this (key part of our social or economic
system) last?” or to put it in a way that has a little more math in it, “how long can this last?” For
example, modern society is a profligate consumer of energy, most of which is supplied by fossil fuels:
coal, oil, and natural gas. Fossil fuel supplies, though of vast and unknown size, are limited. How
long can they last? What’s more, there is a strong scientific consensus that the carbon dioxide gas
(also known as CO2 ) released by burning fossil fuels is affecting Earth’s climate, making it more
unstable. How much more CO2 can we emit before climate instability becomes dangerous? These are
big problems. Bright possibilities for addressing them include renewable energy sources like wind and
solar. No doubt you have heard all of these things before—as well as many other news stories about
“sustainability,” both positive and negative.
We started developing the “Mathematics for Sustainability” course, and writing this book, because
of three convictions:
• Many of the key choices that humans will have to make in the twenty-first century are rooted in
sustainability questions. These include choices that we must make together, as citizens, as well
as choices related to individual lifestyles.
• In a democracy, as many people as possible need to participate in well-informed discussion of

these sustainability questions. They are too important to be left to “experts.”
• We may engage with sustainability questions from a wide variety of perspectives, including
scientific, technological, political, ethical, and religious. For many of these discussions, we
need some knowledge of mathematics in order to participate in a well-informed way.
The aim of this book is to help you, the student, gain that mathematical knowledge and the ability
to apply it to sustainability questions.
You may not consider yourself a “math person.” Your studies may center on music or English or art
or education or architecture or agriculture.1 But if you want to find out for yourself what “the numbers
say”—not just to choose which “expert” you prefer to listen to—then this book is for you. Together,
we will find out how to model sustainability on local, regional, and global scales. We will learn about
1 Students in all these majors have succeeded in the course on which this book is based.
ix
x BEFORE WE BEGIN
measurement, flow, connectivity, change, risk, and decision-making. Some of the topics we discuss
will probably be challenging, perhaps even unsettling. Whatever conclusions you reach, this book will
prepare you to think critically about your own and other people’s arguments and to support them with
careful mathematical reasoning.
As citizens in a democracy, you will ultimately be the ones whose decisions will guide your world
toward a sustainable future. We wish you the very best in your studies, and as you participate in
building the sustainable society of the future.
John Roe, Russ deForest, Sara Jamshidi
August 2017
Sustainability—The Key Idea

In spring 2017, our Earth’s human population surpassed 7 12 billion. Here’s a question. What do you
imagine was the population of Earth two thousand years ago, at the beginning of the Common Era?
Demographers (scientists and historians who study population) obviously don’t know an exact
answer to this question. But they are able to make some good estimates, which are in the range of
200 to 300 million people (Section 8.2). That is to say, the number of people on the whole planet
twenty centuries ago was roughly the same as the number in the United States (U.S.) today. Or, to put
it differently, the population of our Earth has increased by twenty-five times over that period.
That population increase has not been a steady one. Most of the growth has occurred in the last
century. And many other measures of human activity follow a similar pattern. Take a look at the
graphs in Figure 1, which are taken from a book by Will Steffen [308]. These graphs show a pattern
of accelerating increase that mathematicians call exponential growth.2 This is important news: some
good (most societies have regarded large families and increased wealth as good, for example) and
some less so (Steffen’s book includes similar curves about pollution and overuse of resources, which
most would regard as bad). Both “goods” and “bads” have been growing exponentially, especially over
the past two hundred years, since the Industrial Revolution got into gear. Can this pattern continue?
Figure 1: Some measures of the “size” of humanity (from [308]).
Some think so. In July 2015, one presidential candidate declared that his objective for the United
States was “4 percent annual growth as far as the eye can see.” That is about the growth rate for the
2 See Section 3.4 for more about this concept.
TO THE STUDENT xi
curves in Figure 1. Others, though, look at similar data and see it differently. “[These] remarkable
charts. . . ,” writes Gus Speth, former dean of the Yale School of Forestry and Environmental Studies,
“reveal the story of humanity’s impact on the natural earth. The pattern is clear: if we could speed up
time, it would seem as if the global economy is crashing against the earth—the Great Collision. And
like the crash of an asteroid, the damage is enormous.” [305]. “Societies are now traveling together
in the midst of unfolding calamity down a path that links two worlds,” he continues. “Behind is the
world we have lost, ahead is the world we are making. . . . The old world, nature’s world, continues, of
course; but we are steadily closing it down, roping it off. It flourishes in our art and literature and in
our imaginations. But it is disappearing.”
The “old world” that Speth describes is a world in which

Earth appears to be huge, teeming with life, abundant, ex-
hilarating and dangerous. Humanity exists on the margin. In
the “old world,” to ask about humanity’s impact on nature
might seem absurd: much more important to worry about
nature’s impact on human beings (diseases? predators? food
shortages?) By contrast, the iconic image of the “new world”
is the Apollo astronaut’s view of Earth: the “blue marble”
(Figure 2), floating in space, gemlike and beautiful, yet cut
down to finite size by human achievement. In this finite world
it makes sense to ask: How long can we keep growing? Have
we already become too big? Can our complex society remain
diverse, active, and productive for an extended period of time?
Or could we overshoot the limits of our resources and then
decline, as many earlier civilizations have done [96]?
This is the sustainability question from which all other such
questions derive. We can put it another way by thinking of
the successive generations of humanity on this planet. Each Figure 2: The “blue marble.”
generation inherits Earth’s resources from its predecessors and
passes them on to its successors. In a lasting or sustainable
society, each generation would leave the world system in as good a shape as it found it; my
generation’s enjoyment of Earth’s bounty would not deprive the next generation (yours) of the
opportunity for similar enjoyment. Thus we arrive at the famous definition3 given in the Brundtland
Report [255] as early as 1987:
Definition 1
Sustainability is the ability of a social, economic, or ecological system to meet the needs
of the present generation without compromising the ability of future generations to meet
their own needs. A process or practice is sustainable to the extent that it contributes to the
sustainability of the social, economic, or ecological systems in that it is embedded.
It is important to recognize that working for sustainability does not mean just trying to keep things
as they are. “Things as they are” include patterns of privilege and inequality that deprive many
members of the present generation of their basic needs. A sustainable world system must continue
to extend the provision of these needs to a growing share of its population—ultimately, to all—even
3 The Bruntland Report in fact defines “sustainable development”; we have slightly modified its language to arrive at a
definition of “sustainability” itself.
xii BEFORE WE BEGIN
as it also works to ensure that these needs can be provided in a way that doesn’t ask future generations
to pay the bill. This double challenge, we believe, will define the century that you live in.
We should also recognize that questions about sustainability and about the needs of the present
and future generations are inherently value-laden. Particularly when we approach these questions on
a global scale, we should expect to encounter different value judgments concerning how human well-
being is defined and what are the basic needs whose provision should be sustained.
Think about it. . .
The Brundtland definition of sustainability, Definition 1, was written in 1987. Yet such
ideas appear many years earlier in the thought of indigenous peoples around the world. For
instance, the Iroquois Confederacy’s “Great Law of Peace,” which is older than the U.S.
Constitution, contains a clause that is often paraphrased as, “In every deliberation, we must
consider the impact on the seventh generation. . . even if it requires having skin as thick as the
bark of a pine.” Thus, Westerners’ recent “discovery” of the notion of sustainability might
be more properly described as a “rediscovery” of ideas that are rooted in many traditions
(including some of our own). How do you feel about this? Why might Western society have
lost touch with the sustainability idea for part of its history?
Sustainability and Resilience

The time scale on which we ask sustainability questions is a
long one—many generations. Some changes to our world are not
perceptible unless we take such a long view. The rate at which
Earth’s climate is changing, measured by global average surface
temperature, is no more than a couple of hundredths of a degree
per year: of little consequence from one year to the next but (as we
will see) hugely significant on longer time scales. But change does
not always come in tiny steps: our world also experiences sudden,
extreme events. (Think of Hurricane Harvey, which flooded Houston
in August 2017, doing damage initially estimated at $60 billion.)
Extreme events are unavoidable: the question is not how to avoid
Figure 3: Marine rescue operation in Galveston,
them, but how quickly we can bounce back from them. This “self-
Texas, following Hurricane Harvey, August 31,
2017. healing” capacity is resilience.
Definition 2
Resilience is the ability of a social or economic system to absorb and respond to a sudden
shock without damage to its core functioning.
Sustainability and resilience are not the same, but they are closely related: a system that is already
near the limits of sustainability may be pushed to collapse by a shock that a more resilient system
would easily have survived. In the same way, an infection that a healthy person would shrug off may
be fatal for someone whose immune system is already compromised by malnutrition. Many historical
examples can be found in [96].
One way to think about the relationship between sustainability and resilience is through the idea of
control. It is tempting to believe that the more elements of our socio-ecological system we can bring
TO THE STUDENT xiii
under human control, the better we will be able to steer it on a safe path and so the more sustainable
it will be. For example, wildfire management in the Western U.S., and elsewhere, focused for many
years on fire suppression: bringing fires under human control. This seemingly allowed development in
wildfire-threatened areas to be lasting, or “sustainable.” Only recently has it become clear that these
measures, which indeed halted many local fires, also set the stage for huge buildups of fuel that make
the forests less resilient to the possibility of a catastrophic burn. Similar ideas apply to flood control
measures like overflow dams and levees. In Section 4.4 we will develop a mathematical perspective
on tipping points—situations in which a system lacks resilience because a small shock can produce
an overwhelming response.
Ecosystem Services
In Speth’s “new world” there is a tight connection between the economy, environmental quality,
human well-being, and the well-being of the entire community of living and nonliving things that we
call Earth’s ecosystem. This requires a significant change in our thinking. From the perspective of the
“old world,” Earth was simply a source of abundance: nature provided. Our “new world” perspective
requires a deeper understanding of what it is that nature provides, and indeed of the fact that human
activity can no longer be so neatly distinguished from “nature” at all. Economic activity and human
prosperity are embedded in the natural world, and they depend on and are ultimately constrained by
the productive capacity of the whole Earth ecosystem (Figure 4)—which itself is constrained by the
rate at which energy arrives from the Sun.
In Figure 4, we should therefore think of the outer-
most oval, which represents the entire ecosystem of the
Earth, as of more or less unchangeable size. That means
the inner ones cannot grow too much. In other words, Earth’s Ecosystem
the definition of a sustainable society implicitly involves

Human Society
the uncomfortable idea of limits: limits on the “goods”
Sustainability
that are available to each generation and also limits on Economy

the capacity of Earth’s ecosystem to absorb the “bads”
generated by human activities. The recognition that such
limits exist sets the stage for the mathematical tools we
develop throughout this book. We need to be able to
interpret and use quantitative information to assess the
size and nature of such limits and so to reach well- Figure 4: Sustainability and limits. Redrawn from [65].
justified decisions about sustainability questions. If we
lack such tools, we’ll be tempted to believe either that there “are no limits” or that if limits exist, they
are so far off that they need be of no concern to us. That may be true in some cases, but in others we
may find that the limits are startlingly close.
The notion of ecosystem services [342] provides one way to conceptualize this dependence of
human activity on the planet’s ecosystems.
Definition 3
Ecosystem services are benefits that human society obtains from Earth’s ecosystems. These
services range from pollination of food plants and provisioning of freshwater and other
resources to climate regulation, soil production, and recreational opportunities.
Specific ecosystem services are not fixed once and for all: human activity may enhance or degrade
the capacity of Earth’s ecosystems to provide these services. Moreover, many ecosystem services
xiv BEFORE WE BEGIN
(such as clean air, clean water, climate regulation) are public goods—they cannot be “owned” in any
practical sense. While some ecosystem services are local or regional (think of the services provided by
a community park or a state forest), many are global in nature (like the pollution-absorbing capacity of
the atmosphere): effective management and stewardship therefore require international cooperation. In
Chapter 6 we develop mathematical tools to study such cooperation. This will allow us to understand
some of the pitfalls and difficulties in reaching mutually beneficial agreements.
Think about it. . .
Gaylord Nelson, former U.S. senator for Wisconsin and founder of Earth Day, wrote in
2002:
The economy is a wholly owned subsidiary of the environment, not the other
way around.
What do you think he means? Do you agree?
About This Book

A few words about the structure of this book. Part I is the longest part of the text. In it, we develop
mathematical tools and apply them to short examples. It is organized into six chapters corresponding
to key concepts that arise in the mathematical study of sustainability: measuring, flowing, connecting,
changing, risking, and deciding. Here is an overview of those concepts:
Chapter 1: Measuring. In this chapter we discuss how to measure, and how to express how big
some quantity is. From the point of view of sustainability, the kind of answer that matters is often not
some absolute number, but a comparison or a level of importance, so we’ll talk about how we might
judge whether some quantity represents something important, and how we might make a decision on
how to respond.
Chapter 2: Flowing. It’s common to talk about the balance of nature. But this image of a “balance”
can suggest something that is static, unchanging. That is not the way natural ecosystems operate.
Individual components of the system are constantly changing, even as the system as a whole maintains
its equilibrium. For example, consider a mountain lake. The water level in the lake may stay the same,
but the actual water in the lake today is not the same as the water that was there yesterday: new water
has arrived through rain and snowmelt, old water has left through runoff and evaporation. It’s the
balance between these various flow processes that keeps the water level—the stock—constant. This
chapter is devoted to exploring these concepts of flow and stock in detail.
Chapter 3: Connecting. Our lives are more interconnected now than at any time in history, and
not just through social media. Rather than most of our food being grown close to where we live, for
example, we have gotten used to obtaining food from supermarkets that are supplied by a production
and transportation network that reaches all over the globe. Energy supply, too, is a vast and complex
network of tanker routes, pipelines, electrical grid connections, and truck deliveries, to mention only
a few. Human-made networks like these are not the only ones: we have also grown much more aware
of the complexity of the natural networks that connect the web of life on earth. In this chapter we will
study the mathematical language that is used to understand these various kinds of networks.
Chapter 4: Changing. In this chapter we will look at examples in which stock-flow systems are
out of equilibrium—that is, how they respond to change. Although we start by studying a simple
model of continuous growth or decay, from the point of view of sustainability the important questions
arise when growth is limited by some external factor (such as the fact that we live on a finite planet).
We’ll study how the strength of feedbacks governs a system’s response to change, and how this leads
to the key idea of a tipping point—a moment when the system “flips” suddenly to a new state. Some
TO THE STUDENT xv
scientists are concerned that Earth’s climate system might be approaching one or more tipping points.
We’ll ask how it might be possible to tell.
Chapter 5: Risking. Any kind of realistic thinking about sustainability must consider likelihoods or
risks. Nuclear power generation does not produce any greenhouse gas emissions, and it is constant and
reliable, but what about the possibility of a catastrophic accident? This question asks us to balance the
near-certainty of a steady benefit against a small risk of disaster. It is hard to do so without numbers.
How likely are you to die from a nuclear reactor meltdown? The math involved in working with these
sorts of questions is called probability and statistics. In this chapter, we are going to learn about these
techniques and how they can help us make good decisions when faced with limited knowledge and
uncertain outcomes.
Finally, Chapter 6: Deciding is where the rubber meets the road. In the end, you and your
generation are going to have to make some sustainability decisions. These will range from the
personal, through the local, to the national and global. Human behavior does not always follow the
“rational” principles discussed in Chapter 5. Even if mine always did, I am not the only person
involved. Other people’s decisions interact with mine in a complicated way—we are in fact a
network of decision-makers, with no final authority. This decision-making interaction can be studied
mathematically, which we will do. The chapter concludes, however, with an extended reflection
on how our sustainability decisions can never be purely mathematical, but must also engage our
fundamental personal and ethical commitments.
In Part II (Chapter 7) of the book we provide a collection of Case Studies in which we apply
the mathematical tools developed in Part I to answer particular questions related to sustainability and
to explore extended examples. We believe that it is important that you, the student, learn to write
extended pieces of this sort, and a student writing requirement has been a major part of the course
on which this book is based. Why? This is how you build the ability to assess, develop, and present
quantitative evidence in support of your own ideas. These skills are vital to you as a future leader, as
an engaged citizen, and as an effective advocate for the things you care about. We don’t want you just
to learn mathematical techniques but also to be able to incorporate them in extended and persuasive
written pieces—pieces that might be published in a course blog, in a student or hometown newspaper,
or even on a national platform.
Finally, Part III (Chapter 8) of the book contains reference material: suggestions for further
reading, tables of useful data, and the list of figures, bibliography, and index. We suggest checking
Part III regularly, especially if you need some numerical information (such as the heat capacity of
water, or the amount of sunlight that falls on Earth, or the planet’s estimated coal reserves) to help you
answer one of the exercises or formulate an argument in one of your more extended written pieces.
xvi BEFORE WE BEGIN
Online Resources
This book has a website, http://math-for-sustainability.com, as well as an email address for
comments and suggestions, comments@math-for-sustainability.com. On the website you will
find many different kinds of supporting materials:
• Hints and solutions to selected exercises.
• Online calculators and examples. Especially later in the book we will describe many models for
sustainability-related processes that change over time (such models are called dynamic). The
best way to visualize such a model is also dynamic, like a video clip rather than a collection of
snapshots. That can’t be done on paper, but it can be done online, and the website uses several
different modeling tools to help you see the way our dynamic models evolve over time.
• Extended discussions of particular topics (refreshing the case studies in part II of the book,
some of which may become outdated as technology advances or the environment changes).
If you publish a written piece in a newspaper somewhere, please write to us (using the email
address above) and tell us! We’ll be happy to link to your writing from the book’s website.
• Corrections or updates. We’ve tried hard to make this the best book possible. But there are
bound to be some mistakes left. What’s more, some information will simply become outdated.
We’ll keep a list of corrections or updates on the website. Again, if you notice an apparent
mistake, please email us. You’ll be helping many future students by doing so.
Conclusion
We hope that through this book you will gain a clearer under-
standing of the sustainability issues that we humans face and of
some choices that we need to make. But let’s be clear: mathematics
cannot make these choices for us. As remarked above, the choices
human beings make on such fundamental questions reflect their
deepest ethical and personal commitments (compare Figure 5).
What mathematics can do, though, is to inform our choices by
making their likely consequences clearer. It can help us prioritize
issues by ranking them in terms of the relative size of the risk
they pose and the relative severity of their potential outcomes.
Understanding mathematics can help us avoid falling for some
plausible-looking “solutions” that really don’t achieve much or
Figure 5: Pope Francis has said: “These ancient
stories...bear witness to a conviction which we are even harmful. Finally, mathematics carries its own values also,
today share: that everything is interconnected, and like communicating clearly, reasoning logically, and considering
that genuine care for our own lives and our rela- all possibilities. These values, as well as the specific content of
tionships to nature is inseparable from . . . justice
and faithfulness to others” [123].
mathematics, can help us all in the decisions that we will all have
to make together.
TO THE STUDENT xvii
Summary of Ideas: To the Student
• Sustainability refers to the ability of a social or economic system to keep functioning

without degrading its environment—to provide for its own needs and also preserve
the ability of future generations to provide for their needs.
• Resilience refers to the ability of a social or economic system to “self-heal”—to
recover from a disruptive event.
• Ecosystem services refer to the benefits that society receives from Earth’s ecosys-
tems.
• Many questions about sustainability and resilience involve measurement, change,
connection, and risk, all of which can be expressed in the language of mathematics.
• This book introduces some of the mathematical ideas that are helpful in making
decisions that involve sustainability.
xviii BEFORE WE BEGIN
0.2 To the Instructor

If you want to make a course interesting, then you should study something of interest
[356].
This text supports a course that is aimed at college students—many thousands of them, in our large
universities—who would not describe themselves as “mathematicians” or “scientists” but who need
to take at least one course that supports quantitative literacy as part of their degree requirements.
Often such students have found themselves steered into courses in the precalculus sequence: courses
that may be excellent preparation for future scientists and engineers, but that fail to catch a student’s
attention as their last experience of mathematics. One of us sometimes asks such students, “Would
you rather learn the quadratic formula or would you rather save the world?” This book is for those
students who would like to save the world, or at least take a step in that direction. Similarly, it is for
those instructors who would like to “teach as if life matters” [328], or at least take a step toward using
the mathematics classroom to help students think more clearly about some of the issues that are only
going to become of increasing importance over the twenty-first century.
Each instructor will, of course, use this book in the way that they see fit. However, one of
our primary goals is to advance student skill in quantitative literacy. A required student writing
component has played an essential role in accomplishing this goal and has accounted for over one-
third of the total grade in the course as we have taught it. The Association of American Colleges and
Universities provides the following definition for quantitative literacy [249]:
Quantitative Literacy (QL) is a “habit of mind,” competency, and comfort working

with numerical data. Individuals with strong QL-skills possess the ability to reason
and solve quantitative problems from a wide array of authentic contexts and everyday
life situations. They understand and can create sophisticated arguments supported by
quantitative evidence and they can clearly communicate those arguments in a variety
of formats.
Because of its intended audience, the book does not require any mathematics beyond high school
algebra; the most complicated idea that appears is a fourth root, which shows up in a few places in
Section 2.4. In particular, no calculus is required to read and study this book. If the instructor does
know calculus, though, they will find that it provides background to the chapters on “Flowing” and
“Changing,” and may even find it helpful—for themselves, not for the students—to investigate how
our presentation can be translated into the classical language of differential equations.
Nor does the book require (from you, the instructor) a great deal of sustainability-specific
mathematical background. Naturally, the more you know, the more you will be able to help your
students, but in the end what we’re presenting in Part I of the book is a self-contained set of
mathematical techniques, and if you learn about them from the text, that will be fine. Should you
wish to pursue the material further, the reading suggested in Section 8.1 gives some possible starting
points.
The book’s website is at http://math-for-sustainability.com. As well as the student-
oriented material described in the previous section, this website contains additional resources and
suggestions specifically for instructors. These include a quite specific description of the writing
component of the course as we have taught it. You’re welcome to use this model exactly as it is,
to adapt it, or to do something entirely different. We do believe, however, that this book will be
most effective if it is used in conjunction with a requirement for some student response in the form
of extended writing. Our experience suggests that most students are glad to have the opportunity
to integrate their mathematical learning with social and environmental concerns and to express
themselves in this way.
ACKNOWLEDGMENTS xix
0.3 Acknowledgments
A book like this does not come into being without the support and critique of many people. We are
grateful to all the many students who have worked through different versions of the book in their
Math 33 class at Penn State, and for the comments and suggestions they have provided. We’re also
grateful for the advice of those who have read all or part of the manuscript, including Peter Buckland,
David Hunter, Marta Mendoza, Liane Roe, and the anonymous readers who reviewed the manuscript
on behalf of Springer. Numerous colleagues from Penn State and elsewhere have given generously of
their time to make special presentations in one or more Math 33 classes; we are deeply grateful to all
of you. Finally, we especially want to thank Francis Su for the generous and impassioned foreword
that he has contributed to the book.
• John Roe writes: I want to express my thanks to Karl “Baba” Bralich, to Fletcher Harper, to
Byron Smith, to Christopher Uhl, and to Marty Walter, for their various challenges to embrace
this work with my whole self. Marty’s book Mathematics for the Environment [340] helped me
see that something like this volume would be possible. I am deeply grateful to my undergraduate
research assistant for the start of the project, Kaley Weinstein, whose flow of ideas was an
inspiration. And I want to humbly acknowledge the support and love of my whole family, who
now and ever have meant more than I can possibly say.
• Russ deForest writes: Foremost, I would like to express my gratitude to John Roe for inviting
me to be a part of this project, for his kind leadership, and for his deep and personal commitment
to the ideals of general education. I would like to thank Rachel Hoellman for her essential role
as a peer assistant and tutor through multiple iterations and refinements of the accompanying
course, and to Matt Lupold for his current work in this position. Finally, I would like to thank
the many engaged students who make teaching a joyous effort.
• Sara Jamshidi writes: I am inexpressibly grateful to John Roe for giving me an opportunity
to be part of this important project; he has given me an expansive understanding of all that a
mathematician can be and I will carry that with me for the rest of my career and life. I am
thankful for my friends Duane Graysay, Shiv Karunakaran, and Monica Smith Karunakaran,
who have taught me so much about mathematics education. I am also indebted to my partner,
Aleksey Zelenberg, who provided invaluable support and feedback for my work here, and to
my father, Morid Jamshidi, who was the first person to show me that mathematics can be used
in civics and ethics. Finally, I would like to thank you, the reader, for picking up this book
and endeavoring to better understand sustainability through a mathematical lens; I hope this
becomes an empowering tool in your life.
Part I
Fundamental Concepts
C HAPTER 1
Measuring
Think about this quotation from a BBC news broadcast:
Archaeologists announced Friday that they have discovered human footprints in England
that are between 800,000 and 1 million years old—the most ancient found outside Africa,
and the earliest evidence of human life in northern Europe [188].
We know 1 million years is a long time, but “a long time” is

subjective. Your friend might say that she sat in traffic for “a long
time” because she drove 20 miles in one hour, but one hour does
not compare to 1 million years. Now suppose your friend instead
said it took her one hour to travel 3 million centimeters. How do
we interpret that? Was that a large or small distance to traverse in
that amount of time?
The most basic mathematical question we can ask is: how big?
But there is (or ought to be) an immediate comeback in any real-
world situation: compared to what? Is a million a big number?
A million dollars is a lot of dollars for an ordinary person. But
a million molecules of carbon dioxide is a tiny amount for most
purposes, and even a million dollars is a rounding error if you are
thinking about the total of the U.S. federal budget. What about
a million centimeters? Could you walk that far in a day? In a
week? In a month? Not just the numbers but also the units matter:
a million centimeters might be a manageable distance to walk Figure 1: Measuring instruments.
(we’ll see!), but we can be quite sure that a million miles is not.
In this chapter we will review basic information about numbers,
units, measurements, and comparisons. We will study scientific notation, which is a convenient way to
work with very large and very small numbers, and we will look at some of the most effective ways to
communicate numerical information in human terms. We’ll also learn some of the skills of estimation:
how to get a useful rough idea of the size of some quantity, even when an exact answer is not available.
© Springer International Publishing AG, part of Springer Nature 2018 3

J. Roe et al., Mathematics for Sustainability, Texts for Quantitative Critical Thinking,
https://doi.org/10.1007/978-3-319-76660-7_1
4 CHAPTER 1. MEASURING
1.1 Units and Measurement
Objectives
I can identify the number part and the unit part of a physical measurement.
I can keep track of units throughout a calculation.
I can work with pure numbers and their percentage equivalents.
I am familiar with standard units of time, distance, weight (or mass), and temperature.
I can convert the same measurement into different units.
I can communicate the meaning of a measurement by relating it to everyday human
experience.
1.1.1 Number Part and Unit Part

Let’s consider some simple examples of measurements:
(a) The fuel tank of my car holds 13 gallons.
(b) We need 5 pounds of potatoes.
(c) Today’s high temperature is predicted to be 91 degrees Fahrenheit.
(d) The distance by road from New York City to San Francisco is about 2900 miles.
(e) Abraham Lincoln took approximately 2 minutes to deliver the Gettysburg Address.
(f) The Hoover Dam is 726 feet high.
(g) The area of Lake Mead (the lake impounded by the Hoover Dam) is 640 square kilometers.
As you can see from these examples, a physical measurement is made up of two parts:
• the number part—like “13” or “5” or “726”;
• the unit part—“gallons” or “pounds” or “feet.”
It is only the number and the unit part together that make a complete measurement. If my gas tank
has a 13-gallon capacity, and I for some reason decide to measure in teaspoons instead, I could say
that my tank holds about 10,000 teaspoons. Both “13 gallons” and “10,000 teaspoons” are complete
measurements, though the first version is much more useful for most purposes. But it would make no
sense just to say “The capacity of my gas tank is 13.” Without the unit part (gallons), the number part
(13) does not tell us anything.
Critical Thinking
When you read an article that gives numerical information, always ask, “What are the
units?” If no units are given, the information is meaningless.
1.1. UNITS AND MEASUREMENT 5
Problem 1: Identify the number part and the unit part in the following
measurements:
(i) The radius of the Earth is approximately 4,000 miles.
(ii) At full takeoff power a Boeing 747 burns over 3 gallons of jet fuel
per second.
(iii) The current U.S. population is about 320 million.
Figure 2: Boeing 747 at takeoff.
Solution: In example (i) it is clear that the number part is 4,000 and the unit part is “miles.”
In example (ii), the number part is 3, but what are the units? The “3” refers to a rate of fuel
consumption, and the units in which this is measured are “gallons per second,” which we may
abbreviate as “ gal/ sec.” Notice that the units are not gallons alone, or seconds alone, but the
combination “gallons per second.” 1
Example (iii) looks at first as though it does not have any units. To see what the units are, try to
expand the statement a bit: ask yourself, “the current U.S. population of what are we describing?”
Clearly the answer is “human beings,” and that tells us the unit part of our measurement: “human
beings” or “people.”
Remark 1: When describing the unit part of a measurement, we don’t distinguish between singular
and plural. Thus, in both “1 inch” and “2 inches,” the units are “inches”; in “1 person” and “300
people” the units are “people” (or “persons” if you prefer). Some units have standard abbreviations:
for example, “2 inches” can also be written “2 in.”
1.1.2 The Unit-Factor Method

Suppose that my car’s gas mileage is 33 miles per gallon. And suppose also that, as in the first example
above, my car’s tank holds 13 gallons. How far can I travel on a full tank?
You probably know that to obtain the answer, we need to multiply 13 by 33. But did you know that
the multiplication can be done with the units as well? Like this:
33 mi
13
gal
× = 13 × 33 mi = 429 mi.
1gal

Using the abbreviations “mi” for miles and “gal” for gallons, the tank capacity is 13 gal, and the fuel
33 mi
consumption is 33 mi/ gal, which we write as a fraction . When we multiply these two together
1 gal
the gallons cancel, leaving the answer (with the correct units) as 429 mi, that is, 429 miles.
This is an example of the unit-factor method.
Definition 1
Using the unit-factor method, whenever we multiply or divide physical quantities, we must
multiply or divide their number parts and their unit parts.
The unit-factor method is valuable in many kinds of problems. It’s especially helpful when we deal
with problems involving unit conversions, like the one below.
1 You may also see this abbreviated as “ gal sec−1 ” in some books; this notation is not wrong, but we’ll try to avoid it.
Problem 2: There are 1,760 yards in a mile, and there are 36 inches in a yard. Using this information,
express one million inches in miles.
Solution: The key idea is that conversion problems can be expressed in terms of “multiplying by 1”
in different ways—and that multiplying a quantity by 1 does not change that quantity. For example,
the statement “there are 36 inches in a yard,” part of the information given in the problem, can be
re-expressed as
1 yd 36 in
= 1 or = 1. (1)
36 in 1 yd
Both ways of writing the fraction are correct; we have to figure out which way will be more helpful to
us. Now a million inches multiplied by 1 is still a million inches; but if we multiply a million inches
by the left-hand expression for 1 in the display (labeled (1)) above, we get the useful fact that
1 yd 1,000,000 yd
1,000,000 in = 1,000,000
in × = ≈ 27,780 yd.
36in 36

(The symbol “≈” means “is approximately equal to.”) We can use the same idea a second time to
1 mi
convert from yards to miles, using the fraction = 1:
1,760 yd
1 mi 27,780 mi
27,780 yd = 27,780
yd
× = ≈ 15.8 mi.
1,760
yd
1,760
The answer to our problem, therefore, is approximately 15.8 miles. If we wanted to shorten our work,
we could combine two steps into one and write
1 yd 1 mi 1,000,000
1,000,000 in = 1,000,000
in × × = mi ≈ 15.8 mi.
36in 1,760
yd 36 × 1,760

This is also a correct solution.
The unit-factor method is a great help in conversion problems because keeping track of the units
automatically lets us know when to multiply and when to divide. For instance, in the first line of our
solution we multiplied by 1 yd/36 in = 1 to convert our measurement from inches to yards. Only this
form of the conversion factor allows the “inches” to cancel, and that cancellation is the signal that tells
us we are heading in the right direction. Watch how this works in our next example:
Problem 3: Beefy Acres is an imaginary cow/calf farming operation located in the Southeast, used
as an example in a pamphlet published by the Natural Resources Conservation Service (a division of
the U.S. Department of Agriculture) [235]. Beefy Acres has 20 acres of pastureland, and each acre
produces, on average, 9,500 pounds of forage per year. To thrive, each cow requires approximately
17,500 pounds of forage per year. What is the maximum number of cows that can be supported
sustainably on Beefy Acres?
Solution: We have three items of information given to us. Let’s express these in terms of number part
and unit part, where our basic units are “acres” (of pastureland), “cows” (this is a very reasonable
unit for working this problem—units do not have to be just mathy things), “pounds” (of forage; for
historical reasons the abbreviation for pounds is “lb”), and “years.” Then our information is the pasture
lb
area (20 ac), the amount of forage production per acre per year (9,500 ), and the amount of
ac yr
lb
forage consumption per cow per year (17,500 ). We need to combine these quantities using
cow yr
multiplication and division to get an answer whose units are cows; the acres, pounds, and years have
to cancel. There is only one way to do this:
1 cowyr 9,500 lb 9,500 × 20

× 20
ac = cow ≈ 11 cows.

×
17,500
lb 1
ac
yr 17,500

The answer is approximately 11 cows.
How do we arrive at the correct way to arrange this computation? We start by remembering that our
objective is to arrive at a solution whose units are cows. Only one of the data involves the unit “cow”
lb
at all, and that is the grazing requirement 17,500 . But this piece of information has “cow” in
cow yr
the denominator (the “downstairs” part of the fraction), whereas to get an answer in units of cows we
are going to need “cow” in the numerator (the “upstairs” part). So we consider the reciprocal
1 cow yr
.
17,500 lb
This has “cow” in the numerator as we wanted, but it also involves units of pounds and years, which
we will need to cancel by multiplying by other pieces of information. If we multiply by the yield per
acre, the pounds and years will cancel:
1 cow
yr 9,500 lb 9,500 cow
× = .
17,500
lb
1 ac
r
y 17,500 ac
This is a meaningful quantity—it tells us how many cows can graze on any given acreage of land,
with units “cows per acre”—but it is not our final answer, whose units must be “cows.” To cancel the
“acre” unit, we multiply by the land area, thus arriving at the final answer
1 cow
yr 9,500 lb 9,500 × 20
× 20
ac = cow ≈ 11 cows.

×
17,500
lb 1
ac
yr 17,500

1.1.3 Standard Units

We already have some knowledge of the standard units for time and distance, area and volume, and
other common physical measurements. Let’s review them.
Units of Time
Each of us is familiar with the idea of time. In the introduction we saw how sustainability questions
can explicitly involve the measurement of time (“how long can this last?”). Table 1 lists some common
units of time.
Table 1: Units of Time

Unit (abbr) Definition Example
Second (s) – Snapping your fingers
Minute (min) 60 s Singing five verses of “Row, Row, Row Your Boat”
Hour (hr) 60 min About the time you spend during a casual lunch
Day (day) 24 hr Time between one sunrise and the next
Week (wk) 7 day About the time it takes to ship a package by regular mail
from the U.S. to Europe
Year (yr) 365 day The time from one birthday to the next
Century 100 yr About 4 generations in a family (from you to your great-
grandmother)
Units of Length
Units of length come from one of two systems of measurement: the U.S. system and the metric
system (used in much of the rest of the world). We will review both systems of measurement and
discuss how to go back and forth between the two systems.
First, let us list some U.S. measurements of length. These measurements will be more familiar to
you if you grew up in the United States.
Table 2: Units of Length (U.S.)

Inch (in) – About the length of a paper clip
Foot (ft) 12 in About the length of a large human foot
Yard (yd) 3 ft About the width of a single (twin) bed
Mile (mi) 1,760 yd About the distance traveled walking for 20 minutes
If you grew up outside the United States, you will be more familiar with metric units of length.
Some of these are listed in the table below.
Table 3: Units of Length (Metric)

1
Micron ( µm) 1,000,000 m Size of a small bacterium
1
Millimeter (mm) 1,000 m About the width of a pencil tip
1
Centimeter (cm) 100 m About the length of a carpenter ant
Meter (m) – A little more than the width of a single (twin) bed
Kilometer (km) 1000 m About the distance traveled walking for 12 minutes
Notice that the metric units all are related by powers of 10, like 10, 100, 1000 and their reciprocals
1 1 1
10 , 100 , 1000 .
Moreover, the names of the units all have a standardized form, made up of a prefix
applied to the basic unit “meter.” You can find a more extensive list of these prefixes, and the powers
of 10 that they represent, in the “Useful Data” section of this book (Table 2 on page 486).
Problem 4: Use the information in the tables to determine how long would it take you to walk 3
million centimeters.
Solution: From the data in Table 3, we know that it takes about 12 minutes to walk 1 kilometer. We
can express this rate of progress as
12 min
.
1 km
We want to use this information to find out how long it takes to travel 3 million centimeters. Since
our rate-of-progress information uses kilometers rather than centimeters, we convert one to the other,
1
using the conversion information 1 cm = 100 m and 1 km = 1, 000 m provided by the table. As we
learned in the solution to Problem 2 on page 6, the way to do this is to re-express the conversions as
1m
fractions equal to 1 (such as = 1, for example). Thus we can obtain the travel time:
100 cm
1m 1
km
12 min
3,000,000
cm
× × × = 360 min.

100
cm
1000m
1
km
1m
1 ft
1m 1 m2
1 ft 1 ft2
Figure 3: Illustrating the formation of units of area.
Our answer is 360 minutes. That’s a lot of minutes, and it will be easier to understand if we convert it
to hours, using the fact that 60 minutes make 1 hour:
360 × 1 hr = 6 hr.
min
60
min

Based on our work above, then, it would take about 6 hours to walk 3 million centimeters. This is
an example of expressing a measurement in human terms—that is, by reference to familiar human
experience. Instead of the mysterious “3 million centimeters,” we can now communicate the same
information by saying “about six hours’ walk.” This alternative way of expressing our measurement
can be grasped directly, without scientific knowledge.2
Remark 2: We separated out the various conversions when we used the unit-factor method in the
above solution. It would be equally correct, however, to combine them all in one line:
1m 1
km
12
min
1 hr
3,000,000
cm
× × × × = 6 hr,

100
cm 1,000
m
1
km
60
min
as we did in the solution to Problem 3 on page 6.
Units of Area and Volume

Area is a unit of measurement created from length. Units like “square feet” ( ft2 ) and “square meters”
( m2 ) refer to the area made by multiplying the dimensions. Figure 3 is a diagram demonstrating this
idea. From any unit for length, we can get a corresponding unit for area.
Let’s work some problems involving unit conversions for areas. As we’ll see, we will need to take
the squares of the conversion ratios to make the units come out right. This works because conversions
in the unit-factor method are simply ways of expressing the number 1, and the square of 1 is just 1
again.
Problem 5: What is a square foot in square inches?
12 in
Solution: There are 12 inches in a foot; in terms of the unit-factor method, = 1. So how many
1 ft
2
square inches in a square foot, 1 ft = 1 ft × 1 ft? We can work this out by the unit-factor method
12 in 12 in

1 ft × 1 ft = 1 ft × × 1 ft × = 12 in × 12 in = 144 in2 .
1 ft 1 ft
2 We’ll develop this idea further in Section 1.1.5.
2
12 in

A shorter way to approach this calculation is to take the square of the conversion ratio, =
1 ft
(12 in)2
. This is still equal to 1 (because the square of 1 is still 1), so we can write
(1 ft)2
(12 in)2 144 in2

1 ft2 = 1 ft2 × 2
= 1 2
ft × = 144 in2 ,
(1 ft) 1ft
2
getting the same answer as before. Look carefully at both approaches to make sure you understand
why they give the same result.
Problem 6: Express 15 square centimeters in square meters.
Solution: Working as in the problem above,
(1 m)2 1 m2
15 cm2 = 15 cm2 × = 15 cm
2
× = 0.0015 m2 .
(100 cm)2 2

10,000
cm
The answer is 0.0015 square meters.

There are a couple of additional units of area that are worth knowing. These are particularly
associated with agriculture.
Table 4: U.S. Units of Area

Acre 4840 yd2 Roughly the area of a (U.S.) football field, without the
end zones.
Hectare 104 m2 About 2.5 acres
Volume is also a unit of measurement created from length. This time, though, we are considering
cubes instead of squares (Figure 4). Units like “cubic inches” ( in3 ) and “cubic meters” ( m3 ) refer to
measurement of volume. The technique for carrying out conversions is the same.
Problem 7: What is 1200 cubic inches in cubic feet?
1 ft
Solution: We now need to take the cube of the conversion factor = 1. We obtain
12 in
(1 ft)3 1 ft3
1,200 in3 = 1,200 in3 × = 1,200 in
3
× ≈ 0.7 ft3 .
(12 in)3 3

1,728 in
The answer is approximately 0.7 cubic feet.
Example 1: As well as the “cubic” units for volume, traditional systems of measurement have come
up with many other volume units, often specialized to a particular trade or profession: bushels of
wheat, gallons of milk, hogsheads of ale, and so on. Table 5 on the opposite page gives a few such
volume units worth knowing
Problem 8: How many gallons are in a cubic foot?
1m
1 ft
1m 1 m3
1 ft 1 ft3
1 ft
1m
Figure 4: Illustrating the formation of units of volume.
Table 5: Units of Volume

Gallon ( gal) 231 in3 Large container of milk
1
Pint ( pt) 8 gal Glass of beer
Barrel, of oil ( bbl) 42 gal Barrel about 3 ft high, 1 12 ft diameter
1
Teaspoon (tsp) 96 pt The smallest spoon in a set of ordinary silverware
Liter ( L) 1000 cm3 or 0.0001 m3 A little over 2 pints
Solution: One gallon is equal to 231 cubic inches (as defined above), and one cubic foot is equal to
123 = 1,728 cubic inches. Therefore,
1 gal
1 ft3 = 1,728 3
in × ≈ 7.5 gal.
3
231in
Problem 9: Approximately how many cubic centimeters make a teaspoon?
Solution: This problem requires us to relate the U.S. and metric systems of measurement for volume.
We can use the unit-factor method and the information in the table above:
1 1 1 L 1,000 cm3
1 tsp = pt ≈ pt × × ≈ 5 cm3 .
96 96 2 pt 1L
Thus, a teaspoon is approximately equal to 5 cubic centimeters.
Units of Weight or Mass
Table 6: Units of Mass (U.S.)

Ounce ( oz) – About the mass of an AA battery
Pound ( lb) 16 oz About the mass of a small bottle of soda
Ton ( t) 2,000 lb About the mass of a subcompact car
In this text, we are going to use the terms “weight” and “mass” interchangeably. Strictly speaking
this is not accurate—the “mass” of an object refers to the amount of “stuff” (matter) that it is made
of, and “weight” refers to the amount of pull that gravity exerts on that mass. But so long as we stay
on the surface of the earth, gravity is pretty much constant, and therefore the two terms “weight” and
“mass” can be taken to refer to the same thing. As earthbound people, then, we will not sweat the
distinction.
Like distances, masses have two unit systems: a U.S. system and a metric system. If you grew up
in the United States, you may be more familiar with the mass units in Table 6 on the previous page.
But if you grew up outside of the United States, you are likely more familiar with the measurements
in Table 7 instead.
Table 7: Units of Mass (Metric)

Gram ( g) – About the mass of a paper clip
Kilogram ( kg) 1,000 g About the mass of a hardback book
Tonne ( T) 1,000 kg About the mass of a subcompact car
Notice that a U.S. “ton” and a metric “tonne” are not quite the same. But they are pretty close
(within about 10 percent)—close enough that the difference will not matter for most of our purposes.
Unit Conversions
There are many situations in which we may need to convert a measurement from one unit to another.
These conversions could be within one unit system (for example, expressing 3 million centimeters
as 30 kilometers) or from one unit system to another (for example, expressing 30 kilometers as just
under 19 miles).
Here is a table of some useful conversions. (More extensive and precise tables can be found in
Section 8.2.)
Table 8: Unit Conversions

Metric Unit U.S. Conversion U.S. Unit Metric Conversion
1 cm 0.39 in 1 in 2.54 cm
1m 3.3 ft 1 ft 0.3 m
1 km 0.62 mi 1 mi 1.61 km
1g 0.035 oz 1 oz 28 g
1 kg 2.2 lb 1 lb 0.45 kg
1T 1.1 t 1t 0.91 T
1 gal 3.8 L 1L 0.26 gal
We have already seen how the unit-factor method allows us to handle these conversions efficiently.
Let’s do a couple more examples as a reminder.
Problem 10: The weight limit for a checked bag on Untied Airlines is 50 lb. If my bag weighs 32 kg,
will I be able to check it without paying the overweight charge?
Solution: The fact that 32 is less than 50 does not answer this question! We must convert 32 kg to
pounds. To do this, we use the fact (from Table 8) that 1 kg equals 2.2 lb. Using the unit-factor method
2.2 lb
we express this by saying that the fraction equals 1. Therefore,
1 kg
2.2 lb
32 kg = 32
k
g× ≈ 70 lb.
1
kg
It looks as though I definitely will have to pay an overweight charge!
Problem 11: I am pulled over on the freeway for driving at a speed of 100 feet per second. If the
speed limit is 65 miles per hour, was I speeding or not?
Solution: We need to express 100 feet per second in miles per hour. Using Tables 1 and 2 (pages 7–8),
ft ft 1 yd 1 mi 60 sec
60 min
100 × 60 × 60 mi mi
100 = 100 × × × × = ≈ 68 .
sec sec
3 ft 1,760yd 1min 1 hr 3 × 1,760 hr hr
So I was speeding, though not by very much.
The moral is: be mindful of your units! Forgetting to convert to the appropriate unit system is an easy
mistake that can lead to disastrous results—such as the loss of a spacecraft! According to news reports
from the late 1990s, “NASA lost a $125 million Mars orbiter because a Lockheed Martin engineering
team used English units of measurement while the agency’s team used the more conventional metric
system for a key spacecraft operation” [199].
Here’s a slightly different example
Problem 12: Write the height “5 feet 7 inches” in centimeters.
Solution: “5 feet 7 inches” means an addition: five feet plus seven inches. But when we want to add
two quantities, they must be expressed in the same units. In this example, we need to re-express the
five feet in units of inches before we add the seven inches:
12 in

5 ft + 7 in = 5 ft × + 7 in = 60 in + 7 in = 67 in.
1 ft
Now that we have expressed the length in units of inches, we can convert to centimeters:
1 cm
67
in × ≈ 170 cm.
0.39in

Remark 3: A calculator will give 67/0.39 as 171.7048 . . . . So why did we “round off” the answer
above to “approximately 170 centimeters”? To understand this, take a look at the discussion of
precision in Section 1.3.1.
Units of Temperature
Our final example of unit conversions concerns the measurement of temperature. In the United States,
temperatures are measured using the Fahrenheit scale. On this scale, the freezing point of water is
32 ◦ F and the boiling point of water is 212 ◦ F, so that there are 180 Fahrenheit degrees between
freezing and boiling.
In most other countries, the Celsius scale is used. On the Celsius scale, the freezing point of water
is 0 ◦ C and the boiling point of water is 100 ◦ C. Thus there are 100 Celsius degrees between freezing
and boiling.
Remark 4: Converting between temperature scales is more complicated than converting between unit
systems for the other kinds of measurements (length, time, mass, and so on) that we have discussed
so far. The reason is that not only do the units have different sizes (one Celsius degree represents
more of a temperature jump than one Fahrenheit degree) but also the zero points of the two scales are
different. That problem does not come up for other sorts of measurements: zero feet and zero meters
both represent the same length, zero! The differences between Fahrenheit and Celsius are illustrated
in Figure 5.
What’s more, neither the Celsius nor the Fahrenheit scales have their zero points
rooted in any fundamental physical reality. (According to legend, 0 ◦ F was simply
the coldest temperature that Fahrenheit could reach using the technology available
to him at the time.) Modern physics tells us, though, that heat is the result of the
disordered motion of atoms and molecules, and this means that there is truly a
“coldest possible” temperature, when all atoms and molecules would be at rest.
This absolute zero of temperature is much colder than anything in our ordinary
experience: it is about −273 ◦ C, or −460 ◦ F. In physics calculations the Kelvin
temperature scale is used: Kelvin temperature is simply equal to Celsius temperature
plus 273, so that absolute zero is 0 K, water freezes at 273 K, and water boils at
373 K. Table 9 gives the algebraic formulas for converting between these three
temperature scales.
Table 9: Temperature Scale Conversions

◦C ◦F K
Figure 5: Celsius and
Fahrenheit comparison. ◦C C =C C = 59 (F − 32) C = K − 273
◦F F = 32 + 95 C F =F F = 95 K − 460
K K = C + 273 K = 59 F + 255 K=K
For instance, if you want to convert a temperature from Fahrenheit to Celsius, let the Fahrenheit
temperature be F. Look in the table to the intersection of the ◦ F column and the ◦ C row, where you
will find the algebraic formula C = 59 (F − 32). This tells you how to compute the Celsius temperature
C in terms of the Fahrenheit temperature.
Problem 13: Normal human body temperature is about 98 ◦ F. Express this on the Kelvin scale.
Solution: Looking in the table above, we see the formula K = 59 F + 255 to convert from Fahrenheit
to Kelvin. Plugging in F = 98 ◦ F, we obtain
5 × 98
K= + 255 ≈ 55 + 255 = 320 K
9
for the Kelvin equivalent.
1.1.4 Percentages and Other Pure Numbers

We’ve stressed that physical quantities have both a number part and a unit part, but we sometimes
have a use for pure numbers. These are numerical quantities that have no unit. The most familiar
example is a percentage.
Example 2: In 2015 the population of California was estimated to be 39 million people while the
population of the United States as a whole was 320 million people [261]. Let’s express the California
population as a percentage of the total U.S. population. A percentage is a proportion; we divide the
population of California by the total population,
39,000,000 people
12
≈ 0.12 = = 12%.
320,000,000 people
100
The units (people) cancel, leaving us with the pure number 0.12, which we can also write as 12%
(read as 12 percent).
Remark 5: Percent comes from the Latin per centum, meaning out of a hundred; 12 out of every 100
people in the U.S. live in California. Notice how you can think of the identity 0.12 = 12% as another
example of the unit-factor method:
0.12 = 0.12 × 100% = 12%,
since 100% is another way of writing 1, and multiplying by 1 makes no difference!

Remark 6: In the previous example, we used the unit “people,” but we could have chosen “million
people” as our unit without changing the result:
39 (
million
((( people
((
320 (million
((( (( ≈ 0.12 = 12%.
people
If a calculation yields a pure number (or percentage) result, then all the unit parts must cancel. That
means we will get the same answer whatever units we use in the numerator and denominator (as long
as we use the same units for both).
Definition 2
Q1
A numerical quantity without a unit is called a pure number. A ratio of two quantities
Q2
Q1 and Q2 having the same units will produce a pure number.
Example 3: The mathematical constant π ≈ 3.14, which is the ratio of a circle’s circumference to
its diameter, is a pure number. It appears (among many other places) in the formula for the area of a
circle, πr2 , where r is the radius.
Example 4: The conversion factors in Section 1.1.3 are also pure numbers. For example,
1 yd 36 in
= 36

=
1 in 1
in

is a pure number.
Example 5: A concentration is a pure number: when we say “the concentration of oxygen in the
atmosphere by volume is about 21%,” we are referring to the ratio
Volume of atmospheric oxygen
.
Volume of entire atmosphere
The units cancel, giving us a pure number, about 0.21 = 21%.
1
A percentage is a multiple of 100 , as we said earlier. Very small (or trace) concentrations may be
1
expressed not as percentages but as parts per million ( ppm, that is, multiples of 1,000,000 ) or even as
1
parts per billion ( ppb, that is, multiples of 1,000,000,000 ). For example, the present concentration of
CO2 in the atmosphere is about 0.04%. We typically express this in parts per million. To do so, use
the unit-factor method again: 1,000,000 ppm = 1, so
1
0.04% = 0.04% × 1,000,000 ppm = 0.04 × × 1,000,000 ppm = 400 ppm,
100
that is, 400 parts per million (by volume).
Remark 7: In Example 5 on the previous page, we expressed the concentration of atmospheric carbon
dioxide in parts per million by volume. Trace concentrations are also frequently expressed in parts per
million (ppm) or parts per billion (ppb) by weight. In the case of atmospheric CO2 , 400 parts per
million by volume corresponds to about 600 parts per million by weight. The weight of a carbon
dioxide molecule is about 1.5 times the weight of an average air molecule and this accounts for the
difference in these two measures.
For concentrations of CO2 and other trace gasses in the atmosphere ppm is almost always intended
as parts per million by volume, while for concentrations of trace substances in soil and water ppm
is usually intended to mean parts per million by weight. To avoid confusion we will usually use the
abbreviation ppmv when referring to parts per million by volume.
Example 6: Toluene is a byproduct of gasoline production that is used as a solvent in paint thinners
and in industrial applications. Toluene is present in the discharge from petroleum refineries and poses
a concern for groundwater and drinking water supplies. The U.S. Environmental Protection Agency
(EPA) enforces a limit of 1 ppm for toluene in drinking water [20].
1 kilogram (kg) is 1,000 grams and 1 gram is equivalent to 1,000 milligrams (mg). A concentration
of 1 ppm (by weight) is thus equivalent to 1 mg/kg:
1
mg 1 gram
1k
g 1,000,000 ppm
× 1,000,000 ppm = = 1 ppm

× ×
1
k
g 1, 000mg 1, 000
gram
1,000,000
See Exercise 18 on page 60 for examples expressing ppm and ppb in more familiar terms.
Problem 14: Suppose that your electric bill is $70 each month. After following the advice contained
in the EPA’s Energy Saving Tips for Renters [17], you manage to cut your electric bill by 15%. How
much money are you saving each month?
Solution: The amount we save is 15% of $70, that is,
1
15% × $70 = 15 × × $70 = $10.50.
100
Problem 15: Alcohol concentration, ABV, or alcohol by volume, is reported as a percent. Suppose a
particular brand of beer has an ABV of 5%. How much alcohol is in a 12 oz beer of this brand?
Solution: Using the same idea as in the previous solution, we get
1
5% × 12 oz = 5 × × 12 oz = 0.60 oz,
100
or a little more than half an ounce.
There are many circumstances in which a percentage may be the most meaningful way of measuring
some kind of change (for more about why this might be, see Section 4.1.1). For example, changes in
population or the size of the economy are usually reported as percentages.
Think about it. . .
According to recent estimates, roughly 10 million tons of plastic makes its way into the
oceans each year to become “marine plastic debris.” Although the U.S. generates more
plastic waste overall than most other countries, it is responsible for only 1% of plastic
flowing into the oceans each year [170], because a much smaller proportion of U.S. plastic
waste is “mismanaged” than plastic waste generated in less well-off countries. How much
responsibility does the U.S. share, in your opinion, for dealing with the problem of plastic
pollution in the world’s oceans and with the global management of plastic waste more
generally?
1.1.5 Measurements in Human Terms

A measurement, as we have seen, has both a number part and a unit part. But we human beings
are not always able to grasp the significance of very large or very small numbers, or of units of
measurement that don’t relate to a familiar scale. If we can express a measurement in terms of more
familiar quantities, it can help us a lot. Let’s make the following definition.
Definition 3
We say that a measurement is expressed in human terms if it is expressed in a way

that allows us to relate it directly to our shared everyday experience, without requiring
specialized scientific knowledge.
Some units of measurement are intrinsically “in human terms.” For example, one foot originally
was just the length of your foot! Here are two simple rules for keeping measurements in human terms:
Rule 1: Human Terms Measurement Rules
To express a measurement in human terms, try, if possible, to follow both of the rules
below:
• Choose units to keep the number part reasonably close to 1; say, between 0.01 and
1,000.
• Use “human scale” units (like feet, inches, hours).
When these rules both apply, it is not hard to express something in human terms. For instance,
the spacing of the studs in standard house framing (in the U.S.) is 16 inches. That is a small number
and a “human scale” unit. It would be foolish to express this spacing as 406,400,000 nanometers or as
0.0002525 miles. Even though both of these conversions are technically correct, they both violate both
of the rules above: the numbers are huge or tiny (violating the first rule) and the units of measurement
(nanometers, miles) are far from human scale (violating the second).
Problem 16: The mass of a full-grown elephant is about 4 tons. How

should I express this in human terms?
Solution: There is no single correct answer to a question like this.

We could refer to Table 6 on page 11 to say that the mass of the
elephant is the same as that of 4 compact cars. Or we could look
up the average mass of a U.S. American (about 180 pounds; see
Section 8.2) and say that the elephant has the mass of
2,000 lb 1 American
4t× × ≈ 45 Americans.
1t 180 lb
Figure 6: A grown elephant can weigh 4 tons.
Notice that the notion of “human terms” depends explicitly on whose
shared everyday experience is taken as a reference point. In expressing the mass of the elephant in
terms of a number of compact cars, we are assuming that compact cars are familiar and elephants are
unfamiliar. In a different society, where elephants were abundant but automobiles were rare, one might
reverse the process and express the mass of a compact car in terms of a fraction of an elephant—and
this would be a “human terms” measurement too.
Example 7: Here’s another example from a recent real-life sustainability discussion. In August 2015,
President Obama addressed a meeting of Arctic nations in Anchorage, Alaska. In his remarks [244]
he used the following example to illustrate the shrinking of Arctic ice:
Since 1979, the summer sea ice in the Arctic has decreased by more than 40 percent—a
decrease that has dramatically accelerated over the past two decades. One new study
estimates that Alaska’s glaciers alone lose about 75 gigatons—that’s 75 billion tons—of
ice each year.
To put that in perspective, one scientist described a gigaton of ice as a block the size of the
National Mall in Washington—from Congress all the way to the Lincoln Memorial, four
times as tall as the Washington Monument. Now imagine 75 of those ice blocks. That’s
what Alaska’s glaciers alone lose each year.
For anyone who has visited Washington, DC, the image of a giant ice block “the size of the National
Mall, four times as tall as the Washington Monument” is both striking and accessible, and this speech
is an effective example of putting a very large quantity in human terms. For those who aren’t so
familiar with Washington, though, perhaps not so much. As we said in the previous example, the
notion of “human terms” depends very much on whose experience is our reference point.
More difficult situations arise in which we can’t follow both parts of the Human Terms Measure-
ment Rules at the same time. For example, consider the following data relating to the volume of water
on Earth (Section 8.2). “Fresh surface water” refers to fresh water in lakes, rivers, and streams, as well
as in mountain snow, glaciers, ice caps, and ice sheets, but not underground freshwater (as in aquifers)
nor atmospheric freshwater (as in clouds).
Table 10: Volume of Water on Earth

Type of Water Volume (cubic meters)
All Water 1,400,000,000,000,000,000
Fresh Surface Water 100,000,000,000,000
A cubic meter can perhaps be considered as a “human terms” unit (you can envisage a cube of water
one meter on each side, or just think about the amount of water it might take to fill an average hot tub),
so the unit part of these measurements is accessible, but the number part is way out of human scale.
In a situation like this, it is a good idea to ask ourselves just exactly what we are trying to convey in
human terms. Do we want to convey the absolute size of the data—what the total amount of water is,
in some way that a human might be able to grasp? Or are we more interested in the relative size of
the two items—just how much of the Earth’s water is fresh? These are two different questions, and as
we’ll see, they need different kinds of answers.
Example 8: Let’s suppose first that we are trying to convey the absolute size of the measurement
of the total amount of water, 1,400,000,000,000,000,000 cubic meters. This is such a huge number
because the units (cubic meters) are relatively so small. Can we choose a different unit which, though
no longer directly on a human scale, is at least accessible to (some of) our readers’ experience?
Here is one way of doing that. Lake Superior is the largest of the Great Lakes; some readers will
have visited Lake Superior or may even live there. From a reference book, we can find that the volume
of Lake Superior is approximately 12,100,000,000,000 cubic meters. That is,
1 Lake Superior
= 1.
12,100,000,000,000 m3
Using the unit-factor method, the total volume of water on Earth is
3 1 Lake Superior
1,400,000,000,000,000,000
m × ≈ 116,000 Lake Superiors.
12,100,000,000,000
m3
Lake Superior is a pretty huge “unit,” but at least it is somewhat familiar; so here is one answer—the
total amount of water on Earth is a bit over 100,000 Lake Superiors.
Example 9: Another approach to the same problem is one that is useful only for volume (and
sometimes, area) measurements. Imagine all that water rolled up into a single spherical ball. How
big would the ball be?
To use this way of expressing the measurement in human terms, you need to know or remember (or
look up) the formula for the volume V of a sphere in terms of its radius r: that is, V = 43 πr3 . We can
rearrange this using algebra to give r
3 3V
r= .
4π
Thus a sphere of water of volume 1,400,000,000,000,000,000 m3 would have radius
s
3 3 × 1,400,000,000,000,000,000 m3
r= ≈ 690,000 m.
4×π
(Notice, again, how beautifully the unit-factor method works: “cubic meters” ( m3 ) appear under
the cube-root sign; the cube root of m3 is m; so the answer is in meters. Thus, all the water on Earth
would make a ball of radius about 690,000 meters; that is 690 kilometers, or about 430 miles. See
Figure 7 for a graphic illustration of this.
Think about it. . .
A student looking at Figure 7 says, “That can’t possibly be right! After all, we know that
in reality most of the Earth’s surface is covered with water! So how could it all possibly fit
into that tiny droplet?” How would you explain things to this student?
Example 10: Suppose though that we were more inter-

ested in the second question raised above: not so much
conveying the absolute size of these gigantic amounts
of water, but their relative size: roughly how much
of the Earth’s water is fresh? Here, what we want to
communicate is how several measurements are related
to one another, and we do this in the same way as you
would if you were making a map: by rescaling.
To rescale a group of measurements, you multiply
each one of them by the same number, called the scale
factor. For example, if a map is on a scale of one inch to
one mile, the scale factor is
1 in 1
in
1mi
1 y
d 1 ft 1
= × × × = .
1 mi 1mi 1,760
yd 3 ft 12in 63,360
Figure 7: All the water on Earth would fit into a “raindrop” 430
miles in radius. A scale factor like this is a pure number (Definition 2 on
page 15), since all the units cancel. The scale factor was
constructed to convert one mile in real life to one inch
on the map. However, now that we know the scale factor, we can see how any distance in real life will
be represented on the map: just multiply by the scale factor. For instance, 1000 feet (or 12,000 inches)
1
in real life will be represented on the map by 12,000 in × ≈ 0.19 in, or just under a fifth of an
63,360
inch.
In our case we have two quantities to consider in our “map” (the volume of all water, and the volume
of fresh surface water). Let’s figure out a scale factor that will convert the volume of all water (in real
life) to ten gallons (on the “map”). This scale factor will be the tiny number
10 gal 10gal
3.8 L 0.001
m3
= × ×
1,400,000,000,000,000,000 m3 1,400,000,000,000,000,000
m3 1gal
1L
≈ 0.000000000000000000027.
Now, to work out how the volume of fresh surface water (in real life) will appear on our “map,” we
must multiply by the scale factor:
100,000,000,000,000 m3 × 0.000000000000000000027 ≈ 0.0000027 m3 .
Since a meter equals 100 centimeters, a cubic meter equals 1003 = 1,0000,000 cubic centimeters.
Thus our scaled representation of the volume of fresh surface water is about 0.0000027 × 1,000,000 =
2.7 cm3 , or roughly half a teaspoon (see Problem 9 on page 11). To put it another way, about half a
teaspoon out of every ten gallons of water on earth is (surface) freshwater. That is certainly a striking
“human terms” expression of how relatively rare and valuable fresh water is.
Here is one more example.
Problem 17: A domestic electrical outlet can deliver energy at a maximum rate of perhaps 3
kilowatts—that is, 3,000 joules per second. A gasoline pump delivers gasoline at a rate of 1 gallon
every 10 seconds or so, and each gallon of gasoline contains about 120,000,000 joules of energy (see
Table 5 on page 488). Compare the rate of energy delivery by the electrical outlet and the gas pump.
Solution: The energy delivery rate from the gas pump is
120,000,000 J
= 12,000,000 J/sec.
10 sec
The corresponding rate for the electrical outlet is 3,000 J/sec. Thus the ratio of the energy delivery
rates is
12,000,000 J/ sec

= 4,000;
3,000 J/sec

that is, the gas pump delivers energy 4,000 times faster than the domestic electrical outlet. Or to put
the matter in terms of time, the gas pump delivers in one second the same amount of energy that
the electrical outlet takes 4,000 seconds (a little over an hour) to supply. Or again, we can convert
to teaspoons (as in the previous problem) and say that the domestic outlet takes over five seconds to
supply the energy equivalent of a teaspoon of gas. All of these can be thought of as simple rescalings.
A more striking “rescaling” is to express the matter in the following way: if the hose of the gas
pump were shrunk to the diameter of a drinking straw, it would deliver energy at the same rate as the
electrical outlet. The striking image of gasoline dribbling slowly out of a ten-foot hose as thin as a
drinking straw certainly reinforces the contrast between the gas and electric “pumps,” and highlights a
major issue for the acceptance of electric cars: it is extremely difficult to achieve the same “refueling
rates” with electricity as the ones gasoline has gotten us accustomed to. For the details of how the
rescaling is calculated, see Exercise 22 on page 60 at the end of the chapter.
Example 11: The Sagan Planet Walk, a public sculpture in Ithaca, NY,
is a marvelous example of the power of rescaling to convey an important
idea in dramatically visual terms. The Sagan Walk is a scale model,
scale 1 : 5,000,000,000, of our solar system and of the various objects
(the Sun and planets) that are parts of it. It cuts the huge size of the Solar
System down to something that we can appreciate visually.
In Figure 8 you can see the circular opening in the granite obelisk,
almost 11 inches in diameter, that represents the Sun to scale. In each
of the other obelisks in the sculpture, representing the planets, there is a
Sun-sized opening that has a glass window into which the model of the
corresponding planet is embedded. This allows the viewer to experience
directly how much bigger the Sun is than even the largest planets.
You can see the Mercury monolith just above the boy’s hand in
Figure 8, to the right of the Sun monolith. It is just over 10 yards away
from the Sun, and another 10 yards will get you to Venus and another
10 more to our Earth. The first four planets are really close together,
no further than the trees in our picture. What’s more, the models of
the corresponding planets are tiny, the size of a pea or smaller. Then,
suddenly, the scale starts to grow. It will take you ten minutes of brisk Figure 8: Standing by the Sun on the Sagan
walking to get to Neptune from the Sun, passing on the way by Jupiter, Planet Walk.
the largest planet—model size 2.9 cm, greater than an inch but not quite
the size of a Ping-Pong ball. The Planet Walk honors Carl Sagan’s work in science communication by
giving visitors a direct sensory engagement with the size and shape of the solar system.
In 2012 the Planet Walk was expanded. A monolith at the Astronomy Center of the University
of Hawai’i now represents the nearest star to Earth—Alpha Centauri—on the same scale. Imagine
walking the distance from Ithaca to Hilo, Hawai’i (where the Astronomy Center is located), to get
some idea of the difficulties of interstellar travel compared to the journeys of our various spacecraft
within the solar system.
If you want to learn more about the Planet Walk and the rescaling involved, we encourage you to
read a beautiful article by mathematician Steve Strogatz, titled “Visualizing Vastness” [313].
Summary of Ideas: Units and Measurement
• A measurement, such as “3.2 million centimeters” consists of a number part, 3.2

million, and a unit part, centimeters.
• There are two main systems of units: the U.S. system and the metric system. Some
examples of U.S. units are feet, pounds, gallons. Some examples of metric units are
meters, kilograms, liters.
• We can use the conversion tables in this chapter to convert units within each system
and between them.
• To help us with these conversions, we can use the unit-factor method to make sure
that units cancel out. This tells us which numbers to divide and which to multiply: for
example,
12 × 0.62 mi ≈ 7.4 mi.
km
1km

• The ratio of two quantities having the same units or dimensions is a pure number.
Pure numbers can also be expressed as percentages.
• To communicate clearly, it helps to express measurements or comparisons in human
terms. This means that we express them in a way that is relatable to ordinary human
experience without requiring scientific knowledge.
1.2. SCIENTIFIC NOTATION 23
1.2 Scientific Notation
Objectives
I can explain why scientific notation is used.

I can convert numbers between decimal and scientific notation.
I can recognize different ways of writing the same number using scientific notation,
including in standard form.
I can add, subtract, multiply, and divide numbers written in scientific notation.
1.2.1 Scientific Notation

In the last section we dealt with some very big numbers. When numbers become very large or very
small, the standard way of expressing them (called decimal notation) takes up a lot of space and
becomes hard to read. That is because very large and small numbers are mostly made up of zeros!
For example, two large volumes appeared in Example 10 on page 20:
• 1,400,000,000,000,000,000 cubic meters (the total volume of water on Earth), and
• 100,000,000,000,000 cubic meters (the volume of fresh surface water).
What matters most about these numbers is that the first is in the millions of trillions and the second
is in the hundreds of trillions. To see that, however, you have to carefully count the number of zeros
appearing in each expression. The commas help, but couldn’t we find some more straightforward way
to convey this information?
The key to doing so is to use the idea of powers of 10. Remember that 10 raised to a certain power,
say n, means multiplying 10 by itself that number (n) of times. Thus, for example,
101 = 10,
102 = 10 × 10 = 100,
103 = 10 × 10 × 10 = 100 × 10 = 1,000,
104 = 10 × 10 × 10 × 10 = 1,000 × 10 = 10,000,
Another way of saying this is that 10n can be written as the digit 1 followed by n zeros. So, if we are
given the number 109 , we know our number is 1 followed by 9 zeros; that is a billion. Similarly, 106
is a million, and 1012 is a trillion.
The number 7,000,000,000,000 is written as the digit 7 followed by 12 zeros—7 trillion. We can
write that as 7 times 1 trillion, or 7 × 1012 . This technique of using a power of 10 to keep track of the
zeros is known as scientific notation. Here is a definition.
Definition 1
A number is written in scientific notation if it is expressed as the product of two parts:

• an ordinary decimal number called the significand (like 7 in the example above), and
• a power of 10 (like 1012 in the example above). The power to which 10 is raised (the
12 in the example) is called the exponent.
Example 1: The expression below represents the number 120,000,000,000,000 in scientific notation.
1.2 × 1014
significand exponent
The significand in this expression is 1.2 and the exponent is 14.
Remark 1: The same number in the example above could be expressed in several different ways. For
instance
1.2 × 1014 = 1.2 × 10 × 1013 = 12 × 1013 ,
because 1.2 × 10 = 12. By shifting the exponent in this way, we can multiply or divide the significand
by 10—that is, we can move the decimal point to the left or right. Usually we choose to move the
decimal point so that the significand is between3 1 and 10. When this has been done, the number is
said to be expressed in standard form. Thus 1.25 × 1014 and 12.5 × 1013 represent the same number,
but the first expression is in standard form and the second is not.
Conversion from Decimal to Scientific Notation

Let’s look at some examples of converting quantities to scientific notation and standard form.
Problem 1: Write 2,000,000,000,000,000,000 in scientific notation, standard form.
Solution: The number 2,000,000,000,000,000,000 has 18 zeros, so we can write it as 2 × 1018 .
Problem 2: Write 150,000,000,000,000,000,000 m3 in scientific notation, standard form.
Solution: This quantity has a number part and a unit part (see Section 1.1.1). In converting to scientific
notation we re-express only the number part; the unit part should remain unchanged. Now in this
example the number part 150,000,000,000,000,000,000 is 15 with 19 zeros. So we could write it as
15 × 1019 . This is a correct expression, but notice that 15 is not between 1 and 10, so this expression
is not in standard form. Instead, we write the number part as 1.5 × 1020 , which is in standard form.
The answer to the complete problem (including the unit part) is therefore
1.5 × 1020 m3 ,
also in standard form.
3 We allow 1 but not 10 for a significand in standard form: the standard form of 1, 000 is 1 × 103 , not 10 × 102 .
Remark 2: When we express a physical measurement in scientific notation (as in the previous
problem), it is only the number part of the measurement that we are working with. The unit part
remains unaffected, and must appear in our final solution. Thus, in the last example, the solution is
1.5 × 1020 m3 , not just 1.5 × 1020 .
You can think of these calculations with powers of 10 in terms of “moving the decimal point.” First,
imagine your number written in decimal notation (if it is a whole number, imagine a decimal point to
the right of the ones place, so that you would think of 127 as 127.0). Then moving the decimal point
one step to the left corresponds to dividing by 10 (e.g., 12.7 = 127/10). This gives us a simple rule
for putting a number greater than 1 into standard form .
Rule 1
To express a number greater than 1 in standard form, move the decimal point to the
left until you obtain a number between 1 and 10. What you obtain is the significand
(in standard form), and the number of steps to the left that you moved the decimal
point is the exponent.
For example, to express the number 127 in standard form we move the decimal point 2 steps to the
left to get a significand of 1.27 and an exponent of 2:
127.0
giving us 127 = 1.27 × 102 .
2 steps
Problem 3: The distance from the Earth to the Sun is approximately 93 million miles. Convert this to
inches, and express your answer in scientific notation using standard form.
Solution: Using the unit-factor method we can calculate
× 1,760
93,000,000
mi
yd 36 in
× ≈ 5,900,000,000,000 in.
1mi
1
y
d
Moving the decimal point 12 steps to the left gives us 5.9, so the answer in standard form is 5.9 ×
1012 in.
These examples show how to write large numbers using scientific notation. But we can also write
small numbers in the same way. To do this, we need to remember about negative powers of 10. Just as
the positive powers of 10 are obtained by successively multiplying by 10, so the negative powers are
obtained by successively dividing by 10:
100 = 1,
1
10−1 = = 0.1,
10
1
10−2 = = 0.01,
10 × 10
1
10−3 = = 0.001.
10 × 10 × 10
Thus 10−n is
10−n = 0. 00 . . 00} 1.
| .{z
n − 1 zeros
Suppose for example that we would like to write the number 0.000000567 in standard form. We can
write this as 5.67 × 0.0000001, and from our discussion above, 0.0000001 = 10−7 . Thus the standard
form of our number is 5.67 × 10−7 .
Alternatively, we can again think in terms of “shifting the decimal point.” This time, the rule is the
following.
Rule 2
To express a number less than 1 in standard form, move the decimal point to the
right until you obtain a number between 1 and 10. What you obtain is the significand
(in standard form), and minus the number of steps to the right that you moved the
decimal point is the exponent.
In the example above, we move the decimal point 7 steps to the right to get the number 5.67.
Therefore, the standard form is 5.67 × 10−7 :
0.000000567.
7 steps
Problem 4: Write 0.00000000000000232 in standard form.
Solution: To get 2.32, we need to move the decimal point 15 steps:
0.000000000000002325.
15 steps
That gives us 2.32 × 10−15 as the result in standard form.
Conversion from Scientific to Decimal Notation

Up to now, we have converted numbers from decimal notation to scientific notation. Converting in the
reverse direction, from scientific to decimal, is simply multiplying by a power of 10. Again, we can
think of this in terms of shifting the decimal point.
Problem 5: Convert 9.8 × 107 to decimal notation.
Solution: Because we have a positive exponent (7), we know that this is likely to be a large number.
Our solution is
9.8 × 107 = 98,000,000.
We can think of this as taking the number 9.8 and moving the decimal point 7 steps to the right, filling
in zeros as needed.
Problem 6: Convert 98 × 10−7 to decimal notation.
Solution: Because we have a negative exponent (−7), we know that this is likely to be a small number.
Our solution is
98 × 10−7 = 98 × 0.0000001 = 0.0000098.
We can think of this as taking the number 98 = 98.0 and moving the decimal point 7 steps to the left,
filling in zeros as needed.
1.2.2 Calculations with Scientific Notation

Adding, subtracting, multiplying, and dividing numbers in scientific notation is straightforward once
you know the rules. We’ll review these rules and look at some examples.
Remark 3: Most calculators and calculator apps can work with numbers in scientific notation, though
not all offer you a choice about whether to display the final result in scientific or decimal. We’ll show
some examples below. Nevertheless, it’s important to appreciate how these calculations are carried
out, and the best way to do that is to understand the rules for calculating by hand.
Multiplication and division

We’ll begin with the rules for multiplying and dividing with scientific notation. It may seem surprising,
but these are actually simpler than the corresponding rules for adding and subtracting! What’s more,
they are used more often. That’s why we have put them first.
The laws of exponents say that
10a × 10b = 10a+b , 10a ÷ 10b = 10a−b .
For example,
102 × 103 = (10 × 10) × (10 × 10 × 10) = 10 × 10 × 10 × 10 × 10 = 105 .
Thus powers of 10 are multiplied by adding their exponents, and divided by subtracting their
exponents. For numbers in scientific notation this gives us the following rule.
Rule 3
When multiplying (or dividing) two numbers in scientific notation, we multiply (or
divide) their significands and add (or subtract) their exponents.
Example 2: Let’s consider the two numbers 4.23 × 1023 and 9.1 × 106 . First we will multiply them.
What is (4.23 × 1023 ) × (9.1 × 106 )? According to Rule 3, we must multiply the significands and add
the exponents. Therefore, we must do the following calculation:
(4.23 × 1023 ) × (9.1 × 106 ) = (4.23 × 9.1) × 1023+6 = 38.493 × 1029 = 3.8493 × 1030 ≈ 3.85 × 1030 .
Notice how multiplying the significands gives us a number (38.493) that is greater than 10. We shifted
the decimal point to the left in order to get the result in standard form.
Now let’s consider division, (4.23 × 1023 ) ÷ (9.1 × 106 ). According to Rule 3, this time we will
divide the significands and subtract the exponents. Therefore, we do the following calculation:
(4.23 × 1023 ) ÷ (9.1 × 106 ) = (4.23 ÷ 9.1) × 1023−6 ≈ 0.46 × 1017 = 4.6 × 1016 .
Once again we needed to shift the decimal point (this time to the right) in order to get the result in
standard form.
Remark 4: Whatever notation we use, it is important to remember that a÷b 6= b÷a. So when dividing
numbers in standard form, be careful to carry out the operations in the correct order!
Now, let us look at some more worked problems.
Problem 7: Calculate (9.5 × 108 ) × (4.1 × 108 ).
Solution:
(9.5 × 108 ) × (4.1 × 108 ) = (9.5 × 4.1) × 108+8 = 38.95 × 1016 ≈ 3.9 × 1017 .
Problem 8: Calculate (3.2 × 104 ) ÷ (4.5 × 108 ).
Solution:
(3.2 × 104 ) ÷ (4.5 × 108 ) = (3.2 ÷ 4.5) × 104−8 ) ≈ 0.71 × 10−4 = 7.1 × 10−5 .
If units are involved, we multiply the number parts according to Rule 3 on the previous page, and
the unit parts according to the unit-factor method (Section 1.1.2).
Problem 9: According to [136], a cow emits about 0.3 kilograms of methane per day.4 There are
thought to be about 1.2 × 109 cows on Earth. Compute the total mass of methane emitted by cows
over one year, expressing your answer in scientific notation.
The mass of the entire atmosphere is about 5 × 1018 kg (see Section 8.2). Methane lasts about 8
years in the atmosphere before breaking down to other gases. What proportion (by mass) of the whole
atmosphere is made up of bovine methane emissions?
Solution: We solve the first problem using the unit-factor method:
kg days kg kg
× (1.2 × 109 ) ≈ 130 × 109 = 1.3 × 1011 .

0.3 cows
× 365

cows

days
yr yr yr
Bovine methane emissions are about 1.3 × 1011 kg/ yr.

For the second problem, the atmosphere contains about 8 years worth of bovine methane. That is,
8 yr × (1.3 × 1011 ) kg/ yr ≈ 10 × 1011 kg = 1012 kg.
Since the mass of the whole atmosphere is about 5×1018 kg, the proportion of bovine methane (which
is a pure number) is
1012
kg 1
= × 10−6 = 0.2 × 10−6 = 2 × 10−7 ,
5 × 1018kg
5
or 0.2 parts per million. Though this is a small number, it is in fact a substantial fraction of the total
amount of methane in the atmosphere (which is about 1 part per million by weight).
Addition and subtraction

Now we will think about adding and subtracting numbers in scientific notation. There is one situation
in which we can do this directly:
4 Often referred to as “cow farts” online, though [136] showed that the great majority of the methane is burped rather than
farted.
Rule 4
In order to directly add (or subtract) two numbers in scientific notation, they need
to have the same exponent. Once two numbers have the same exponent, we add (or
subtract) their significands.
The two numbers below can be added automatically because they have the same exponent, 8.
(2.32 × 108 ) + (3.1 × 108 ) = (2.32 + 3.1) × 108 = 5.42 × 108 .
But you won’t always be adding two numbers with the same exponent. Suppose, for example, that
you need to compute the following sum:
(2 × 109 ) + (4.3 × 1010 ).
These numbers do not have the same exponent and therefore cannot be added together directly. We
must rewrite one number to have the same exponent as the other. Let’s rewrite 2 × 109 to have an
exponent of 10. We’ve already seen how to do this in the previous section; let’s express it by rules.
Rule 5
• For every increase in the exponent, we have to move the decimal point in the
significand to the left.
• For every decrease in the exponent, we have to move the decimal point in the
significand to the right.
This means that 2 × 109 = 0.2 × 1010 . Now we can add using Rule 4:
(0.2 × 1010 ) + (4.3 × 1010 ) = 4.5 × 1010 .
Alternatively, we could have rewritten 4.3 × 1010 , moving the decimal point in the other direction to
write 4.3 × 1010 as 43 × 109 . If we add using this approach, we get another situation in which we can
apply Rule 4:
(2 × 109 ) + (43 × 109 ) = 45 × 109 .
This is equal to 4.5 × 1010 , the same answer as before. Notice that depending on how we choose to
perform the calculation, we may need to re-express our final answer to make sure that it is in standard
form.
Critical Thinking
What does the method we’ve illustrated here have in common with the method we used to
solve Problem 12 on page 13?
Here are some more worked examples.
Problem 10: Calculate 9.5 × 108 + 4.1 × 108 .
Solution:
9.5 × 108 + 4.1 × 108 = 13.6 × 108 = 1.36 × 109 .
Problem 11: Calculate 8.1 × 107 − 3 × 106 .
Solution:
8.1 × 107 − 3 × 106 = 8.1 × 107 + 0.3 × 107 = (8.1 − 3) × 107 = 7.8 × 107 .
Problem 12: Calculate 5.75 × 1027 + 2 × 1025 .
Solution:
5.75 × 1027 + 2 × 1025 = 5.75 × 1027 + 0.02 × 1027 = (5.75 + 0.02) × 1027 = 5.77 × 1027 .
Problem 13: Calculate

4.23 × 1023 − 9.1 × 106 .
Solution:
4.23 × 1023 − 9.1 × 106 ≈ 4.23 × 1023 .
Although we might consider 9.1 × 106 = 9,100,000 to be a large number, it is tiny compared to
4.23 × 1023 , which is
423,000,000,000,000,000,000,000
A value in the millions has little impact on a number that is in the thousands of sextillions. So the
subtraction is a negligible change.
Think about it. . .
A student suggests that the rule for adding two numbers in scientific notation should be
“add the significands and add the exponents.” How will you convince this student that he is
wrong? Avoid appeal to authority (“The textbook says that you should do it this other way”)
and try to find an approach, maybe by means of examples, that helps the student understand
why the correct rule makes sense, whereas his proposal does not.
Using a Calculator
Most calculators or calculator apps allow you to enter numbers in scientific notation. Typically, you
should look for a button labeled “E” or “EE.” Some calculators may have a button labeled “Exp”
instead.5 You can enter a number in scientific notation like this:
significand EE exponent.
For example, Figure 9 shows the result of entering 1.56 × 109 into a smart phone calculator app. In
this particular app, rotating the phone to the vertical position changes the display to scientific notation,
“1.56e9,” as shown in the same figure. It also changes the buttons available on the calculator.6
(a) The result of entering “1.56 EE 9” (b) The result of rotating

the phone to the vertical
position
Figure 9: A smart phone calculator app. Turning the phone on its side changes the display and the set of buttons available.
Another smart phone app, shown in Figure 10 on the next page, displays both what is entered
(below the white line) and the result (above the white line). We’ve entered the number 97.5 × 1015
using “97.5 E 15,” which displays as 97500000000000000 (shown on the left). We can switch to
scientific notation by pressing “FIXED” in the display, which changes the mode from “FIXED” to
“FLOAT” (shown in the center).7
Once you know how to enter numbers in scientific notation, you can do arithmetic on them
with your calculator in the ordinary way. We’ve used the previous calculator app to work out
5.7 × 10−8 × (2.6 × 1012 ); the result is shown in Figure 10 on the next page (on the right).

5 The calculator included with Windows 10 uses “Exp” for this key. However, this is a terrible choice, since many other
calculators, as well as advanced mathematics textbooks, use “exp” for the exponential function y = ex where e ≈ 2.718 is an
important mathematical constant. If you’re going to use Windows 10’s calculator, you will have to get used to this (and the
exponential function won’t appear anywhere in this book, which reduces the possibility of confusion), but nevertheless, watch
out if you switch to another calculator app or program.
6 This is the case with many calculators available on smart phones. If you are using your phone as a calculator, try turning
the phone on its side and see whether there are more or different buttons available.
7 “Fixed” is short for fixed-point notation. The decimal point remains in its usual place. “Float” is short for floating-point
notation. We use powers of 10 to move the decimal point to a place that makes the number easier to read (putting the number
into standard form).
Figure 10: A smart phone calculator app. To enter numbers in scientific notation, use the E button. Results can be displayed
in the usual way (“FIXED” mode, shown on the left) or using scientific notation by choosing the “FLOAT” mode (center and
right).
Summary of Ideas: Scientific Notation
• Scientific notation is a concise way of representing quantities that are very large or
very small.
• A number is represented in scientific notation as the product of two numbers, a
significand (between 1 and 10 if the number is in standard form) and a power of
10, written 10n for some integer n called the exponent. For example, 2.32 × 10107
is an expression in scientific notation, standard form; the significand is 2.32 and the
exponent is 107.
• To write a quantity that has both a number part and a unit part in scientific notation,
re-express only the number part; the unit part remains unaffected.
• For every increase (decrease) in the exponent, we have to move the decimal point in
the significand to the left (right). For example, 4.3×103 can be rewritten as 0.43×104
or as 43 × 102 .
• When multiplying (dividing) two numbers in scientific notation, you must multiply
(divide) the significands and add (subtract) the exponent. For example, (6 × 102 ) ÷
(3 × 107 ) = 2 × 10−5 , since 6 ÷ 3 = 2 and 2 − 7 = −5.
• When adding (subtracting) two numbers in scientific notation, you must first make
their exponents the same and then add (subtract) their significands.
1.3. ESTIMATES, PRECISION, AND ORDERS OF MAGNITUDE 33
1.3 Estimates, Precision, and Orders of Magnitude
Objectives
I know what is meant by the precision of a measurement claim.

I can understand and use significant figures as a measure of precision.
I can express the result of a calculation with appropriate precision, taking into account
the precision of the data used in the calculation.
I can say when two quantities have the same the order of magnitude.
I can make order-of-magnitude estimates of quantities that are relevant to sustain-
ability.
1.3.1 Precision
Example 1: An old story [345] tells of a museum guard whose

post is next to a dinosaur skeleton. One day, a kid asks the guard
how old the skeleton is. He responds, “It’s 70,000,003 years old.”
Puzzled about the “3,” the kid asks the guard how he knows.
“Well,” he replies, “when I started work I was told that the skeleton
was 70 million years old. . . and that was three years ago.”
The 70-million-year figure was some expert’s best estimate.
Perhaps what the expert meant was “between 65 and 75 million
years,” in which case we’d say that the precision of her estimate
was plus or minus 5 million years. Maybe the estimate was more
specific than that—perhaps it was only plus or minus 500,000 Figure 11: Dinosaur skeleton.
years. But surely it was not specific enough for an additional 3 years to make any significant
difference! The guard’s mistake arose from a misunderstanding of precision: by using all the digits of
the number 70,000,003, he gave an answer that seemed much more precise than was in fact possible.
Example 2: Here is another example that will help us come to grips with this notion of precision.
Consider the following statements about the Earth’s current human population.
(a) The population of Earth is 7,188,895,672.
(b) The population of Earth is about 7 billion, plus or minus a billion.
(c) The population of Earth is between 109 and 1010 .
(d) The population of Earth is 6.4 × 107 , to 2 significant figures.8
Each of statements (a)–(d) makes a claim that Earth’s human population is within a certain range of
values. In statements (b) and (c) the range of values is given explicitly, while in statements (a) and (d)
the range of values is implied. A statement of this kind, about a physical or environmental quantity, is
what we call a measurement claim.
8 Significant figures will be explained in more detail later in this chapter.
Definition 1
A measurement claim, about some physical or environmental quantity, is the claim that
the true value of that quantity lies within a certain range.
When someone makes a measurement claim we often pay most attention to the center of the range
of values (“Polls show Clinton ahead by three percentage points”). Often the width of the range
receives less attention (Footnote: “The margin of error is plus or minus five percentage points”), but
it is equally important. That width is what we call the precision of the measurement claim.
Definition 2
The precision of a measurement claim, about some physical or environmental quantity,

describes the width of the range of values within which the true value of that quantity is
claimed to lie. This width can be described explicitly, or can be expressed implicitly using
significant figures or in other ways.
Thus, of the four statements (a)–(d) above, statement (a) is the most precise, and statement (c) is
the most imprecise. The others are intermediate in their level of precision, with (a) being more precise
than (d), which is more precise than (b), which is more precise than (c).
Notice that precision is still a claim! A highly precise measurement claim does not for that reason
have to be true (nor does a highly imprecise claim have to be false). In fact, of the statements (a)–(d)
above, which are actually true? At the time of writing, (b) and (c) are true, while (a) and (d) are false.
The highly precise statement (a) may have been true for an instant sometime in 2014, but it is now
false; the moderately precise statement (d) has not been true for thousands of years.
Remark 1: Notice that statements (a)–(d) also provide examples of several different ways of
describing the precision of a measurement claim. You could explicitly give a range of values for
the measurement (example (c)); you could specify a single value with a possible error “plus or minus”
(example (b), and also the language we imagined our expert using in the dinosaur example above);
or you could use the notion of significant figures (example (d)), which we’ll discuss in a moment.
Example (a) does not include a specific statement of precision, but since it specifies a whole number
of discrete objects without mentioning possible error, its implied claim is to complete precision—“this
exact number is the population of the Earth.” It is difficult to imagine how anyone could accumulate
sufficiently precise data to make such a claim, even for single instant.
Remark 2: You may read, in some texts, about a distinction between the words precision and
accuracy. For instance, consider again the examples (a)–(d) above. Even though both (a) and (d)
are false, we might want to say that (a) is more accurate (that is, closer to the truth) than (d).
For another example, consider the Volkswagen emissions scandal [72]. Many models of VW diesel
automobiles were equipped with special software that detected when EPA-mandated emissions tests
were being run, and if so, altered the engine’s settings to make it behave more “virtuously” than
normal. Measurements from these tests could then have been highly precise, but they would not have
been accurate. In this text, we will speak mostly of precision, and we won’t use the word “accuracy”
in this specific, technical sense.
Critical Thinking
Open a newspaper today and find some examples of measurement claims. Which (if any) of
them include an explicit description of the level of precision that they claim? For those
that don’t, what seems to be the implied level of precision? Do you think that these
measurements are, in fact, as accurate as the claims made for them? Why or why not?
1.3.2 Significant Figures

One way to describe the precision of a measurement claim is to use the concept of significant figures.
Consider again the example of the dinosaur skeleton. If we know that its age is 70 million years
“plus or minus five million,” that tells us that the true age is nearer to 70,000,000 than to 60,000,000
or to 80,000,000 or to any other number given by one digit followed by a string of zeros. We would
say that the 70-million-year estimate is accurate to one significant figure.
Or if we learn from census data that the population of Pennsylvania, on April 1, 2010, was
12,702,884 plus or minus 40,000, then we know that the true population was closer to 12,700,000
than to any other number given by three digits followed by a string of zeros. We would say that the
count of 12,700,000 is accurate to three significant figures.
These examples involved integers (whole numbers). To understand the concept of “significant
figures” when decimals are involved, it helps to use our work with scientific notation. Our examples
can be written
70,000,000 = 7 × 107 , 12,700,000 = 127 × 105 .
Notice that we have written both numbers so that the significand is an integer. When this is done, the
number of digits in that integer is the number of significant figures.
Definition 3
An approximation to a number x, using n significant figures, is the n-significant-figure

number that is closest to x. (We also can say that the approximation is obtained by rounding
off x to n significant figures.)
Problem 1: Write an approximation to 83,333,333,334.4 in standard form, using one significant

figure.
Solution: The leading digit in the given number is 8. Because we are told to use only one significant
figure, the subsequent digits (the 3’s and the 4’s) are simply place holders—the only information that
they provide is how far to the left of the decimal point the 8 occurs. In fact, all these subsequent digits
may as well be zeros. In other words, we can write
83,333,333,334.4 ≈ 80,000,000,000 = 8 × 1010 .
If we were asked to write an approximation to the same number using two significant figures, we
would keep the first two digits (the remainder being place holders) and write
83,333,333,334.4 ≈ 83,000,000,000 = 8.3 × 1010 .
Look carefully at the next example to see how we may sometimes need to take the size of the first
“placeholder” digit into account also.
Problem 2: According to the 2010 census, the U.S. population on April 1, 2010, was 308,758,105.
Find an approximation to this, using three significant figures.
Solution: According to our rules, a number with three significant figures can be expressed in scientific
notation with a significand that is a three-digit integer. So we write the given number in scientific
notation with a significand that has three digits before the decimal point:
308,758,105 = 308.758105 × 106 .
Then we choose the nearest integer to the significand. Here, since 0.758105 is greater than 0.5, we
“round up”: the nearest integer significand is 309, and our answer to a precision of 3 significant figures
is 309 × 106 (it would also be correct to express this in standard form as 3.09 × 108 , or to write it out
in long form as 309,000,000).
We can express the technique that we used in the previous solution as a rule.
Rule 1
To approximate a datum x to n significant figures, first express x in scientific notation

in such a way that the significand has n digits before the decimal point, then replace
the significand by the nearest integer to it.
When you make a measurement claim, you should approximate it to a number of significant figures
that reflects the precision of your claim. What does this mean? I just measured the depth of the
computer keyboard on which I’m typing this manuscript, and got an answer of 16.4 centimeters.
Then I asked Google what that is in inches, and got the answer 6.456693 inches. Does that mean
that I should quote 6.456693 inches as the depth of my keyboard? No! In the first place, I am not
the world’s most careful measurer. Let’s say that my original measurement of 16.4 cm was within
0.5 cm of the true value. That means it has two significant figures. Then I should not claim a greater
degree of precision in the measurement after I have converted it to inches than it had before. Thus, the
appropriate way to express my measurement claim in inches would be “the depth of my keyboard is
6.5 inches (to two significant figures).”
Sometimes, people make measurement claims without specifically stating the number of significant
figures involved. What to do in this situation? We try to interpret the claim as “economically” as
possible, which is to say that we remove all the trailing zeros (zeros to the right that are followed only
by more zeros) and leading zeros (zeros to the left that are preceded only by more zeros). The number
of digits that are left is the implied number of significant figures.
Rule 2
If a measurement claim is made without specifically stating a number of significant

figures, the default assumption is that as few as possible of the digits are significant:
that is, the number of digits left after leading and trailing zeros have been removed.
The following examples show how this works out in practice:
Problem 3: Recent annual wheat yields in Bangladesh are stated to be 1,370 pounds per acre. How
many significant figures are claimed by this statement?
Solution: The final 0 is a trailing zero; the other three digits are significant. Thus the default
assumption is that our original claim was made to three significant figures.
Problem 4: The manufacturer claims that a new hybrid automobile obtains 61.07 miles per gallon.
How many significant figures are implied by this claim? Do you believe that the manufacturer can, in
fact, measure gas mileage to this level of precision?
Solution: There are no leading or trailing zeros, so the manufacturer’s original claim was expressed
with four significant figures. Notice that the zero between the 1 and the 7 is significant.
It is highly unlikely that the manufacturer, or anyone else, can measure gas mileage to the implied
precision of one part in ten thousand. Even if such a precise measurement could be carried out under
highly controlled conditions, such a level of precision would be irrelevant to highway driving. The
manufacturer’s statement is probably just trading on the idea that a more complicated-looking number
is somehow more impressive or “scientific.”
Remark 3: You might have noticed what seems like an ambiguity in our definitions. For instance, how
many significant figures has the number 200? We can write it as 2×102 (one significant figure). But we
could also write it as 20 × 101 (looks like 2 significant figures), or even as 200 × 100 or 2,000 × 10−1
(three or four significant figures)! Rule 2 tells us that we take the least number of significant figures
that makes sense: one significant figure in this example. If we want to express greater precision than
that, we need to add extra explanation. For instance, “there are 200 teaspoons in a liter, to 2 significant
figures,” means that there are between 195 and 205 teaspoons in a liter. (The exact figure is 202.884)
Significant Figures and Calculations

An important fact about calculations is this: you cannot get a more accurate result out of a calculation
than the accuracy of the data you put into it. (As the computer scientists say: “Garbage in, garbage
out”!). In terms of significant figures this translates to the following rule of thumb:
Rule 3
• In multiplying or dividing two or more numbers, the number of significant figures in

the result is the smallest number of significant figures in any piece of the input data.
• The result of a calculation should be approximated to the appropriate number of
significant figures (given in the previous bullet point).
• If a calculation involves intermediate steps, don’t approximate the intermediate steps.
Only approximate when you have reached the final result.
Thus, for example, if we multiply one number (a) that is given to five-figure accuracy by another
number (b) that is given only to two-figure accuracy, the result (a × b) will have only two-figure
accuracy, and we should round it off to express this. (There is also a rule for how to deal with
significant figures when adding or subtracting, but it is more complicated—just as, on pages 27–
29 in the previous section, Rules 4 and 5 are more complicated than Rule 3—and we are not going to
worry about it.)
Remark 4: Using a calculator can easily tempt one to express something with inappropriate precision.
For example, suppose that next time I fill my gas tank, the odometer indicates that I have driven 316.2
miles since my last fill-up, and the gas pump shows that I used 9.12 gallons. Using my calculator, I
obtain
316.2
= 34.6710526
9.12
(and I can compute more decimal places if I like). But it would be ridiculous to quote my gas mileage
in this form, which suggests that I measured all the quantities involved to a precision of 1 part in a
billion! The input data claim 4-figure precision (the odometer reading) and 3-figure precision (the gas
pump measurement), so it would be appropriate to express the result to 3-figure precision at most,
34.7 miles per gallon. In fact, you might prefer to use the nearest whole number and say, “My car gets
about 35 miles to the gallon.”
Definition 4
Mathematicians use the symbol ≈ to signify that two quantities are approximately equal—
equal within the level of precision appropriate to the context. Thus, using this symbol, we
could write the mileage calculation of Remark 4 as
316.2
≈ 35,
9.12
adding the phrase “to 2 significant figures” if we needed to make it clear what level of
precision we were using.
Problem 5: A local lake has a surface area of 1.273×106 m2 and an average depth of 5.1 m. A polluter
dumps a truckload (7,300 gallons) of contaminated wastewater into the (previously unpolluted) lake;
each gallon of the wastewater contains about 2 grams of arsenic. Assuming that the contaminated
wastewater is uniformly dispersed throughout the lake, find the resulting concentration of arsenic
in the lake water, expressed9 in micrograms per liter (µg/ L). Be sure to express your answer with
appropriate precision.
Solution: The concentration of arsenic in the lake will be the total weight of arsenic deposited, divided
by the total volume of lake water. Using the unit-factor method, we calculate
2g 106 µg
Weight of arsenic = 7,300
gal
× × = 1.46 × 1010 µg.
1gal
1
g
Because this is an intermediate result, we do not approximate it yet. Similarly,
1,000 L
Volume of lake = (1.273 × 106 )
m2 × (5.1
m) × = 6.4923 × 109 L.
1m3

Again, this is an intermediate result. Our final answer will be obtained by dividing the weight of
arsenic by the volume of the lake, and then approximating to the appropriate number of significant
figures. How many significant figures are appropriate? The data have 4 significant figures (lake area),
2 significant figures (lake depth), 2 significant figures (truckload volume), and 1 significant figure
(level of contamination of the wastewater). According to Rule 3 on the previous page, we should
approximate the result to the smallest of these, i.e., 1 significant figure. Thus we write
9 This “concentration” is a weight divided by a volume, so it is not a pure number. Compare Example 5 on page 15;
where is the important difference?
1.46 × 1010 µg
Arsenic concentration = = 2.2466 . . . µg/ L ≈ 2 µg/ L.
6.4923 × 109 L
The answer is 2 micrograms per liter, to the appropriate level of precision, which is 1 significant
figure. (For comparison, in 2001 the EPA set a maximum level of 10 micrograms per liter for arsenic
in domestic drinking water [12].)
1.3.3 Orders of Magnitude

One or two significant figures give a useful level of precision. In many of the calculations that we will
do in this course, one or two figure precision will be quite sufficient. Quite often, indeed, the quantities
that we are trying to understand cannot be measured with greater precision than this.
But there are other situations in which even one significant figure represents a greater level of
precision than we can attain—or than it is useful to work with. Is there a notion of precision still more
wide-ranging than that expressed by “one significant figure”? Indeed there is: in fact we have seen an
example in Statement (c) of Example 2 on page 33. The level of precision expressed here is called an
order of magnitude. Let’s make a definition:
Definition 5
Let A and B be two numbers, or two physical quantities measured in the same units. We
say that A and B have the same order of magnitude if when we divide them, the result is
between 10−1 and 10,
1 A
< < 10.
10 B
It does not matter which quantity we pick as A and which as B: the definition gives the same
result either way.
Problem 6: Do the numbers 7 and 28 have the same order of magnitude?
Solution: Yes. Dividing the larger number (28) by the smaller (7) yields 4, which is between 10−1
and 10.
Problem 7: Does 1 million feet have the same order of magnitude as 5 kilometers?
Solution: In order to answer this question, we need to express both quantities in terms of the same
unit. We choose to use kilometers as the common unit, so we must express 1 million feet in kilometers.
Using the conversion factors from Tables 2 on page 8 and 8 on page 12, we get
1 yd 1mi 1.61 km
106 ft × ×

3 ft 1760
yd
×
1 ≈ 305 km.
mi
Thus the ratio
106 ft 305 km

≈ ≈ 61
5 km 5
km

is greater than 10, so the answer is no: the two lengths are not of the same order of magnitude.
Problem 8: According to Vaclav Smil, a professor at the University of Manitoba, the total mass of
all the termites on Earth is at least 500 million tonnes. In terms of order of magnitude, how does this
compare with the mass of all the humans on Earth?
Solution: Let’s try to estimate the mass of all humans on Earth. We’ll assume 100 kg for the mass
of a human, which is something of an overestimate but gives an easy calculation. The total mass of
humans is equal to the number of humans times the mass of each one. Let’s include the conversion
factor from kilograms to tonnes as well, so as to get a final answer in tonnes:
7 × 109 × 100
humans
k
g 1T
= 7 × 108 T.
× 103

1human kg

Divide this by 5 × 108 T of termites, and you’ll see that we get a number between 10−1 and 10. That
means the two masses have the same order of magnitude. When all insect species are taken into
consideration, the total mass of insects is thought to be of greater order of magnitude than that of
humans—in fact, several tonnes of insects per human!
Orders of magnitude are closely related to scientific notation,
the topic that we have studied in Section 1.2. When we represent a
number in standard form, the most important information about that
number is the power of 10, or the exponent. In fact, if we completely
neglect the significand and simply take the “nearest” power of 10,
we will get a number that has the same order of magnitude as the
original one. Thus, “order of magnitude” calculations simply involve
manipulating powers of 10.
Remark 5: You might ask which power of 10 is “nearest” to a given
significand. The rule we use is that significands (in standard form)
beginning with 1, 2, 3 are “nearest” to 1, while significands beginning
Figure 12: Planes at San Francisco Airport. with 4 and greater are “nearest” to 10. So, for instance, the “nearest”
power of 10 to Earth’s human population of 7 billion is 1010 , not 109 ,
whereas the “nearest” power of 10 to the mass of a flea (estimated at
2.2 × 10−4 g) is 10−4 g.
Estimating Orders of Magnitude

The order of magnitude of a quantity is a very crude approximation to that quantity. You might be
surprised to hear, though, that their very crudeness makes orders of magnitude extremely useful. There
are two reasons for that:
• Orders of magnitude are easy to estimate. We already observed this in Problem 8 on the
previous page where we used the easy order-of-magnitude estimate 100 kg for the mass of an
average human being. We know that this number (about 220 pounds) is a bit of an overestimate,
but it’s certainly closer than 10 kg or 1000 kg, so it has the right order of magnitude.
• If two quantities have different orders of magnitude, then that difference is pretty significant. If
the quantities are related to sustainability questions, an order of magnitude comparison can give
us important information on which to prioritize. For example, an article in the San Francisco
Chronicle [269] reported that the installation of solar panels at San Francisco International
Airport’s Terminal 3 “is expected to reduce energy use at the airport by 15 percent” within a few
years (Figure 12). That’s true—if you neglect the energy used by the actual airplanes arriving
and departing from the airport, which are, after all, the reason for the airport’s existence! The
amount of energy used by these airplanes is about three orders of magnitude greater than the
savings from installing solar panels.
Think about it. . .
When you read that energy use by airplanes at San Francisco Airport is much greater than
the savings coming from installing solar panels at the airport, what is your reaction?
(a) Installing solar panels at an airport is simply a distraction from the much larger
question of the fossil fuel consumption (and greenhouse gas emissions) created by
our desire for air travel.
(b) Why are you so negative about a well-intentioned effort? Every little bit helps.
(c) The comparison is not appropriate. The demand for air travel comes, through the
market, from thousands of consumers. The San Francisco Airports Authority is one
public utility making its contribution to save energy and fight climate change.
(d) Something else.
Whatever alternative best describes your reaction, try to imagine the best arguments that
someone else could make for the other alternatives. How would you respond to them?
What matters most in understanding a quantity is its order of magnitude. And, as we said
above, orders of magnitude are relatively easy to estimate. In their book Guesstimation [345],
authors Lawrence Weinstein and John Adam give two simple ideas for generating order-of-magnitude
estimates. Here they are.
Rule 4: Order-of-Magnitude Estimation
Idea 1: Write down the answer, straightaway. In other words, come up with a “reasonably
close” solution. Remember we are only trying to find the nearest power of 10.
Idea 2: If you can’t estimate the answer straightaway (in other words, if Idea 1 fails), break
the problem into smaller pieces and estimate the answer for each one. You only need
to estimate each answer to within a factor of ten. How hard can that be?
Here are some examples.

Problem 9: Find an order-of-magnitude estimate for the number of clowns who can fit inside a VW
Beetle.
Solution: This is a problem we can solve using our imagination directly (Idea 1). Clearly, 4 clowns
will fit into a Beetle, and 40 clowns will not. Since our answer must be a power of 10, bigger than 4,
and smaller than 40, our order-of-magnitude answer is 101 .
Problem 10: Find an order-of-magnitude estimate of the volume of gasoline burned in a year by a
typical U.S. driver.
Solution: This is probably not something we can imagine directly. But we can break the problem
down into two simpler ones (Idea 2) that we can solve using our imagination and experience:
• How many miles are driven per year by the typical driver?
• What is the typical gas mileage?

How many miles a year does the average American drive? Is it 12,000? 15,000? 30,000? Notice that
all of our guesses have the same order of magnitude. While we might not agree what number is a
good guess, we can agree that the order of magnitude should be 104 . Similarly, what about the typical
gas mileage? You may know cars that get 20 miles per gallon while others get 30 miles per gallon.
Regardless, the order of magnitude is 101 . Therefore, our order-of-magnitude answer is
× 1 gal = 103 gal.

104
mi
101mi

Our estimate is 1,000 gallons of gasoline per driver per year. In this example we can compare our
back-of-the-envelope estimate to a more accurate figure. According to the U.S. Energy Information
Administration [6], the U.S. consumed about 3.3 × 109 barrels of highway gasoline in 2015
(remember, a barrel is 42 gallons). According to the Department of Transportation [111], we may
estimate10 approximately 2.2 × 108 licensed drivers in that same year. Dividing, we get
3.3 × 109 × 42 gal ÷ (2.2 × 108 drivers) = 630 gal

bbl
1bbl
driver
which is fairly close to our order-of-magnitude number.
Problems that require you to estimate orders of magnitude are sometimes called Fermi problems,
after the physicist Enrico Fermi, who was famous for the skill and speed with which he could
make estimates of this sort (see the video [220]). Such estimates are a great way to understand the
world around you from a simple yet quantitative perspective, whether you use that skill to work on
sustainability or on other issues. According to the authors of [345],
You can also use these tools to further your career. Many top companies use estimation
questions in job interviews to judge the intelligence and flexibility of their applicants.
Leading software firms, management consultants, and investment banks. . . ask questions
such as What’s the size of the market for disposable diapers in China? How many golf
balls does it take to fill a 747? and How many piano tuners are there in the world?
Companies use these questions as an excellent test of the applicants’ abilities to think on
their feet and to apply their mathematical skills to real-world problems.
We are not necessarily encouraging you to become an investment banker, but this is a good moment
to remember that the skills we are learning through this book have a very wide range of applications!
You can get much more practice in making these estimates at the Fermi Questions website [217]. Here
is one more example from [345]:
Problem 11: Find an order-of-magnitude estimate for how many people are airborne (as commercial
airline passengers) over the United States right now.
Solution: Again, our imagination probably doesn’t give us any direct access to this number. But what
we can do without too much trouble is to think about our own experience. How many flights did you
take in the last year? How long did they last, in total? Then we argue as follows (Idea 2): if your
flying is representative of the average American, then the fraction of your time that you spent in the
air is the same as the fraction of everyone else’s time that they spend in the air, which is in turn
equal to the fraction of the U.S. population that is in the air right now. Now we know that the U.S.
population is about 3 × 108 , and I might guess that I spend 12 hours (half a day) in the air each year.
Imagining myself as representative of the average American, I estimate that the average American
10 The most recent year for which the OHP provides data is 2009; this estimate assumes that the growth in the number of
drivers continued for the next years at the same rate as in the previous decade.
spends 21 /365 ≈ 10−3 of his/her time airborne. Therefore, I estimate that the number of airborne
Americans right now is
(3 × 108 ) × 10−3 ≈ 3 × 105 ,
or, to the nearest power of 10, just 105 . That is a bigger number than I might have expected!
Summary of Ideas: Estimates, Precision, and Orders of Magnitude
• The precision of a measurement claim expresses how close the measurement is

claimed to be to the true value of the quantity measured.
• One way of describing precision is in terms of the number of significant figures.
• When we multiply or divide quantities, the number of significant figures in the result
is the least number of significant figures among the quantities we multiplied or
divided.
• Another (lesser) level of precision is described by order of magnitude. Two
quantities (with the same units) A and B have the same order of magnitude if
1 A
< < 10.
10 B
• An Order-of-magnitude estimate for a quantity refers to the nearest power of 10 to

that quantity.
• Order-of-magnitude estimates can often be obtained easily by following Rule 4 on
page 41.
1.4 Communicating Quantitative Information
Objectives
I can understand when and why I might want to communicate quantitative informa-
tion to nonmathematical readers or listeners.
I can distinguish between verbal, tabular, and graphical ways to communicate
quantitative information, and I know when each is effective.
I can design tables to be clear and readable.
I can identify when it is appropriate to use a line graph, a bar chart, a scatter plot,
and a pie chart to represent a set of data.
I can design each of the above four types of graphical presentation to be clear and
readable.
I am aware of some of the ways graphical presentations can be used to misrepresent
data. I will avoid them in my own work and be alert for them in the work of others.
By the time you have finished this book, you will be able to carry out quite sophisticated calculations
about growth, pollution, planetary climate, risk and risk management, decisions, and payoffs. The day-
to-day work of a professional scientist in one of these areas—an atmospheric physicist or an ecological
economist or a statistician—will involve still more complicated calculations of the same general kind.
In the end, though, all the analysis and calculation ought to lead a democratic society to some
decisions. Do we build this pipeline? Do we sign this treaty? Do we subsidize this innovative industry
(and if so, by how much)? Do we build a nationwide, interconnected energy grid, or would a
more decentralized system be safer or cheaper? And as society tries to make these decisions, the
scientists and mathematicians need to communicate quantitative information in a way that makes its
implications clear and easy to grasp. That is what this section is about.
We will think about three ways of communicating such information: verbal (using words),
numerical (using numbers in a text or table), and graphical (using charts or graphs).
1.4.1 Verbal Communication

Sometimes, the quantitative data to be presented are so straightforward that a summary in words
is enough to vividly convey the underlying information. This is often the case when what we wish
to convey is a single “data point” rather than a comparison (between multiple data points) or a trend
(information about how something varies over time). We saw a striking example of this in Example 10
on page 20. After some calculation, we reached the conclusion that “out of every ten gallons of water
on Earth, just under a teaspoon is fresh.” No numbers or charts are needed to explain this; it stands
by itself as a reminder of how rare a resource fresh water is, even though water itself is abundant on
Earth and indeed covers well over half the planet’s surface.
Here is another, similar example.
Problem 1: Geologists tell us that our present-day oil reserves were formed from the bodies of tiny
marine organisms that became trapped in ocean-floor sediments during the Mesozoic period, roughly
100 million years ago. In roughly 250 years of industrial civilization, humanity has used about one-
quarter to one-half of these reserves. Find a vivid verbal way to contrast the length of time that it took
our oil reserves to be formed and the speed at which we are burning them up.
1.4. COMMUNICATING QUANTITATIVE INFORMATION 45
Solution: If we assume (perhaps optimistically) that humanity has consumed only 41 of the Earth’s oil
reserves in 250 years, then the length of time to consume the entirety of those reserves would be
250 yr ÷ 14 = 250 yr × 4 = 1,000 yr.
Since the time to form the reserves is 100,000,000 yr, and
100,000,000 yr
= 100,000,
1,000 yr
we could say “humanity is burning up Earth’s oil reserves a hundred thousand times faster than they
were formed.” That is pretty striking! But let’s look for a way to illustrate it in still more direct terms—
in a way that does not involve the big number 100,000 (which is too large for many of us to wrap our
heads around).
Remember, there is no one right answer here. But a way of expressing the information that may
be very evocative involves rescaling (see Example 10 on page 20 for this concept) the times involved
by a factor of 100,000,000. Thus the 100,000,000-year period to form the oil reserves rescales to 1
1
year, and the 1,000-year period to burn them up rescales to 100,000,000 of 1,000 years. We’ll convert
the answer to minutes:
1 365day
24 hr 60 min
× 1,000yr × ≈ 5 min.

× ×
100,000,000 1
yr 1
day 1hr

Thus, if the time to form all our oil reserves is represented by a year, the time during which humanity
is expected to consume all of them is represented by five minutes.
Every Fourth of July, State College, PA (the home of Penn
State), holds the largest all-volunteer fireworks display in the
United States. Planning for the Central Pennsylvania 4th Fest
begins on July 5th the year before, so it is not inaccurate to
say that the Fest takes a year to set up. It usually takes 30–45
minutes to shoot off the 12,000 or so firework shells that make
up the display. That makes an impressive, indeed overwhelming,
amount of noise, light, and smoke.
Imagine now that, perhaps owing to a software error, all the
shells are shot off in five minutes11 instead of 45. You are
basically looking at a nonstop explosion for five minutes. But
you are also looking at an analogy for the extraordinary rate at
Figure 13: Fireworks.
which humanity is consuming its fossil fuel reserves. A year to
set up, five minutes to burn. This picture of a crazy, out-of-control fireworks display provides a vivid
way of representing the numerical data about oil consumption. And it invites the right question: No-
one thinks of a fireworks display as anything other than a brief, spectacular excitement—it lights up
the night for a moment, but darkness rolls back again soon enough. What might make us think, then,
that by our discovery of oil and other fossil fuels we have driven darkness away forever?
Human Terms Again

Did you recognize what we were doing in the previous example? We were simply following the
methods laid out in Definition 3 on page 17 and the subsequent discussion to express the age of our oil
reserves in human terms. Specifically, we were using the rescaling method to translate the comparison
between two large quantities—the age of the oil reserves and the length of time it may take to consume
them—into another comparison that it is easier for us to grasp: the comparison between the one-year
setup time for the fireworks show and the five minutes it takes to shoot off all the shells at once.
Here is one more example, which relates to plastic pollution in our oceans.
11 Something like this actually happened in San Diego in 2012.
Example 1: Consider the Great Pacific Garbage

Patch (GPGP). If you know a little bit about ocean
currents, then you probably know that they move in
circular patterns called gyres. (There’s a bunch of
reasons for this—the basin topography (the shape of
the land under the ocean), the rotation of the Earth,
the differences in temperature and saltiness of the
different parts of the ocean—are three of the most
important.) When garbage (especially plastic) makes
its way to the ocean, these gyres pool it in particular
locations. The biggest of the gyres is the North Pacific
Gyre, and the pool of our garbage in this gyre has
become infamous as the Great Pacific Garbage Patch
Figure 14: Ocean currents and garbage.
(see Figure 14, and [167] for more information). It
has been estimated that the area of the GPGP is
between 700,000 and 15,000,000 square kilometers.
How can we express these measurements in human terms? Because the areas involved are so large,
we can’t express them with “human size” units and small numbers—that is, we can’t follow both
parts of Rule 1 on page 17. But we can try to make the number part more humanly accessible by
choosing a measurement that will be more familiar to the reader and is of somewhat similar size to
the measurement we are trying to express. That is what we did by choosing the “volume of Lake
Superior” as a “unit” in Example 8 on page 19.
How might this work? Let’s take as our “unit” the area of the commonwealth of Pennsylvania.
(Pennsylvania happens to be our state, so it is reasonable for us to hope that at least some students
have traveled the state and have gained a feeling for the amount of land that it contains. But if you
are using this book in a different state, of course you should substitute that state for Pennsylvania and
change the numbers accordingly. The point is to have a large area that is a little familiar at least.)
Let’s do some calculations. We find online that the area of Pennsylvania is 119,283 square
kilometers. How does that compare to the size of the GPGP? We’ll consider both the “low end”
and “high end” estimates of its area.
Solution: We will start with the area of the GPGP and estimate how many Pennsylvanias we could fit
in that area. We will first compute the low estimate and then the high one:
1 Pennsylvania
700,000 km2 × ≈ 5.87 Pennsylvanias;
119,283 km2
1 Pennsylvania
15,000,000 km2 × ≈ 126 Pennsylvanias.
119,283 km2
From our two calculations, we see that the area of the GPGP is roughly between 6 Pennsylvanias
and 126 Pennsylvanias. That’s a very large range of large areas! But by expressing the area in terms
of “Pennsylvanias,” we have made the huge size at least a bit easier to grasp. It would be completely
safe to say, “An area of the Pacific Ocean many times larger than the state of Pennsylvania—perhaps
even a hundred times larger—is covered with plastic trash. What do you think about that?”
1.4.2 Tables
Simple verbal communication, as we have discussed in the previous section, can be used effectively
to convey numerical data, provided that only one or two numbers are involved. For example:
In a survey, 27 percent of students identifying as male said that they worked out regularly
(twice or more a week), compared to 34 percent of students identifying as female.
Or this:
At the end of 2015 the U.S. had an installed solar photovoltaic generating capacity of
approximately 25 gigawatts (GW) peak. The installed wind power generating capacity at
the same time was 75 GW peak.
But if three or more numbers are involved, a text including

them all will feel cluttered and will not be easy to understand. Source Peak Capacity (gigawatts)
For instance, suppose we also wanted to show the installed Natural Gas 504
power generating capacities sourced by coal, natural gas, Coal 305
Nuclear 104
hydropower, and nuclear fission. Writing all these down in a
Hydro 79
single sentence would be unwieldy and unhelpful. It is better Wind 74
to display the data in a table, as in Table 11. Solar PV 22
Remark 1: Notice that Table 11 contains only a couple of Table 11: U.S. installed electric generating capacity by
dividing lines—one beneath the first (title) row and one sepa- source in 2015 (gigawatts) [7].
rating the columns. Especially if you often use a spreadsheet
program like Excel, it is tempting to put each piece of data
into its own “box.” Don’t do this! The extra row and column dividing lines will actually make your
table harder, rather than easier, to read. If you want to help the reader’s eye move across the rows, an
alternating, gentle shading is usually more effective than heavy grid lines. For example, see Table 12,
which shows the steady increase in U.S. gas output over the past ten years.12
Year Q1 Q2 Q3 Q4
2005 4595 4652 4432 4370
2006 4474 4591 4693 4743
2007 4644 4784 4856 4980
2008 4999 5059 5001 5098
2009 5192 5162 5148 5120
2010 5115 5240 5404 5555
2011 5415 5665 5789 6031
2012 5937 5898 6091 6105
2013 5879 5999 6158 6168
2014 6077 6345 6585 6719
2015 6606 6736 6885 6804
Table 12: U.S. quarterly natural gas production (billions of cubic feet) [8].
We will summarize “best practice” for using tables in a series of rules:
12 This is mostly the result of hydraulic fracturing (“fracking”), a new technology that has enabled us to extract gas from
deposits that were previously thought to be inaccessible.
Rule 1: Rules for Tables
• When you want to present three or more pieces of numerical information, consider
using a table rather than a verbal presentation.
• Presenting information in a table is most helpful when the reader will need to look
up individual values. (If trends or overall comparisons are of more interest, consider
a graphical presentation instead.)
• Since most people find it easier to see patterns in numerical data by reading down
columns rather than across rows, it is a good idea to plan your table so that the most
important patterns appear in the columns.
• The level of precision used for data presented in a table should not exceed the
minimum needed to communicate effectively with your readers. See the discussion
in Section 1.3.1, as well as Remark 2.
Remark 2: When we present numerical information in the form of a table, the question of precision
is bound to arise (see Section 1.3.1). Remember, the level of precision claimed by a numerical
measurement describes the width of the range (around the claimed value) within which the true value
of the measured quantity should lie. Thus, if I say (as in the last line of Table 12 on the previous page)
“U.S. natural gas production in the first quarter of 2015 was 6606 billion cubic feet,” I am implicitly
claiming a level of precision of plus or minus a billion cubic feet (compare Rule 2)—that is, I am
saying that the actual production level was somewhere in between 6605 and 6607 billion cubic feet.
In Section 1.3.1 we talked about how important it is, especially in environmental calculations, not to
claim a greater level of precision than we can in fact justify.
1.4.3 Graphical Communication

Along with verbal and numerical (or tabular) ways of communicating mathematical information, we
need also to be familiar with graphical communication. Using graphical communication, we encode
numerical values as visual objects, often presented relative to one or two axes. A graph, therefore, is
a visual display of quantitative information. There are many different kinds of graphs and charts. We
will consider four in this section: a bar chart, a pie chart, a scatter plot, and a line graph. Let’s think
about the two kinds of charts (bar charts and pie charts) first.
These charts are used when we are dealing with categorical information—data that is presented
in a variety of distinct categories. The information in Figure 11 on the previous page, for example,
is categorical—the energy source is categorized into coal, natural gas, nuclear and so on. We can
represent this data in the form of a bar chart by drawing six horizontal bars, of the same width but of
different lengths in proportion to the various measures of generating capacity. Thus we obtain a bar
chart that might look something like Figure 15 on the opposite page.
Remark 3: One can make bar charts with either horizontal or vertical “bars”; use whichever approach
is most convenient. A bar chart with vertical “bars” is sometimes called a column chart or “column
bar chart.”
In a bar chart, the lengths (or heights in the vertical case) of the bars represent the size of the
numerical data being conveyed. An alternative way of representing similar data is by a pie chart: in
this case, the numerical values are represented by the angles of different sectors rather than by the
lengths of different bars. Figure 16 on the opposite page represents the same data in pie chart form.
Solar 22
Wind 74
Hydro 79
Nuclear 104
Coal 305
Natural Gas 504
0 50 100 150 200 250 300 350 400 450 500 550
Peak Capacity (Gigawatts)
Figure 15: Bar chart representation of the data in Table 11 on page 47 on U.S. peak generating capacity from various sources
(gigawatts).
Natural Gas
504
22
Solar
74
305
Wind
79
Coal 104
Hydro
Nuclear
Figure 16: Pie chart representation of the data in Table 11 on page 47 on U.S. peak generating capacity from various sources
(gigawatts).
In both bar and pie chart representations, the areas of the bars or “pie pieces” represent the size of
the relevant numerical data. Because in a pie chart, the total area of all the pieces has to add up to one
whole pie, this kind of chart is particularly helpful in giving a visual impression of what proportion
of the total a particular category makes up. For instance, from the pie chart above you can quickly
see that coal, nuclear and hydro together account for about half of U.S. (peak) electricity generating
capacity.
Remark 4: It is helpful to use software such as Excel to prepare charts of this sort—of course
you can draw them by hand, but if a computer can draw them quickly and accurately instead, why
bother? One problem with using software, though, is that many programs offer a wide range of purely
cosmetic options—making the bars or pies “3-dimensional,” adding images or shading or shadows,
and so on. Most of these “effects” do not improve communication, but rather the reverse; this is
particularly so for 3-dimensional effects, which can confuse our understanding badly, for reasons
we will see in a moment. Therefore, even if you have access to software that enables you to add
“effects” to your charts, we encourage you not to do so. Remember, our desire is simply to achieve
clear communication; nothing more than that.
Bar and pie charts are used when we are relating categorical data (like the different types of
electricity generation) to numerical data. By contrast, scatter plots and line graphs are used when
we are relating two different sets of numerical data. For example, we could consider the information
in Table 12 on page 47 as representing the relationship between two different sets of numerical data—
one set representing the year and quarter, and the other set representing the natural gas production in
that quarter. Figure 17 is a scatter plot showing the data from this point of view.
Remember, this plot shows exactly the same data as in
Table 12. Compared to the table, it has some obvious advantages
6,000 and disadvantages. If you want to know the exact amount of
natural gas produced in the third quarter of 2011, the table is
more useful to you. But if you want to estimate, overall, how
Production
4,000
fast U.S. natural gas production has been increasing over the past
decade, the graph will be much more helpful. It will even allow
2,000 you to “eyeball” the answer to a question like “is the increase in
natural gas production leveling off?”—it clearly looks from the
0 graph as though the answer is “no” (at least over the range of
2004 2006 2008 2010 2012 2014 2016
dates that it covers), whereas it would take some thinking even
Year
to figure out how to express the question in terms of data in the
Figure 17: Scatter plot of data from Table 12 on table, let alone how to answer it!
page 47 on U.S. natural gas production (billions of
cubic feet per quarter). The final kind of graphical representation that we mentioned
is a line graph. You can think of this, if you like, as a special
kind of scatter plot in which the points are so close together that
they make up a smooth curve. The kinds of “graphs” you learned about in high school algebra (“Draw
the graph of y = 2x − 7”) are line graphs. Here is another example, the so-called Keeling curve. This
has been called one of the most important geophysical records ever made [224].
Figure 18: The Keeling curve, which shows atmospheric CO2 concentrations in parts per million by volume (ppmv).
The Keeling curve summarizes measurements of the carbon dioxide concentration (see Example 5
on page 15) of Earth’s atmosphere, made continuously since 1958 at the Mauna Loa Observatory
in Hawai‘i under the direction of C.D. Keeling. The red curve shows the actual measurements,
which fluctuate according to seasonal changes in global vegetation; the black curve is an average
that smooths these seasonal variations out. This graphical presentation makes it vividly clear that
carbon dioxide levels in our atmosphere are rising quite rapidly, and it can be shown that this rise
corresponds very accurately with the increasing rate at which humans are releasing carbon dioxide by
burning fossil fuels. The problems that this may give rise to, and the possible ways these problems
might be addressed, will occupy our attention in several later sections of this book. For the moment, we
simply want you to observe how effective graphical presentation of this data is—much more effective
than a corresponding table.
Think about it. . .
Notice that the y-axis of the Keeling curve does not start at zero, but at about 320 parts per
million by volume—the level of atmospheric CO2 when Keeling began his measurements.
How would the appearance of the curve differ if it was plotted using axes where the y-axis
did begin at zero parts per million? Which possible graphic (the version in Figure 18, or
the replotted version with y-axis starting at zero) seems more informative to you? More
accurate? Read ahead to the next section for more discussion of these issues.
Remark 5: To learn more about the skillful use of graphical presentation, and about how such
presentations can mislead (the subject of our next section!) we recommend Edward Tufte’s classic
book [325].
1.4.4 How Presentations Can Mislead

Visual, graphical communication is incredibly powerful. A well-constructed graphic can allow the
human eye to take in, in an instant, the significance of a mountain of data that might be impossible
to comprehend in numerical form. Software companies and consultants thrive in the area of big data
visualization, creating graphical tools that allow companies and researchers to get a visual picture of
the huge amounts of information that their efforts have generated.
But just because graphics can be so powerful, they can also be powerfully misleading. In this section
we’ll review a couple of the more common ways in which graphics can mislead, whether by accident
or design. We want to help you recognize misleading graphics when you encounter them, and also
to help you avoid creating graphics that fail in their purpose of communication by misleading other
people. Here are some key problems to look out for.
Example 2: Questions about the choice of origin
Go back and look again at Figure 17, showing the change in
7,000
U.S. natural gas production over time. In that figure, we started the
y-axis at zero. What happens if we instead start the y-axis about at
the lowest production figure (as we saw was done with the Keeling 6,000
Production
curve; see Question 1.4.3)? We get the result in Figure 19.

Both Figures 17 and 19 show the exact same data, with the
same accuracy, and yet there is no question that the second figure 5,000
gives a much more dramatic impression of growth than the first
one does. Is this a misleading impression? At the very least, it has
the potential to mislead. The percentage increase in gas production 2004 2006 2008 2010 2012 2014 2016
over the period covered is a little over 50% . It seems that Figure 17 Year
represents this accurately, whereas Figure 19 suggests a much
Figure 19: Scatter plot of the data from Table 12 on
more dramatic increase. page 47 on U.S. natural gas production (billions of
Political news releases often give us much more egregious cubic feet per quarter).
example of the same kind of thing. Figures 20 and 21 on the next
page are fairly recent ones. In both cases, the changes involved are
made to look much more dramatic than they really are by choice of origin on the y-axis. The first
figure doesn’t even specify the origin, though by examining the other data we can deduce that it is
at about 70% . In the second figure, the origin is at 94 million. Sure, in both cases the numbers have
increased over time, but we can be pretty confident that those who created the graphics have chosen
the y-axis origin with intent to make the increase look as dramatic as possible.
Figure 20: Obama administration (2015) graphic of high school graduation rates.
Figure 21: Senate Budget Committee (2012) graphic of welfare rolls.
104
How can we avoid this kind of problem? Some people propose a rule that the
y-axis should always start at zero. But this is too general. There are plenty of
Temperature (degrees F)
102 examples for which the zero point is not relevant, or even is arbitrary, whereas the
changes in the y-value are extremely significant.
100
Consider, for instance, a chart of someone’s body temperature over time, such
98 as might be recorded in a hospital. If my temperature chart over a couple of days
0 10 20 30 40 50
Hour looks like Figure 22, we can definitely say that I am very sick; I have a fever
Figure 22: Body temperature
that is spiking rapidly and has reached the level (over 104 ◦ F) that is considered
plot. extremely dangerous. But if we plotted the same data with a baseline of zero, this
important information would be obscured (Figure 23).
In this case, starting the y-axis at zero has led to a decrease in the effectiveness
of communication. Zero, in this case, does not reflect anything relevant to the
100
80
question at hand. In fact, we have already seen that zero degrees Fahrenheit is
60
simply the coldest that Fahrenheit could achieve (see Remark 4 on page 13) using
40
the technology of the early eighteenth century in Europe.
20
A similar point, by the way, applies to the Keeling curve. We asked what the
0
0 10 20
Hour
30 40 50
significance might be of the fact that the Keeling curve is not based at zero. Here
again, the answer is that what is significant is not how the CO2 measurements
Figure 23: Body temperature
plot starting at zero.
are related to zero but how they are related to a “normal” level; we want to see
whether the planet is “running a fever” as measured by carbon dioxide levels that
are far above normal. The usual presentation of the Keeling curve takes the levels
when Keeling started his measurements as “normal.” (In fact, human-generated emissions had already
had a measurable effect by that point, but much smaller than the effect recorded by Keeling between
the 1960s and the present day.)
Figure 24: Prevalence of West Nile virus. Avoid the use of three-dimensional charts.
Even though starting the y-axis at zero should not be considered an absolute rule, it is a good policy
in cases in which the zero level is significant, as it is, for example, in Figures 20 and 21.
Example 3: Problems with three-dimensional plots

We said above that three-dimensional plots are particularly liable to be perceived wrongly, whether
by accident or by intent on the part of the plot designer. Let’s look at some examples.
As you can see, Figure 24 is built on a map of the continental United States and purports to track the
incidence of the West Nile virus in different states for one year. But what is it that actually represents
the number of West Nile cases? The height of the box—or its volume? The eye is naturally drawn to
the volume of the box, so that a “prism” of given height above Texas looks much more significant than
one of the same height above Pennsylvania, but it is not clear from the captions whether that is in fact
how the chart is to be read. (These kinds of confusion between height and volume are very common
when people read “3-dimensional” charts.) Perspective makes the areas of the more northern states
look smaller in any case—another opportunity for confusion. Finally, some “prisms” are partly hidden
behind other prisms. All in all, not a good representation of an interesting set of data. The author is
really trying to convey a relationship between two numerical quantities (distance from the Mississippi
and West Nile prevalence), so a clearly labeled scatter plot would be much more informative.
Critical Thinking
We have cited 3-dimensional plots as one example in which sophisticated data visualiza-
tions, made possible by technology, can obscure the actual information being presented—
whether accidentally or malevolently. Can you think of other examples in which new ways
to communicate information sometimes work to obscure it?
Perspective errors, which we mentioned above, are front and center in the next chart (Figure 25
on the next page). The two pie charts show the exact same data on U.S. energy annual consumption
by each source as a percentage of the total. But in the left-hand pie chart, the “pie” has been made
three-dimensional, then flattened out and displayed in perspective. The effect is to make the “pie
segments” nearest to the viewer, especially the red segment corresponding to natural gas, look bigger
relative to the others. Without the reported percentages, the viewer, who naturally perceives areas,
could be left with the misleading impression that natural gas accounts for the largest share of our
Petroleum Natural Gas Coal Renewables Nuclear
8.6%
8.6%
36.9% 10.5% 36.9%
10.5%
14.6%
29.3%
14.6%
29.3%
Bad Good
Figure 25: U.S. annual energy consumption by fuel source as a percentage of total (2016). Avoid using 3-dimensional charts.
energy consumption. The right-hand chart, which is an ordinary two-dimensional pie chart, correctly
shows that the largest share is made up by petroleum, and by a sizable margin.
Example 4: “Cherry-picking” (showing only a subset of the data) You will sometimes encounter
graphical presentations that mislead by ignoring or excluding data that contradicts a desired message
or conclusion. When you present data demonstrating a trend or correlation it is essential to include all
relevant data, not just a carefully selected subset of the data that is most favorable to a predetermined
conclusion.
As an extreme example, consider the following graph showing average temperatures in Pennsylva-
nia over time.
90
80
70
60
50
0 1 2 3 4 5
Time interval
Figure 26: Average temperatures in PA plotted against time.
Wow! So much for global warming, huh? What these figures demonstrate is a rapid and definitive
cooling trend. But no doubt you have guessed the trick: the “time intervals” on the x-axis are months,
and the plot runs from July to December. No great surprise to learn that it is colder in winter than
in summer, and no long-term conclusions about global warming can possibly be drawn. You might
say that the long-term trend, whatever it is, has been “contaminated” by the short-term cycle of the
seasons.
Most attempts to refute global warming by cherry-picking data are not as foolish as this (though
Senator Inhofe’s stunt presenting a snowball on the Senate floor, Figure 28 on the opposite page,
perhaps comes close). But more sophisticated cherry-picking is surprisingly common. In 2012
Britain’s Daily Mail newspaper published a front-page article entitled “Global warming stopped 16
years ago,” [275] and accompanied it by a chart similar to that shown in Figure 27.
Global Temperature Anomaly 1998–2014
0.8
Temp Anomaly ◦ C
0.6
0.4
0.2
98
00
02
04
08
10
12
14
06
19
20
20
20
20
20
20
20
20
year
Figure 27: Global monthly mean temp anomaly1997–2014, data from [119].
Certainly, this chart tells us that one data set showed a very small
change in surface temperatures over the period indicated. But as
it happens, there was some cherry-picking going on here, some
inadvertently. The starting point of the chart was an unusually hot
year (because of the El Niño phenomenon, a periodic weather
cycle in the Pacific Ocean). And it turns out, though of course
the report’s authors could not have known this at the time, that
the end point of the chart was unusually cool (the next El Niño
was about to get going, and the successive years 2013–2016 have
each been notably warmer than the previous one). We can put this
figure in context by looking at a larger data set. Figure 29 on the
next page plots the global mean monthly temperature anomaly
from the year 1880 through 2016. The small data set included
in Figure 27 is indicated by the yellow line. In this larger data
set, the so-called “global warming pause” is barely visible. We’ve
highlighted a longer “pause” from 1945 to 1980 in orange. We are Figure 28: Sen. J. Inhofe carries a snowball into the
Senate to “refute” global warming.
reporting the anomaly in average surface temperatures, but most
of the warming that Earth experiences is absorbed by the oceans.
The reasons for these apparent pauses have to do with cycles of ocean warming and cooling, such as
the El Niño phenomenon; it turns out that even a 35-year time period may not be long enough to avoid
“contamination” of the long-term trend by other, shorter-term variable phenomena.
Question 1: What do you mean by the word anomaly in the previous paragraph and on the y-axes of
Figures 27 and 29 on the next page?
Answer: The anomaly in some quantity is the amount by which it differs from its normal or expected
value. The global temperature anomaly therefore measures how much the whole planet is overheating
or “running a fever”.
Global Temperature Anomaly 1880–2016

1.5
1
Temp Anomaly ◦ C
0.5
−0.5
00
20
40
0
80
00
20
60
8
19
19
19
18
19
20
20
19
year
Figure 29: Global monthly mean temp anomaly 1880–2016, data from NASA [119].
Critical Thinking
Cherry-picking data is very tempting, especially if you think that you “know” what the right
or virtuous answer has to be. To avoid it, try to cultivate the ethics of a scientist, whose
job is to tell the story “as it is” in the most accurate way possible; not to manipulate the
presentation of information so as to steer people to a predefined conclusion. This viewpoint
is sometimes called “skeptical” or “critical.” At the time of writing, the website Skeptical
Science [336] is a useful source of this kind of “skepticism” in matters related to global
warming. (But treat its claims skeptically, of course!)
Example 5: Fallacies of multiplication

Our final example of misleading communication is a purely verbal one, which comes about when
we put together a great many small quantities to obtain a large one. For example, you may have heard
the term energy vampires used to refer to appliances that continue to draw electrical power even when
they are not obviously being used. Microwave ovens, network routers, TVs, stereos, wireless phones,
and so on can all be energy vampires.13 Often the cell phone charger is singled out as a particularly
notorious example.
Problem 2: A public service announcement says, “Unplug your cell phone charger when it is not in
use! If everyone did this, we could save enough electricity to power half a million homes!” Assuming
that this is true, what percentage of your overall electricity use is represented by your cell phone
charger?
13 For definitive information, see the measurements carried out by the Lawrence Berkeley Laboratory at [190].
Solution: Half a million homes certainly sounds like a lot, but then there are a lot of homes in the
United States. Let’s unpack the numbers a bit. According to the Census Bureau there are something
like 130 million households in the USA. Half a million is about 0.4 percent of all U.S. households
500,000
(that is, 130,000,000 × 100%). So an equivalent way to express the claim made by the advertisement
is to say that cell phone chargers represent less than half of one percent of U.S. household electrical
consumption—or that (if you are “average” in this respect) your cell phone charger represents less
than half of one percent of your electrical energy consumption. It is good to save energy, but one half
of one percent is not a large saving.
In other words, we’ve obtained what sounds like an impressive total saving by multiplying a small
saving (from unplugging your cell phone charger) by a large number (the number of households in
the USA). But we have forgotten that the total energy consumption of each household must also be
multiplied by that same large number. Once we have done that, we realize that we have not made such
a big difference, relative to total energy consumption, as we might have thought.
Think about it. . .
David MacKay [202] writes about claims like the one about cell phone chargers:
“But surely, if everyone does a little, it’ll add up to a lot?” No. This if-everyone
multiplying machine is just a way of making something small sound big. The if-
everyone multiplying machine churns out inspirational statements of the form
“if everyone did X, then it would provide enough energy/water/gas to do Y,”
where Y sounds impressive. Is it surprising that Y sounds big? Of course not.
We got Y by multiplying X by a big number—the number of people involved.
If everyone does a little, we’ll achieve only a little. We must do a lot.
Do you agree with MacKay? Why or why not?
Summary of Ideas: Communicating Quantitative Information
• Three ways of communicating quantitative information are verbal, tabular, and

graphical.
• Verbal communication is most effective when only one or two pieces of information
need to be shared. “Human terms” comparisons enhance its effectiveness.
• Tabular communication is effective when three or more data values are to be presented
and where it is important that the reader be accurately aware of the individual values.
• Principles for designing effective tables can be found in Rule 1 on page 48.
• Graphical communication is effective to demonstrate trends and comparisons in a
way that can be grasped visually.
• Bar charts and pie charts are used to share categorical data, that is, data that is divided
into different types or “categories.”
• Scatter plots and line graphs display the relation between two types of numerical data.
• Graphical (and other) means of communication can mislead. Avoid misplaced
origins, abuse of three-dimensional effects, cherry-picking data, and fallacies of
multiplication.
• All these means of communication are intended to enhance citizen understanding and
thus the effectiveness of democratic decision-making.
1.5. EXERCISES FOR CHAPTER 1 59
1.5 Exercises for Chapter 1
8. Alaska Wildfires. In 2015, a below normal winter

1.1. UNITS AND MEASUREMENT snowpack and relatively dry summer contributed to an
above average summer wildfire season in Alaska in
1. Identify the number part and the unit part in which 5.1 million acres burned. During the first week
the measurements below. Which of the measurements of August 2015, 246 active fires were increasing the
would you say are expressed in “human terms”? area burned by 86 thousand acres per day.
Convert the total acres burned to square kilometers
(a) The distance around the equator is about 24,000
miles. and convert the additional acres burned per day to
square kilometers per day. Put these into human terms
(b) The lower 48 states receive, in total, about 1,400
by comparing with other geographical areas.
cubic miles of rainfall per year.
(c) Each year, the lower 48 states receive enough
9. In early May of 2016, the city of Fort McMurray,
rain to cover them to a depth of about 2 12 feet.
Alberta was evacuated because of an encroaching wild-
(d) Penn State (University Park) produced about fire. On the morning of May 5, the fire was described
15,000 tons of solid waste in 2012. as being 25% larger than the area of Manhattan Island,
(e) The average amount of solid waste produced at which has an area of 88 square kilometers. Estimate
University Park each day weighs as much as ten the size of the fire and compare this with a familiar ge-
elephants. ographical area (such as your campus, city, or county—
(f) My height is approximately 0.0011 nautical that is, if Manhattan Island isn’t already familiar!).
miles.
(g) My height is approximately 6 feet. 10. What is 4,000,000 millimeters in miles? Round
(h) Worldwide consumption of jet fuel is roughly your answer to the nearest whole number.
5,000,000 barrels each day.
11. Which is larger, 3 cubic feet or 2 square meters?
(Does the question even make sense?)
2. In what other ways could you express 3 million
centimeters in human terms?
12. If we wish to convert cubic inches into cubic
centimeters, what fraction should we multiply by?
3. The total rainfall in the State College, PA, area
averages 39 inches per year. If my house has a roof area
13. Finn wishes to convert 300 ◦ F into Celsius. He
of 650 square feet, how much rainwater (in gallons) has
does the following calculation:
fallen on it since it was built, 18 years ago?
5
(300 ◦ F − 32 ◦ F) = 167 − 32 = 135 ◦ C.
9
4. Use the unit-factor method to express 1 million
seconds in days. What, if anything, is wrong with Finn’s calculation?
5. Make a table of definitions and examples for units 14. Express yesterday’s minimum and maximum
of area, like the one in Table 2 on page 8 for units of temperatures for your region in the Fahrenheit, Celsius,
length. Try to come up with your own examples of each and Kelvin scales
area.
15. The daytime surface of Mercury reaches a tem-
6. An acre is defined to be 4840 square yards. How perature of 700 kelvin. Express this temperature in both
many acres in a square mile? Celsius and Fahrenheit.
7. About how many cubic yards are in 3 cubic 16. The American Dental Association recommends
meters? that you brush your teeth for two minutes twice a
day [302]. Calculate the amount of time you spend
brushing your teeth as a percentage of your total time, under fixed conditions (constant length of pipe,
assuming you follow this recommendation. constant pressure difference) is proportional to
the fourth power of the pipe’s diameter. Using
this fact, figure out by what factor one should
17. In 2017 the concentration of carbon dioxide in
change the diameter of a pipe in order to change
the atmosphere was around 405 parts per million by the flow rate by a factor of 4,000.
volume (ppmv). Describe what this means in your own
(b) A standard gas pump hose has an internal diam-
words, and express 405 ppmv as a percentage. eter of about 34 inch. Using the result of (a), find
the diameter of a hose of the same length that
18. In Exercise 4 on the previous page you used the would deliver gas (and, therefore, energy) 4,000
unit-factor method to show that one million seconds times more slowly.
is about 11.5 days. Thus, we can think of 1 part per
Your answer should be approximately the diameter of
million (1 ppm) as corresponding to 1 second out
a drinking straw.
of 11.5 days. Use this idea to express each of the
quantities listed below as 1 part per million (1 ppm)
and 1 part per billion (1 ppb) of quantities that are in 23. An electrician might be able to install a specialist
familiar terms. electrical outlet in your home that would deliver power
at the rate of 20 kilowatts: see the picture below. Redo
(a) 1 second,
the calculations of Exercise 22 using this outlet instead.
(b) 1 minute, How much difference does it make?
(c) 1 inch,
(d) 1 teaspoon (tsp) (1 gallon is 768 teaspoons).
19. The diameter of the Earth is about 8,000 miles. In

a scale model, the Earth is represented by a sphere the
size of a basketball, diameter 9 12 inches. How far away Figure 30: Five-pronged industrial AC plug.
is the Moon in the scale model? (The actual distance
from Earth to the Moon is about a quarter of a million
miles.) What size of sphere represents the Moon? (The 1.2. SCIENTIFIC NOTATION
actual diameter of the Moon is about 2,200 miles.)
24. Convert the following to scientific notation in
Does this correspond to the ball used in any game that
standard form:
you know?
(a) 75,000, (e) 0.0000002319,
20. The International Space Station (ISS) orbits (b) 2,100,000,000, (f) 23.189,
about 250 miles above Earth. Since the year 2000, no (c) 0.007, (g) 0.1056,
humans have traveled much farther from Earth than the (d) 17, (h) 0.0156.
ISS. Where would the ISS be situated in the model
described in the previous exercise?
25. Convert the following to decimal notation:
21. In 2015, Penn State’s University Park campus (a) 2 × 1010 , (d) 885 × 1011 ,
diverted 9000 tons of waste from landfills through (b) 3.1 × 10−9 , (e) 4.783 × 10−1 ,
recycling and composting programs. Convert this to
(c) 6.66 × 102 , (f) 1.01 × 10−2 .
tons per day and express the result in human terms,
using a unit that is easy to visualize.
26. Let a = 1.27 × 10−6 and b = 8.9 × 10−7 . Calcu-
22. Look back at Problem 17 on page 20. In its late the following and express your answers in scien-
solution, we calculated that the ratio of the energy tific notation:
delivery rates of the gas pump and the electrical outlet (a) a + b, (c) a × b,
was about 4,000.
(b) a − b, (d) a ÷ b.
(a) It is a law of physics (Poiseuille’s law) that the
rate of flow of a fluid through a circular pipe
27. Let x = 7.55 × 1018 and y = 2.1 × 1019 . Calculate make tea each year? Use scientific notation to express
the following and express your answers in scientific your answer.
notation:
(a) x + y, (c) x × y, 33. The Moon orbits the Earth once every 27.3 days.
How many orbits has the Moon made during your
(b) x − y, (d) x ÷ y.
lifetime? Suppose that the orbit of the Moon is circular,
with radius r = 250,000 miles. Then, during each
orbit, the Moon travels 2πr miles (the circumference
28. Assume that there are 250 million cars on the of the circle) in its journey around Earth. How far has
road in the U.S., that on average each car gets 25 miles the Moon traveled in this way during your lifetime?
to the gallon and drives 10,000 miles per year, and that Express your answer using scientific notation.
each gallon of gas burned generates 20 pounds of CO2
(carbon dioxide). How many tons of carbon dioxide are
emitted by all these vehicles over the course of a year? 1.3. ESTIMATES, PRECISION, AND
(Use scientific notation.) ORDERS OF MAGNITUDE
34. Without consulting any references, estimate the

29. In 2015, in a scandal affecting half a million
distance (in a straight line) from your present location
vehicles in the United States, Volkswagen was caught
to Des Moines, IA. What do you believe is an appro-
using software to cheat on emissions testing for diesel
priate level of precision for your estimate? Now use an
vehicles. VW diesels were found to emit about 1 gram
online mapping tool to check your answer. Was your
of nitrous oxide per kilometer driven (about 40 times
estimate correct to within the precision you specified?
the legal limit). In a vehicle with the software, the
emission control systems were fully activated only
when the vehicle was being tested. If each car is driven 35. A human being needs approximately 12 kilo-
20,000 kilometers per year (on average), estimate the calories of food energy per pound of bodyweight in
mass of the nitrous oxide pollution that is emitted order to sustain life for a day. The planetwide average
in one year from these 500,000 “dirty diesels.” (Use bodyweight of a human is about 140 pounds (Table 6
scientific notation). on page 489). The total human population of the planet
can be estimated as 7.5 × 109 people. Estimate the
number of kilocalories of food energy required each
30. A U.S. one dollar bill is 0.0043 inches in thick-
day by the entire human population, being sure to
ness. Convert this to meters, and express the result in
express your answer with appropriate precision.
scientific notation.
36. The meltdown at the Fukushima Daiichi nuclear

31. The salary of a member of Congress is $174,000 power plant, triggered by the Tohoku earthquake and
per year, a figure that (at the time of writing) has tsunami of March 11, 2011, was one of the most severe
been constant since 2009. Suppose that a member of nuclear accidents of modern times. In Iitate, Fukushima
Congress was paid at this rate from January 1, 2009, prefecture, outdoor radiation levels of approximately
until the day that you work this problem, and that the 30 microsieverts per hour were measured immediately
total salary payment was converted into dollar bills. following the accident. Suppose that (contrary to gov-
Using the result of the previous exercise, estimate the ernment advice) a resident of Iitate stood in the open
air continuously for 3 days following the accident.
thickness (in meters) of the resulting pile of dollar
Estimate the total amount of radiation this person
bills.
would receive.
A CT (computed tomography) scan of the abdomen
32. The current population of Britain is about 65 and pelvis can deliver a dose of approximately 20 mil-
million people. Assume that on average, each Briton lisieverts of radiation (1 millisievert = 1,000 microsiev-
drinks 3.5 cups of tea per day, each containing 0.2 erts). Compare the order of magnitude of radiation
kilograms of water that must be heated from supply received by the patient undergoing the CT scan and the
temperature (about 15 ◦ C) to boiling point (100 ◦ C). It hypothetical resident of Iitate whose radiation dose you
takes about 4200 joules of energy to heat 1 kilogram of calculated above.
water through 1 ◦ C. How many joules do Britons use to
37. “In an attempt to run a marathon in under 2 hours, is correct, how much did your fingernails grow during
Eliud Kipchoge ran a 26.2 mile marathon course in the time it took to complete this problem?
2 hours and 25 seconds, corresponding to an average
speed of 13.0547001903 miles per hour.” 3.56 × 106
What, if anything, is wrong with the preceding 42. Compute . Express the result in stan-
4.64 × 1011
statement? dard form with the appropriate number of significant
figures.
38. An aging underground natural gas storage facility
in Porter Ranch, near Los Angeles, leaked 1.1 × 105 43. Particulate matter is a form of air pollution con-
pounds of methane per hour between October 2015 sisting of small particles that are harmful when inhaled.
and February 2016, when the leak was finally plugged. PM2.5 is the abbreviation for fine particulate matter,
Convert this leakage rate to cubic feet of methane per smaller than 2.5 microns in size. Use scientific notation
year. You may use the fact that 1 cubic foot of methane to express 2.5 microns in meters to 2 significant figures.
weighs approximately 4.5 × 10−2 pounds at standard Follow-up questions: As of February 2017, there are 20
conditions for pressure and temperature. counties in the U.S. that have failed to meet EPA air
quality standards for PM2.5 pollution. These are called
39. Approximate the following quantities to the num- nonattainment areas. Do you live or go to school in
ber of significant figures specified. Write your answer a nonattainment area? What are the health effects of
in scientific notation, standard form: PM2.5 exposure?
(a) 467.967 to 2 significant figures;
44. Look in the “Useful Numbers” section (Sec-
(b) 299,792,458 to 3 significant figures. (This is the
tion 8.2) to find the area of the Earth’s oceans, and the
speed of light in meters per second.)
volume of the Greenland ice cap.
(c) 0.0000000059055118 to 2 significant figures. If the entire Greenland ice cap were to melt into the
(d) 15,157,486,080 to 1 significant figure. (This ocean, estimate the resulting sea level rise.
number was found online as the result of some-
one’s calculation of the distance, in inches, from 45. In 2016, the United State generated 35.5 terawatt-
the Earth to the Moon. It is a good example hours (TWh) of electrical energy from large-scale solar
of an excessively precise measurement claim.
power plants. Use scientific notation to express this
According to NASA, the distance from the Earth
in kilowatt-hours (the standard unit used in measuring
to the Moon varies by almost 10 percent over the
course of a month, because the Moon’s orbit is residential electrical energy consumption). 1 kilowatt-
not a perfect circle. So most of those impressive- hour is 1,000 watt-hours. 1 terawatt-hour is 1012 watt-
looking figures are. . . not significant.) hours. Express your answer to 2 significant figures.
46. Use scientific notation to express the following

40. Approximate the following quantities to the num- quantities in the specified units with the specified
ber of significant figures specified (you may need to number of significant figures:
search for the information online). Write your answer
in scientific notation, standard form: (a) Width of a human hair (100 microns), in inches,
to 1 significant figure.
(a) The current population of the United States to 2
(b) Diameter of tobacco smoke particles (0.01 mi-
significant figures.
crons), in meters, to 1 significant figure.
(b) The global population to two significant figures.
(c) Average length of an E. coli bacterium (0.002
(c) The U.S. federal debt to 2 significant figures. millimeters), in meters, to 1 significant figure.
(d) Last year’s U.S. federal budget to 3 significant
figures.
47. Is the average length of an E. coli bacterium of
the same order of magnitude as the width of a human
41. It is estimated that your fingernails grow at a rate hair? (See Exercise 46)
of 1.2 nanometers per second. Express this quantity
in units of meters per second, using two significant 48. A car is lowered (carefully) into an empty
figures and scientific notation. Assuming this estimate backyard swimming pool. If the pool is then filled with
all the gasoline that the car will burn over its working 55. Give an order-of-magnitude estimate of the num-
lifetime, will the gasoline cover the car? (If you try this ber of Rubik’s Cubes that could fit in an Olympic size
at home, be sure not to strike a match.) swimming pool.
49. Penn State’s Beaver Stadium has a capacity of 56. The ambient level of atmospheric mercury is 2
approximately 106,000 spectators. Estimate the num- nanograms per cubic meter. Estimate the amount of
ber of Porta-Potties needed to cater to all the tailgaters mercury inside your classroom (assuming the ambient
on a day when Beaver Stadium is filled to capacity. level of atmospheric mercury).
57. Estimate the number of unused AA batteries in

50. Penn State’s University Park campus serves
all homes in the United States right now.
roughly 50,000 students, faculty, and staff. Estimate the
number of plastic water bottles that are discarded on the
University Park campus in one year. (You can look this 58. Give an order-of-magnitude estimate for the total
figure up, but the point of this problem is to make a amount of water used to flush toilets in the United
reasonable estimate based on your own experience and States during one day.
things that you already know.)
59. Your local news station is on the scene of a crash
on I-95 involving a tractor trailer hauling unminted
51. Total U.S. federal expenditures for fiscal year pennies to the U.S. Mint in Philadelphia. The news
2014 were 3.50 trillion dollars (to 3 significant figures). desk urgently needs an estimate for the total number
Suppose that for fiscal year 2015, federal expenditures of pennies scattered over the northbound lanes of I-95.
increased by 396 billion dollars over the 2014 baseline. The legal weight limit for a tractor trailer on I-95 is
Using the appropriate precision, give your best estimate 40 tons (you can assume the truck was at this limit);
for 2015 total federal expenditures. one penny weighs 2.5 grams.
(This describes a real event that occurred on Septem-
ber 8, 2016. A newscaster reported that there were a
52. The total area of the U.S. is 3.8 × 106 square “gajillion pennies” on I-95.)
miles, and its population (at the time of writing) is
3.19 × 108 people. Calculate the average number of
60. Estimate the total amount of blood (in gallons)
people per square mile, expressing your answer, with
donated in the U.S. each year. After making your
appropriate precision, in ordinary (not scientific) nota-
estimate, research statistics on blood donation. How
tion.
close is your estimate? What assumptions did you
make about blood donations and how did these differ
53. Suppose that the average home in Centre County, from what you found in your research?
PA, has a roof area (viewed from above) of 1600 square
feet. Centre County receives about 40 inches of rain
or equivalent precipitation (snow in the winter) per 1.4. COMMUNICATING QUANTITATIVE
year. Estimate the total amount of precipitation, in INFORMATION
gallons, that falls on a Centre County roof during a
year. According to the EPA Partnership Program Water 61. In this section we have talked about verbal, tab-
Service, the average American family of four uses ular, and graphical ways to communicate quantitative
1.5×105 gallons of water per year in their home. Is this information. Do you think these are the only possibil-
figure of the same order of magnitude as the volume of ities? Try to name some other ways of communicating
rain you calculated above? this kind of information, especially some that modern
digital technology has made possible.
54. Estimate the height (in miles) of a stack of $100
bills equivalent to the net worth of the wealthiest person 62. It is estimated that U.S. consumers used, and
on Earth. The thickness of a $100 bill is approximately threw away, 16 billion disposable paper coffee cups
4 × 10−3 inches. in 2006, and that this number had increased to 23
billion by 2010. Each cup weighs approximately 0.7
ounces. What do you think is the most effective way to temperature. How does this relate to the discussions in
communicate this information? Prepare a chart, table, Section 1.4.4 and Example 2 on page 51?
or verbal presentation that you believe will bring home
to the viewer the amount of waste involved here. 67. Pick up a copy (or visit the website) of your
favorite major newspaper and search for a story that
63. It is estimated that the preindustrial level of CO2 , includes quantitative information (if visiting a website,
that is, the amount before the Industrial Revolution you might try a topic-based search on “Environment”
started the large-scale consumption of fossil fuels, may or “Energy” or “Science”). Discuss how the quan-
have been about 270 parts per million by volume titative information is presented. Is the information
(ppmv). Replot the Keeling curve on a scale that begins presented clearly? Does it connect with and support the
with 270 ppmv on the y-axis. Does the result convey a overall story?
significantly different message?
68. In 2016, the United States produced 315 million
64. The following table (see [79]) shows the annual tons of bituminous coal, 338 million tons of sub-
sales of hybrid (gasoline-electric) vehicles in the U.S. bituminous coal, 73 million tons of lignite coal, and
for years from 2000 to 2013.
1.6 million tons of anthracite coal. Present this data as
Year Sales a table and in graphical form.
2000 9,350
2001 20,282 69. Think of at least three questions related to
2002 36,035 Exercise 68 that would require additional (or different)
2003 47,600 data, and then try to answer them. For example, you
2004 84,199
might ask, “How does the coal produced in the U.S. get
2005 209,711
used?” The U.S. Energy Information Administration is
2006 252,636
2007 352,271 an excellent source of data for answering your ques-
2008 312,386 tions. If you look for data on coal at the EIA website,
2009 290,271 take note of the various ways they present quantitative
2010 274,210 information.
2011 268,752
2012 434,498
70. The EPA estimates that 258 million tons of
2013 495,771
municipal solid waste were generated in 2014 [15]
Represent this information in graphical form, using and 34.8% of this waste was recycled or composted.
what you feel is the most appropriate kind of chart Figure 10 on page 426 shows the makeup of mu-
or graph. What do you notice? Can you suggest any nicipal solid waste that was recovered (recycled or
explanation for this? composted), and Figure 15 on page 431 shows the
makeup of waste that was discarded (landfilled). Use
the information in these figures to discuss the potential
65. In a poll of voters in Florida carried out some
for increasing the recovery rate of municipal solid
time before the 2016 election, 45 percent supported
waste.
the Democratic candidate, 40 percent supported the Re-
publican candidate, 5 percent supported the Libertarian
candidate, 2 percent supported the Green Party can- 71. According to the EPA, 34.8% of the municipal
didate, and the remainder supported other candidates solid waste generated in 2014 was recovered (com-
or were undecided. Represent this information in a bar posted or recycled). In 1960, the recovery rate was
chart and in a pie chart. Which representation is better under 6%. Assuming you could access similar data for
at communicating the information here? all years in the time-period 1960–2014, what way(s) do
you think would be most effective for conveying how
66. If you search for information on the Earth’s global this rate has changed over time?
surface temperature, you may notice that scientists
often present the temperature anomaly (the departure 72. The EPA presents information on composting and
from the long-term average) in place of the actual recycling in both percentage terms and in terms of total
weight. Discuss situations in which each method of 75. On a webpage about water use [315], the U.S.
reporting might be more effective than the other. Geological Survey presents the following table on wa-
ter use of the top states as a percentage of the national
total. Discuss the usefulness of presenting information
73. Average household water use for indoor fixtures
about water use in this way.
by category is shown in the table below (as reported in
the Water Research Foundation’s 2016 Residential End State Percentage of Total
Use Study [93]). Present this data in a graphical form. Domestic Use
California 15%
Fixture Gallons per day Texas 8%
Bath 3.6 Florida 6%
Clothes Washer 22.7 New York 6%
Dishwasher 1.6 Illinois 4%
Faucets 26.3
Leaks 17.0 76. During President Obama’s first term, gasoline
Other 5.3
prices increased from a national average of $1.83 per
Shower 28.1
gallon in January 2009 to a national average of $3.96
Toilet 33.1
in May 2011, an increase of over 115%. This statement
is factually correct but misleading. Why might this be
so? Historical data on gasoline prices is available from
74. The study on residential water use referenced in
the U.S. Energy Information Administration.
Exercise 73 updates a previous study done in 1999.
Discuss how you might present information comparing
household water use in 1999 with household water use 77. Refer back to Figure 29 on page 56. Can you
in 2016. You can then refer to the study to see how this identify other intervals of time where, if viewed in
was done. isolation, it would appear that global temperatures are
not changing over time?
C HAPTER 2
Flowing
“Everything is flowing,” said the Greek philosopher

Heraclitus of Ephesus, over 2500 years ago. “You cannot
step twice into the same river.” Heraclitus’s insight is a
key one in understanding the economic and ecological
systems within which we have to ask and answer the
sustainability question. Even those things that seem un-
changing or “in balance” are so because of a tradeoff of
changes in different directions. In normal circumstances
your blood does not seem to change much from day
to day, and yet 100 million red blood cells are being
replaced every minute!
It’s common, especially in environmental contexts, to
talk about the “balance of nature.” But as we wrote in
the introduction, this image of a “balance” is too static
to incorporate the message that “everything is flowing.”
We need to replace it with the more complex image of
a system that is in dynamic equilibrium—one that in- Figure 1: Heraclitus, painting by Johannes Moreelse.
volves various stocks (total quantities or accumulations
of something, like the total amount of water in a lake or
the number of cartons of ice cream at the grocery store) and various flows (processes that change
the amount of a stock, like a stream flowing into the lake or water evaporating from its surface, or
customer purchases of ice cream, deliveries, and spoilage). Stocks and flows are the basic building
blocks of systems theory [214], and part of the aim of this chapter is to introduce you to the language
of systems theory and also to a simple way of drawing diagrams that represent stocks, flows, and their
interconnections. We’ll introduce these diagrams in Section 2.1 along with the general language of
systems theory, and then elaborate them in Section 2.3 where we will begin to learn how to calculate
the equilibrium state that a stock-flow system will end up in.
We’re also going to take some time to look in more detail at one particular set of stocks and flows
that could be the most fundamental of all: these are stocks and flows of energy. In Section 2.2 we’ll
look at energy stocks and flows in general, learning about the units that they are measured in and why
they are vital to every discussion of sustainability. Western-style society is fundamentally dependent
on the availability of large supplies of energy, and yet the means by which we currently obtain
them conflict with basic sustainability principles. We’ll also learn about the laws of thermodynamics,
which impose basic constraints on processes that convert heat to and from other forms of energy.
Then in Section 2.4 we’ll zoom right out and consider the giant stock-flow system represented by
considering our entire planet as holding a stock of heat, with an inflow of heat from the Sun and an

https://doi.org/10.1007/978-3-319-76660-7_2
68 CHAPTER 2. FLOWING
outflow via radiation to space. Understanding this system lets us see why it is that the Earth’s present
temperature is “just right” to support human life. More ominously, it also helps us understand how
human actions (especially those related to our current ways of obtaining energy) might affect this
comfortable equilibrium value.
2.1. STOCKS, FLOWS, AND EQUILIBRIUM 69
2.1 Stocks, Flows, and Equilibrium
Objectives
I can define stocks and flows.

I can represent a stock-flow model by a system diagram.
I can identify inflows and outflows associated with a given stock.
I can calculate the net flow associated with a stock.
I can recognize when a system is in equilibrium.
I can compute residence times in an equilibrium system.
I understand the role of models in the scientific method.
2.1.1 Stocks and Flows

In Example 1 on page 46, we discussed the Great Pacific Garbage Patch (GPGP), one of five huge
ocean gyres that are filled with plastic and other debris. But the GPGP does not exist in isolation. It is
part of a system that includes
• Human beings, whose plastic waste ultimately ends up in the GPGP;
• Ocean currents, which create a vortex into which waste is swept;
• Sunlight, which helps break down plastic into tiny pieces;
• Marine animals, which ingest microscopic amounts of plastic and pass it on up the food chain;
• and probably many more components. How many others can you think of?
It is a vital insight of ecology and systems theory that in order to understand whether something
is “sustainable,” it is not enough to look at that thing by itself. We have to understand the system in
which it is embedded. As the early American naturalist John Muir wrote, “Whenever we try to pick
out anything by itself, we find it hitched to everything else in the Universe” [229, Chapter 6].
In this unit we will begin to look at some basic ideas about systems, focusing on two key notions:
stock and flow.
Definition 1
A stock is an element of a system that measures the quantity or accumulation of some kind
of resource or material or information that has built up over time.
The stocks in a system are what you see if you take a snapshot of the system at some instant of time.
The total amount of plastic waste in the GPGP is a stock. The amount of water in a bathtub at some
moment is a stock. So are the population of the United States, the reserves in an oilfield, the money
in my bank account, the amount of carbon dioxide in the atmosphere. A stock need not be a physical
thing: the amount of information in a library, or your own supply of hope that the world can become
a better place, can be thought of as stocks.
Definition 2
A flow is some process that changes the quantity of a stock over time.
Flows are processes like births and deaths, filling or emptying of reservoirs, deposits or withdrawals
in a bank account, emissions or capture of CO2 . “A stock, then,” writes Donella Meadows, “is the
present memory of the history of changing flows within the system” [214, page 18]. In a system in
which the quantity of plastic in the GPGP is a stock, flows might include the inflows of plastic waste
(deposited by human beings and carried by ocean currents) and outflows (perhaps other currents,
photodegradation, and marine animal ingestion). If we want to know whether the problem of the
GPGP is going to get worse or is eventually going to go away, then we need to understand the balance
between these various flows and the resulting overall effect on the stock of plastic waste.
We used the words “inflow” and “outflow” just now. Let’s define them.
Definition 3
An inflow for a given stock is a flow that tends to increase the quantity of that stock. An
outflow for a given stock is a flow that tends to decrease the quantity of that stock.
Remark 1: Whether a particular flow is an inflow or an outflow depends

on which related stock we are considering. When the university pays me
my salary, that process is an inflow to my bank account (my stock of
money) and an outflow from the university’s. Conversely, when I write
the university a check for my child’s tuition, that is an outflow from my
account and an inflow to theirs.
Let’s look at some more examples of stocks and corresponding flows.
Example 1: A bathtub (see Figure 2) can be considered part of a system
with one associated stock (the total amount of water in the tub), one inflow
Figure 2: A bathtub is a standard example
of a stock-flow system. (of water via the faucet), and one outflow (via the drain). In this case the
stock of water is measured in volume units, like gallons or liters. The
flow, being a rate at which the stock is moving, is measured in stock units divided by time units—like
gallons per minute, or perhaps liters per second. We can represent this simple stock-flow system by a
system diagram, like the one in Figure 3.
Faucet Water in tub Drain
Figure 3: System diagram for the bathtub.
In such a diagram, stocks are shown as “containers” or boxes (green), and flows are shown as
“pipes” or arrows (blue) that lead from and to the stocks. The “spigots” (blue with red “handles”) on
the flow arrows stand for whatever process controls the rate of the corresponding flow. Finally, the
“cloud” symbol (gray) signifies a source (arrow coming out of the cloud) or sink (arrow going into
the cloud) of material that is outside the scope of the system currently being discussed.
Remark 2: A “cloud,” which indicates a stock “outside the system,” is not an absolute thing, but is
relative to the specific system that we are trying to understand. Later, we might perhaps return and
consider that specific system (maybe call it “A”) as a part of some larger system “B.” For instance, if
we were looking at a wastewater recycling system for a home in an arid climate, the “cloud” to which
the drain leads in Figure 3 might be replaced by another “stock,” the contents of a wastewater holding
tank. On a larger scale, the stock-flow system for the whole house is part of the larger system that
represents the hydrological cycle for the whole neighborhood, which itself is part of the water cycle
for the whole planet.
Think about it. . .
The second of Barry Commoner’s “four laws of ecology” (from [68]) is “Everything must
go somewhere. There is no away.” What do you think this slogan means? Relate it to the
discussion of “clouds” in the preceding remark. Do you agree that “there is no away? ”
Charge in battery Current flow
Figure 4: Stock-flow system for a primary cell, such as an AA battery.
Example 2: Consider a system that includes an AA battery as an energy source. If this is a non-
rechargeable type,1 then the system diagram for the battery (by itself) is very simple (see Figure 4).
In this instance, the flow of electrical current out of the battery is measured in amperes (sometimes
abbreviated to “amps”) or milliamperes (10−3 of an ampere). An ampere is a measure of the rate
of flow of electrical charge—it is in fact a speed, even though it does not have a “per hour” or “per
second” in its name to remind you of that fact. Correspondingly, the electrical capacity of the battery,
the stock that corresponds to the flow of current, is measured in ampere-hours—current multiplied
by time. For example, we would expect a battery with a capacity of 1500 mAh (milliampere-hours),
which is fairly typical for an AA, to be able to supply 150 mA for ten hours, or 30 mA for fifty hours,
or 500 mA for three hours, and so on.
Remark 3: The “cloud” in Figure 4 could be replaced, on a larger view, by a diagram containing
the other components of the electrical system that the battery supplies. However, it is useful to look
at the very simple diagram in Figure 4 by itself. A diagram that looks like this—with outflow but
no inflow—will arise whenever we are dealing with a nonrenewable resource, a finite quantity of
some commodity that we can use or mine or capture while it exists, but that is not replenished to a
significant extent on any human time scale. The oil in a particular oilfield, the mineral deposit in a
mine, even the water in a deep aquifer like the Ogallala aquifer [196] that supplies irrigation for much
of the farmland of the Great Plains, can be regarded as nonrenewable resources. When the battery has
run down, the oil wells have run dry, or the mineral deposit has been mined out, their usefulness is
exhausted. Abandoned mines and oil fields around the world testify that this is a real possibility.
1 Also sometimes called a single-use battery or primary cell.
2.1.2 Equilibrium
Example 3: Figure 5, which is taken from NASA’s Earth Observatory, illustrates the carbon cycle
of planet Earth. This is in fact a system diagram with stocks and flows, though the graphics are more
lively than the spigots and pipes that we have been drawing so far.
Figure 5: The Earth’s carbon cycle from [271]. The white numbers in the figure refer to stock levels, measured in gigatons
(billions of tons) of carbon. The yellow numbers refer to natural flows, in billions of tons of carbon per year. The red numbers
refer to human-caused flows, or human-caused changes to the natural flows.
The figure contains seven different carbon stocks, as well as many different flows between them. It
is much more typical for a system diagram to contain multiple stocks and flows like this; the single-
stock diagrams we looked at in the previous two examples were especially simple ones.
Problem 1: From the carbon cycle diagram, compute the total inflow of carbon into the atmosphere,
and the total outflow, in the absence of human-caused contributions.
Solution: The diagram contains three natural inflows of carbon to the atmosphere, labeled “plant
respiration” (60 GT per year), “air-sea gas exchange” (90 GT per year), and “microbial respiration and
decomposition” (60 GT per year). (For this calculation we are ignoring the red figures like “human
emissions.”) These total 210 GT per year.
There are also two natural outflows: “photosynthesis” (120 GT per year) and “air-sea gas exchange”
(90 GT per year). These also total 210 GT per year. Thus, in the absence of human-caused
contributions, the total carbon stock in the atmosphere does not change over time.
The balance between inflows and outflows that we see in this example is so important that there is
a special name for it.
Definition 4
A stock in a system is in equilibrium if the total amount of that stock does not change over
time. It is the same to say that the stock is in equilibrium if its total inflows are exactly equal
to its total outflows. (We can also say that a system is in equilibrium if all its stocks are in
equilibrium.)
Remark 4: Flows give us information about how we expect a system to change in the future. Above,
we defined the outflow and inflow. When we do calculations, however, it is helpful to define the net
flow:
net flow = inflow − outflow.
A stock is in equilibrium if its net flow is zero. If the net flow is positive, then the stock is growing.
This means that more is added than removed. If the net flow is negative, then the stock is shrinking.
This is important information; it can tell us how (if at all) we should change our behavior. For example,
suppose your bank account has a stock of $10. Should you be concerned? If you know the daily net
flow is positive then no. You may not be rich, but you are living within your means and saving money.
Alternatively, if you have a lot of money in your bank account but your net flow is negative, then
eventually you will have a problem. It would benefit you to spend less (or earn more).
The equilibrium of a stock is a static equilibrium if both inflows and outflows are zero (think of
a bathtub full of water, with the faucet shut off and the drain closed). The equilibrium is a dynamic
equilibrium if total inflows and total outflows are equal but not zero (now think of the bathtub, again
full of water, but with the drain partially open and the faucet running, in such a way that inflow and
outflow balance out). Over a sufficiently long period of time, many natural systems tend to settle down
to a dynamic equilibrium. The atmosphere is an example.
The carbon cycle diagram (Figure 5) points us toward a discussion of one of the most pressing
sustainability issues of the present time, that of anthropogenic (that is, human-caused) climate
change. Briefly, the carbon in the atmosphere is almost all stored as carbon dioxide (CO2 ). Carbon
dioxide is a “heat-trapping” gas, so the amount of carbon dioxide in the atmosphere affects the Earth’s
ability to retain heat and thus the long-run temperature that we expect to experience at the Earth’s
surface. Later, we will study in more detail what observations tell us about the concentration of
atmospheric CO2 (see Figure 18 on page 50) and its heat-trapping effects.
Remark 5: As its chemical formula CO2 indicates, carbon dioxide is put together from carbon and
oxygen, in the ratio of one atom of carbon to two atoms of oxygen. Because an oxygen atom is slightly
heavier than a carbon atom, this means that slightly less than one-third2 by weight of a given amount
of CO2 is actually made up of carbon. To put it the other way around, completely burning 1 kilogram
of carbon produces 3.67 kilograms of carbon dioxide. When speaking about stocks or flows of carbon
it is very important to be clear whether you are talking about the total weight of carbon (as the NASA
diagram, Figure 5, does) or the total weight of carbon dioxide.
Constructing models
It is helpful to be able to construct and diagram simple stock-flow models of environmental processes.
What this means is to identify the most important stocks in the problem and the flows between them.
In the environmental science literature these models are often also called “box models,” because one
can think of each stock in the model as the contents of a “box.”
2 The exact figure is 27% by weight.
Depending on the level of analysis, stock-flow models can range from extremely simple (just one
stock and one flow) to fiendishly complex. The more complicated a model is, the more aspects of
reality it can represent, but the harder it will be to understand. In this course we will focus for the
most part on models that are not too complicated. The author of a standard textbook on environmental
modeling [156] puts it this way:
I believe it is preferable in environmental analysis to develop relatively simple, tractable
models rather than complex ones. . . The advantage of being able to “tinker” mentally
with a simple, penetrable model, and thus explore the consequences of a variety of
assumptions, outweighs in most cases the greater realism that might be attained with
a more complex model.
Problem 2: Develop a stock-flow model for the progression of undergraduate students through a
four-year university.
Solution: There are many levels of detail at which one could treat this problem. The simplest would
be to consider all enrolled undergraduate students as making up a single stock. Students are added
to this stock by the admissions process, which is a flow (units: students per year) and they leave by
graduation which is also a flow. So the stock-flow diagram is like that for the bathtub (Figure 3 on
page 70), with a single stock, a single inflow, and a single outflow.
Of course this model is oversimplified. (Every model is oversimplified: the trick is to find the
simplest model you can while still keeping the key features of the modeling situation.) Two ways in
which it is oversimplified are:
(a) It does not take into account that some students leave without graduating (for any of a variety
of reasons).
(b) It does not take into account that a newly admitted student must pass through a series of stages
before being eligible to graduate: as far as our model is concerned, a student could be admitted
to the general “stock” on Monday and graduate on Friday.
Trying to include these factors could lead to a more realistic model that divides the student body
into four “stocks”: freshmen, sophomores, juniors, and seniors.3 Students need to pass through these
stages in succession, and only members of the senior class graduate. There are always more levels
of detail that could be added. It is always possible to make a more complicated model, but the extra
complication does not always produce extra understanding. For many purposes, the simple “bathtub”
model will suffice.
Problem 3: Radon-222 is a radioactive gas that is naturally generated in soils that contain salts of
uranium. Radon can enter a home by diffusion through the foundation slab, and once in the home
it can leave either by venting to the outside air, or by radioactive decay to polonium-218. Draw a
stock-flow model for the radon concentration in a home, based on this information.
Solution: This is a model with one stock (the amount of radon in the home), one inflow (diffusion
through the foundation), and two outflows (radioactive decay and venting). The total quantity of radon
in the house might be expressed macroscopically (grams of radon gas) or microscopically (number of
atoms of radon): either is fine, as long as all units are consistent. See Figure 6 on the opposite page
for the stock-flow diagram.
Because radon is breathed deep into the lungs with the inspired air, and because its α-radiation is
particularly damaging to lung tissue, radon pollution is a major public health issue in the United States.
3 Models like this, which subdivide a population according to age, often arise in demographics (the study of human and
other populations); the way in which the population is subdivided into age cohorts is called the demographic profile. See Case
Study 7.4.
Diffusion Venting to
Radon in house atmosphere
thru slab
Radioactive
decay
Figure 6: Simple stock-flow model of radon concentration in a home.
The U.S. Environmental Protection Agency (EPA) estimates [16] that radon exposure is the second
biggest source of lung cancer deaths in the U.S. after smoking, being responsible for something like
20,000 deaths per year. New homes in high radon areas are typically equipped with active ventilation
systems that extract the gas from underneath the foundation slab and vent it to the outside air before
it can enter the home and do harm.
Remark 6: Polonium is also a highly radioactive element, though it is less dangerous than radon
because it emits a different kind of radioactivity and is a solid rather than a gas. However, a more
thorough modeling of radiation risks would have the “radioactive decay” spigot running to a second
stock, the stock of polonium, which itself would then decay and run to a third stock, and so on. . . You
can see how these models can quickly become complicated!
2.1.3 Residence Time

Imagine a stock-flow system in equilibrium. If we consider a particular
stock S there are two important numbers related to it:
• The quantity of S that is present in the system. (Because we

are assuming that the system is in equilibrium, this quantity is
a constant—it does not change with time.)
• The total flow rate of S in the system. (We can consider either
the total inflow to S or the total outflow from S; because we are
assuming equilibrium, these must be the same.)
Example 4: Consider a college as a stock-flow system, with a “stock”

of students flowing in through admission and flowing out through
graduation. The equilibrium “stock” might be 20,000 students, and the
Figure 7: A student ready to graduate. “Res-
flow rate (admissions or graduations) might be 4,000 students per year. idence time” in college is the average length
of time from matriculation to graduation.
In general, the “stock” will be measured in certain units (students in
the above example) and the “flow” will be measured in “stock” units
divided by “time” units (students per year in the example). If we divide the stock by the flow we will
therefore get a time, which is called the residence time associated with the particular stock:
Definition 5
The residence time associated with a stock that is at equilibrium in a stock-flow system is
Level of the stock
.
Flow rate of the stock through the system
The residence time is the average amount of time that a unit of the stock will spend between
inflow and outflow.
In the example above, the residence time of the students is 20, 000 (students) divided by 4, 000
(students per year), which equals 5 years. Using the unit-factor method we would write this calculation
as follows:
1 yr
20, 000
students × = 5 yr.

4, 000
students

That doesn’t mean that each student will take exactly this length of time to graduate: some will
graduate in 4 years, some in 6 or more. Five years is the average amount of time spent by a student as
an undergraduate at this college.
Problem 4: We open the faucet of the bathtub of Example 1 on page 70, to deliver 2.2 gallons of water
per minute, and we also open the drain. The tub fills until the rate of outflow via the drain equals the
rate of inflow via the faucet. At this point of dynamic equilibrium, the tub contains 24 gallons of water.
Find the residence time of water in the tub.
Solution: The residence time is obtained by dividing the stock level by the flow rate
1 min
24gal
× ≈ 11 min.
2.2
gal

This tells us that, on average, a molecule of water spends about 11 minutes in the bathtub before going
down the drain. However, any individual water molecule arriving through the faucet may run down
the drain instantly, or may stay in the tub for hours!
Remark 7: In Problem 4 we did not specify where the water came from, or where it went; this
information is not part of the concept “residence time in the bathtub.” We could even imagine that the
waste water is recirculated to the faucet via some kind of pump. In that case the same water would
stay in the bathtub forever. Nevertheless, we would still say that the residence time was 11 minutes:
that would be the average length of time a water molecule spent in the tub before being recirculated
via the pump. Be sure you have this picture clear before moving on to the next example.
Problem 5: Estimate the residence time of carbon in the atmosphere, using data from the NASA
carbon cycle diagram (Figure 5 on page 72).
Solution: Not all the numbers in the diagram are needed for this calculation. What matters is the total
stock of carbon in the atmosphere (about 800 gigatons, from the diagram) and the flow rate in (and
out) of the atmosphere (about 210 gigatons per year, as we calculated in answering Problem 1 on
page 72). Thus the residence time is about 800/210 ≈ 4 years.
Question 1: We’ve just shown that the residence time for carbon in the atmosphere is about 4 years;
yet carbon dioxide emissions from the burning of fossil fuels and emissions from other human sources
are expected to impact the climate for thousands of years to come. How can these two things be
simultaneously true?
Answer: This is a good question and a common point of misunderstanding. The answer requires us
to think carefully about the difference between the flow of carbon through the atmosphere as part of
the carbon cycle and an increase in the stock of carbon in the atmosphere (e.g., from the burning of
fossil fuels).
Consider once again the picture of the recirculating bathtub, with residence time 11 minutes. If I
pour in 12 gallons of extra water, the level in the bathtub will rise because the tub now contains 36
gallons rather than 24. This rise will not go away after 11 minutes—there is nowhere for it to go. We
will simply have a larger volume of water in total. The residence time tells us how long, on average, a
water molecule must wait before being recirculated.
The same principle applies to the carbon cycle. Adding extra carbon to the atmospheric stock (by
burning fossil fuels) makes this and all the connected stock levels rise. And these stock levels will
remain elevated for hundreds of thousands of years. The 4-year residence time does not tell us when
excess carbon will “disappear” from the system. It merely tells us how fast the excess (and other)
carbon is circulating around the cycle.4
2.1.4 Models and the Scientific Method

We have just introduced the word model, and we will use it often throughout the rest of the book.
This word has a special meaning in science. Let’s define it.
Definition 6
In science, a model is a simplified representation of a complex system that can be used to

explain or predict its behavior.
A globe is a simple example of a model. If you want

to know why the shortest route from Chicago to Beijing
passes more or less directly over the North Pole, a globe
will make the reason very plain. But scientific models are
usually not physical objects such as globes. Instead of being
built by putting together physical components, they are built
by putting together ideas. For instance, the models we have
discussed in this chapter are built by putting together the basic
ideas of stock and flow, and representing their connections by
means of a system diagram (see Section 2.1.1).
Scientists use models for two related reasons: to under-
stand a complex system, and to predict its future behavior.
Consider for example Problem 3 on page 74, which asks us to Figure 8: A globe.
develop a stock-flow model of radon concentration in a home.
In answer, we have so far only drawn the system diagram (Figure 6 on page 75), which describes the
model at the most basic structural level. To develop the model further, we would need to include
some equations describing the flow rates, as we will do in later sections beginning with Section 2.3.
Having done that, we could match the radon concentrations suggested by our model to the actual
values measured in a variety of houses. If we obtain a good match, that suggests that we’ve achieved
the first objective of modeling—understanding the system being modeled, in this case the flow of
radon through a home.
4 Over very long time scales, carbon is pulled from the atmosphere through a process involving rock-weathering and ocean
sedimentation, eventually forming limestone. See [208] and also Exercise 37 on page 179.
A F 6 3
Figure 9: The Wason four-card task.
What if we do not obtain a good match? Then the model must be modified, or perhaps rejected
entirely. The scientific method demands that all models must pass the test of reality; when applied
to a real situation, they must describe it accurately. Scientists “stress test” their models by applying
them to many different real situations; the more such tests a model passes, the more confidence we
can have in it. (Section 5.4 describes this process of assimilating new information in greater detail.)
Question 2: “Stress testing” means that having formed a model—a picture of how reality works—we
should look for evidence that might disconfirm it, which might prove our model incorrect. Many of us
find it intuitively difficult to do this; we are biased to look for evidence that will confirm our ideas. A
famous demonstration of this is the so-called Wason 4-card task [341]. You are shown four cards on
a table as in Figure 9. You want to test the following model: if there is a vowel on one side of the card,
then there is an even number on the other side. (You could think of this as a stock-flow model if you
like, with input stock the letters recorded on one side of the card, the flow being some process that
converts numbers to letters, which are the output stock recorded on the other side. And your model is
that if vowels go in to the process, then even numbers come out.) Question, then: Which cards should
you flip in order to “stress test” your model as quickly as possible?
Answer: You should flip the cards labeled A and 3. The model predicts that the A card should have
an even number on the other side. But the model also predicts that the 3 card should have a consonant
on its other side. Indeed, if you flipped the 3 card and saw a vowel, the proposed model would be
invalidated—because it claims that a vowel “input” should always produce an even “output.”
Many people faced with the Wason task immediately suggest flipping the A card and the 6 card.
But flipping the 6 card cannot disconfirm the model. If you flipped the 6 and saw an E, that would
agree with the model, true. But if you flipped the 6 and saw an F, that would not disconfirm the model:
the model does not say anything about the “output” that consonants may produce. The Wason task is
contrived, of course, but it is helpful in thinking about the scientific mind-set of always stressing your
model as much as possible.
Suppose though that a model (say the radon model that we were talking about before) has passed
many “stress tests.” Then we can cautiously begin to use it for the second function described above,
that of prediction. In the case of the radon model, we might use it to predict the radon concentration in
a new house before it is built, based on information about the house’s construction, the local geology,
and so on. We can have some confidence that the predictions will be accurate, because they are based
on a model that has been tested in a wide variety of real situations. Again in the radon example,
a builder might use the information provided by the model to determine whether to install a radon
mitigation system during construction. (This will help save money and time—if it is later discovered
that a mitigation system is needed, retrofitting one in an already-built home is going to be much more
complicated and expensive.)
Stocks, Flows, and Public Discussion
While the distinction between stocks and flows may seem clear, public discussion on human-caused
climate change and other topics often confuses them. Thus, such discussions present an important
opportunity for critical thinking. Here are some examples.
Critical Thinking
Consider the following quotation from Representative Michele Bachmann from her 2009
Earth Day speech on the House floor. Can you identify any stock/flow confusion in this
statement?
“What part of human activity creates carbon dioxide? If carbon dioxide is a
negligible gas and it’s only three percent of Earth’s atmosphere, what part is
human activity?
“Human activity contributes perhaps three percent of the three percent. In other
words, human activity is maybe 3 percent contributing to the 3 percent of
carbon dioxide that’s in Earth’s atmosphere. It’s so negligible—it’s a fraction
of a fraction of a fraction of a percent—that it can hardly be quantified.”
It is indeed true that human-caused flows of carbon are small compared with natural flows (see
Figure 5 on page 72 again). However this ignores the fact that the natural inflows and outflows are
equal and therefore lead to no change in the stock (dynamic equilibrium). The human-caused flows,
even if small, disturb this equilibrium and cause an increase in the stock of atmospheric carbon dioxide
which accumulates over time. This increase can be quantified and, as we have seen (Figure 18 on
page 50), can be measured quite directly.
Critical Thinking
Consider the following quotation from President Barack Obama from a press conference in
2012. Can you identify any stock/flow confusion in this statement?
“Now, in my first term, we doubled fuel efficiency standards on cars and trucks.
That will have an impact. That will take a lot of carbon out of the atmosphere.”
Increasing fuel efficiency standards will mean that cars and trucks will generate less carbon dioxide
per mile. In other words, the rate of flow of CO2 into the atmosphere will decrease. But that is not
the same as “taking carbon out of the atmosphere” (reducing the stock). The stock will continue to
increase, just less quickly than it otherwise would have.
Think about it. . .
The systems describing natural processes tend to be cycles (like Figure 5 on page 72);
the systems describing human activities tend to be one-way (like Figure 4 on page 71),
beginning from some usable resource and ending with useless waste. This fundamental
difference in system structure underlies every question about the ‘sustainability’ of human
society. Do you agree? Why or why not?
Summary of Ideas: Stocks, Flows, and Equilibrium
• A stock is the quantity of some resource or material existing at some point in time,
for example, the volume of water impounded by a dam, measured in gallons.
• A flow is a process that changes a stock over time. An inflow increases the stock and
an outflow decreases it. Flows are measured per unit time: the inflows and outflows
for the dam might be measured in gallons per hour.
• The net flow is the difference between total inflows and total outflows.
• A stock is in equilibrium if its net flow is zero.
• A dynamic equilibrium is one in which the inflows and outflows are not zero
individually (but they still balance out).
• The residence time associated with a dynamic equilibrium is equal to the stock
volume divided by the flow rate.
• In science, a model is a simplified representation of a complex system.
• The scientific method requires a model to be tested by accurately describing real
situations.
• It is important to exercise critical thinking to avoid confusing stocks and flows and
to evaluate public discussion involving them.
2.2. ENERGY STOCKS AND FLOWS 81
2.2 Energy Stocks and Flows
Objectives
I can define work, energy, and power.

I am familiar with the standard units of energy and power.
I know why energy is central to thinking about sustainability.
I can describe the first and second laws of thermodynamics.
I can define the efficiency of an energy system.
I know the Carnot limit on the efficiency of a heat engine.
I can apply these ideas to evaluate energy systems in the home.
In the last section we learned about the general concepts of stocks and flows. Some of the most
important stocks and flows in our society are those of energy in its various forms. In this section we
will focus on understanding energy: what it is, where it comes from, how it flows, and why it matters.
2.2.1 Work, Energy, and Power

Have you ever used a rowing machine? (You can find them in most gyms alongside other aerobic
machines, though they tend to be less popular—perhaps because they are more demanding—than
treadmills and ellipticals.)
Figure 10: Rowing machine. Image courtesy of Concept 2.
Rowers call the machine an “erg,” which comes from the Greek word for “work,” ergon. And that
three-letter combo “ERG” is the also part of the word “enERGy.” In fact, energy can be defined simply
as “the ability to do work.”
Energy can come in many forms. The most straightforward is mechanical work, which is what the
athlete is doing during her “workout.”
Definition 1
Mechanical work is done when something (like the handle of the rowing machine) is
moved against a resisting force. The amount of work is the product of the force and the
distance moved.
With each stroke, the handle of the rowing machine is moved about 3 feet against a resistance of
maybe 50 pounds of force. This means that about 3 × 50 = 150 foot-pounds of work is done. Foot-
pounds is a unit of energy, though as we will see there are other units that are more common and
useful.
Definition 2
Energy is the capacity to do work.
As we said, energy comes in various forms. Let’s hope the athlete on the rowing machine had a
good breakfast. By doing so, she stored up chemical energy in her body. During her workout, she
is converting that chemical energy to mechanical energy in moving the handle of the machine. The
handle then spins a resistance fan inside the body of the machine, stirring and warming the nearby
air. The mechanical energy is converted to heat energy (and a little sound energy as well). In fact,
the whole system (athlete plus rowing machine) is a stock-flow system of the kind we looked at in the
previous section, as shown in Figure 11.
Chemical energy Workout Mechanical Energy Fan Friction
Body heat Waste Heat
Figure 11: Energy stock-flow diagram for athlete and rowing machine.
The chemical energy stored in the athlete’s body is transformed to mechanical energy in the rowing
machine, which is transformed to heat by the fan. Notice that we have also shown a direct transfer of
chemical energy to heat. When you work out, you warm up!
Remark 1: Why have we shown the heat as a “cloud” rather than as another stock? As we will see a
little later, this kind of “waste” heat is an example of low-grade energy—it is there in the room, but
it is virtually impossible to gather it together for a useful purpose. That’s in contrast to the athlete’s
chemical energy (which can lift weights or move the rowing machine) or even the spinning flywheel’s
mechanical energy (which could be used, for example, to generate electricity—in fact, the rowing
machine’s on-board computer is powered in exactly this way.) For more about this see Section 2.2.4
below, “The Laws of Thermodynamics.”
Rule 1: The Money Analogy
Metaphorically, we can think of energy as a kind of money that allows us to “buy” an

action (some useful work). To achieve a desired result (move the car, boil the water,
light the room) we have to “pay” for it with energy. After we have done that the energy
still exists, but it is no longer “in our wallet”—that is, it is no longer available for us
to use as we wish. Throughout this unit, you are encouraged to use this metaphor to
better understand the physics of energy.
Suppose the athlete makes 30 strokes per minute. Each stroke transfers about 150 foot-pounds
of energy, so that her workout is transferring energy at a rate of 30 × 150 = 4, 500 foot-pounds per
minute. This rate of flow is the power of the energy transfer:
Definition 3
Power is the rate of flow of energy.
Just as “foot-pounds” is not the most common unit of energy, “foot-pounds per minute” is not the
most common units of power. In a moment we will review alternative units for measuring energy and
power.
Different Forms of Energy

If we think of energy as a kind of “money” (as was suggested above) then our civilization is the
“biggest spender” there has ever been. Since Thomas Newcomen invented his “atmospheric engine”
in 1712 to convert heat to mechanical energy that could be used to pump water out of a mine, the
story of industrial society has been the story of more and more ingenious ways to convert energy from
one form to another, and to use this energy to accomplish tasks that human muscle power could never
achieve. Here is an incomplete list of some of the forms of energy that we have used and some of the
ways in which we have been able to interconvert them.
• Chemical Energy. This is energy that is stored in the bonds of chemical compounds. The
energy contained in fuels that we burn, like gasoline (fuel for an automobile) or food (“fuel” for
the human body), is chemical energy. The burning (oxidation) of these fuels breaks the chemical
bonds and releases energy, which then flows into another form such as heat or mechanical work.
• Mechanical Energy. This is the energy that is stored by virtue of the motion or position
of a heavy object. The spinning flywheel of the rowing machine stores mechanical energy.
Mechanical energy is sometimes divided into kinetic energy (energy of motion) and potential
energy (energy of position). For example, if I climb up to a high-dive board, I gain potential
energy (I have done work against the force of gravity to climb up). When I dive, I am losing
potential energy as I fall, but I am moving faster and faster—gaining kinetic energy. When I
splash into the pool, my kinetic energy is transferred to the pool water.
• Electrical Energy. Electrical energy is what flows when we plug an appliance into a wall outlet
or connect it to a battery. Electrical energy has always been around in the form of lightning,
but its development as a usable means of energy transmission resulted from the discovery of
the relationship between electricity and magnetism, and the consequent invention of machines
like the dynamo (for converting mechanical energy into electrical energy) and the motor (for
converting electrical energy into mechanical energy).
• Radiant (Electromagnetic) Energy. This is the energy transmitted by electromagnetic radi-

ation, including radio waves, microwaves, infrared (heat) radiation, light, X-rays, and so on.
Nearly all the energy on earth originates, in one way or another, from the Sun, and energy
is transmitted from the Sun to the Earth by electromagnetic radiation. We now have many
technologies for manipulating electromagnetic energy: electric lights, microwave ovens, X-ray
machines, radio transmitters and receivers, among many others.
• Nuclear Energy. Nuclear energy is stored in the bonds between the particles of the atomic
nucleus. These bonds are much shorter and “tighter” than chemical bonds, and the energies
associated with them are proportionately greater. The energy that powers the Sun is generated
from nuclear reactions. Managing nuclear energy on Earth (for non-destructive purposes) has
presented many challenges but has been successful in some contexts. For instance, the nation
of France derives about 75% of its electrical energy from nuclear power.
• Thermal (Heat) Energy. It is an important fact that heat is also a form of energy, produced by
the disordered motion (kinetic energy) of the particles that make up an object. In the example
of the diver above, the ordered (one-directional) motion of the diver is transferred to motion of
the pool water which soon becomes disorderly; in fact, it becomes heat.
It is (usually) easy to convert other forms of energy into heat. For example,
suppose that a climber is rappelling down a cliff. Her potential energy is
converted to heat, as she descends, by the friction of the rope through the
rappel device. After a speedy rappel, you need to be careful with the rappel
device. A skinny device, like the one shown in Figure 12, can easily get hot
enough to burn your fingers or, in the worst case scenario, to melt the sheath
of the rope.
It’s also possible to convert heat to other forms of energy. For instance,
in a nuclear power plant, nuclear energy is converted to heat; the heat then
boils water to make steam, which turns a turbine (mechanical energy), which
in turn drives a dynamo to produce electrical energy. But in contrast to the
ease with which we can turn other forms of energy into heat, turning heat
Figure 12: Rappel device. back to other forms of energy is subject to strict limits imposed by the laws
of thermodynamics, which we will look at in Section 2.2.4.
2.2.2 The Centrality of Energy
Think about it. . .
We said earlier that our society is a “big spender” of energy. Do you see this as a problem
or an opportunity?
It is perhaps enjoyable to be a big spender, and it may empower you to do a lot of good. But you
had better be sure you have enough money in the bank to maintain your big-spending lifestyle. In the
same way, we could ask ourselves whether the world has enough energy in its “energy bank” to keep
up our current “energy lifestyle.”
Table 1: Amounts of Energy
Process Energy used or

released
Phone charger, plugged in but not charging, for 5 seconds 1J
Heating water for a cup of coffee 105 J
Food energy in large cookie that you eat with cup of coffee 106 J
Chemical energy in one gallon of gasoline 108 J
Heat produced by campus steam-heating plant in one day 3 × 1012 J
Thermonuclear explosion 1015 J
U.S. electrical energy generated by wind power during 2016 8.2 × 1017 J
U.S. electrical energy generated from all sources, same period 1.3 × 1019 J
This is, of course, a sustainability question. In fact, energy resources and energy use are central
to any discussion of sustainability. There are at least four reasons for this: energy is essential and
nonsubstitutable, and the way we obtain most of our energy is nonrenewable and produces pollution
that disrupts the climate. Let’s unpack these points:
• All the processes of industrial society rely on the availability of energy. Manufacture,
transportation, food systems, security, agriculture, education, communication. . . all (in their
present forms) require continual inputs of energy at a level much greater than human beings
could provide by muscle power. Ample supplies of energy are therefore essential to the way
society is presently ordered.
• Energy is not substitutable. Economists use the word “substitution” to say that one good can
take the place of another: if I don’t have an apple, I can eat a banana; if I run out of tea, I
can drink coffee (or perhaps beer!). Substitution can take place within the energy sector (e.g., a
power plant can burn natural gas instead of coal). But the laws of thermodynamics tell us that
there can be no substitute for energy as a whole.
• The majority of our energy needs are presently met by burning fossil fuels such as coal and
oil. Fossil fuels store the energy of ancient sunlight that fell on Earth millions of years ago, but
they are a finite resource. It seems likely that we may approach the limits of these resources,
especially of easily extracted oil and natural gas within a human time scale (a few generations).
Whether or not this is the case, there is no doubt that our present use of fossil fuels is a
nonrenewable, one-way process—just like drawing power from an AA battery (Figure 4 on
page 71), though on a much larger scale.
• The burning of fossil fuels also releases carbon dioxide into the atmosphere. This gas remains
in the atmosphere for thousands of years, and the amounts that have been released by fossil
fuel burning so far are already sufficient to affect the Earth’s climate (see Question 1 on
page 76). Scientists believe it is very probable that continued fossil fuel burning will have
highly disruptive and negative effects on the climate and on the planet’s capacity to support life.
For all of these reasons, sustainability questions often boil down to questions about energy. How
can we use less energy? How can we store it? Where can we find renewable sources of energy, which
will not run out and which won’t produce greenhouse gases?
2.2.3 Units of Energy and Power

We are going to learn about the units that are used to express energy and power. Remember from
Section 1.1.3 that in the U.S. two different unit systems are commonly used: the metric system and
the U.S. conventional system. The most important energy measure is the one that belongs to the metric
system. It is called the joule (denoted by the letter J).
Since the fundamental metric units are meters (for distance), kilograms (for mass), and seconds (for
time), the joule must be defined in terms of these. In fact, a joule can be defined as the kinetic energy of
a two-kilogram mass moving at a speed of one meter per second. However, a formal definition of this
sort is not too important to us. Much more relevant is to have an idea of how large various everyday
quantities of energy are, in joules. Some examples are contained in Table 1 on the previous page.
As you can see from the table, one joule is a rather small quantity of energy. Because the joule is a
metric unit, we can make larger units by adding the standard prefixes. Kilojoules (103 J), megajoules
(106 J) and even gigajoules (109 J) are useful.
Remark 2: The energy (about 10 joules) needed to raise 1 kg through 1 m, which is the same as the
energy released when 1 kg falls through 1 m, is a measure of the strength of the Earth’s gravity. It is
sometimes called the gravitational constant and denoted by g.
The U.S. conventional system does not have a single standard unit of energy. Instead, it has a variety
of different units that were developed and used for different purposes. Some of these units are listed
in the table below, together with their approximate equivalents in joules, which allow you to convert
energy measurements from conventional units to metric units or vice versa.
Table 2: Conventional Units of Energy
Unit Equivalent Definition

Foot-pound 1.4 J Work done in moving a distance of one foot against
resistance of one pound
British thermal unit (Btu) 1050 J Energy needed to heat 1 pound of water through 1
degree Fahrenheit
Food calorie (Cal) 4200 J Energy needed to heat 1 kilogram of water through 1
degree Celsius
Kilowatt-hour (kWh) 3.6 × 106 J Energy that flows for 1 hour at a rate (power) of 1
kilowatt
Quad About 1018 J A quadrillion British thermal units (used only for
national-scale data)
Finally, let us discuss units of power. Power is the rate of transfer of energy—it is the rate of flow
corresponding to the energy stock. The fundamental metric unit of power will therefore be a flow rate
of one joule per second. This unit is so important that it has a special name.
Definition 4
A power (rate of flow of energy) of one joule per second is called one watt (W).
You are probably familiar with the way light bulbs and other electrical appliances are rated in terms
of the number of watts they consume. The computer on the rowing machine can show the athlete how
many watts of energy she is expending. The rate of production as well as consumption of energy can
be measured in watts. For example, the electrical power output of the Hoover Dam generators is about
2 gigawatts (2 × 109 W). A large coal-fired or nuclear generating station has a similar output.
Figure 13: The Hoover Dam (Ansel Adams photo).
Problem 1: The human body is continuously generating heat energy by “burning” food (we need to
do this to stay warm and alive). Estimate the average power output of a U.S. adult human being, in
watts.
Solution: The simplest way to figure this is to think about how much we eat. From the information
in Table 6 on page 489, an average U.S. adult consumes about 2,450 calories per day, most of which
ends up as heat. Now “calories per day” is already a unit of power (it is a measure of energy divided
by a measure of time), so all that we need to to is to apply the correct conversion factors (using the
unit-factor method) to change the unit to watts (that is, joules per second).
2450
Cal
4200 J 1 day
1
hr J
≈ 120 = 120 W.

× × 24 ×
1day
1
Cal hr 3600 sec
sec
Think about it. . .
The power output of one human being is of order 102 W overall. But each person in an
industrial society disposes of a much greater amount of power through our various machines
and mechanical energy sources—perhaps something like 4 × 103 W of useful energy on
average in the U.S. [191]. Another way to put this is that each U.S. person continually
makes use of the power that forty or so human beings could generate using their muscles.
In a way, each of us disposes of the power of forty or so “energy servants.” (Imagine, for
instance, how many people it would take to push your car along the road at highway speed!)
In preindustrial societies, having many servants was almost the only way to multiply your
personal energy output: only monarchs and people of great wealth, if even they, would have
had access to the level of such output that we now take for granted.
The BBC program “Bang Goes the Theory” took this idea of human “energy servants” literally by
having a team of cyclists on stationary bikes generate all the electrical energy needed for one person
for one day. The most stressful moment occurs when their subject decides to take an electrically
powered instant-heat shower. This consumes a lot of energy rather quickly (in other words, it requires
a lot of power), and as a result it’s “all hands on deck” for the cycling team. See the video at [39].
Remark 3: Remember that the 120-watt figure is a 24-hour average: the power output of the body
will be higher during vigorous exercise and lower during sleep or rest. Even during exercise, not all of
the body’s power output will be available as mechanical power (that is, available to do work); some
will inevitably be lost as heat. Modern competitive cyclists carefully measure their mechanical power
output [195]; an elite cyclist might produce 350 watts for a 1-hour period of intense exercise.
2.2.4 The Laws of Thermodynamics
Heat is a special form of energy. As we have seen in examples,

whenever we do work, some, and eventually all, of the energy that is
expended ends up as low-grade heat, a form in which we can no longer
exploit it. On the other hand, high-grade heat is a significant component of
many of our energy production processes. In eighteenth-century Britain,
James Watt burned wood or coal to boil water and used the steam pressure
to do mechanical work. The fuel in a modern nuclear power plant is very
different, but it is still used to produce high-grade heat, which is then
converted to useful energy by boiling water and using the resulting steam
to drive turbines. This two-way relationship between heat and other forms
of energy is therefore involved in all the energy processes of our modern
world—both those that generate useful energy and those that consume it.
The laws of thermodynamics govern the two-way relationship between
energy and heat. They set limits on what we can and can’t do with
energy. One thing these laws tell us is that the universe does not provide
a “free lunch” in the energy sector: to the disappointment of generations
Figure 14: Perpetual Motion by Norman of inventors, there is no way to build a “perpetual motion machine”—
Rockwell. the name given to a mythical gadget that would go through a cycle (like
turning a wheel) allowing energy to be continually extracted and yet return
to its starting position at the end of the cycle. See Figure 14 for Norman Rockwell’s vision of one such
disappointed inventor (his “invention,” called the overbalanced wheel, has in fact been “discovered”
several times in the fruitless history of perpetual motion).
The first law of thermodynamics is also called the law of conservation of energy. It states that
energy cannot be created or destroyed, merely transformed from one form to another. This idea is
implicit in the way we model energy systems as stock-flow systems: the total amount of energy does
not change, it just flows from one stock to another.
Rule 2: First Law of Thermodynamics
• Energy can be neither created nor destroyed, although it can be changed from one
form to another.
• Heat is a form of energy.
Remark 4: Energy can be neither created nor destroyed—that is what scientists mean by saying that
energy is “conserved.” But when environmentalists or people in the media discuss the importance
of “conserving energy,” they are talking about something different: keeping energy in a usable form.
When, for example, cars or light bulbs produce motion or light from chemical or electrical energy, that
energy ultimately ends up as low-grade heat: still there, but no longer usable by us. So “conserving
energy” is about keeping as much energy in a usable (high-grade) form as possible.
The money analogy is helpful here. If you use money to buy a meal or a textbook or a gallon of
gas, that money has not disappeared, but it is no longer usable by you; it is in the seller’s pocket. If
your bank charges you a fee every time you use your credit card, that money becomes unavailable to
you too—it is a “loss,” something like the waste heat from a car or from human exertion. It wasn’t
destroyed; it was transferred to the bank and is no longer available to you. Conserving energy (in the
ordinary day-to-day sense) is like reducing expenses and transaction fees and keeping as much money
in the bank as possible.
As far as the first law of thermodynamics is concerned, all forms of energy are on the same level.
The distinction between “high-grade” and “low-grade” energy is the province of the second law of
thermodynamics. This has a precise mathematical formulation [113], but for now we’ll be content to
state the second law in a quite informal way:
Rule 3: Second Law of Thermodynamics
• Energy comes in high-grade and low-grade forms.

• In any energy transformation, the overall grade goes down.
Let’s say a bit more about this business of high-grade versus low-grade. First we agree to classify all
kinds of energy except heat energy (that includes mechanical, electrical, electromagnetic, chemical,
nuclear energy, and so on) as high-grade.
To understand the “grade” of a quantity of heat energy is more complicated.
The key principle is this: there needs to be a temperature difference between an
object and its surroundings in order for heat energy contained in the object to
do useful work. One of the clearest examples of this is given by thermoelectric
materials [159]: materials that can convert heat directly into electricity. In order
for a thermoelectric material to function, it needs a temperature difference
between its “hot side” and “cold side,” and the amount of energy produced
depends on this temperature difference. See Figure 15, which shows the nuclear-
thermoelectric power system for NASA’s Mars rover undergoing tests at Idaho
National Laboratory. The heat on the “hot side” of the thermoelectric material
is provided by the radioactive decay of plutonium, and the “cold side” is at the
temperature of Mars’s atmosphere. If both sides were hot, the device would not
work.
Plutonium-powered space vehicles are rather exotic, but James Watt’s steam
engine illustrates the same principle. The “hot side” is provided by the furnace, Figure 15: Nuclear thermoelec-
which turns water into steam. The “cold side” is provided by the atmosphere, tric power system for Mars
where steam condenses back to water. We make this a definition: rover.
Definition 5
High-grade heat energy is contained in an object that is much hotter than its surroundings.
The heat energy in an object that is at a similar temperature to its surroundings is low-
grade.
Example 1: Imagine a cup of hot coffee sitting on my desk. The coffee is at a temperature of maybe
80 ◦ C and the room temperature is 20 ◦ C. Relative to its environment, the coffee represents heat energy
of a reasonably high grade. It can do some work (such as warming me up). Now leave the coffee there
for an hour. At the end of that time, the coffee has cooled down to room temperature (and the room
has warmed up a tiny bit). The total amount of heat in the room is still the same, but the ability to do
work has disappeared. We say that the heat energy is now in a low-grade form. Remember, the second
law says that the “overall grade” is reduced in any energy process; so the cooling of my cup of coffee
is an example the second law of thermodynamics.
Remark 5: You might ask: What about an object that is much cooler than its surroundings? It is
certainly possible to extract useful energy from this situation: we could, if we wanted, reverse the
roles of “object” and “surroundings” and say that the heat energy in the surroundings is high-grade
relative to the object. However, this situation seldom occurs in reality, so we won’t worry about it in
this book.
Efficiency
Think again about the athlete on the rowing machine. Only some of her chemical energy was converted
into the mechanical work of moving the handle of the machine. The rest was dissipated in low-grade
heat. It is a consequence of the second law that all energy processes operate in this way. The energy
output as useful work is only a fraction of the energy input. That fraction is called the efficiency.
Definition 6
The efficiency of an energy process is the quotient

Useful power output
Power input
It can be expressed as a fraction, decimal, or percentage. Note that efficiency is a pure
number (Definition 2 on page 15).
Efficiency depends on what we are trying to produce, of course. If your desired output is low-grade
heat, your efficiency can be 100%! Efficiencies of actual processes vary widely. Table 3 lists ranges
of efficiency for different types of passenger cars.
Vehicle Type Efficiency
Gasoline 12% to 30%
Hybrid 21% to 40%
All-Electric 72% to 94%
Table 3: Efficiency of passenger cars by type. Data from the U.S. Department of Energy [331].
Critical Thinking
“One of these things is not like the other.” Conventional and hybrid cars are powered by
gasoline, but fully electric cars are powered by electricity, which must itself be generated
from other sources—often from burning fossil fuels. Table 3 does not take into account the
efficiency of generating electricity from its primary sources. Do you find this misleading?
Try to figure out how the numbers would change if this information were also included in
the calculation.
The engine in a gasoline-powered car is a complicated machine, but from the point of view of
thermodynamics it, like the steam turbine or the thermoelectric generator, is simply a device for
turning heat (generated by the exploding gasoline-air mixture in the cylinders) into usable energy
(mechanical in this case). There is a general name for such devices:
Definition 7
A heat engine is any energy process that produces usable energy out of the temperature
difference between a hot object and its cooler surroundings.
Our discussion of the second law of thermodynamics (Rule 3 on page 89) has so far taken a
qualitative form: the greater the temperature difference, the higher the “grade” of the heat energy
in the hot object and so the more efficiently we can expect to extract useful work from it. Sometimes,
though, we need to know the quantitative form of this law: exactly how does the efficiency depend on
the temperatures? The answer is called the Carnot limit:
Rule 4
Consider a heat engine that has a “hot side” and a “cold side.” Let Th equal the
temperature of the hot side of the engine and Tc the temperature of the cold side,
both measured in kelvins (see page 13). Then the efficiency with which the engine
converts heat energy into useful work can be no greater than the fraction
Th − Tc
,
Th
which is called the Carnot limit for the given temperatures Th , Tc .
The Carnot limit is the theoretical maximum possible efficiency: actual efficiencies will be less.
Problem 2: What is the Carnot limit on the efficiency of a coal-fired power plant? Assume that
the superheated steam enters the turbines at a temperature of 800 K. (This is a reasonable figure in
practice.)
Solution: From the data of the question we have Th = 800 K. The “cool side” temperature Tc will be
the temperature of the cooling water, roughly equal to the surface temperature of the Earth, which is
about 15 ◦ C = 288 K, let us say (near enough) 300 K. Thus the Carnot limit on efficiency is
800 K − 300 K 500 K

= ≈ 0.6 = 60%.
800 K 800 K
This is a theoretical figure: the actual conversion efficiency will be less. In practice, efficiencies of 40
to 50 percent can be attained.
Now imagine a one-gigawatt power plant operating with 40 percent efficiency. The rated figure
(one gigawatt) is the electrical output of the plant, so the thermal power input must be 1/0.4 = 2.5
gigawatts: the remaining 1.5 gigawatts of thermal input is wasted as low-grade heat. That heat has
to go somewhere! Often it is used to evaporate water in huge cooling towers like those shown in
Figure 16.
Problem 3: How much cooling water each day will the

one-gigawatt power plant require if it is cooled by water
evaporation?
Solution:
In order to solve this problem we need an extra item of
physical information, namely, how much heat energy it takes
to evaporate a certain quantity of water. This figure is called
the latent heat of evaporation of water and is approximately
equal to 2 × 109 joules per cubic meter (see Section 8.2.3).
Figure 16: Cooling towers. Let us now compute the daily amount of waste heat energy in
joules: that will equal the waste-heat power in watts (joules
per second) multiplied by the number of seconds in a day:
1.5 × 109 J 3600 sec

hr ≈ 1.3 × 1014 J.

× × 24
sec 1 h r

Dividing by the latent heat of evaporation gives us the amount of water used in cubic meters:
1 m3
1.3 × 1014 J × ≈ 6 × 104 m3 .
2 × 109 J
The volume of an Olympic swimming pool is about 2 × 103 m3 , so our power plant needs to evaporate
about one Olympic pool of water every hour in order to remain operational.
Where might the electrical energy from that power plant go? Some
of it may be used to light your home. This is also an energy conversion
process (electrical energy to light energy) whose efficiency can be
measured according to Definition 6 on page 90.
Problem 4: Estimate the efficiency of a “traditional” incandescent bulb
(the kind invented by Thomas Edison in 1879).
Solution: Edison’s idea was to use electricity to heat the bulb’s filament
enough that it would glow (we’ll discuss this process of thermal
radiation in greater detail in Section 2.4). Anyone who has put their
Figure 17: Incandescent bulb. hand near an incandescent bulb knows that this is not very efficient: a
great deal of heat is wasted for the amount of light that is generated.
Still, you might be surprised by the actual numbers.
Let’s consider a 100 watt bulb as an example. According to figures from the General Electric
Company, such a bulb produces 1690 lumens of light output. A lumen is a measure of light intensity
Figure 18: Energy flows in the U.S. in 2016.
as perceived by the eye, and we can convert to watts of light energy by using the luminous coefficient
which is approximately 683 lumens per watt (see Case Study 7.7). The efficiency of the bulb is
therefore
Light power output 1690 lumens 1W

≈ 0.025.

= ×
Power input 100W 683 lumens

The efficiency of the bulb is about 2 12 percent. We’ll look at alternative, more efficient kinds of lighting
in Case Study 7.7.
Example 2:
A great many of the concepts that we’ve introduced so far are illustrated in the flow chart (Figure 18)
produced by the Lawrence Livermore National Laboratory [191]—see the bibliography for a link to
an enlargeable graphic. We see, for instance, that the total U.S. energy consumption for the year 2016
was 97.3 quads—remember from Table 2 on page 86 that a “quad” is a quadrillion Btu, or about 1018
joules. Of this, 30.8 quads went to “energy services,” that is, useful energy, and 66.4 quads went to
“rejected energy,” that is, waste. The overall efficiency of U.S. energy use is therefore
30.8
× 100% ≈ 32%.
97.3
Looking at the left of the chart, we see that fossil fuels (coal, oil, natural gas) account for 14.2 +
35.9 + 28.5 = 78.6 of the total 97.3 quads, or about 81% (notice also, by the way, that the next biggest
source of energy is nuclear power). As explained in Section 2.2.2, fossil fuel consumption is deeply
embedded in our present way of life, but in its present form it is not sustainable. That fundamental
tension is something that humanity will have to resolve very soon.
Think about it. . .
One way in which we can try to reduce our energy dependence is to make our energy-using
devices more efficient. But does technological progress toward increased energy efficiency
always reduce overall energy use in the long term? Consider the following example: the
earliest computers were power-guzzling monsters. A modern computer uses only a few
watts (and has many more capabilities too). But this efficiency gain has made computer
ownership much more widely accessible than in the days of the ENIAC. The growth in the
number of computers more than outweighs the power savings: computers in total use much
more energy now than in their early days. This effect is called the Jevons paradox [173].
Can you think of other examples?
Summary of Ideas: Energy
• Work is done when something is moved against a resisting force. Energy is the
ability to do work. It is measured in joules (J).
• Energy comes in a variety of forms, including mechanical energy, electrical energy,
chemical energy, nuclear energy, and heat (thermal) energy.
• Energy cannot be created or destroyed, but it can flow from one form to another (first
law of thermodynamics). Power is the rate of flow of energy. It is measured in watts
(W). One watt is a rate of flow of energy of one joule per second.
• Energy comes in high-grade and low-grade forms. In any energy process, the overall
grade goes down (second law of thermodynamics).
• The efficiency of an energy process is the percentage of energy input that is converted
to a usable output form.
• The Carnot limit gives a theoretical maximum efficiency for a heat engine (a process
that produces high-grade energy from the difference in temperature between two
objects). It equals (Th − Tc )/Th , where Th is the “hot side” temperature and Tc the
“cold side” temperature, both measured in kelvins.
• Modern society depends on steady flows of high-grade energy. This supply is
essential (we can’t do without it) and nonsubstitutable (there isn’t an alternative
to it).
• The majority of our present energy supplies come from burning fossil fuels. These are
nonrenewable and climate-changing. Developing renewable alternatives is a high
priority for sustainable living.
2.3. CALCULATING EQUILIBRIUM STATES 95
2.3 Calculating Equilibrium States
Objectives
I understand that flow rates (in a stock-flow model) can depend on stock levels.
I know how to represent these dependencies by control arrows in a system diagram.
I can compute equilibrium levels in models with one or two stocks.
I can apply these ideas to calculations about home heating and insulation.
I know what is meant by a parameter in a stock-flow model and how to represent it
in a system diagram.
Up to this point in our discussions of stock-flow models, we have treated the stock levels and flow
rates as given or measured quantities. For example, in the NASA carbon cycle model (Figure 5 on
page 72), all the flow rates were given as part of the data.
But as things change, the stock levels and flow rates can change too. In many circumstances, the
flow rates actually depend on the stock levels. For example, you may have noticed5 that the more
money you have, the faster you seem to spend it! In other words, the rate of flow of money out of your
account depends on the stock of money in there already.
This relationship between stocks and flows makes it possible to develop a mathematical under-
standing of stock-flow systems. In this unit we’ll begin this study by learning how to figure out the
equilibrium levels of stock in the system.
2.3.1 Models with a Single Stock

Let’s start by revisiting the “bathtub” example, Example 1 on page 70. If we open the faucet to fill
the tub, water will flow in at a constant rate. If we open the drain as well, water will flow out, but the
rate at which it flows out depends on the depth of water in the tub: the deeper the water, the higher the
pressure, and therefore the faster the flow. We can indicate this relationship in a stock-flow diagram
by a control arrow.
In Figure 19 on the next page, we have redrawn the stock-flow diagram for the bathtub system,
adding a new level of detail. We are now using two types of arrows. The ones we have seen before
are the flow arrows (the thick blue arrows in the diagram), which denote flows related to stocks in the
model. In our example, there is an inflow of water to the bathtub through the faucet, and an outflow of
water down the drain. The new element that we have added is a control arrow (the thin red arrow in
the diagram): arrows of this kind denote relationships by which a stock level (green “box”) controls a
flow rate (blue “spigot”). In our bathtub model, there is one control arrow running from the “water in
tub” stock to the “drain” spigot. This indicates that the outflow rate through the drain is governed by
the water level in the tub. Qualitatively, the more water in the tub, the faster it drains. Now it’s time to
get quantitative, and put some numbers to this example.
Problem 1: Water flows into a bathtub at a constant rate of 2.2 gallons per minute.6 The outflow rate r
is related to the depth of the water, h (in inches), by r = Ch, where the constant C equals 0.46 gallons
per minute per inch. At what depth will the water in the tub be in equilibrium?
5 If you have not noticed this, congratulations on your self-discipline!

6 This is the maximum flow rate permitted for a shower by the Federal Energy Policy Act of 1992.
Blue arrows
denote flows
Red arrow
denotes a control
Figure 19: Stock-flow model of the bathtub example. The thick blue arrows denote flows. The red control arrow indicates a
relationship between the water level and the outflow through the drain.
3
Inflow
Flow Rate (gallons per minute)
Outflow
0
0 1 2 3 4 5 6
Water Depth (inches)
Figure 20: Equilibrium for the bathtub model.
Solution: First, let’s think what the outgoing flow rate equation is saying. If there is no water in the tub
(h = 0), then there is no outgoing flow (r = 0) either. The equation tells us that the outgoing flow rate
increases in proportion to the depth of water in the tub: 0.46 gal/ min when h = 1 in, 0.92 gal/ min
when h = 2 in, and so on. The equilibrium depth will be the value of h for which the net flow is zero,
that is, the inflow and outflow are equal. The inflow is 2.2 gal/ min and the outflow is Ch gal/ min, so
this lets us set up an equation:
2.2 gal 0.46 gal
= · h.
1 min 1 min × 1 in
We can solve this by multiplying by the inverse:
1
min
× 1 in 2.2 gal 2.2
in ≈ 4.8 in.

· =
0.46
gal
1min
0.46
Thus, when h ≈ 4.8 in, the tub is in equilibrium. Let’s take a moment to verify our work:
0.46 gal gal
r= · 4.8
in ≈ 2.2 .
1 min × 1 in min

Remark 1: Instead of solving this problem using algebra, we could solve it by drawing a graph, like
the one in Figure 20. On the x-axis we have put the stock level (as measured by the water depth).
The y-axis measures flow rates, and we plot the inflow (in purple) and the outflow (in orange) as two
separate graphs. (Notice that the purple line is horizontal, because the inflow rate does not depend on
the water depth.) The point where the two lines cross gives the water depth at equilibrium—about 4.8
inches, as we said before.
Remark 2: Notice that in this problem we measure the stock (of water) by looking at its depth in the
bathtub rather than its volume. We can do this because the volume of water in the tub is equal to the
depth multiplied by a constant, the cross-sectional area of the tub. A quantity (like the depth in this
example) that “stands in” for a stock measurement in this way is called a proxy for that measurement.
Problem 2: In the above example, the bathtub has a cross-sectional area of 8 ft2 . How many gallons
of water will the tub contain when in equilibrium?
Solution: The water in the tub at equilibrium is a “box” of base area 8 ft2 and depth 4.8 in. The volume
of such a box is its base area times its depth. To calculate the volume in cubic feet, we need first to
express the depth in feet rather than inches:
1 ft
4.8
in × = 0.4 ft.
12in

The volume of the water when the system is in equilibrium is then
8 ft2 × 0.4 ft = 3.2 ft3 .
We need to convert this to gallons. From Problem 8 on page 10, we know that one cubic foot is
approximately 7.5 gallons. Therefore, the volume of water in the tub at equilibrium is
7.5 gal
3.2 ft3 = 3.2 ft3 × = 24 gal.
1 ft3
There are about 24 gallons of water in the tub at equilibrium. Notice, though, that since this is a
dynamic equilibrium, it is not the same 24 gallons all the time! In fact, at a flow rate of 2.2 gallons per
minute, the 24 gallons of water is completely replaced every 11 minutes or so. This is the residence
time (Definition 5 on page 76) of the water in the tub.
Here is an example with more environmental content.

Problem 3: A certain lake contains 10 million cubic meters of
water. Water flows into the lake at a rate of 30,000 cubic meters
per day; 10,000 cubic meters per day leave by evaporation, and
20,000 cubic meters per day leave via an outgoing river. A
factory on the lake begins to dump 0.3 tonnes per day of a highly
soluble and well-mixed pollutant into the lake water. Find the
equilibrium amount of pollutant in the lake (in tonnes), assuming
that the pollutant does not codistill (that is, it does not leave the
water along with the evaporating water vapor).
Remark 3: A pollution problem like this involves two materials:

Figure 21: Lakeside factory.
a pollutant and a substrate (here, the lake water) with which
it is mixed. In calculations, a key role is often played by the
concentration of the pollutant, that is, the fraction
Quantity of pollutant
Concentration = .
Quantity of substrate
Since the two quantities have the same units, the concentration is a pure number (see Example 5
on page 15). Our problem specifies that the pollutant is well mixed, meaning that the concentration
of pollutant in any sample of lake water is the same as that in the whole lake. In particular, the
concentration of pollutant in the outgoing water is equal to the concentration in the whole lake. The
solution (below) depends on this idea.
Solution: Suppose that P is the total tonnage of pollutant in the lake. Then the concentration of
pollutant in the lake water is given by dividing the tonnage of pollutant by the tonnage of lake water,
that is,
P
Concentration = 7 ,
10 T
since the mass of one cubic meter of water is one tonne. The amount of pollutant leaving with the
outgoing flow is therefore
P 2 × 104 T P
× = .
107 T 1 day 500 day
In equilibrium the outgoing flow of pollution must equal the incoming flow, so
P
= 0.3 tonnes per day,
500 day
and thus P = (500
day)
× (0.3 T/
day)
= 150 T.
Problem 4: Compute the residence time of the pollutant in the “polluted lake” problem, Problem 3
on the previous page.
Solution: We already computed that the equilibrium stock of the pollutant is 150 tonnes, and its flow
rate is 0.3 T/ day. Using Definition 5 on page 76, the residence time is
150 T
= 500 day.
0.3 T/ day
Because we know that on average, the pollutant stays in the lake for this long, we can set a realistic
timetable for cleanup efforts. Even if the pollution source is shut down tomorrow, it will take a year
or two before the lake returns to normal. We cannot realistically expect to change things in a week!
This is one of the most important applications of the residence time notion—it tells us the shortest
time scale on which change may be possible.
Think about it. . .
Once upon a time, the motto for much industrial waste management was “dilute and
disperse.” That is, pollution would be released into systems in which it had a short local
residence time—it would be quickly washed away. What do you think of “dilute and
disperse” as a pollution management strategy? What information is the residence time not
giving us?
Remark 4: You’ll notice that we can also get the 500-day residence time figure another way, by
looking at the water rather than the pollutant: there are 107 m3 of water in the lake, and water leaves
via the river at 2 × 104 m3 day−1 , so the residence time is
1 day
107
m3 × = 500 day.
2 × 104 m3

This is to be expected: since the water and the pollutant are well mixed, the length of time that a
quantity of the pollutant resides in the lake is the same as the residence time of the water that it is
mixed with.
Space Heating
41.5%
6.2%
34.6% Air Conditioning
17.7%
Appliances, Electronics,
and Lighting
Water Heating
Figure 22: Energy consumption in U.S. homes by end use (2009).
2.3.2 Heating and Insulation

Of all the energy used in the U.S. residential sector, the largest share (more than 40%, see Figure 22)
goes toward heating our living spaces [5]. In this section we’ll see how to use a stock-flow system to
model the process of heating a home. This gives us a tool to understand how much energy we could
save by improving home insulation or setting back the thermostat at night.
First, let’s develop a little background information about heat flows. The laws of thermodynamics
(Section 2.2.4) tell us that heat energy tends to flow from hotter to cooler objects. There are three ways
in which this transfer of energy can occur: through conduction, through convection, and through
radiation.
• Conduction occurs when heat energy passes through a material. A

closed thermos of hot coffee will cool by conduction as heat passes from
the coffee through the container to the cooler surroundings. Conduction of
heat energy through the structure of a house will appear as a flow in our
stock-flow model for home heating.
• A home can also lose heat by convection. This occurs when heat energy
is carried on currents of air (or of another fluid). If a child leaves the door
open on a winter day and a parent exclaims, “You’re letting the heat out!”
the parent is complaining about heat loss by convection. Our simple model
does not take account of convection losses.7 Figure 23: Thermographic image,
showing heat loss through uninsulated
• Finally, a warm object emits heat by radiation at an intensity that walls.
depends on its surface temperature (we’ll discuss the mathematics of thermal
radiation in Section 2.4). Direct heat losses from a home by radiation are negligible: heat has to pass
through the house structure by conduction before it can be radiated away from the outside surface.
This radiated heat can be measured by thermal imaging (see Figure 23). Such an image, made as
part of a home energy audit, allows the technician to “see” which parts of the house structure are
allowing the most heat to pass through by conduction. These “leaky” areas can then be targeted for
better sealing or insulation upgrades.
After learning about the different ways that heat flows, let’s return to the question of how to model
home heating:
7 Such losses can be quite significant, especially in an older home where draughts of air leak past poorly sealed door or
window frames.
Problem 5: A home is heated by a furnace that is controlled by a thermostat. Heat is lost through
the structure of the house by conduction. The rate at which heat is lost depends on the temperature
difference between the inside and outside. Regarding the amount of heat in the home as a stock, draw
a stock-flow diagram for this problem. Use control arrows to indicate how the stock controls the flows
in the model.
Solution: A stock-flow diagram is shown in Figure 24. The amount of heat in the home is the stock.
The inflow is the heat from the furnace and the outflow is the heat that is lost through the structure of
the house, primarily through the walls, windows, and roof. We measure the amount of heat in the home
using the indoor temperature as a proxy (Remark 2 on page 97), analogous to the way we measured
the stock of water in the bathtub using its depth.
Thermostat Heat in home Heat loss
Figure 24: Stock-flow diagram for a model of home heating.
The thermostat turns the furnace “on” when the temperature in the house falls below a set level, and
“off” when the temperature rises above that level again. In other words, the stock of heat in the home
(as measured by its proxy, the indoor temperature) controls the inflow of heat from the furnace. This
is indicated by the control arrow from the stock to the “thermostat” spigot in our model. It was also
stated in the problem that the rate of heat outflow is determined by the difference between the inside
and outside temperatures. The outside temperature is not something we can do anything about, but
the inside temperature is (again) our proxy for the stock of heat in the home. Thus, the heat stock also
affects the outflow rate, and we indicate this relationship with another control arrow, this time from
the stock to the “heat loss” spigot.
In Problem 5 we stated that the rate of heat loss depends on the temperature difference between
the inside and outside of the home. In order to calculate with our model, we need to know how this
dependence should be expressed. We know from everyday experience (on a cold day, you need to wear
a thicker jacket) that the extra information we need is some measure of the quality of the insulation
provided by the house structure: the better the insulation, the less heat will leak out. Insulation quality
is measured by a number called an R-value:
Definition 1
The R-value of a building component is a measure of “thermal resistance”—how effectively

the component impedes the passage of heat. The higher the R-value, the better the
insulation, and therefore the less energy your home is going to need. If ∆T is the temperature
difference between the warm and cold sides of a wall, the rate of heat loss is given by the
following equation:
1
Rate of Heat Loss = ∆T × Area of wall × .
R-value
In the United States, R-values are usually expressed in units of “hours times degrees (F) times
square feet per Btu” or hr ◦ F ft2 / Btu. That’s quite a mouthful. Let’s work through an example to see
how the units work out.
Problem 6: An exterior wall in a particular house is 160 ft2 in area and has an R-value of
12 hr ◦ F ft2 / Btu. Calculate the rate of heat loss when the indoor temperature is 70 ◦ F and the outside
temperature is 40 ◦ F.
Solution: We are going to apply the equation given in Definition 1. First note that we want to use
the difference in temperature, ∆T = 70 ◦ F − 40 ◦ F = 30 ◦ F. Also notice that the equation contains
the inverse R1 of the R-value. If the R-value is 12 hr ◦ F ft2 / Btu, then its inverse R1 = 12
1
Btu/ hr ◦ F ft2
(when using the unit-factor method, if we invert a quantity then we must also invert its units). Now
let’s make the calculation:
1
Rate of heat loss = ∆T × Area ×
R-value
2 1 2
= 30 ◦ F × 160
ft × Btu/ hr ◦ F
ft
12
= 400 Btu/ hr.
We calculated the rate of heat loss in the units “Btu per hour.” These are, not coincidentally, the
same units used (in the United States) to measure the rate of heat production from a home furnace.
If you want to convert to watts, you need the conversion factor 1 W ≈ 3.4 Btu/ hr, which you should
have calculated in Exercise 25 on page 119 for this chapter. (You can also look up the unit conversions
in Section 8.2.)
Remark 5: If you look at the rating plate for your home furnace
(Figure 25), you may see two Btu per hour figures: the heating
input, which specifies the rate at which the furnace uses fuel,
and the output capacity, which specifies the rate at which the
furnace delivers heat to the home. It’s the second of these figures
that is used in our home-heating calculations. The difference
between them is the amount of heat that goes to waste through the
furnace exhaust; their ratio measures the efficiency (Definition 6 on
page 90) of the furnace.
Now we will see how to calculate equilibrium values in our
stock-flow model of home heating. Here is an example: Figure 25: Furnace rating plate.
2
Problem 7: A particular home has 3500 ft of wall, roof, and
window area and an overall R-value of 7 hr ◦ F ft2 / Btu. The home has a furnace whose output capacity
is 34,000 Btu/ hr. The homeowner wants to maintain an indoor temperature of 70 ◦ F. What is the
lowest outside temperature at which it will still be possible to maintain the desired indoor temperature?
Solution: This is an equilibrium calculation. The question asks: what will be the value of the heat
stock in the house, as measured by the difference in temperature with the outside, ∆T , when the heat
flow into the house from the furnace running at full blast is in equilibrium with the heat loss through
the structure?
We can set up a dynamic equilibrium and solve for ∆T in the same way we found the equilibrium
depth of water in the bathtub model on page 96. The inflow from the furnace is 34,000 Btu/ hr. Set
this equal to the outflow using the equation for the rate of heat loss from Definition 1:
1
Heat Inflow from Furnace = ∆T × Area × .
R-value
Putting in the values from the problem and calculating, we get
2 1 Btu
34000 Btu/ hr = ∆T × 3500
ft × ,
7 hr ◦ F 2
ft
Btu
that is, 34000 Btu/ hr = ∆T × 500 .
hr ◦ F
1 hr ◦ F
Multiplying each side by 500 , we get
Btu
Btu
1 hr ◦ F
34000

× = ∆T,

hr
500 Btu
and doing the division on the left side gives us the final answer,
68 ◦ F = ∆T.
Our answer, ∆T = 68 ◦ F, gives the difference between the indoor and outdoor temperatures when the
furnace is running at full power. The lowest outdoor temperature that will allow the desired indoor
temperature of 70 ◦ F to be maintained is then
70 ◦ F − ∆T = 70 ◦ F − 68 ◦ F = 2 ◦ F.
If the outdoor temperature falls below 2 ◦ F, the house will get colder than desired. If the outdoor
temperature is above 2 ◦ F, the furnace will cycle on and off (governed by the thermostat) to keep the
indoor temperature at the desired level.
Parameters
As we saw in the preceding calculation, the outside temperature plays an important part in our home
heating model. You might wonder, though, how it fits into the structure of system diagrams as we
have been discussing them so far. The outside temperature is not a stock or a flow in the model, but it
nevertheless influences how the model works. Such a quantity is called a parameter.
Definition 2
A parameter in a stock-flow model is a quantity that is not itself a stock or a flow in the
model but that affects one (or more) of the flow rates in the model in some specific way.
If we want to consider the effect of a parameter in our model, we can represent it in a system diagram
by a special box, as in Figure 26 on the opposite page. This is exactly the same as the earlier Figure 24
except that we have added a parameter box (the yellow box with wavy edges, sometimes called
a “flag” because of its shape) to represent the outside temperature parameter, and another control
arrow, from the “outside temperature” parameter box to the “heat loss” spigot, to indicate that outside
temperature affects the rate of heat loss.
If the parameters of a system change, the equilibrium value can change too (in the home heating
example above, if the outside temperature falls to 0 ◦ F, the equilibrium indoor temperature will fall to
68 ◦ F). We’ll study this question—how do equilibrium values respond to a change in parameters?—in
Chapter 4.
Outside temperature
Figure 26: Stock-flow diagram with outside temperature as a parameter.
2.3.3 Models with More Than One Stock

So far in this section, we have calculated equilibrium levels in models with a single stock. Similar
ideas can be used to do equilibrium calculations in models that have two (or more) different stocks.
In general such problems will lead to simultaneous equations, with one unknown for each stock
level. However, some simpler problems of this sort can be solved more directly. Here are a couple
of examples.
Problem 8: (taken from [156, Problem II.A.7].) Ethane is a constituent of natural gas, and emissions
from natural gas wells constitute the only significant source of ethane in the atmosphere. In a study
conducted in 1979 it was found that the concentration of ethane in the (atmosphere of the) northern
hemisphere was about 10−9 (1 part per billion) and the concentration in the southern hemisphere was
about 5 × 10−10 (half a part per billion). Because of the geographic distribution of natural gas wells,
ethane emissions occurred (at the time of the study) almost entirely in the northern hemisphere, at a
rate of about 3 × 106 tonnes per year. Estimate the flow of ethane across the equator, in tonnes per
year.
Ethane in Flow across Ethane in Losses from

Inflow N Hemisphere equator S Hemisphere S Hemisphere
Losses from
N Hemisphere
Figure 27: Two-box model for hemispheric ethane flow, after Harte [156].
Solution: This is a two-box model, with one “box” representing the stock of ethane in the northern
hemisphere and one box representing the stock in the southern hemisphere (Figure 27). As shown in
the model, all the ethane enters via the northern hemisphere, but it leaves partly from the northern
and partly from the southern hemispheres. The flows labeled “losses” in the model are supposed to
represent all the processes by which ethane can leave the atmosphere, including chemical reaction and
deposition to the surface of the Earth with rain or snow.
As we saw in Problem 3 on page 97, we expect the rate of loss of ethane by these processes
to be proportional to its concentration. Since it is given that the concentration of ethane in the
northern hemisphere is twice that in the southern hemisphere, the loss rate of ethane from the northern
hemisphere must equal twice the loss rate from the southern hemisphere. So, if we denote the unknown
loss rate from the southern hemisphere by the letter s, then the loss rate from the northern hemisphere
will be 2s.
Now assume that the system is at equilibrium. Then the inflows and outflows for each hemisphere
must balance. The inflow to the southern hemisphere is the flow across the equator, so this too must
equal s. Now the total outflow from the northern hemisphere equals the losses for that hemisphere
plus the diffusion across the equator, that is s + 2s, which equals 3s. This outflow from the northern
hemisphere (3s) must equal the inflow to the northern hemisphere, which is given us in the problem,
3 × 106 tonnes per year. Thus
3s = 3 × 106 T/ yr,
and so s, the flow across the equator, is about 1 × 106 T/ yr, or 1 million tonnes per year.
Problem 9: The floor of a poorly insulated attic has an R-value of 10 hr ◦ F ft2 / Btu. The homeowner
plans to reduce heat loss through the attic by adding a layer of fiberglass batts, which have an R-
value of 20 hr ◦ F ft2 / Btu, on top of the existing floor. What will be the overall R-value of the new
“sandwich” floor system, consisting of the old floor with the insulating batts laid on top?
Solution: To figure the R-value of the combined floor system, we want to know how fast heat will
flow through a certain area A of floor, given a certain temperature difference between “inside” and
“outside.” We know from our previous home heating calculations that we can represent a problem like
this in stock-flow terms by taking the “inside temperature” as a stock. The key to this problem is to
introduce a second stock, the “intermediate temperature,” meaning the temperature in the middle of
the “sandwich,” above the old floor but below the new insulating batts. Now we can make a stock-flow
diagram:
Inside Flow through old Intermediate Flow through new

Temp, Tin insulation Temp, Tmid insulation
Furnace Outside Temp, Tout
Figure 28: Stock-flow diagram for two-layer insulating sandwich.
The model shows heat flowing from the furnace to the interior of the house (kept at temperature Tin ,
through the old insulation to the middle layer between old and new (at temperature Tmid ), and then
through the new insulation to the “outside” (actually the attic space) at temperature Tout . We know
the R-values of the two insulating layers, so let’s use the basic equation for R-values (Definition 1 on
page 100). We get
1
Flow through old insulation = (Tin − Tmid ) × A × .
10
1
Flow through new insulation = (Tmid − Tout ) × A × .
20
But in equilibrium, the net flow (Remark 4 on page 73) of any stock must be zero. Applying this
principle to the “middle layer” stock, we find that the flow through the old insulation and the flow
through the new insulation must be equal. Thus
1 1
(Tin − Tmid ) × A × = (Tmid − Tout ) × A × .
10 20
The unknown factors A cancel, and we find that the temperature difference Tmid − Tout across the new
insulation must be twice the temperature difference Tin − Tmid across the old insulation. But then the
temperature difference across the whole “sandwich” (old and new together), that is Tin − Tout , must
total three times the temperature difference Tin − Tmid across the old insulation. That is to say, 3(Tin −
Tmid ) = (Tin − Tout ), or (Tin − Tmid ) = (Tin − Tout )/3. Therefore, the flow rate through the insulation
“sandwich” (which is the same as the flow rate either through the old insulation or through the new
insulation—they are the same) is given by
1
Flow through insulation = (Tin − Tmid ) × A ×
10
Tin − Tout 1
= ×A×
3 10
1
= (Tin − Tout ) × A × .
30
Comparing this with the fundamental equation from Definition 1 on page 100 again, we see that the
R-value of our “sandwich” is 30. That, then, is the answer to our problem.
Critical Thinking
No doubt you have noticed that the value 30 that we obtained for the R-value of our
“sandwich” is simply the sum of the R-values of the two layers, 10+20 = 30. Do you expect
that this will always happen—that the R-value of a “sandwich” will always be the sum of the
R-values of its layers? It certainly sounds plausible, but one numerical coincidence should
not be enough evidence to convince you! Take a look at Exercise 64 at the end of the chapter
for more information.
Summary of Ideas: Calculating Equilibrium Levels
• In stock-flow systems, the flow rates may depend on the stock levels.
• In a system diagram, a control arrow is used to indicate that a flow rate depends on
a stock level.
• If we know how flows depend on stocks, we can set up an equation or equations that
allow us to calculate the equilibrium level(s) of the system.
• A proxy for a stock level is another variable that is directly related to that level and
can be used to quantify it. For example, the temperature inside a house is a proxy for
the amount of heat the house contains.
• Three ways in which heat can flow are conduction, convection, and radiation.
Simple stock-flow models of domestic heating take account of conduction only.
• The R-value of a building material measures its resistance to heat flow by conduction.
The higher the R-value, the more effective the insulation.
• A parameter in a stock-flow model is some quantity that is not itself a stock or flow
in the model but that nevertheless affects one or more flow rates in the model.
2.4 Energy Flows in the Climate System
Objectives
I can explain the purpose of the IPCC’s work.

I can distinguish between climate and weather.
I can calculate the amount of energy radiated from a heated object, using Stefan’s
law.
I can define the albedo of the Earth and understand how it affects the climate.
I can compute the equilibrium temperature for a planet such as Earth, using a model
that neglects the effect of the atmosphere (a so-called naked planet model).
I know how the greenhouse effect causes the Earth’s temperature to be warmer than
the naked planet model predicts.
I understand that the greenhouse effect is essential to life on Earth, but that changes
to the atmosphere that vary the strength of the greenhouse effect may be harmful.
I can compute with a simple model of the greenhouse effect.
2.4.1 The Climate Question

The Intergovernmental Panel on Climate Change (IPCC) was set up in 1988 at the request of
the member governments of the United Nations, to “assess scientific, technical and socioeconomic
information relevant for the understanding of human-induced climate change” [254]. Over the past
quarter century, the IPCC has attempted to inform the world community about an unexpected
sustainability risk: the possibility that human-generated carbon dioxide, an invisible gas emitted
whenever fossil fuels are burned, might be being released into the atmosphere in sufficient quantity
to significantly affect the Earth’s climate. An international body was chartered to investigate the issue
because the atmosphere has no borders: emissions of carbon dioxide have a worldwide effect. What’s
more, these emissions are long-lasting as well as global: the effects of the fuels we burn today may be
felt for hundreds or thousands of years. And finally, almost everybody in the Western world generates
carbon dioxide emissions in ways (transportation, for example, or using electrical devices powered
by fossil-fueled generating stations) that seem bound up with the structure of our lives. This is not a
problem that can be blamed on some small minority of “polluters”: the people who are going to have
to fix it are the same people as the people who are causing it. These three characteristics (universal
involvement, global extent, long-term nature) have led some [193] to call climate change a “super
wicked” sustainability problem.
The IPCC does not carry out scientific research on its own, but every few years it has summarized
the work of scientists working on climate and related issues in its Assessment Reports. Thousands of
scientists and experts in other fields contribute to these reports as authors, editors, and reviewers. The
most recent of their reports is the Fifth Assessment Report [253], published in 2014. Here is one of
the key paragraphs from the Summary for Policymakers in the Fifth Assessment Report:
Continued emission of greenhouse gases will cause further warming and long-lasting
changes in all components of the climate system, increasing the likelihood of severe,
2.4. ENERGY FLOWS IN THE CLIMATE SYSTEM 107
pervasive and irreversible impacts for people and ecosystems. Limiting climate change
would require substantial and sustained reductions in greenhouse gas emissions which,
together with adaptation, can limit climate change risks.
The models that the IPCC uses to reach conclusions like this are enormously complex and
sophisticated, but they are based on some quite simple ideas. In this section we are going to explore
the world of climate modeling by constructing two highly oversimplified stock-flow representations
of our planet’s climate. The first model—we will call it the “naked planet” model—oversimplifies
to the extent of neglecting the Earth’s atmosphere altogether. We will find that this model does not
give a realistic result: a “naked” Earth would be too cold to sustain life. This focuses our attention
on the role of the atmosphere, and we will incorporate this into our next model, a two-stock model
that tries to represent the way the atmosphere “blankets” the Earth and keeps us warm. This is the
famous greenhouse effect, and we’ll see that our two-stock model does an okay job of representing
it. Finally, we’ll ask how strong the greenhouse effect is in reality. This parameter—the strength of
the greenhouse effect—is the one that we humans are busy modifying by our emissions of carbon
dioxide and other greenhouse gases. In later chapters, we’ll explore the range of possible effects this
may have.
Climate and Weather
People often ask how the IPCC (or anyone else) can make predictions about the climate in future
years. After all, no one can forecast the weather more than ten days ahead at most! But weather and
climate are not the same thing.
It’s an old saying that “Climate is what you expect; weather is what you get.” In other words,
“climate” refers to a long-term average, and “weather” to fluctuations about that average. Think about
it this way. Each day there is a certain chance that the temperature will be between 30 and 40 degrees,
a certain chance that it will be between 40 and 50, a certain chance that it will be between 50 and
60, and so on. You could imagine these outcomes as being determined by the roll of a die. From day
to day, they are unpredictable. But in the long term, if you roll a die many times, then you expect
that roughly one-sixth of the outcomes will be “1”s, one-sixth will be “2”s, and so on (this is called a
statistical claim about the behavior of the die; we will study such claims in greater detail in Chapter 5).
In the same way, scientists can make a statistical claim what the climate is likely to be in the future (the
long-term average) without being able to make detailed predictions about the weather (the individual
rolls of the die).
We can take the analogy further. Suppose you rolled a die 1,000 times and it came up with a six on
500 of them. You would suspect that the die is weighted! Climate change is like this: we humans are
“weighting the die” whose rolls determine the day-to-day weather, making some extreme outcomes
more likely than our experience up to now has led us to expect.
Climate of course comprises many factors. Two of the most important (because of their effect on
agricultural production) are surface temperature (the air temperature as measured by a thermometer)
and precipitation (rain and snowfall). Of these two, surface temperature is easier to understand and
measure, and also is more directly affected by human-caused changes. For this reason, discussions
of climate change often focus on surface temperature—indeed, the phrase “global warming” is
sometimes used instead of “climate change.” It’s important to understand, though, that the climate
is a complex system with many moving parts. In any particular area the most significant effects of
climate change might be through drought, or flooding, or sea-level rise, or changes in animal migration
patterns, or any one of a host of other things, rather than the direct result of changing temperatures.
Think about it. . .
Commenting on climate change, Rick Perry (then governor of Texas, and now U.S. secretary
of energy) said in August 2011:
“Yes, our climates change. They’ve been changing ever since the Earth was
formed.”
It is true that the Earth’s temperature has changed substantially over geological time. There
are periods when the climate has been significantly hotter than it is now, and also periods
when it has been significantly colder. However, modern-day humans have not been around
for any of them, and probably would not have survived if they had. During the time that
agricultural civilization has existed, the climate has been remarkably stable (within a degree
or two of its present-day temperature).
Do some background research. Learn about periods when the climate has been both
significantly colder and significantly hotter than now. What do most scientists believe were
the causes of these changes? Is it rational to argue that since nonhuman causes can bring
about climate change, therefore human causes cannot?
2.4.2 Radiant Energy

We said in the previous section that one of the key constituents of the climate is the (average) surface
temperature of the Earth. The current average surface temperature of our planet is about 288 K,
comfortably above the freezing point of water at 273 K. That’s a far cry from the temperature of outer
space, about 2 K! Some astronomers speak of the “Goldilocks zone” for a planet—warm enough that
water does not freeze to ice, cool enough that water does not boil to steam. Only in this narrow range
could complex life (as we presently know it) exist. What is keeping planet Earth in the Goldilocks
zone?
The answer, of course, is heat from the Sun. How is that heat getting here? In Section 2.3.2 we
learned about the three ways in which heat can be transmitted from one object to another: conduction,
convection, and radiation. Of these, only radiation is capable of transmitting energy across the vacuum
of outer space. Radiation (to be precise we should call it electromagnetic radiation) is a self-
sustaining oscillation of electric and magnetic fields in space: it travels at a huge speed (roughly
3 × 108 m sec−1 , or 186, 000 mi sec−1 ) and can transmit energy from one place to another, even
through a vacuum.
Definition 1
Energy transmitted by electromagnetic radiation is called radiant energy.
Electromagnetic radiation comes in many different forms: radio waves, microwaves, infrared,
ordinary light, ultraviolet, and X-rays are all examples of electromagnetic radiation. These forms
are distinguished by their wavelength, which is the distance between one “crest” and the next of the
oscillating electric field. The list above goes from longer to shorter in terms of wavelength: at one
extreme, radio waves can have wavelengths of many meters or even kilometers, and at the other, X-
ray wavelengths can be on the order of nanometers (10−9 meters). In the middle, the visible light
spectrum extends from about 700 nanometers (red) to about 400 nanometers (blue). Notice that the
shorter the wavelength, the bluer the light. It is also true that the shorter the wavelength, the more
energetic the radiation: radio waves pass through our bodies all the time without having any effect,
while even a brief exposure to high-energy X-rays can be damaging to the structure of living cells.
There are several ways that radiant energy may be generated, but only one is relevant to the climate
discussion. This is heat or thermal radiation.
Definition 2
Thermal radiation is the radiant energy emitted by a physical object when it is heated.
Imagine heating a piece of steel bar in the flame of a blowtorch. As the steel heats up, it begins
to glow: first cherry red, then orange, yellow, and eventually (if you can get it hot enough) bright
blue-white. Holding up your hand, you can feel the heat radiating off the metal. The steel is giving off
thermal radiation, which is traveling through space until it is interrupted by your hand. The radiation
is deposited in your skin, heating it up and giving you a sensation of pleasant warmth—or of pain if
you get too close. We can see from the example of the steel bar that as the temperature increases, two
things change about the thermal radiation that is emitted: first, the total amount of radiation increases,
and second, the wavelength decreases (the radiation becomes bluer). There is a mathematical formula
called Stefan’s law for the total amount of radiation emitted by an “idealized” physical object at
various temperatures.
Rule 1: Stefan’s Law
• The amount of thermal radiation emitted by each square meter of the surface of an
“idealized” physical object at temperature T kelvins is
σ T 4,
where the constant σ , called Stefan’s constant, is approximately equal to 5.7 ×

10−8 W/ m2 K4 .
The symbol σ is a Greek letter. It is pronounced “sigma.”

Remark 1: An “idealized” physical object in this sense is also called a blackbody, and Stefan’s law
is also called the blackbody radiation law. Most condensed matter (solids, liquids) follows this law
to a reasonable accuracy.
Example 1: The Sun is modeled by a sphere of radius 7 × 108 m and surface temperature 5, 800 K.
What is the thermal power output of the Sun, according to Stefan’s law?
Solution: Stefan’s law tells us that the power output per square meter is
W W W
σ T 4 = 5.7 × 10−8 × (5, 800 K)4 ≈ 5.7 × 10−8 × (1.13 × 10154
K ) ≈ 6.5 × 107 2 .
m2 K4 2
m K
4 m
To get the total power output we must multiply by the surface area of the Sun, which is8 4πR2 , where
R is the radius and π ≈ 3.14; that is,
8 The area of a sphere of radius R is equal to 4πR2 .
Some incoming
radiation is reflected
directly
Outgoing radiation from all of

Earth’s surface (less hot, longer
Incoming radiation from wavelength)
Day
Sun (hotter, shorter Night
wavelength)
Figure 29: The naked planet model. Notice that the wavelength of the incoming radiation is governed by the surface temperature
of the Sun (relatively high, hence short wavelength), whereas the wavelength of the outgoing radiation is governed by the surface
temperature of the Earth (relatively low, hence long wavelength).
4 × π × (7 × 108 m)2 ≈ 6 × 1018 m2 .
Multiplying these two numbers gives us the power output
W
6.5 × 107
m2
× 6 × 1018 ≈ 4 × 1026 W.
m2

That is a lot of power.
Remark 2: As we mentioned above, the “color” or wavelength of thermal radiation also depends on
the temperature of the emitting body (the hotter the object, the shorter the wavelength and the “bluer”
the radiation). For an idealized physical object, this dependence is also given by a mathematical
formula called Planck’s law. But in our discussion we won’t need the exact mathematical formula,
just the simple “hotter means bluer” idea.
2.4.3 The Naked Planet Model

Now that we have Stefan’s law available, we can set up our simplest climate model. It’s going to be a
stock-flow model in which the “stock” is the total amount of heat on Earth, as measured by the average
temperature as a proxy. There is one inflow—incoming radiant heat from the Sun—which arrives at a
constant rate. And there is one outflow—outgoing radiant heat from the Earth—which leaves at a rate
that depends on the Earth’s temperature. At equilibrium, these two will balance, giving us an equation
that we can solve to find the Earth’s temperature. It’s just like the model we used for the temperature
of a house in the previous section (incoming heat from the furnace, at a constant rate; outgoing heat
through the structure, dependent on the temperature) but on a much grander scale. See Figure 29.
We are making many simplifying assumptions. Some of the most important are the following:
• This is a “single stock” model: there is just one stock variable, the “average temperature.” The
real temperature of the Earth varies both in time (days are warmer than nights, summers than
winters) and in space (the poles are colder than the equator; continental climates are more
extreme than maritime). For a first approximation, we are neglecting all these variations.
• We are also neglecting the atmosphere, the envelope of gas that surrounds our planet. (That is
why we call this the “naked planet” model.) We’ll see when we do the calculations that this
assumption leads to a conclusion that is far from what we actually observe. This is because
the atmosphere has a major effect on the climate, by “trapping” some of the outgoing radiant
energy and not letting it escape from the planet. In fact, it is only because of this greenhouse
effect that the Earth is in the Goldilocks zone, allowing us to live here. We’ll do the math for
that in the next section.
Here are some symbols that we will use in the calculations for this model.
Table 4: Notations for Simple Climate Model
Symbol Name Description

R Radius of planet For the earth, this is about 6.4 × 106 m.
α Albedo Fraction of incoming solar energy that is immediately re-
flected back to space.
L Solar constant Rate of incoming solar energy, measured as the number of
watts per square meter of area directly facing the Sun.
σ Stefan’s constant Appears in the thermal radiation law; approximately 5.7 ×
10−8 W/ m2 K4 .
T Temperature of planet This is the “absolute” temperature, measured in kelvins.
π Pi Ratio of the circumference of a circle to its diameter; approx-
imately 3.14.
Most of these we have seen before, but let’s just talk about the new ones, L and α:
• The solar constant L is the amount of solar radiation that would fall on a 1 m2 panel that was
square-on to the Sun’s radiation (e.g., horizontal at the equator at noon at the spring equinox).
For the Earth, the solar constant is approximately 1, 350 W/ m2 .
• The albedo α is the fraction of the incoming solar radiation that is directly reflected back to
space, as indicated by the thin orange arrows in Figure 29. If the planet were a perfect mirror,
a giant disco ball in space, then the albedo would be 1; if the planet were perfectly black,
absorbing all the energy that lands on it, the albedo would be 0. A realistic value for the Earth
is about α ≈ 0.3; most of the energy reflection that contributes to α is from clouds, which are
white (at least on top!) and some is from reflective areas of the Earth’s surface (ice caps, snowy
mountains, deserts).
Problem 1: Calculate the solar constant for the Earth, using the information about the Sun’s total
power output (Example 1 on page 109) and the fact that the radius of Earth’ s orbit is about 150
million kilometers.
Solution: From Example 1 on page 109, the total power output of the Sun is approximately 4 ×
1026 W. At the radius of Earth’s orbit, this amount of power is spread over a sphere of radius 150
million km = 1.5 × 1011 m. The area of this sphere is 4π(radius)2 , that is,
2
4 × π × 1.5 × 1011 m ≈ 3 × 1023 m2 .
This gives the value of the solar constant:
4 × 1026 W
≈ 1, 300 W/ m2 .
3 × 1023 m2
This was an approximate calculation, but we arrived pretty close to the correct value of 1, 350 W/ m2 .
Short-wave Planet Long-wave

radiation temperature T radiation
Figure 30: The naked planet model as a stock-flow system.
Let’s now set up the stock-flow equilibrium for the naked planet model, whose system diagram is
shown in Figure 30. The inflow of heat to the planet is given by the product of three quantities:
• The solar constant L, in watts per square meter.
• The “square-on” area of the Earth that is visible to the Sun, that is, the area of a circular disk of
radius R, which is πR2 . (A sphere looked at “square-on”—think of looking at the full moon in
the sky—looks exactly like a circular disk of the same radius.)
• A factor (1 − α), where α is the albedo: α represents the fraction of incoming solar energy that
is reflected without being absorbed, so (1 − α) is the fraction that is absorbed as heat.
In summary,
Inflow = πL(1 − α)R2 .
Notice that this inflow does not depend on the temperature T . The outflow of heat, however, does
depend on the temperature. It is given by Stefan’s law as the product of σ T 4 and the area of the whole
planet, which is 4πR2 . Thus
Outflow = 4πσ T 4 R2 .
At equilibrium, the inflow and the outflow must balance. Putting these two quantities equal to each
other gives the equation for equilibrium:
πL(1 − α)R2 = 4πσ T 4 R2 .
Conveniently, the factors of πR2 can be canceled, so that 4σ T 4 = L(1 − α), or (solving for T )
r
4
L(1 − α)
T= . (2)
4σ
Problem 2: Find the temperature prediction of the naked planet model for the Earth.
Solution: As we have mentioned, the solar constant L for the Earth is about 1, 350 W/ m2 and the
albedo α is about 0.3. Thus the predicted temperature is
v
4
!
W 0.7 m2 K4
u
4
p
4.2 × 109 K4 ≈ 255 K.
u
T = t 1, 350

× ≈

m
2 4 × 5.7 × 10−8 W

You can confirm this graphically, if you want, by plotting the graphs of the incoming and outgoing
energy flows (in watts per square meter) as a function of temperature. This is done in Figure 31 on
the opposite page, and you can see that the orange (incoming) and red (outgoing) curves intersect at a
temperature value of approximately 255 K.
This answer is well below freezing (273 K), and a lot colder than the correct value9 of about 288 K.
So what went wrong? As we’ll see in the next section, we got an answer that was too cold because we
neglected to consider the effect of the atmosphere.
9 See the IPCC report [253], Annex III (Glossary) to Report WG1, entry Energy balance.
300 Inflow
Outflow
Power (Watts/m2 )
200
100
0
0 50 100 150 200 250 300
Temperature (Kelvins)
Figure 31: Equilibrium in the naked planet model occurs at about 255 K.
Remark 3: We have focused our attention, naturally enough, on the planet Earth. But the “naked
planet” model can be applied to any planet, once we know the solar constant (which can be calculated
from the distance from the Sun, as in Problem 1 on page 111 above) and the albedo. In Exercise 68
for this chapter we ask you to calculate the “naked planet” predictions for the surface temperatures of
Venus and Mars. As was the case for Earth, these predictions are too cold—a little too cold in the case
of Mars, and much too cold in the case of Venus, whose actual surface temperature is hot enough to
melt lead. Again, atmospheric effects are the reason for this discrepancy.
2.4.4 The Greenhouse Effect

The temperature of a greenhouse (like the one pictured in
Figure 32) also results from a balance between incoming heat
(from solar radiation) and heat loss. The greenhouse warms
up because its glass, which is fairly transparent to incoming
solar radiation, is much less transparent to the reradiated heat
from the soil and plants in the greenhouse.10 Much of that
reradiated heat is therefore trapped inside the greenhouse,
where it makes things warmer than they would be otherwise.
American scientist Eunice Newton Foote (1819–1888) pre-
sented a paper to the American Association for the Advance-
ment of Science in 1856 “on the Circumstances Affecting the
Heat of the Sun’s Rays” [116]. It was not an easy time to
be both a woman and a scientist—the AAAS had been all Figure 32: A greenhouse.
male until a few years before, and Foote was apparently only
its second female member. She did not, or perhaps was not
permitted to, present her paper in person, and instead it was read on her behalf by Professor Joseph
Henry.11 Foote wrote regarding carbon dioxide, “if, as some suppose, at one period in its history the
air had mixed with it a larger proportion [of CO2 ] than at present, an increased temperature. . . must
10 It should not be a surprise that various materials can be transparent to some wavelengths of electromagnetic radiation but
not others. For example, radio waves (longer wavelength) pass fairly easily through the walls of your apartment, whereas light
waves (shorter wavelength) do not.
11 According to a contemporary account, Henry prefaced his reading by “a few words. . . to the effect that science was of no
country and of no sex.” True enough, but allowing Eunice Foote to read her own paper seems to have been a step too far for the
worthies of the AAAS.
necessarily have resulted.” This appears to be the first time that it had been suggested, on the basis
of experiment, that CO2 in the upper atmosphere could “trap” heat on Earth in the same way that its
glass walls “trap” heat inside the greenhouse. At any rate, Eunice Foote anticipated the better-known
and more systematic experiments of Irish physicist John Tyndall by three years.
The Swedish scientist Svante Arrhenius (1859–1927) was the first to attempt to calculate the
magnitude of this “greenhouse effect,” obtaining results that are remarkably accurate even by today’s
standards. He wrote,
To a certain extent the temperature of the earth’s surface, as we shall presently see, is
conditioned by the properties of the atmosphere surrounding it, and particularly by the
permeability of the latter for the rays of heat.
Using satellite measurements, it is now possible to directly “see” the greenhouse effect, by observing
the amount of Earth’s heat radiated from the upper atmosphere and comparing it with the amount
leaving the surface: the difference between these quantities is the amount of heat trapped by the
greenhouse effect.
Remark 4: Arrhenius also proposed the idea that industrial emissions of heat-
trapping gases such as carbon dioxide might be big enough to alter the strength
of the greenhouse effect and thus change the climate. We will come back to this
later. Right now, though, notice that the greenhouse effect is, overall, definitely
good news for us humans. Without any greenhouse effect at all, the Earth would be
frozen solid!12
So how shall we model the greenhouse effect? It arises from the interaction of
radiant energy with atmospheric gases (especially triatomic gases like water vapor
and carbon dioxide), and a full-scale climate model has to incorporate a lot of detail
about the physics of these interactions. That’s too hard for us even to begin at this
point. We’ll consider a very crude “greenhouse” model called the glass ball model.
Figure 33: Svante Arrhenius. Here, we model the atmosphere as a single layer, like a sphere of glass
surrounding the planet. We assume that
• The atmosphere is transparent to the incoming solar radiation (which you’ll remember arises
from the hot surface of the sun, so has a relatively short wavelength). That incoming radiation
passes straight through the atmosphere without any impediment, and lands on the planet’s
surface, warming it up as before.
• The atmosphere is opaque to outgoing radiation (which comes from the much cooler surface
of the planet, so has a relatively long wavelength). All the outgoing radiation from the planet’s
surface is reabsorbed by the atmosphere.
If we’re assuming that all the outgoing radiation from the planet’s surface is reabsorbed by the
atmosphere, how can any energy get out of the model at all? Why doesn’t it just heat up for ever?
The answer is that the atmosphere itself (the “glass ball”) warms up and radiates thermal energy. The
thermal energy radiated by the atmosphere travels both upward and downward (because the “glass
ball” has an inside and an outside). The inward-traveling energy heads back to the planet’s surface
again, but the outward-traveling energy heads out to space. This is the outflow from the whole system
that brings it into equilibrium.
In other words, the system diagram for the glass-ball greenhouse model is a two-stock one like
Figure 34 on the opposite page.
Example 2: Find the temperature prediction of the glass ball greenhouse model for planet Earth.
12 See Problem 2 on page 112.
Short-wave Planet
radiation temp Tp
Flow from atmosphere to planet

Flow from planet to atmosphere
Flow from atmosphere
to space
Atmosphere Long-wave
temp Ta radiation
Figure 34: The “glass ball” greenhouse model as a stock-flow system.
Solution: We will measure all the energy flows per unit area (in other words, divided by the factor
4πR4 , which we saw canceled out in the naked planet model). The place to begin is by looking at the
model as a whole—both stocks together. The energy inflow to the model as a whole is the blue arrow
on the left, which is given by net solar radiation corrected for albedo 14 L(1 − α), just as in the naked
planet model. And the outflow from the model as a whole is the red arrow on the right, σ Ta4 . So this
gives us the energy balance equation
L(1 − α)
σ Ta4 = ,
4
which can be solved to tell us Ta . Notice that this is the same equation as we got for the Earth
surface temperature in the naked planet model. Now, however, it gives us Ta , the temperature of the
atmosphere.
Now that we’ve got this equation, let’s look at the energy balance for the atmosphere (the lower box)
by itself. The inflow is given by the orange-captioned arrow, in the middle of the diagram, showing
heat flow from the Earth to the atmosphere, that is, σ Tp4 . The net outflow from the atmosphere is
the sum of the two red-captioned arrows, one indicating heat radiated by the atmosphere to outer
space and the other indicating heat returned by the atmosphere to the Earth’s surface. These both have
magnitude σ Ta4 , so their sum equals 2σ Ta4 . Equating inflow with outflow, we get
L(1 − α)
σ Tp4 = 2σ Ta4 = .
2
Solving for Tp , we get
r
4
L(1 − α)
Tp = .
2σ
Plug in the numerical values for planet Earth to get
v
4
!
W 0.7 m 2 K4
u
4
p
8.3 × 109 K4 ≈ 300 K.
u
Tp = t 1, 350

× ≈

m
2 2 × 5.7 × 10 −8 W

This is now a bit too hot, but it is much closer to the correct value of 288 K than the prediction of the
naked planet model is.
Critical Thinking
In the glass ball model, Tp > Ta . Is this what you would expect from the metaphor of the
greenhouse layer as an “insulating blanket”?
Remark 5: The glass ball model is extremely crude and oversimplified. However, the basic structure
of the model—that the upper atmosphere is cooler than the surface of the Earth, that it radiates
heat both outward to space and back inward to the surface, and that the temperature of the upper
atmosphere, not the surface, is what counts when we look at radiation to space—is still present in
much more accurate models [265], including those used by the IPCC. The last point, that we “see” a
cooler temperature than that of the surface when we look at radiation to space, can now be confirmed
by direct measurements from satellites that look down at the atmosphere from above.
2.4.5 Varying the Strength of the Greenhouse Effect

The Earth’s real greenhouse blanket is not an imaginary ball of glass, but a column of gases of variable
composition and thickness, in which each gas absorbs and emits radiation at specific frequencies.
Crucially, the real greenhouse effect is not just “off” (no glass ball) or “on” (glass ball); its strength
can vary according to the amount of greenhouse gases in the atmosphere. To understand this, we
therefore need a model in which the greenhouse effect has variable strength.
Short-wave Planet Long-wave

radiation temperature T radiation
GHE strength, g
Figure 35: Stock-flow climate model with variable greenhouse effect.
Without getting involved in the sophisticated physics of modeling the atmospheric column
(see [265]), we can make a very simple model with a variable-strength greenhouse effect as follows.
Notice that the formulas for the Earth’s temperature in the naked-planet and glass-ball greenhouse
models are related by a factor of 2 underneath the fourth root sign:
r r
4 4
L(1 − α) L(1 − α)
Tnaked = , Tglass ball = .
4σ 2σ
This is because the rate of heat loss from the Earth’s surface in the glass ball model is 12 × 4πR2 σ Tp4 ,
exactly half what the “naked planet” model would predict. It is natural to introduce a parameter g
(Definition 2 on page 102) into the problem, a pure number g between 0 and 1, in such a way that the
heat loss equals
4πR2 σ Tp4
.
1+g
Thus g = 0 represents no greenhouse effect—the naked planet model—and g = 1 represents the “full
on” greenhouse effect predicted by the “glass layer” model. The system diagram for our modified
greenhouse model, incorporating the special flag-shaped box for the parameter g that measures the
strength of the greenhouse effect, is given in Figure 35, and its predicted temperature is
r
4
L(1 − α)(1 + g)
Tp = .
4σ
The actual present temperature of the Earth corresponds to a g-value of roughly 0.65.
Remark 6: If you look closely at Figure 35, you’ll notice one extra feature that we haven’t described
so far. The minus sign attached to the control arrow from the parameter box for g to the “long-wave
radiation” spigot indicates that the effect of this parameter on the quantity it controls is negative: if the
parameter gets bigger, the flow rate that is linked to it must be reduced. We will study the importance
of these signs on control arrows at much greater length in Chapter 4 when we get to look at the
different types of feedback (positive and negative) that may occur in a stock-flow system, and at its
sensitivity to small shocks that may change a parameter slightly.
Summary of Ideas: The Climate System
• Radiant energy is transmitted by electromagnetic radiation.

• The radiant power output of an “ideal” heated object is given by Stefan’s law as σ T 4
watts per square meter, where σ is Stefan’s constant.
• Hotter objects produce shorter-wavelength (bluer) radiation.
• The Earth’s climate results from an equilibrium between incoming radiant energy
from the Sun and outgoing radiant energy from the warmth of the Earth itself.
• The greenhouse effect traps some of the Earth’s radiated heat beneath a layer of
atmospheric gases. This makes the planet warmer than it otherwise would be.
• Without any greenhouse effect the Earth would be frozen. However, scientists are
concerned that human activity may be increasing the greenhouse effect and changing
the climate.
• The IPCC is a United Nations body that was chartered in 1988 and is charged with
regularly reporting on the effects of human-induced climate change,
7. Do the same calculation as in Problem 1 on

2.1. STOCKS, FLOWS, AND EQUILIBRIUM page 72, but incorporating the human-caused contribu-
tions (red figures) as well. Is the system in equilibrium
now? What do you expect will happen to the stock of
1. What do you think are reasonable values for the atmospheric carbon?
amount of water in a full bathtub (in gallons), the
maximum rate of inflow from the faucet (in gallons per 8. Refer back to the carbon cycle depicted in Figure 5
minute), and the maximum rate of outflow through the on page 72. The “human emissions” label depicts a
drain (also in gallons per minute)? Explain how you flow of carbon into the atmosphere. Where is this
arrive at these values. carbon flowing from (primarily)?
2. Suppose a shower head pours 1.5 gallons of water 9. In Example 4 on page 75 we saw that a simple
per minute into the tub. The drain has a clog and allows stock-flow model treating all enrolled students as a
for only 0.5 gallons of water per minute to exit the tub. single stock is sufficient to investigate the average time
What is the net flow that applies to the tub? a student spends as an undergraduate. Develop some
questions that could be investigated with a model that:
3. Can small adjustments to a flow lead to large (a) divides the student body into two stocks: part-
changes in a stock? Explain why or why not. time and full-time students.
(b) divides the student body into four stocks: fresh-
4. Instead of measuring the current capacity of a men, sophomores, juniors, and seniors.
battery (in amp-hours) we could measure its energy
capacity (in watt-hours). The energy capacity (in watt-
hours) is equal to the current capacity (in amp-hours) 10. Consider a freshwater aquarium with a stock
multiplied by the battery voltage (in volts). The voltage of 45 gallons of water and a recirculating pump with
of an AA battery is 1.5 volts. Estimate its energy a flow rate of 150 gallons per hour (gph). Calculate
capacity. the residence time of the water in the tank. Explain
How much does a single AA battery cost? Find the meaning of the residence time in this case (see
the cost of the energy delivered by an AA battery, in Remark 7 on page 76).
dollars per kilowatt-hour. (Remember, the prefix “kilo”
denotes one thousand, so one kilowatt-hour equals
1000 watt-hours.) 11. Diagram a stock-flow model for the temperature
Electrical energy from a wall outlet costs about 0.1 in a room in winter, including heat supplies from the
dollars per kilowatt-hour. How many times more ex- heating system and heat losses to the outside.
pensive is “battery energy” than “wall outlet energy”?
Why are batteries used if they are so much more 12. There are about 1013 m3 of water in the atmo-
expensive? sphere, and worldwide annual precipitation is about 5×
1014 m3 / yr. Calculate the approximate residence time
5. Draw a system diagram for a rechargeable (rather of water in the atmosphere, expressing your answer in
than single-use) battery. appropriate units.
6. So-called renewable energy sources (wind, solar, 13. Sulfur dioxide is an atmospheric pollutant that
hydroelectric, and so on) depend ultimately on the flow is a major component of smog. L. Barrie and R. Hoff
of energy that the Earth receives from the Sun. “But,” studied sulfur dioxide pollution in the atmosphere of
argues a student, “scientists say that in 6.5 billion years the Alaskan Arctic. They determined that the residence
or so, the Sun’s energy stock will be exhausted. So solar time of sulfur dioxide in the atmosphere (at a certain
energy, and everything that depends on it, are nonre- season) was approximately 3 weeks, and that the level
newable too.” Comment on the student’s argument. of sulfur dioxide in the atmosphere was approximately
2.8 micrograms per cubic meter. Using this informa-
tion, estimate the net rate of inflow of sulfur diox- 19. Draw a stock-flow diagram in which the stock is
ide into a layer of atmosphere 100 meters thick and the number of guests currently staying at a particular
covering 24,500 square kilometers (the area of Denali hotel. Suppose that during the peak season the number
National Park). Note: 1 microgram = 10−9 kilograms. of guests at the hotel is in dynamic equilibrium, with
60 guests arriving and departing daily, and the average
stay for a guest being 3 days. How many guests are
14. It is estimated that approximately 1011 kilograms
staying in the hotel each night during the peak season?
of sulfur are added to Earth’s atmosphere every year
by natural sources of sulfurous gases (primarily sulfur
dioxide, SO2 , and hydrogen sulfide, H2 S ). Sulfurous 20. During the morning rush hour in New York City,
gases are in dynamic equilibrium in the atmosphere traffic averages 6 miles per hour along a particular 3
with an average residence time of approximately 4.4 mile stretch of multi-lane highway (in one direction).
days. Estimate the total mass of sulfur contained in the If at any one time there are 2200 cars moving along
atmosphere in the form of these gases. Why do you this 3 mile stretch of road, about how many cars are
think the residence time in this question, referring to passing through this section of road each hour?
sulfurous gases worldwide, is so much less than the
residence time for similar gases in the question before? 21. A supermarket sells 6 boxes of a particular cereal
each hour, and a box of cereal typically sits on the
shelf for 5 hours before being sold. How many boxes of
15. Draw a stock-flow diagram for the money in a cereal are on the shelf, if the shelf stockers are keeping
checking account. the cereal on the shelf in dynamic equilibrium?
16. Soils contain stocks of nutrients for plants. 22. An estimated 300,000 deer are harvested in
Farmers apply fertilizers to soils to increase the nu- Pennsylvania each year by hunters and another 100,000
trients available to crops. One essential nutrient used are killed in collisions with vehicles or by other means.
in commercial fertilizers is phosphorus. Draw a stock- If the average lifespan of a deer in Pennsylvania is 4
flow diagram for the phosphorus content of the soil on years, estimate the total deer population of Pennsylva-
a plot of farmland. nia.
17. The Ogallala aquifer is a groundwater source 2.2. ENERGY STOCKS AND FLOWS
that underlies portions of eight Great Plains states.
The USGS estimated that total water storage in the 23.
Ogallala aquifer is approximately 3.6 × 1012 cubic (a) Identify the different kinds of energy, and the
meters. Annual withdrawals of water from the aquifer energy flows, that are involved in the operation
are approximately 3 × 1010 cubic meters. Assume that of a conventional automobile. Draw a stock-
60 percent of the water withdrawn returns to the aquifer flow diagram.
via recharge and irrigation return flow. Use this infor- (b) Now what if the automobile is a hybrid, like a
mation to estimate the length of time until the aquifer Toyota Prius? Draw a stock-flow diagram in this
case.
becomes exhausted.
24. Read about the history of the quest to build a

18. The seeming conflict between the short residence
perpetual motion machine (Figure 14 on page 88), and
time of carbon in the atmosphere and the long-term
effects of human-caused CO2 emissions is the source find the mistake in a few historical examples. A good
of an argument used by “climate skeptics” against the place to start is the Lock Haven University perpetual
notion of climate change. motion page at [298].
Follow up on our answer to Question 1 on page 76
by visiting the Skeptical Science website where you can 25. Express an energy flow rate of 1 watt in terms of
read more about the distinction between residence time Btu (British thermal units) per hour. You will need the
and the increase of atmospheric carbon dioixde [203]. deﬁnition of a British thermal unit from Table 2 on
Explain this distinction in your own words. page 86.
26. Estimate the annual cost of energy for an electric do you expect that solar will be contributing? Use the
clothes dryer. The dryer consumes 3.4 kW and runs 5 data from the LLNL energy flow chart [191].
hours per week. Assume that electrical energy costs 13
cents per kWh. 34. The flow chart [191] headlines U.S. energy
consumption as 97.3 quads per year. This is a rate
27. A newspaper article reports on the solar panels of flow of energy (energy divided by time), that is, a
that my neighbors have installed on the roof of their power. Express this power in terawatts (one terawatt
home. The reporter writes, “The solar panels will equals 1012 watts).
generate 10,000 kilowatts of electricity every year.”
What, if anything, is wrong with this reporting? 35. As of this writing you can find an interactive
chart showing energy flows in the United States at
energyliteracy.com [138]. Explore this interactive flow
28. Before it was breached in 2011, the Condit chart. “Mousing over” a particular node in the chart
Dam on the White Salmon River in Washington State will highlight all the flows going through that node and
fed a hydroelectric power plant. Water from the dam provide more detail about that node in text. Try to use
descended 51 meters to feed the turbines, and the peak the chart to answer the following questions:
flow rate was 1.4 × 105 cubic meters per hour. Peak (a) When we break down energy use in single-
electricity generation was 14.7 megawatts. What was family homes, the single largest energy use is
the efficiency of electricity generation at the Condit for what?
Powerhouse? (See Remark 2 on page 86 for informa- (b) What percentage of the energy used in highway
tion about the energy released by a certain mass falling transportation is used for trips that are 5 miles
through a certain distance, the gravitational constant.) or less?
(c) What percentage of the total value of electrical
29. Using data from the LLNL energy flow energy produced is lost in transmission and
chart [191], calculate the percentage of U.S. electricity distribution (that is, in moving the electrical
production that is generated by nuclear power. energy from where it is generated to where it
is used)?
30. Using data from the LLNL energy flow

chart [191], find out what percentage of overall U.S. 36. As of this writing you can find an interactive
energy production comes from renewable sources (not chart showing energy flows in the United States at
including nuclear energy). energyliteracy.com [138]. Explore this interactive flow
chart. “Mousing over” a particular node in the chart
will highlight all the flows going through that node and
31. At the Manapouri hydroelectric station on the
provide more detail about that node in text. Use this
South Island of New Zealand, water from Lake Man-
chart to formulate and answer several questions about
apouri drops through 230 meters before being dis-
energy use in the United States.
charged to the sea at Doubtful Sound. The maximum
flow rate of water through the plant is 510 cubic
meters per second. Calculate the maximum amount of 37. At Iceland’s Kárahnjúkar Hydroelectric Plant,
power that could be generated at Manapouri. (Use the water is diverted from the Jökulsá á Dal river and
gravitational constant). decends 600 meters through underground water tunnels
to the power station. The maximum flow rate is 145 cu-
bic meters of water per second. The plant’s maximum
32. Using data from the LLNL energy flow power output is 690 MW. Estimate the efficiency of the
chart [191], estimate the efficiency of electricity gen- plant (the maximum power output as a percentage of
eration in the U.S. the total energy released by the water falling 600 meters
at maximum flow).
33. The solar energy sector is presently experiencing
very rapid growth. Suppose that solar energy produc- 38. In 2016, utility-scale solar plants in the United
tion in the U.S. doubles in size every two years. Ten States produced about 36 terawatt-hours of electrical
years from 2016, how much energy (in quads per year) energy. A large-scale coal plant can provide 3,000
megawatts of power if operating at maximum capacity. these cities is 165 miles. I drive a Toyota Prius, which
How long would such a plant need to operate at full weights about 1 12 tons and has a normal fuel economy
capacity to produce 36 terawatt-hours of electrical of 50 miles per gallon. It also has a fuel consumption
energy? meter, which is accurate to 0.1 miles per gallon. Should
I expect to see a difference between the fuel economy
39. One kilogram of bituminous coal contains about measured on the “downhill” trip from State College to
30 megajoules of energy. A coal train car holds about Baltimore and on the “uphill” trip from Baltimore to
110 tonnes of coal. As of this writing the United States State College? If so, estimate that difference.
electrical grid has about 6 gigawatt-hours of grid-tied
battery energy storage. About how many train cars of
coal is this equivalent to?
2.3. CALCULATING EQUILIBRIUM
STATES
40. An electric water heater consumes 3.8 kW of
power and converts it to heat. How long will it take 46. Suppose a person has a retirement investment
the water heater to heat an 80 gallon tank of water from portfolio that consistently earns 4.5% annually. In re-
10 ◦ C to 50 ◦ C? tirement she intends to draw $36,000 each year from
this account. Calculate the (nonzero) equilibrium value
41. Suppose that one weekend day a hiker exerts of the investment portfolio.
200 watts of power while ascending 300 meters on a
steep trail without a pack. The hiker repeats this trip 47. Suppose that the account in Exercise 46 instead
the following weekend but is carrying an additional 10 earns 6.5% annually. Use the plot of earnings and
kilograms in a pack. How much longer should the hiker withdrawals shown in Figure 36 on the next page
expect her ascent to take assuming her power output to estimate the equilibrium value of the investment
remains the same? (Use the gravitational constant.) portfolio.
42. One kilogram of gasoline contains about 4.2 × 48. Water flows into a large tank at a constant rate of
107 J of chemical energy and has a volume of 0.37 10 gallons per minute. There is an outgoing flow rate r
gallons. A car with a gas mileage of 28 mpg drives related to the height of the water in the tank (in feet) by
through town at 45 mph. Calculate the car’s rate of r = Ch, where C is a constant:
consumption of chemical energy in watts.
C = 1.25 gal/ ft min.
43. Express the fuel economy of the car in Exer- (a) Draw a stock-flow diagram for this problem.
cise 42 in units of watt-hours per mile. Why might it be Include control arrows that indicate how the
useful to use watt-hours instead of gallons of gasoline stock influences the flow rate(s).
in discussing car energy use? (b) At what height will the water level in the tank
be in (dynamic) equilibrium?
44. Compare the results from Exercise 43 with the
efficiency of
49. Each spring the Porcupine caribou herd migrates
(a) an all-electric vehicle (look up the ratings in to its calving grounds on the Arctic coastal plain in
watt-hours per mile of one or more electric
northeastern Alaska and northwestern Canada; 36,000
vehicles).
surviving calves are added to the herd. If the average
(b) a person traveling by bike (a person traveling on residence time (i.e., lifetime) of a caribou in the Porcu-
a bicycle over flat ground can travel 8 mph with pine herd is 5 years, and assuming the herd size is in
a power output of about 100 watts).
equilibrium, what is the size of the herd?
45. For a medical appointment, I need to drive from 50. Draw a stock-flow diagram for the Porcupine
State College, PA to Baltimore, MD and back. Balti- caribou herd in Exercise 49. Include control arrows that
more is close to sea level; State College is about 1100 indicate how the stock influences the flow rate(s).
feet above sea level. The distance by road between
50
Earnings
Withdrawals
Flow rate ($1,000s per year)

40
30
20
10
0
0 100 200 300 400 500 600 700
Account Value ($1,000s)
Figure 36: Figure for Exercise 47 on the previous page.
51. A soluble pollutant is dumped into a lake at the it takes to heat the house to 68 ◦ F when the outside
rate of 1.5 tonnes per day. The late has a volume of temperature is 32 ◦ F.
600 million cubic meters and the daily flow rate of
water into the lake is 500,000 cubic meters. The daily 56. In the house of Exercise 55, the homeowner
flow rate of water out of the lake is 250,000 cubic decides to replace all the windows with double-paned
meters, and 250,000 cubic meters of water leave the ones having an R-value of 2.1. Assume that the average
lake per day by evaporation. Assuming the pollutant outside temperature for the month of January is 32 ◦ F.
is well mixed and evaporates with the water, what is How much energy, in Btus, will the homeowner save
the equilibrium level of pollutant in the lake, measured over the month by replacing the windows?
in tonnes (the mass of one cubic meter of water is 1
tonne)?
57. Suppose that the furnace in the home of Exer-
cise 55 has a rated heat output of 40, 000 Btu/ hr. What
52. How would the answer to Problem 3 on page 97 is the lowest outside temperature at which it will still be
be affected if the pollutant codistills (and thus leaves possible to maintain the desired interior temperature of
the lake by evaporation as well as by river flow)? 68 ◦ F? Give two answers, first for the original windows,
and then after window replacement as in Exercise 56.
53. The effectiveness of an insulating material is
sometimes expressed in terms of the “U-factor,” which 58. Suppose that a poorly insulated house has an
is just the inverse of the R-value: U = 1/R, R = 1/U. overall R-value of 7 hr ◦ F ft2 / Btu. The house has a
What are the units in which the U-factor is measured? furnace that can deliver 34, 000 Btu/hr and the total area
Does a large U-factor indicate that a material is a good of the walls, windows, and roof is 3500 ft2 .
insulator or a bad one?
(a) Draw a stock-flow diagram for this problem.
Include control arrows that indicate how the
54. A section of exterior wall has an R-value of stock influences the flow rate(s).
19 hr ◦ F ft2 / Btu and an area of 200 square feet. When (b) What is the minimum outdoor temperature at
the interior temperature is 70 ◦ F and the exterior tem- which this furnace could maintain an indoor
perature is 30 ◦ F, how many Btu per hour are lost temperature of 70 ◦ F?
through the wall section?
59. An electric water heater is set to heat the water

55. An older home has 2, 800 ft2 of total wall and
to 130 ◦ F. The water heater is in a basement with a
roof area with an R-value of 10 hr ◦ F ft2 / Btu, and temperature of 65 ◦ F. The water heater is insulated with
300 ft2 of single-pane windows with an R-value of an R-value of 5 hr ◦ F ft2 / Btu and has a total surface
0.9 hr ◦ F ft2 / Btu. Find the amount of Btu per hour area of 36 ft2 .
Estimate the monthly cost of the heat energy that is

lost through the walls of the water heater if the cost of 2.4. ENERGY FLOWS IN THE CLIMATE
electricity is $0.14 per kWh. SYSTEM
65. Use Stefan’s law to estimate the thermal power

60. Suppose you live in a two-story house with a total radiated by a cube of metallic zirconium, 12 centime-
roof, wall, and window area of 3000 square ft. The R-
ters on each side, heated to a temperature of 1,700 K.
value of the home (averaged over windows, walls, and
(This is still well below the melting point of zirconium,
roof) is 15 hr ◦ F ft2 / Btu.
a metal that is used in the cladding for fuel rods in
(a) Calculate the flow of heat energy through the
nuclear reactors.)
house in Btu per hour on a cold night when
the outside temperature is 20 ◦ F and the inside
temperature is 70 ◦ F. 66. An incandescent bulb is modeled as a sphere of
radius 3 cm having a temperature of 600 K. Approxi-
(b) If you throw a party, about how many people
need to be in the house to maintain equilibrium mately how much power does it emit as radiant energy
(at 70 ◦ F) if the heat is turned off? Assume that according to Stefan’s law?
a person standing around mingling radiates heat
energy at a rate of 100 watts. 67. As presented in Section 2.4.3, Earth’s albedo is
α = 0.3. Explain the meaning of this parameter in your
own words. What would it mean if α were to increase
61. Consider a house that has 3500 ft2 of roof,
or decrease?
wall, and window area and an average R-value of
16 hr ◦ F ft2 / Btu. Using an average outdoor nighttime
temperature of 20 ◦ F, calculate the total energy savings 68. Compute the temperature predictions of the naked
in Btu from setting back the thermostat from 70 ◦ F planet model for Venus (L = 2, 600 W/ m2 and α =
to 60 ◦ F for 8 hours each night during the month of 0.7) and for Mars (L = 600 W/ m2 and α = 0.15).
January. Make the (unreasonable) assumption that the
house temperature is 60 degrees the moment you set the 69. The real-life temperatures of Venus and Mars are
thermostat temperature to 60 degrees, and similarly for approximately 700 K and 240 K respectively. How do
when you set it to 70 in the morning. By how much these compare with your calculations in the previous
does the result change if the outside temperature is exercise?
40 ◦ F, instead of 20 ◦ F?
70. In this exercise we’ll estimate human body
62. Refer back to the carbon cycle depicted in temperature (at the skin). Body temperature can be
Figure 5 on page 72. The “human emissions” label modeled as an energy balance between the radiant en-
ergy emitted from the body based on skin temperature
depicts a flow of carbon into the atmosphere. Draw a
(as estimated by Stefan’s law) and a combination of the
stock-flow model with two stocks that shows this flow.
inflow of radiant heat energy from the surrounding air
and energy produced by the body (from the conversion
63. Find the residence time of the pollutant when the of the chemical energy in food).
lake in Exercise 51 is at equilibrium. (a) In Problem 1 on page 87 we estimated that
a human has a power output of 140 watts. If
64. Suppose that a “sandwich” is made of layers of the human body has a surface area of 1.5 m2 ,
two materials, one having R-value R1 and the other estimate the power output H of a person in
having R-value R2 . It is indeed true that the overall R- watts per square meter (the units that appear
value of the sandwich will always be R1 + R2 . Follow in Stefan’s law). We’re interested in estimating
through the argument used in solving Problem 9 on how much this power output will raise the body
page 104 to show this, replacing the numerical R- temperature of a naked human above the tem-
values in that example with the algebraic symbols R1 perature of the surrounding air, assuming that
and R2 . there is no heat loss by convection (airflow).
(b) Suppose that the room temperature is 20 ◦ C.
Find T1 , the equivalent of this temperature in
kelvins. Use Stefan’s law to estimate the inflow
of radiant energy, σ T14 watts per square meter,
to an object in the room (for example, a human that was twice the distance from the Sun as the Earth
being) from its surroundings. is. What would its solar constant be?
(c) The outflow of heat energy from our human
experimenter is given by σ T24 watts per square 76. Recall that the solar constant for Earth is 1350
meter, where T2 is skin temperature (in kelvins). watts per square meter. Jupiter is about five times as far
In equilibrium we therefore have away from the Sun as the Earth is. Estimate the solar
constant for Jupiter.
σ T14 + H = σ T24 ,
where σ T14 was calculated in part (b) and H was 77. Consider the naked planet climate model pre-
calculated in part (a). Estimate T2 . This is skin sented in Section 2.4.3. On the basis of this model,
temperature. what would happen to the Earth’s temperature (mea-
(d) Convert the result from (c) (which is in kelvins) sured in kelvins) if the planet were somehow moved
to degrees Celsius and degrees Fahrenheit. twice as far away from the Sun as it now is (with
all other parameters remaining the same) (also see
Exercise 75)?
71. Review Exercise 70 on the previous page. Sup-
pose that instead of sitting in a room that is 20 ◦ C, you
are outside on a cloudy autumn day and the outdoor 78. The newly-discovered planet Krypton travels in
temperature is 5 ◦ C. You are probably “bundled up” to an orbit between Earth and Venus, where the solar
protect yourself from the chilly weather. constant is 2000 watts per square meter. The albedo
(a) Thinking of your outer layer of clothing as your of Krypton is 0.2. What temperature does the “naked
“skin” (the surface exposed to the surrounding planet” model predict for the surface of Krypton?
air), describe how the act of wearing warm
clothing affects your “effective skin tempera- 79. Globally, fossil fuel energy use is about 5 ×
ture.” Explain this in relation to Exercise 70 on 1020 joules per year. Burning these fuels releases, on
the previous page. average, about 70 grams of CO2 for each megajoule of
(b) Suppose the Sun comes out from behind the energy produced.
clouds. Describe how this changes the inflow of Calculate the overall production of CO2 from fossil
radiant energy to your body? fuels, in kilograms per year.
80. The weight of the Earth’s atmosphere is about 6 ×

72. Find the predictions of the “glass ball” green-
1018 kg. Use the result from Exercise 79 to calculate
house model for Venus and Mars and compare these to
annual CO2 emissions as a percentage of the weight of
the known surface temperatures on Mars and Venus.
the atmosphere.
73. According to the nuclear winter scenario, much 81. In Problem 2 on page 112, we found that if
discussed in the 1980s, a nuclear war might throw up so the Earth had no atmosphere, the naked planet model
much dust that the atmosphere would become opaque predicts that Earth’s surface temperature would be
to both short- and long-wave radiation—so that all the about 255 K. The actual temperature is 288 K, about
incoming solar energy would be deposited in the upper 33 K higher (or 33 ◦ C higher). The greenhouse effect
atmosphere rather than on the Earth’s surface. Adapt is responsible for this difference. Assuming that CO2
the glass ball greenhouse model to this scenario. What accounts for 20% of the greenhouse effect, how much
does it predict would be the surface temperature of the of this temperature difference is atmospheric CO2
Earth during a nuclear winter? responsible for?
74. The solar constant L is approximately 1,350 watts 82. The annual change in CO2 concentration is
per square meter. Express the solar constant in Btu per currently about 0.5% per year. Using the results from
square foot per day. Exercise 81, estimate the number of years it would take
for CO2 emissions to produce a 1 ◦ C temperature rise?
75. The solar constant for a particular planet depends Make clear what assumptions you are using to derive
on its distance from the Sun. Suppose a planet existed your answer.
83. As mentioned following Remark 4 on page 114, 84. Refer back to Figure 18 on page 50, which shows
water vapor is also a greenhouse gas. In fact, water concentrations of carbon dioxide in the atmosphere.
vapor is responsible for about 70% of the greenhouse Notice that although the level rises over time, there is
effect. A warmer atmosphere holds more water vapor; also an annual oscillation. What is a possible cause for
to increase the water vapor in the atmosphere, we must this oscillation?
first warm the atmosphere. Based on your understand-
ing from this section of how changes in CO2 affect 85. Visit the Skeptical Science website [336] and pe-
Earth’s temperature, describe how changes in CO2 may ruse their list of “Most Used Climate Myths”. Choose
affect the strength of the greenhouse effect due to water a “climate myth” for further investigation. Summarize
vapor. (This is an example of a feedback loop, a concept both the “myth” and “what the science says” in your
that will be discussed much more fully in Section 3.3.) own words.
C HAPTER 3
Connecting
It is a truism that we live in a world that is highly connected. A hundred years ago, your great-
grandmother’s “social network” might have comprised a few dozen friends and acquaintances, all of
whom lived in the same community, together perhaps with a few distant family members with whom
communication was by letter: slow and not always reliable. Now, your friend group on Facebook
probably has hundreds of members and may already be spread across several continents. What
difference has this made to the world? Has it enriched our experience by opening us to a previously
unheard-of range of perspectives, or has it made our lives shallow at the same rate as it has made them
broad?
Mathematically, a network is made up of nodes, or vertices,
and links, or connections between these vertices (see Fig- G
ure 1). A “social network” provides a simple example, with

the nodes denoting people and the links denoting some kind
of social connection (like Facebook friendship in the example E
above). But social networks are only one manifestation of the
highly connected nature of our lives. Our economic system
has similarly grown in networked complexity. Rather than
most of our food being grown close to where we live, we
have gotten used to obtaining food from supermarkets that
are supplied by a production and transportation network that D
C
reaches all over the globe. The same goes for clothing and all F
the other basic necessities of life. Energy supply, too, is a vast
and complex network of tanker routes, pipelines, electrical
grid connections, and truck deliveries, to mention only a few. A B
The same goes for water and waste disposal.
The way we learn and transmit information has also be- Figure 1: A network.
come networked. Gone are the days when every American’s
understanding of current events came from the voice of Walter Cronkite, “the most trusted man in
America,” signing off with his signature phrase “And that’s the way it is.” Now we have access to a
multitude of conflicting voices with a click of a mouse. Even Wikipedia, regarded by many users as
the closest modern equivalent to a single authoritative source of knowledge, is edited anonymously
over the web by millions of contributors, who do not have to pass any kind of test; the underlying
belief is that “the truth will win out” in such an open system. (Whether this is in fact so can itself be
studied—under certain assumptions—by the mathematical tools of network theory.)
But beyond and behind all these human-conceived networks, and partly as a result of the attention
paid to them, we have grown more aware of the complexity of the natural networks that connect the
whole web of life on Earth. The Earth’s carbon cycle (Figure 5 on page 72) is a network of stocks

https://doi.org/10.1007/978-3-319-76660-7_3
128 CHAPTER 3. CONNECTING
and flows; a more precise study than that very schematic picture would produce a more complicated
network but would not change this basic fact. In a similar way, one can produce diagrams of the
Earth’s water cycle, its nitrogen cycle, its phosphorus cycle, and many other features of the ecosystem
in which we live. Even the key word “cycle,” which we just used repeatedly, is one that refers to a
certain type of network structure—a structure that, it has been argued, is critical in understanding the
difference between sustainable and unsustainable processes (see page 80).
In this chapter we’re going to introduce some of the mathematical vocabulary and ideas that can be
used to study networks. We’ll begin by thinking of a network statically, as a fixed, abstract structure
of nodes and connections (Section 3.1); then in later sections we will move on to thinking of networks
dynamically, as locations for change and evolution (like the stocks and flows of the previous chapter)
or of competition and power struggles (a theme we’ll return to in Chapter 6). If you are interested in
studying networks and their many applications in more depth than we have time for here, a wonderful
text is [100].
3.1. NETWORKS AND CONNECTIONS 129
3.1 Networks and Connections
Objectives
I understand the mathematical notion of a network, made up of nodes connected by

links.
I can recognize examples of networks in many aspects of daily life.
I can compute distances in a network, and can recognize when two nodes are
connected.
I understand what is meant by a component and a giant component.
I understand why triadic closure is to be expected in many networks, and why it
leads to the strength of weak ties in social networks.
I can describe the clustering coefficient as a measure of triadic closure.
I can describe the resilience measure β in qualitative terms, and compute it in simple
examples.
3.1.1 Basic Ideas of Network Theory

The objects we are going to study in this chapter are called networks. Networks come in many
different kinds, some more complicated than others; but the structure that they all have in common
is a set of objects called nodes and a set of connections called links that “join” two different nodes
together. For example, the network of Figure 1 on page 127 has seven nodes (represented by the
black dots and labeled A–G) and eleven links (represented by the blue lines that join different nodes
together).
A link is specified by listing (in either order) the two nodes that it joins; in more mathematical
language, a link is a two-element subset of the set of nodes. This gives us the following definition.
Definition 1
A network consists of two sets: a set N of nodes and a set L of links, where each link (that
is, each element of L) is a two-element subset of N.
This is the kind of definition that perhaps only a pure mathematician could love. We will seldom
need to refer to it; we’ll content ourselves with the intuitive understanding of networks that we obtain
by looking at pictures such as Figure 1. There is one aspect of these pictures that we should clear
up straightaway, though. In Figure 1, no links (as drawn on the page) crossed each other; to put it
another way, the only places where links could meet were nodes. That’s a nice property to have, and it
certainly helps clarify the pictorial representation, but it might not always happen: what, for instance,
are we to make of a network like the one diagrammed on the left-hand side of Figure 2 on the next
page?
The answer is this: since this network has only four nodes (labeled A, B, C, and D), and links are
completely specified by the nodes they join (here the links are AB, AC, AD, BC, BD, and CD), the
A B A B
D C D C
Figure 2: Two representations of the same network.
apparent “point of intersection” between AC and BD is not part of the network. You should think of
the links AC and BD as passing “through” or “over” each other without meeting. To put it another
way, the network diagrammed on the left-hand side of Figure 2 is exactly the same as the one on the
right-hand side, where none of the links are represented as intersecting except at nodes; the price to
pay, though, is that one of the links is represented by a curve rather than a straight line.
Definition 2
Two nodes in a network are neighbors if they are joined by a link. The degree of a node is
the number of neighbors that it has.
In the network represented by Figure 2, any two nodes are neighbors. The network of Figure 1 on
page 127 contains several pairs of nodes that are not neighbors: for example, A is a neighbor of B and
C but not of any other nodes; D is a neighbor of B, C, E, and G, but it is not a neighbor of A or F.
For many people, a familiar example of a network is the Facebook “friend network,” whose nodes
are Facebook users and where two nodes are linked if the corresponding users are “friends” on
Facebook. In this network, your neighbors are your friends; but not everyone on Facebook is your
friend, and not all of your friends are also friends of one another (see Figure 3 on the opposite page,
which shows the friendship links among one of the author’s Facebook friends, omitting their links to
the actual author). Most real-world networks are like the Facebook network in this respect—more like
Figure 1 than Figure 2.
Of course, you might want to know not only who are your friends, but also who are your friends-
of-friends, and who are their friends, and so on. This leads to the idea of a path:
Definition 3
A path in a network is a finite sequence of nodes, each of which (except the first) is a
neighbor of the preceding one. (That is, nodes A0 , . . . , An constitute a path if there is a link
between A0 and A1 , between A1 and A2 , between A2 and A3 , and so on.) The length of the
path is the number n of links that it contains (notice that this is one less than the number of
nodes).
For example, Figure 4 on the opposite page shows two paths from A to E in the network of Figure 1.
The red path has length 2, and the green path has length 5.
Figure 3: One author’s Facebook neighborhood.
If Figure 4 represented an actual transportation network, and each link took 1 hour to traverse, you
probably would follow the red path rather than the green one if you wanted to get from A to E. The red
path is the shortest route you can follow between those two points. That leads us to the next definition.
Definition 4
A network is connected if there is some path between any pair of nodes. The distance
between two nodes in a connected network is the length of the shortest path that joins them.
G
For instance, the distance between A and E is 2, as our picture
shows.
Problem 1: Compute the distance between nodes A and G in the E
network of Figure 4.
Solution: The answer is 3. There are several possible paths of

length 3 between A and G (two examples are A–C–E–G and A–
B–D–G). There are no possible paths of shorter length: by going D
one step from A you can reach B or C, and then by going one more C
F
step (for a total of two steps) you can reach D, E, or F, but not G.
Remark 1: The small-world theory is the idea that the distance A B

between any two randomly chosen people in the global friendship
Figure 4: Two paths from A to E in the network of
network will tend to be small; six is the number often quoted, Figure 1 on page 127.
as in the title of the 1993 movie Six Degrees of Separation. The
theory, and the number “six,” originates from the research of social
scientist Stanley Milgram and others [324] in the 1960s. Their
experiment involved asking randomly chosen people to get a letter to a specified “target” person
by forwarding it to someone they knew on a first-name basis and who they thought would be “closer”
to the target than they were; the recipient received the same instructions, so that in the end, if the
letter arrived at all, it would have passed along a path in the friendship network, whose length would
give an estimate for the distance between the source and target in that network. In Milgram’s initial
experiments he found that the median path length for those letters that arrived was 6; hence the term
“six degrees of separation.” But many letters failed to arrive at all (other attempts to replicate the
study failed because so few letters arrived that no conclusion could be drawn). Judith Kleinfeld [183]
criticized the methodology of the original Milgram study and suggested a different metaphor for the
global friendship network:
Rather than living in a “small, small world” we may live in a world that looks a lot like
a bowl of lumpy oatmeal, with many small worlds loosely connected and perhaps some
small worlds not connected at all.
What Kleinfeld is suggesting is a network that indeed contains “small worlds,” that is, tight groups
with close connections, but that has loose connections (or perhaps none at all) between the different
groups. We will return to this idea when we explore the strength of weak ties on pages 136 and
following, as well as in Section 3.2 on page 140) with the notion of homophily.
Definition 5
If a network is not connected, then starting from any given node A, there are some nodes
that are “reachable” from A by a path and others that are not. The set of all nodes that are
reachable from A is called the component of the network that contains A.
A disconnected network will be divided into several compo-

nents; all the nodes in a given component are reachable from one
another, and no two nodes in different components are reachable
from one another.
Remark 2: Many large, complex networks, although they are not
connected, have a single component that is much bigger than all
the others and in fact contains nearly all, or at least a significant
fraction of, the nodes. (For example, one would expect this to be
true for a friendship network like that in Figure 3 on the previous
page; see [100, pages 27–28] for some intuition about why this is
likely to be so.) This single, very large component is called the
giant component of the network. The assertion that nearly all the
Figure 5: A network with several components, important networks in our world possess such a giant component
colored differently. Blue might be considered a is another way of describing the highly connected nature of our
“giant component.”
world that we mentioned in the introduction. For good or ill,
lives and systems around the world are tied together, sometimes
in completely unexpected ways. The decline of India’s vulture
population provides an extraordinary example.
3.1.2 The Indian Vulture Crisis

Diclofenac is an anti-inflammatory drug belonging to the same class as the over-the-counter medicines
ibuprofen and naproxen. It is prescribed for fever, inflammation, or pain such as that caused by
arthritis. It can be used in a similar way in veterinary medicine—that is, to treat animals. Beginning
in the 1990s, diclofenac became widely used in India to treat the symptoms of pain or inflammation
associated with diseases of cattle. This treatment led to a steep rise in human deaths from rabies and
also from attacks by leopards. The unexpected links here are a striking example of the connected
nature of our world.
The key link in the chain is provided by the Indian vulture (Figure 6).
The majority of the population of India follow the Hindu religion, to which
the cow is a sacred animal. Cows may not be consumed by humans for
meat, and from ancient times, when a cow died, its body would be left
in the open to be fed upon by vultures. These scavengers constituted a
highly effective “burial system” for cow carcasses, and conversely, human
agricultural activity supported a large vulture population (estimated to be
of order 108 vultures in the 1980s).
This system began to break down in the 1990s and beyond as the vulture
population suddenly (and at first inexplicably) collapsed. There are now
only a few thousand of most vulture species alive in the wild in India, and
conservationists consider some species to be “functionally extinct.” The
Figure 6: Indian vulture.
culprit, it was eventually discovered [242, 141], was diclofenac. Diclofenac
is highly toxic to vultures, leading to death from kidney failure even at the
very small amounts that a vulture will ingest from feeding on a cow that has recently been treated
with a veterinary dose of the drug. This, it was found, was the cause of the catastrophic decline in the
Indian vulture population.
Such a huge change, though, has to have effects else- Leopards

where in the web of life. The number of cow carcasses
being left out to decay did not change much, but now there Dogs Vultures
were far fewer vultures to feed on them. Other scavengers
filled the gap: most notably, feral dogs, whose population
has greatly increased. Feral dogs carry rabies, and a greater
population of feral dogs leads to more interactions between Humans
these dogs and humans, and therefore to more rabies cases Cows
(there are more than 30,000 deaths from rabies in India
each year). What is more, the increased population of feral Figure 7: Predation links in the Indian vulture crisis. The
solid arrows show direct predation; the dashed arrow repre-
dogs also means an increase in the population of predators sents the effect of dogs on humans via the transmission of
higher up the food chain, such as leopards, which prey rabies.
on the dogs; an increase in the leopard population leads
to more interactions between leopards and humans, some-
times with fatal consequences for the humans involved. A
long and expensive list of consequences for an inappropriate choice of medication for cows!
We can diagram the relationships involved here by a network like Figure 7. Notice that the links
in this network have a direction represented by the arrows (vultures eat cows, but cows don’t eat
vultures).
Definition 6
A network in which each link is given a direction (such as the predation network of Figure 7)
is called a directed network.
Remark 3: The stock-flow systems that we examined in Chapter 2 provide other examples of directed
networks. The nodes of the networks are stocks, clouds, or spigots, and the directed links are the
various flows. As we will see later in this chapter, this relationship makes directed networks natural
settings in which to understand dynamic processes—those that change with time.
Figure 8: Aftermath of Smokey Spur fire, Wisconsin.
3.1.3 Measuring Neighborliness
Imagine the aftermath of a devastating forest fire (Figure 8). At first it looks as though everything
is dead—from the earth to the trees. We know that eventually the forest will regrow. But it will not
grow back right away into the same forest that it was before. After those burnt trees fall, they will
not immediately be replaced by new trees. Instead, new species will arrive in an orderly process
called succession: first annual plants, then perennial plants and grasses, then small shrubs, then larger
shrubs, then softwood trees, and finally hardwood trees. The stable web of species that marks the end
of the process of succession—not just the great hardwood trees themselves, but the whole complex of
plant and animal species that survive and provide resources (and prey!) for one another in the forest’s
understory—is called the climax community.
One of the first biologists to study succession in detail was Frederic Clements (1874–1945). He
interpreted the climax community as the “best” response to local microclimate: unlike the simple
weeds and grasses that are the first arrivals, the climax community, Clements posited, is complex,
balanced, densely interconnected, and (therefore) resistant to change. In the language of Definition 2
on page viii, the climax community is resilient. This idea that resilience is related to complex
interconnectivity has survived the rise, fall, and (partial) rise again of Clements’s ideas in his specific
field of plant ecology. For example, the forerunner of today’s Internet was a network of Defense
Department computers intended as a prototype for a communications system that would be resilient
in the face of the most extreme stress imaginable—a nuclear attack. The network’s designers realized
that one way to achieve such resilience was to make the network highly interconnected, with many
different paths that any given “packet” of data might traverse to reach its destination. This principle is
still at work in the Internet today, making it resilient at least to occasional hardware failures: messages
are automatically rerouted around a broken piece of equipment until it can be fixed.
This discussion makes it clear that resilience is, at least in part, related to network interconnectivity.
How are we to measure that, though? The mathematics of networks is a developing field, and new
ideas for measuring resilience are still being proposed. In the rest of this section we will take a look
at two ideas for measuring connectedness and resilience. The first of these, the clustering coefficient,
goes back to the early days of network theory, The second, which we will call the resilience measure,
has only very recently been developed [126]. Both, though, are attempts to quantify the “level of
interconnectedness” of a network. One important difference is that the clustering coefficient is a
local measure, attempting to quantify interconnectedness near a specific node, whereas the resilience
measure is a global quantity, attempting to assign a numerical value to the resilience of the entire
network.
Triadic Closure and the Clustering Coefficient

Looking at pictures of real-world networks like the one in Figure 3 on page 131 shows us that
such networks often consist of “clusters,” tightly connected groups that are much less strongly
connected to one another. (In the actual network of Figure 3, one such cluster represents the author’s
mathematical colleagues, another his rock-climbing friends, yet another the group of friends who are
into sustainability, faith, and justice issues, and so on.) A simple way to measure such clustering is to
think about the notion of triadic closure.
Definition 7
A node in a network exhibits triadic closure if every two neighbors of that node are also
connected to each other.
In other words, the node representing you in your social network has triadic closure if all of your
friends are also friends of one another. This example illustrates that there are often pressures that push
a network toward triadic closure: your friends are more likely to meet one another (at your parties
perhaps?) and so to become friends themselves. But it also illustrates that complete triadic closure is
a rather rare thing. That encourages us to measure how close a node gets to triadic closure.
Definition 8
The clustering coefficient of a node is the number of actual links among the node’s
neighbors divided by the number of possible links.
Problem 2: Consider the network of Figure 1 on page 127. Compute the clustering coefficient of node
G.
G
Solution: In Figure 9 we have redrawn Figure 1 to illustrate how the
calculation of the clustering coefficient goes. The links involving node
E
G are shown in red, so that we can see that G’s neighbors are nodes E,
D, and F. Among those nodes, there is one actual link, between E and
D, which we have shown in green. There are two other possible links—
links that could exist, but are not actually part of the network—among
these three nodes, namely between E and F and between D and F. We’ve D
indicated these possibilities by the green dashed lines. Since there is one C
F
actual link among the three possible links between G’s neighbors, the
clustering coefficient of G is 13 .
A B
Remark 4: We don’t actually need to count all the possible links among Figure 9: Figure 1 on page 127, redrawn.
neighbors in the way we did in this example. If d is the number of
neighbors of a node (that is, its degree), then the number of possible
links among those neighbors is given by the formula 12 d(d − 1). In our example G has three neighbors,
so d = 3 and
1 3×2
d(d − 1) = = 3,
2 2
corresponding to our hand count of the number of possible links among G’s neighboring nodes. Using
the formula saves a lot of time, especially if d is large.
C
F
D A
Figure 10: Weak ties link different clusters in a network.
Problem 3: In the same network compute the clustering coefficient of node C.
Solution: C has four neighbors, A, B, D, and E, among which there are three links, AB, BD, and DE.
With d = 4 the formula gives us
1 4×3
d(d − 1) = =6
2 2
possible links among C’s neighbors. Thus the clustering coefficient is
Number of actual links 3 1
= = .
Number of possible links 6 2
Strong and Weak Ties

To proceed further with our understanding of resilience, let’s think about another feature of social
networks. Some of our friendships are stronger than others. Some of your Facebook “friends” may
be people you interact with every day; others may be contacts from an earlier time in your life who
you have not sent a message to, let alone seen, in decades. If we wanted, we could imagine assigning
each link in a network a numerical measure of strength, ranging from 1 (or 100%) for “as strong as
can possibly be” to 0 (or 0%) for “not there at all.” But we can get quite a way with the simple idea of
dividing links into “strong” and “weak.”
In 1974, social scientist Mark Granovetter published a landmark study called “Getting a Job” [137].
In this study, he interviewed 282 men in the U.S. who had recently obtained a new job to find out
how they discovered the opportunity that they had filled. He found, as earlier studies had also done,
that many people learned this information through personal connections—through their friendship
network. The new observation that he made, however, was that most interviewees described the
connections through whom they had learned about their new job openings as “acquaintances” rather
than “close friends.” In our language above, the weak friendship links were actually better sources
of new-job information than the strong ones. This observation has become known as “the strength of
weak ties.” But why should it be true?
The answer depends on the “lumpiness” of social networks that Kleinfeld describes in the quotation
on page 132. It comes about like this: Suppose that you (person A) have two friends (B and C). The
stronger your friendship connections are with B and C, the more likely it is that they will meet, that
you will want to introduce them to each other, and so on; in other words, the more likely it is that
the triadic closure property will lead to them becoming friends themselves. Strong connections lead
to triadic closure, which leads to tightly knit little groups—leaving the weak connections to perform
the vital function of bridges1 between these tight groups. Thus a social network—and this is true for
many other kinds of network also—ends up looking something like the network of Figure 10, where
strong ties are symbolized by thicker lines and weak ties by thinner ones. As you can see, the network
contains three distinct “clusters,” and the “bridges” between the different “clusters” in the network are
all weak ties.
Now suppose that you are person A in the figure and you are looking for a job. Which connection in
the network is most likely to open up a new and different possibility for you? You are tight with your
friends B, C, D, and E—but exactly for that reason, you already know most of the things they know,
and the chances they have access to something completely new to you are low. It’s different with your
connection to F. They may be someone you don’t see often—but they have access to a completely
different cluster of information from that shared by you and your friends. If you’re looking for a job,
it is likely that you’ll remember your old acquaintance F and call them up, and there’s a much better
chance that they can open a new door for you.
The Resilience Measure

Having understood the key concepts of triadic closure and link strength, we are now able to come to
grips with the measure of resilience proposed by Gao and coauthors [126]. They identify a universal
parameter, which they call β , which (under certain assumptions) will predict the resilience of complex
directed networks having links of varying strengths. Their claim, which is supported by a range of
examples, is that this single quantity β provides a good summary of the resilience of all the complex
behavior that may be associated with the entire network.
Critical Thinking
How likely do you think it is that a single number can measure the slightly vague concept
of resilience? What kind of assumptions might be used to reach such a conclusion? Take a
look at the paper [126] if you have access to the journal Nature (you may find it hard going).
Are you more convinced now, or less?
The resilience measure β is given by an equation
β = A + HS,
where we have to define the three quantities A, H, and S. The first one (A) is the easiest to define. It
measures the average strength of the connections in the network. In a network of the simplest kind
we have been discussing, where links are either “on” or “off,” A is simply the average degree of the
nodes—the sum of the degrees of all the nodes in the network, divided by the number of nodes.2
The measure A therefore gives expression to our natural understanding that the more complex the
interlinking in the network, the more resilient it will be. (Note that in Figure 11 on the next page, the
authors of [126] use the notation hsi for what we have called A.)
1 There is a formal, mathematical definition of this idea, but we won’t get into it; the intuitive picture is clear enough.
2 Ina more complex situation in which different links have different strengths, we take the average total strength of the
nodes instead.
Figure 11: Explanation of the three components of the resilience measure β : from Gao et al. [126].
The other two quantities, H and S, also have specific mathematical definitions, but rather than going
into those in detail, let’s simply say what it is they measure. The quantity H measures what the authors
call the heterogeneity of the network. If H is small, the nodes tend to be of the same or similar total
degrees (or total strengths). If H is large, the nodes are of widely varying total strengths. You might
wonder whether heterogeneity makes a positive or negative contribution to resilience. The answer
depends on the third term, S, the symmetry measure. In a directed network (one in which the links
have a direction or “arrowhead”; see Definition 6 on page 133) the notion of “total degree” or “total
strength” of a node splits into two: the “incoming” degree or strength (counting incoming links only)
and the “outgoing” degree or strength (counting outgoing links only). The quantity S measures the
extent to which “incoming total strength” and “outgoing total strength” match up with each other. So
for an ordinary “undirected” network, incoming and outgoing strengths match perfectly, and S = 1.
This means that H makes a positive contribution to β : the more heterogeneous, the more resilient. But
for a highly asymmetric network, S may become negative. That means that H now makes a negative
contribution: a less heterogeneous structure (other things being equal) will be more resilient. This
two-way influence of heterogeneity on resilience (depending on symmetry) is one of the surprising
results of the analysis in [126].
Remark 5: Here is the actual process for computing the resilience measure β . It is computed directly
from the network: the division into three components A, H, and S results from the mathematical
analysis.
(a) Make a table with four columns and as many rows as there are nodes in the network you are
studying.
(b) Put the names or labels of the nodes in column 1.
(c) In each row, column 2, list the total in-strength of the corresponding node (the sum of the
strengths of the links pointing inward to that node). If the links don’t have specific strengths,
then list the number of links pointing inward to the node (sometimes called the in-degree).
(d) In column 3, list the total out-strength of the corresponding node (the sum of the strengths of the
links pointing outward from that node). If the network is undirected, like many of our examples,
this will be the same as column 2.
(e) Finally, in column 4, list the product of the entries in columns 2 and 3 of the same row.
Once this is done, take the sums of the entries in columns 2, 3, and 4. The sums of columns 2 and 3
should be identical (even if the network is not symmetric—can you see why?). The resilience measure
β is given by the formula
Total of column 4
β= .
Total of column 2 (or 3)
By the way, the “average strength” A that we mentioned above is just the total of column 2 (or 3)
divided by the total number of nodes.
Table 1: Calculation of β in Figure 1 on page 127
Node In-strength Out-strength Product

A 2 2 4
B 4 4 16
C 4 4 16
D 4 4 16
E 3 3 9
F 2 2 4
G 3 3 9
Totals 22 22 74
Example 1: Let’s go back to our old friend, the network of Figure 1 on page 127. We think of this as
an undirected network, without specific strengths. The table entries in the second and third columns
are just the degrees of the nodes. What we get is shown in Figure 3.1.3. The average strength A is
equal to 22/7 ≈ 3.1, and the resilience coefficient β is 74/22 ≈ 3.4. Notice that β is only a little bit
greater than A—which is to say, the heterogeneity H makes a relatively small contribution to β . The
authors of [126] suggest that a β value above 6 or so is needed to qualify a network as “resilient.”
Summary of Ideas: Networks and Connections
• A network is a system of nodes, which may or may not be joined by links.

• Networks model aspects of many social, ecological, and human-designed systems.
• A path is a sequence of adjacent links. Two nodes are connected if they have a path
in common; in that case their distance is the length of the shortest path joining them.
• A component of a network consists of all the nodes that are connected to a given
node.
• In more complex examples, the links may have specified directions (directed
network), or varying strengths.
• Triadic closure is the propensity for neighbors of one node also to be neighbors of
one another. The extent of triadic closure is measured by the clustering coefficient.
• The resilience measure β for a network can be computed in terms of the total
degrees, or total strengths, of its nodes, It is made up of average node strength,
symmetry, and heterogeneity measures.
3.2 Networks and Behavior
Objectives
I can describe the information-deficit model of science communication, and can

point out some of its limitations.
I can use networks to model communication among individuals or social groups.
I can describe the bystander effect and use a simple network model to partially
explain it in terms of an information cascade.
I am ready to move from bystander to ally if I find myself in a situation that demands
this.
I understand the notion of homophily in networks, and appreciate how homophily in
social networks can impede the communication of sustainability messages.
I can explain how homophily may arise from the pressure toward structural balance
in networks.
3.2.1 The Information-Deficit Model

In 1998, a British doctor, Andrew Wakefield, claimed to have found evidence connecting the MMR
vaccine, which is usually administered to children around the age of one year, with the later
development of autism spectrum disorders. The vaccine causes children to develop immunity to
measles, mumps, and rubella, diseases that formerly killed significant numbers of children, and which
had been thought to be almost eradicated by vaccination. But the suggestion of a link with autism
set off widespread alarm among parents: vaccination rates (in both the U.K. and the U.S.) declined
sharply, and outbreaks of measles and mumps began to occur again, sometimes leading to deaths.
Meanwhile, Wakefield’s work was exposed as fraudulent, and large-scale studies looking at thousands
of children showed no evidence of any correlation between the MMR vaccine and autism. Public
health experts and medical associations continued to strongly endorse vaccination.
In early 2015, nearly a hundred people across the U.S. and Mexico were affected by an outbreak
of measles that began at Disneyland. Such an outbreak would have been next to impossible if
vaccination rates had remained as high as they had been ten years previously, but California health
authorities reported that “34 of the 42 California patients whose vaccination status is known were
unvaccinated for measles.” The next month, a YouGov poll [226] found that only 57% of all Americans
support requiring children to receive vaccinations against childhood diseases. Support for mandatory
vaccination was far less among younger people (only 42% of under-30s), and the discredited theory
that vaccines can cause autism appeared to be the reason why.
Doctors and public health officials were troubled and frustrated by the persistence of antivaccine
beliefs. Shortly after the Disneyland outbreak, Dr. Robert Wah, past president of the American
Medical Association, released a message to his fellow physicians [339]. It reads in part
As physicians, we know the importance of keeping each individual patient as healthy as

possible. With the issue of vaccinations getting increased attention in political discourse,
now is a good time to review the AMA’s policies so you can talk to your patients using
scientific facts. . .
3.2. NETWORKS AND BEHAVIOR 141
We know that vaccinations are safe and effective. We know their benefits far outweigh any
risks. And we know that we must encourage our patients to listen to the science and facts
behind this issue.
Dr Wah’s message is a striking example of the information-deficit model
of science communication. According to this model, there is a group of
experts who know the “scientific facts” about some matter at issue (here,
the effectiveness and safety of vaccines). And there is a much larger group,
the general public (“our patients,” in Dr. Wah’s letter) who, sadly, do not
know these facts—if they did, they would surely behave differently. The key
problem, as this model sees it, is that the public suffers from an information
deficit, and this deficit can be corrected by “experts” who supply the missing
information. If “unscientific” behavior (e.g., vaccine refusal) persists, it
can only be because the expert information was not communicated clearly,
convincingly, or comprehensively enough.
The information deficit model has an obvious, “commonsense” appeal,
because it rests on a basic truth: it is hard to make wise decisions about some-
thing you just don’t understand. What’s more, the model is very attractive to
many professionals who work in the field of sustainability. “If only” (they
might say) “people understood the facts about global energy use, or about
human-caused climate change, or about aquifer depletion, or (you can insert
your own favorite issue here), then they would surely take action to avoid the Figure 12: Vaccination.
risk.” And so, every year, experts redouble their efforts to communicate to
us the facts about these and other questions. With what result? Take climate
change as an example. In the decade between 2006 and 2016, Americans received thousands of pieces
of professionally sourced information about climate change, ranging from the highly technical (e.g.,
the 2007 and 2015 Assessment Reports of the IPCC) to the popular (e.g., the movie An Inconvenient
Truth, and the PBS series Earth: The Operators’ Manual [23]), all reflecting the expert consensus that
human activity is causing Earth’s climate to grow significantly warmer (see Section 2.4 for more about
this). Yet according to a Pew Research poll [61], the overall percentage of Americans agreeing with
the statement “The Earth is growing warmer, mostly because of human activity” changed hardly at all
over those ten years (then and now, about 50%). Expertly sourced information is clearly important;
but contrary to the predictions of the information-deficit model, it is not the only thing that matters.
If it was, the deluge of information on climate change that Americans have received over the last ten
years would surely have moved the needle, if only a little!
Critical Thinking
To what extent do you think that this book itself depends on an information-deficit model
of science communication? Do we (the authors) seem more concerned with giving you
specific information, or with giving you the tools to analyze information, wherever it may
come from?
Appealing as it is, the information-deficit model does not accurately describe how people form
their opinions about scientific findings, or even about what they allow to count as “scientific.” Expert
testimony is only one factor in a process that also involves one’s own moral foundations [147] as well
as the understandings of one’s friends and neighbors. This is a situation in which research in social
science (for example [175]) is needed to help workers in physical science communicate what they
know—knowledge that may be vital to human flourishing in the future.
In this section, we are going to focus on just one aspect of such social science, the aspect suggested
by the phrase “friends and neighbors” above. We don’t come to our beliefs alone; we come to them
as part of a social network. What can we learn from network theory, then, about the transmission and
sharing of beliefs, and about our sadly evident tendency to polarize into groups for which holding the
“right” belief is an important marker of inclusion?
3.2.2 The Bystander Effect

Have you ever saved someone’s life in an emergency? Rescued someone from drowning, performed
CPR or the Heimlich maneuver, or something like that? If you have done so, or if you have been
rescued in a situation like that, try to remember some details. How many other people were present—
not just “in the same room,” but actively aware of and relating to the situation—when the intervention
took place?
Definition 1
The bystander effect is the well-documented observation that “the more bystanders witness
an emergency, the less likely it is that the victim will receive help from any of them.”
The bystander effect seems quite paradoxical: surely having another potential rescuer on the scene
should make it more likely that I get rescued, not less? In this section we will see how a simple model
of decision-making in networks can help explain the bystander effect, and we’ll use that model to
suggest some specific advice that could help you if you’re ever unfortunate enough to need help from
passersby in an emergency.
In [89], social psychologists John Darley and Bibb Latané devised an experiment to measure the
strength of the bystander effect. In their experiment, college students were recruited to participate in
a “discussion group” about personal issues. The students were told that to maintain anonymity, all the
members of their “group” would be in separate rooms and that they would communicate only through
microphones and earpieces. In fact, there was no discussion group; the other voices were prerecorded.
But the students who participated did not know this, and crucially, different participants were led to
believe that their discussion groups were of different sizes. Some thought that they were having a one-
on-one conversation with a single anonymous partner; some thought they were in a group of three;
some of four, five, or six.
One of the prerecorded voices was that of a student who was prone to seizures. The experimenters
write:
In the discussion, the [student prone to seizures] spoke first, saying that he found it
difficult to get adjusted to New York City and to his studies. Very hesitantly, and with
obvious embarrassment, he mentioned that he was prone to seizures, particularly when
studying hard or taking exams. Then the other people, including the real subject, took
their turns and discussed similar problems (minus, of course, the proneness to seizures).
The real subject talked last in the series, after the last prerecorded voice was played.
The experimenters arranged that a couple of minutes into the next interaction, it would appear to
the experimental subject that the first student was indeed having a seizure. He choked, called out for
help, and expressed fear for his life. What the experiment was actually measuring was how long it
would take for the subject to ignore the experimental protocol, leave the room, and look for someone
who could assist the person apparently in need. Would the subject run quickly for help, or sit tight,
thinking that “someone else” would take care of the problem?
The outcomes of Darley and Latané’s experiment were striking. In Table 2, the left-hand column
(“apparent group size”) represents the total number of people that the subject believed were
participating in the group. In other words, an “apparent group size” of 2 means that the subject
believed that they and the victim were the only participants; an “apparent group size” of 3 means
that the subject believed that they were participating with two others, one of which was the victim;
and so on. The figure in the right-hand column is the percentage of subjects who sought help within
a six-minute time period. The results show that the subject’s willingness to intervene dropped sharply
as the apparent group size increased.
Table 2: Result of Bystander Experiment
Apparent Group Size Percentage Response

2 85
3 62
6 31
Various explanations have been proposed for the bystander effect. When one reads about notorious
cases, such as the death of Wang Xue in Guangdong province in October 2011, one might be tempted
to say that the bystanders in these situations were pathologically indifferent. But in the Darley–Latané
study and others conducted since, no significant psychological difference was found between those
who intervened and those who did not. Other explanations speak of the “diffusion of responsibility” in
a group situation—it may be clear that “action must be taken,” but nobody knows whose responsibility
it is to implement this “must.” Other explanations again focus on how we evaluate the information
provided by the actions other people are already taking; this leads us to the concept of an information
cascade.
Definition 2
An information cascade occurs when someone’s response to a situation depends more on

their perception of other people’s response to the situation than to the evidence that the
situation itself presents.
Example 1: As a simple example of an information cascade, imagine that you are fond of Mexican
food and that you arrive in a new town one evening where there are two Mexican restaurants on the
same block. Restaurant A has a noisy line out the door; Restaurant B seems quiet and at most half-
full. You have no other knowledge about which restaurant is better. Which will you choose? Many
people will reason as follows: the locals all seem to favor Restaurant A—that must be why it’s so
busy—and I expect that means it is better. So I should join the line for Restaurant A as well, even if I
have to wait longer. The trouble with this reasoning is that the people already waiting for Restaurant
A may be there because it is a better restaurant—but they may also be there because it happened to be
a bit busier that night and they followed your line of reasoning themselves! And if the next arrivals
reason as you did, the line for Restaurant A will keep growing longer and longer. That is the reason
for the term cascade—at some point the self-reinforcing effect of following other people’s responses
overwhelms any actual difference in the quality of the two restaurants.
Let’s consider how this principle might be applied to the bystander effect. A group of bystanders
(B1 , . . . , Bn ) are faced with an apparent emergency—a victim (V ) who is in trouble in some way,
let’s imagine someone suffering a medical emergency. The bystanders and the victim constitute an
impromptu “social network,” and within that network each bystander is asking the same question: “Is
Figure 13: The bystander effect.
this a real emergency? Do I need to intervene?” To answer that question, each bystander consults all
the information available to them, and, critically, this information includes their perceptions of other
bystanders’ responses as well as their perception of the victim’s own situation.
Oversimplifying things a great deal, let us model each bystander’s response as follows. Each
bystander Bi has two choices: to intervene or not to intervene. We’ll give Bi an “intervention score”
of 1 if they intervene and 0 if they don’t. In deciding whether to intervene, Bi tries to figure out “how
sure” s/he is that an emergency is taking place. To do that, Bi averages the intervention scores of all
their neighbors in the social network (including the victim with an intervention score of 1, meaning
that they seem to need an intervention). If the resulting average is greater than some threshold ti set
by Bi ’s personal ethical code, he or she will intervene.
B4 , 0.7 B3 , 0.6 B4 , 0.7 B3 , 0.6
V V
B1 , 0.4 B2 , 0.5 B1 , 0.4 B2 , 0.5

(a) In a “tight” network, no one intervenes (b) Removing one connection leads to a “looser”
because no one else is intervening. network in which intervention takes place.
Figure 14: The bystander effect.
Consider the examples in Figure 14. The victim is represented by the red node, and there are four
bystanders B1 , . . . , B4 whose thresholds for intervention are respectively 0.4, 0.5, 0.6, and 0.7. In the
left-hand example, each bystander is connected to (that is, can see) two other bystanders as well as
the victim. Starting from a state in which there is no intervention, each bystander averages the signals
that they are receiving: a 1 from the victim, and two 0’s from the two other bystanders whom they can
see, meaning an average of 13 ≈ 0.33. Since this is less than any bystander’s threshold for intervention,
none will take action; the victim’s calls for help will go unheeded.
Things become very different if we imagine that B1 and B4 cannot see one another (everything else
being the same as in the first example). This apparently minor change makes an enormous difference
to the outcome. Consider bystander B1 first. They receive two signals: a 1 from the victim and a 0
from B2 , averaging to 0.5. Since this is greater than their threshold of 0.4, B1 will intervene.
Now consider B2 . With B1 intervening, they are receiving two “1” signals (from V and B1 ) and
one “0” (from B3 ). The average of these is 23 ≈ 0.67, greater than their threshold of 0.5. So B2 will
intervene. Once B2 intervenes, the same reasoning will lead B3 to intervene as well. Finally, even B4
will be receiving two “1” signals and—despite his relatively high threshold of 0.7—will intervene as
well. The victim is saved! (So long, that is, as B1 –B4 don’t trip over each other in their haste to help.)
Remark 1: We see from the example above that a reduction (even a temporary one) in the strength of
the bystanders’ connections with one another can increase the victim’s chances of receiving aid. This
is the network-theoretic reason behind an important piece of advice that you might want to remember
if you are ever unlucky enough to be the victim in an emergency situation: Don’t just call out “Help!”
to everyone around. Single out a particular person—it’s often suggested that you identify someone
you don’t know by pointing to them and describing their clothing—and make it clear to that person
that s/he is the one you are counting on to intervene. “You, sir—yes, you, in the green hoodie—please,
get out your phone and call 911. I need an ambulance. You’re my only hope.” The only way that Green
Hoodie Guy will know is if you tell him.
B4 , 0.7 B3 , 0.6
B1 , 0.4 B2 , 0.5
Figure 15: Restructuring the social network by appealing to a specific bystander.
The way this works is by restructuring the bystanders’ social network until it looks more like
Figure 15. You see that Green Hoodie Guy turns out to be B4 , the one with the highest threshold for
intervention. Nevertheless, because he has been singled out, his attention is now redirected entirely
to the victim. This is expressed by the single-headed arrows in Figure 15: though bystanders B1 and
B3 are looking at him, he is no longer looking to them, and the only signal he is receiving is from
the victim. This signal value of 1 overrides even B4 ’s relatively high threshold for action, and he
intervenes. You can check that B4 ’s intervention will be followed up by B1 , B3 , and B2 in another
classic “information cascade.”
Remark 2: We have focused on the mathematical and network-theoretic aspects of the bystander
effect because that is what this book is about. We shouldn’t leave the subject, though, without pointing
out that there are ways to prepare yourself to be ready to respond when you meet someone in need—
to make the move “from bystander to ally,” as Holocaust scholar and ethicist David Gushee has put
it [143]. It helps simply to be aware of the bystander effect. It helps to be observant of situations
around you, to pay mindful attention. It helps to have specific skills that you are confident in and can
apply (like up-to-date CPR certification, for example). Many schools and colleges offer training in
bystander intervention to any students willing to sign up for it: do you know whether yours does, and,
if it does, have you taken the class? And yes, it helps if you are a person of compassion and courage,
and a person who has learned through deep connection and relationships how we all need some help
now and again. Those qualities will also help you avoid being a “bystander” to the Earth itself, which
may yet become a victim of our collective need for “more.” See Section 6.4.
Our analysis of the bystander effect is an example of the use of network theory to study information
transmission and decision-making processes. We’ll return to this topic in Sections 6.2 and 6.3.
3.2.3 Homophily and Structural Balance

A few pages ago (page 141), we cited evidence from Pew Research suggesting that the average
proportion of Americans agreeing with the statement “The Earth is growing warmer, mostly because
of human activity” remained at roughly 50 percent throughout the decade 2006–2016. Yet all our
readers know that this average figure conceals deep divisions between different groups of Americans.
According to the Pew data [61], the percentage agreement with the statement varied according to
political affiliation as shown in Figure 16.
Liberal Democrat 66
Moderate Democrat 49
Moderate Republican 41
Conservative Republican 21
0 10 20 30 40 50 60 70 80 90 100
Percentage Agreement
Figure 16: Percentage agreeing with “The Earth is growing warmer, mostly because of human activity” by self-described
political affiliation [61].
We are so used to this “political” division over the basic data about climate change that we may
forget how surprising it is. What could account for so wide a variation in acceptance of something that
the information-deficit model would describe as a “scientific fact”? One possible answer can be ruled
out straightaway. It may be tempting (especially for liberals who also believe in the information-deficit
model) to attribute this variation simply to scientific ignorance: “conservatives are less well educated
scientifically. . . when they do get educated, they will become more liberal.” But the Pew survey also
included a test of general scientific literacy. The results of this test were almost uncorrelated with the
response to the question on global warming (in other words, someone’s scientific literacy seemed to
have almost no effect on their beliefs about global warming). The only effect the survey found—small,
but measurable—was not comfortable news for believers in the information-deficit model. The data
showed that a higher level of scientific literacy seemed to make people slightly more entrenched in
whatever position their political affiliation would predict.
To understand a little bit about how this might come about, let’s go back to the previous section.
We observed the tendency of social networks to aggregate into tightly connected “clusters” linked
by “bridges.” Look at Figure 10 on page 136 again, and now recall our discussion of the bystander
effect. Imagine person A trying to figure out what they think about human-caused global warming.
Climate experts perhaps play the role of the victim V in the bystander effect model, shouting loudly
“Emergency! Action needed now!” But as we saw in the bystander effect model, person A is inevitably
going to calibrate their response not only by the signal they are receiving from V but also by the
“signals”—the level of response—they perceive from others in their social network, especially the
closely clustered friends B, C, D, and E. The bystander effect model suggests that the whole social
cluster is likely to respond or not respond together to the climate news. Some clusters will deny the
statement, “The Earth is growing warmer, mostly because of human activity,” and they will do so as
a group—perhaps matching the “conservative Republican” group in the Pew survey. Other clusters,
perhaps matching the “liberal Democrat” group in the survey, will agree with the statement, and they
too will do so as a group. What’s more, disagreement on this question may act as a source of tension
in relationships, in the end reducing the number of positive connections between distinct clusters.
This is a problem on multiple levels—not only does it reduce the chance of people’s being exposed
to information that might lead them to change their minds, it also reduces the number of “weak ties,”
which, as Granovetter’s study showed (page 136), are particularly effective for positive social purposes
like helping people find a new job.
The phenomenon we have been describing has a special name:
Definition 3
Homophily refers to the tendency for strongly clustered members of a social network to
have similar beliefs or concerns.
Table 3: Homophily on Global Warming Concern
Group Percentage Response

Level of Concern None Low Medium High
Overall sample 2% 10% 36% 53%
Concern among Friends

Friends’ Concern
Own Concern None Low Medium High
None 5% 11% 37% 47%
Low 2% 12% 36% 50%
Medium 1% 8% 34% 57%
High 1% 5% 27% 67%
Remark 3: It is a testable question whether a given social network exhibits homophily regarding
some specific belief or concern (such as, say, concern about global warming). Itkonen [168] tested
this specific question (homophily regarding global warming) on university students in Finland, where
strictly political divisions are not as rancorous as they presently are in the United States. To do this, the
researchers programmed a Facebook survey application that (with participants’ permission) asked a
series of questions about environmental and other issues (including measures of concern about global
warming) and also collected a list of the survey user’s Facebook friends. The survey was initially
sent to students and faculty members at the University of Helsinki, but the initial recipients were also
encouraged to invite their friends to participate, and the Finnish public broadcasting company ran a
story about the survey with a link attached. Participation was limited to Finnish nationals. In the end,
the survey involved 5,205 participants.
Overall, 53% of participants expressed “very serious” concern about global warming. But among
those who stated they had “very serious” concern, 67% of their friends also had “very serious”
concern; and among those who were “not at all” concerned, only 47% of friends were had “very
serious” concern. More detailed information is contained in Table 3 (adapted from Table 1 of [168]).
For example, the overall percentage of the population reporting “no concern” about global warming
was 2%, but among those who themselves reported “no concern,” 5% of their friends reported
“no concern” also. These results indicate a certain degree of homophily among the respondents:
respondents tended to “cluster” with friends who were like-minded on the issue of global warming.
Think about it. . .
One could explain a finding such as that of the Finnish study in two ways: Maybe like-
minded people tend to cluster together, or maybe people who are already clustered together
(that is, already friends) tend to adopt the same opinions. Which explanation do you find
more plausible? Or can you think of an alternative to both?
Harary’s Theorem
The Finnish study went on to analyze this homophily using more sophisticated measures like the
clustering coefficient of Definition 8 on page 135. It is clear that the homophily they found, though
real, does not reflect the high level of polarization that sadly exists in the U.S. at the time of writing. To
understand this higher degree of polarization it may be helpful to turn to an analysis by Harary [152].
Harary imagined a social network in which links are divided into two kinds: positive, where the two
people who are linked reinforce one another’s beliefs, and negative, where they challenge one another.
We might, with a bit of exaggeration, call them “friend” and “foe” relationships. Remember now the
triadic closure property (Definition 7 on page 135), which tends to lead to the formation of triangles in
a social network. With his notion of two kinds of relationships, positive and negative, Harary arrived
at a division of these triangles into two types, which he called “balanced” and “unbalanced.” See
Figure 17.
C C C C
+ + − − + − − −
A + B A + B A + B A − B
(a) Balanced triangles. (b) Unbalanced triangles.
Figure 17: Harary’s triangles.
“Balanced” triangles, for Harary, are those that are “comfortable” for the participants—those that
exhibit little social strain. On the left-hand side of Figure 17 we see two types of balanced triangles:
either A, B, and C all have positive relationships (clearly comfortable), or A and B have a positive
relationship and both agree that C is their “common enemy.” On the right-hand side of Figure 17 we
see two types of unbalanced triangles. If A has positive relationships with B and C, but they have a
negative relationship with one another, then A will experience cognitive dissonance: should she try to
reconcile B and C? Or should she become the enemy of one to try to preserve the others’ friendship?
Similarly, if all three relationships are negative, there will be a perennial temptation for two of A, B,
and C to “gang up” against the third. Harary posited that a social network of this kind would try to
remove unbalanced triangles, “either by changing the sign of some edges or by expelling some nodes.”
Of course, Harary’s idea—that social networks tend to resist the formation of unbalanced
triangles—would be of little consequence if the network were so sparse that there were few or no
triangles at all. In his initial work, therefore, Harary considered the opposite extreme: a complete
network, that is, one that exhibits 100 percent triadic closure—any triangle that could exist actually
does exist. A complete network is called balanced if all its edges are either positive or negative and
all its triangles are balanced.
How could such a network exist? One way would be for the nodes to be divided into two factions—
we might as well call them the “red” and “blue” factions—so that every two nodes of the same color
have a positive relationship, and every two nodes of a different color have a negative relationship.
Harary gave a mathematical proof that this is in fact the only way that a balanced, connected complete
network can be formed. In other words, he proved the following theorem.
Theorem 1: Harary’s Theorem
The nodes of a balanced connected complete graph are necessarily divided into two factions,
in such a way that all relationships between nodes of the same faction are positive, and all
relationships between nodes of different factions are negative.
The pressure for balance therefore leads to the most extreme degree of homophily imaginable.
Think about it. . .
Dan Kahan [175] has written: “Nothing any ordinary member of the public personally
believes about the existence, causes, or likely consequences of global warming will affect
the risk that climate change poses to her, or to anyone or anything she cares about. Nothing
she does as a consumer, as a voter, as a contributor to political campaigns and causes,
or as a participant in public conversation will be of sufficient consequence on its own to
have any impact. However, if she forms the wrong position on climate change relative to
the one [held by] people with whom she has a close affinity—and on whose high regard
and support she depends on in myriad ways in her daily life—she could suffer extremely
unpleasant consequences, from shunning to the loss of employment. Because the cost to her
of making a mistake on the science is zero and the cost of being out of synch with her peers
potentially catastrophic, it is indeed individually rational for her to attend to information
on climate change in a manner geared to conforming her position to that of others in her
cultural group.” Do you agree with Kahan’s reasoning here? Why or why not? Are there
other factors to take into account when considering how someone might make this decision?
Remark 4: For anyone concerned with the future of our society, Harary’s theorem must be a
depressing result. It suggests that the polarization into two mutually incomprehensible factions that
America experiences today is the result of some kind of mathematical necessity, not of historical,
political, or economic evolution. And yet: No theorem is better than its assumptions (“garbage in,
garbage out”; see page 37). By reviewing the assumptions of Harary’s theorem we may be able to plot
an escape route from the apparent prison it contrives for us.
• The theorem assumes a complete graph—every triangle is closed. I can push back against this
by leaving triangles open. My friend S likes Rush Limbaugh. Triadic closure encourages me to
close the triangle and have a connection (positive or negative) with Rush myself. If it’s negative,
I may lose my friendship with S, who challenges me to think in interesting ways. If it’s positive,
many of my other friends like N and Y will experience cognitive dissonance and I may lose
them. But if I resist the pressure for triadic closure, I don’t have to construct a “connection”
with Rush with either sign. And how real would such a connection be anyhow? Better to keep
the real, messy human connections that I have with N, S, and Y.
• The theorem assumes that every connection is either completely positive or completely negative.
I can push back against such a one-dimensional view of human interaction. My life would
be diminished without my connections with N, S, and Y, not to mention D, J, and K. We
disagree about many matters that are deeply important to all of us: questions of faith, political
philosophies, how we approach life—and yet we all agree that our relationships matter a great
deal to us, that the connections we have are among our most meaningful. Positive or negative?
The question makes no sense. Perhaps we should give up the idea that we have to ask it.
Summary of Ideas: Networks and Behavior
• The information-deficit model suggests that a lack of expert knowledge is the only
hindrance to public understanding of science.
• The bystander effect is the observation that the more bystanders witness an
emergency, the less likely it is that the victim will receive help from any of them.
• The bystander effect may be modeled in terms of an information cascade among the
network of bystanders.
• Preparation, including formal “bystander intervention training” courses, can increase
one’s ability to move from bystander to ally if the situation demands it.
• Homophily is the tendency for members of a dense cluster in a social network to
share similar opinions.
• A network with positive and negative edges exhibits structural balance if each
triangle has an even number of negative edges.
• Harary’s theorem states that a complete structurally balanced network must be
divided into two factions. However, this conclusion can be avoided if the hypotheses
of completeness (100% triadic closure) or signedness (each edge is either positive or
negative) do not apply.
3.3. FEEDBACK AND DYNAMIC NETWORKS 151
3.3 Feedback and Dynamic Networks
Objectives
I can distinguish between equilibrium and nonequilibrium behavior in stock-flow

systems.
I understand that stock-flow systems that are not in equilibrium can be studied by
dynamic modeling.
I can recognize feedback loops that control the behavior of dynamic models.
I know that feedback loops can give rise to many different types of dynamic
behavior.
I can distinguish between negative (stabilizing) and positive (amplifying) feedback
loops using the feedback rule of signs.
I can understand the possible effect of delay in feedback loops.
I can recognize growth, decay, oscillation, overshoot and collapse, and chaos as
possible behaviors of dynamic models.
3.3.1 Stock-Flow Systems and Dynamic Modeling

In the previous section, especially in thinking about the bystander effect, we have developed the idea
that a network need not be thought of simply as a static arrangement of nodes and links—it can be
a dynamic system, showing behavior that evolves as each node is affected by the behavior of the
other nodes. We have already studied a class of networks that hold the potential for this kind of
dynamic (that is, time-varying) behavior: the system diagrams associated with the stock-flow models
of Chapter 2. A system diagram is a network: the nodes are the various stock boxes, spigots, clouds,
and parameter flags, and the links connecting them are the different kinds of arrows (flow arrows and
control arrows). The dynamic possibilities depend on the control arrows. For example, if there is more
water in the bathtub, the outflow of water from the tub will be greater; the more rabbits there are in
a population, the more rapid the inflow of baby rabbits; the larger the stock of greenhouse gases in
the atmosphere, the less is the net outflow of radiant heat from the Earth’s surface. But so far we have
studied the behavior of these networks only in equilibrium—when stocks and flows are not changing
(Definition 4 on page 73). Now we are going to remove that assumption.
Remark 1: Many real-life systems are not, in fact, in equilibrium—their stocks are far from
constant. The Earth’s human population is growing, the population of Atlantic cod has declined
dramatically,3 the stock of carbon dioxide in the atmosphere is growing steadily but with a superposed
annual variation (see Figure 18 on page 50) coming from the growth of vegetation in the Northern
Hemisphere. To understand the behavior of nonequilibrium systems, where the stocks can change
3 See [246]. A quotation from this overview of the Atlantic cod: “U.S. fishermen have been harvesting Atlantic cod since
the 17th century. Cod was said to be so abundant then that you could almost walk across the ocean on their backs. Fisheries
for the species were booming, too—in fact, cod was one of the most lucrative products traded during colonial times. More
recently, New England groundfish such as cod had their heyday in the 1980s and early 1990s. Unfortunately, under this high
fishing pressure throughout the latter part of the 20th century, U.S. stocks of Atlantic cod came close to commercial collapse in
the mid-1990s.”
over time, we need a more sophisticated modeling approach called dynamic modeling or system
dynamics.
The dynamic (changing) structure of a stock-flow system away from equilibrium is governed by
feedback loops, which are mechanisms whereby the level of a stock can control the rate of a flow
into or out of that stock. Feedback loops may cause a system to settle comfortably down into an
equilibrium state (Section 2.3); but they may also cause all kinds of other behavior, including growth
or decay, “flipping” from one equilibrium state to another, oscillation, or overshoot and collapse.
4.8
Remark 2: The word dynamic indicates that something is
4
changing. In Definition 4 on page 73, we spoke of dynamic
equilibrium in which the stock levels remain constant but the
Depth (inches)
3
flows are nonzero. (In dynamic equilibrium, even though the stock
2
level stays the same, the actual physical “stuff” that makes up the
stock is constantly changing—in the bathtub example, new water
1 is constantly arriving through the faucet, old water is constantly
running away down the drain.) Dynamic modeling, which we are
0 now considering, is more general: we allow for the possibility that
0 10 20 30 40 50 60
Time (minutes)
stock levels may be changing as well.
Example 1: By definition, dynamic modeling involves studying
Figure 18: Dynamic behavior of bathtub model.
stocks that vary with time. Mathematicians represent this kind
of information in a familiar way—by a graph. Think about the
bathtub model again (Example 1 on page 95). Imagine that with
the tub empty and the drain open, we turn on the faucet to the flow rate of 2.2 gallons per minute. The
amount of water in the tub will then gradually increase from zero until it levels out at the equilibrium
depth of 4.8 inches. A graph of water depth against time will look something like Figure 18: the water
level increases fairly rapidly at first, but then levels off. The graph gives us a complete picture of the
behavior of the bathtub system, showing the exact way in which the water levels off as a function of
time.
Remark 3: In Problem 4 on page 76 we computed the residence time of the water in the bathtub
(once equilibrium is reached) to be 11 minutes. A quick approximation is that a process of this kind
will effectively reach equilibrium within five residence times. You can see from the graph that after
55 minutes, the depth is very close to its equilibrium value.
This kind of simple leveling off is not the only way that a system can approach equilibrium. Two
other possibilities are shown in Figure 19 on the opposite page. On the left is a model showing
oscillation about equilibrium: the system alternates between being above and below the equilibrium
level. On the right is a model showing overshoot and collapse: a stock level expands above the
equilibrium, but in doing so it degrades the system in some long-term way, so that when equilibrium
is finally attained it is at some lower level than was originally possible. Neither of these behaviors is
characteristic of bathtubs, but both are seen in biological populations.
When we study graphs like this, their overall shape is often more important than the precise numbers
that are involved. For instance, if we know that a system can exhibit the overshoot-and-collapse
behavior, then we might want to take steps to control the stocks involved before they get far above
equilibrium levels. If we know instead that the characteristic behavior of our system is steady approach
to equilibrium, such effort might be less important.
Remark 4: Several software packages offer the capability to construct dynamic models—often
using a “drag and drop” interface to position the various stocks, flows, spigots, and controls—and
then to generate the results of the model (curves like those in Figures 18 and 19) in real time.
A simple, free package of this sort is called Insight Maker [335]; this runs in any web browser
that supports JavaScript, so it can be used on all kinds of platforms. On the online resource site
1,000
60
Population (thousands)
Population (thousands)
800
40 600
400
20
200
0 0
0 20 40 60 0 20 40 60 80 100
Time (years) Time (days)
(a) Behavior of a stock undergoing oscillation. (b) Behavior of a stock undergoing overshoot and
collapse.
Figure 19: Examples of oscillation and overshoot and collapse.
(http://math-for-sustainability.com) that goes with this book, you can find examples where
we have used Insight Maker to illustrate the dynamic behavior of some of the models presented here.
3.3.2 Feedback
Let’s revisit the stock-flow model we developed for home heating in Example 5 on page 100 and
think about how the temperature of your home or apartment is governed on a cold winter’s day. The
apartment contains a stock of heat energy. We measure this stock using the indoor temperature as a
proxy. There is continual outflow of heat through the structure to the outside. As heat flows out the
temperature falls. Eventually, the temperature falls low enough that a thermostat activates, firing up
the furnace. The furnace blows warm air into the apartment, increasing the temperature. We call this a
feedback loop: the stock of heat controls the thermostat, and the thermostat affects the inflow of heat
to that stock.
We saw in this example that the stock of heat also affects the outflow: the rate of heat loss through
the structure is proportional to the difference in temperature between the inside and outside. The
cooler your apartment gets, the less rapidly heat energy will flow out of it. This is another feedback
loop. It is also the reason why you will save on your energy bills by lowering the thermostat; less heat
energy will be flowing to the outside, and therefore to maintain equilibrium, less heat energy will need
to flow in from the furnace. Look at Figure 20 on the next page and identify each of these feedback
loops.
Definition 1
In a stock-flow system, a feedback loop exists whenever the level of a certain stock directly
or indirectly affects one of the inflows or outflows to that same stock.
Let’s take a moment to review what the different arrows in Figure 20 represent. The heavy blue
arrows indicate the inflow of heat from a “cloud” (actually, from the furnace) and the outflow of heat
from the living space to a “cloud” (actually, to the outside). The inflow and outflow arrows each have
a spigot controlling the flow: the inflow spigot is the thermostat; the outflow spigot is the rate of heat
loss. You can imagine, for example, that opening a window on a cold day might further open the “heat
loss” spigot, increasing the rate of heat loss.
Figure 20: Stock-flow model of heat in home. There are two feedback loops.
The two thin red arrows pointing from “heat in home” to each flow spigot are the control arrows
introduced in Section 2.3. The control arrow running from the “heat in home” stock back to the
“thermostat” spigot indicates that the heat in the home, as measured by the temperature, controls the
heat flow from the furnace. In this case it determines whether the furnace is on or off. The control
arrow running from the “heat in home” to the “heat loss” spigot indicates a relationship between the
level of heat in the home and the rate that heat is lost to the outside.
Notice that the stock-flow diagram in Figure 20 doesn’t indicate how the stock level affects each
flow rate. Think again about what happens when the indoor temperature goes up. Will this tend to
increase or decrease the heat flowing in from the furnace? Will it tend to increase or decrease the
rate of heat loss? As we hinted at the end of the last chapter (Remark 6 on page 117), we can label
each control arrow with either a + sign or − sign to indicate how a change in the stock affects each
flow; we will then be able to see how each feedback loop affects the overall system.
Here are some additional examples of feedback loops.
Example 2: Your body’s regulation of blood sugar levels is an example of a feedback loop. The
pancreas contains sensors (called the islets of Langerhans) that monitor glucose levels in the blood.
When blood glucose levels drop, these release a hormone, glucagon, which signals to the liver to
release more sugar from its reserves. In the other direction, when blood glucose levels rise, a different
hormone, insulin, is released which tells the liver to extract glucose from the blood and store it. Thus,
the stock of blood glucose controls the hormone levels, and the hormone levels affect the inflow and
outflow of glucose to the stock.
Example 3: Bacteria are single-celled organisms that reproduce asexually by division (splitting in
two). Each parent bacterium divides into two “children.” For the bacterium E. coli, a common cause
of foodborne illness, this division process takes place roughly every 20 minutes. We could model
the number of bacteria by a stock-flow process in which the rate of inflow of “child” bacteria is
proportional to the number of “parent” bacteria already present. This is another a feedback loop.
“Newborn”
Stock of bacteria
bacteria
Figure 21: Feedback loop for bacteria population.
There is a very important difference between the feedback loops in Figure 20 and in Figure 21. Both
feedbacks in the home heating model work to stabilize the stock that they govern. If the temperature is
“too low” (less than the thermostat setting), the feedback involving the thermostat will try to increase
it. If the temperature is “too high” (greater than the outside temperature), the feedback involving “heat
loss” will try to decrease it. The two feedback loops have different “set points,” but they both have the
same property that if the room temperature deviates from the set point, they “push back” and try to
move it in the opposite direction to the deviation.
Definition 2
A feedback loop that tries to move a stock level in the opposite direction to its deviation
from some set point is called a stabilizing or negative feedback.
Things are very different with the feedback for the bacteria 600
population. Here the “set point” for the feedback is zero: a population
of zero is in equilibrium. But as soon as the population gets above
zero, and the bacteria begin to reproduce, the effect will be to increase
Number of bacteria
400
the bacterial numbers still further—to move them further away from
the set point. Then, of course, we will have still more bacteria, and
we will move away from the set point still faster! Eventually some 200
environmental constraint will kick in (maybe the bacteria will run out
of nutrients, or poison themselves with their own waste products);
0
but until it does, the bacteria will experience a population explosion 0 0.5 1 1.5 2 2.5 3
Time (hours)
(Figure 22). This is a characteristic outcome of the kind of feedback
we see in this example. Figure 22: Growth of bacteria population, start-
ing with a single cell.
Definition 3
A feedback loop that tries to move a stock level in the same direction as its deviation from
some set point is called an amplifying or positive feedback.
Remark 5: The behaviors we see in Figure 22 and Figure 18 on page 152 are characteristic of systems
with a single stock and a single feedback, positive in the case of Figure 22 and negative in the case of
Figure 18. We call these patterns of stock behavior growth and decay, respectively, and we will have
much more to say about them in Section 3.4. (Notice that in the second example the word “decay”
refers to how the system settles down to a fixed level or “set point,” 4.8 in in this example; it is not
necessary that the “fixed level” be zero.)
In many cases it may be clear from “common sense” whether a particular feedback loop is
amplifying or stabilizing. But we can always tell by marking the feedback loop with + or − signs
and “counting around the loop” according to the following rules.
Rule 1: Feedback Rule of Signs
• Mark a + sign on each flow arrow (blue) that represents an inflow to the stock that
you are considering, and mark a − sign on each flow arrow that represents an outflow
from that stock.
• Mark a + sign on each control arrow (red), going from a stock to a spigot, if an
increase in the stock leads to an increase in the corresponding flow; mark a −
sign on such a control arrow if an increase in the stock leads to a decrease in the
corresponding flow.
• To determine whether a given feedback loop is positive or negative, count the total
number of − signs on all the arrows around the loop. If this total is even, the feedback
is amplifying (positive); if the total is odd, the feedback is stabilizing (negative).
For example, marking the arrows according to the feedback rule of signs in the stock-flow diagram
for the home heating system gives us the diagram in Figure 23. Both feedback loops have a total of
one minus sign and are therefore stabilizing, as we said before.
inflows are outflows are
marked with a + marked with a −
+ −
− +
− indicates an + indicates an
increase in stock will increase in stock will
decrease the flow increase the flow
Figure 23: System diagram for heat in home, signs marked. We mark the red thermostat control arrow with a − because an
increase in temperature tends to decrease the inflow of heat (by shutting off the furnace). We mark the red control arrow to
“heat loss” with a + because an increase in temperature increases the rate of heat loss. Both feedback loops are stabilizing.
Problem 1: Use the feedback rule of signs to confirm that the feedback loop for the model of bacteria
in Example 3 on page 154 is amplifying.
Solution: We label the blue flow arrow with a + because it is an inflow. We label the control arrow
with a + because an increase in the bacteria population will tend to increase the inflow of new bacteria.
There are two + signs and no − signs; this confirms that we have an amplifying feedback.
Problem 2: A short while after drinking a cup of coffee there is a certain concentration of caffeine
in your bloodstream. The caffeine is metabolized (i.e., processed and excreted from your body) at a
rate that is proportional to the stock of caffeine in your bloodstream. Draw a stock-flow diagram, label
the flow and control arrows according to the feedback rule of signs, and identify any feedbacks as
amplifying or stabilizing.
Caffeine in −
Metabolism
bloodstream
Figure 24: Stock-flow model for metabolism of caffeine in the body.
Solution: The problem says that a stock of caffeine is already present in your body when the modeling
begins. The blue flow arrow is labeled with a − because it is an outflow. The control arrow is labeled
with a + because metabolism occurs at a rate proportional to the stock of caffeine. Thus an increase
in the level of caffeine will increase the outflow through metabolism. Counting signs around the loop
we get one − sign, so this is a stabilizing feedback.
Let’s apply these ideas to a more complicated example.
Problem 3: Electric blankets made for double beds have two temperature controllers, one for each
side. Each partner can turn the heat on their side of the bed up if it is too cold, and down if it is too
hot. Each is therefore part of their own personal stabilizing feedback loop.
President Jimmy Carter tells the following story in his autobiography [60, page 74]:
During each of the increasingly cold winter nights we argued about the temperature of
our electric blanket. Whenever I said it was too warm, Rosalynn said it was too cold, and
vice versa.
One morning I returned from an overnight trip to New York, and she met me at the door
with a warm hug and a smile. “I think our marriage is saved,” she said. “I just discovered
that the blanket controls were on the wrong side of the bed, and each of us was changing
the temperature of the other’s side.”
Before Rosalynn Carter’s discovery, what kind of feedback was the Carters’ electric blanket
demonstrating?
Solution: Let’s draw the stock-flow diagrams: first for things as the blanket manufacturer intended,
and second for the situation that the Carters had accidentally set up by interchanging the controllers.
In Figure 25(a) on the next page you see a simplified version of the system as it was intended—two
independent negative or stabilizing feedback loops like the left-hand one in Figure 23. (To keep things
a bit simpler, we left out the “heat loss” part of the picture, which isn’t relevant to this particular
discussion.)
But after the controllers are exchanged, things look different! Figure 25(b) shows the new setup.
In place of two stabilizing feedback loops, we have a single loop that involves two positive and two
negative signs. The total number of negative signs is even, so this is a positive or amplifying feedback.
Any small deviation from comfort was bound to be increased, even as both Jimmy and Rosalynn
behaved “rationally,” as it seemed to them, in trying to correct the problem as they saw it.
+ +
Controller Jimmy Controller Jimmy
−
−
+ +
Controller Rosalynn Controller Rosalynn
−
−
(a) Domestic bliss for the Carter family. (b) What happens after switching controllers.
Figure 25: Stock-flow diagrams for the Carters’ electric blanket.
Think about it. . .
Notice that in Problem 3 on the previous page, it would make no sense to ask, “Who was
to blame for the problem—Jimmy or Rosalynn?” The problem was inherent in the way the
system was structured, it was not the “fault” of any of the individual participants. Can you
think of similar examples in the environmental or sustainability realm?
Including a Parameter
In Definition 2 on page 102 we introduced the notion of a parameter for a stock-flow model—an
exogenous quantity (that is, a quantity that is not itself one of the stocks or flows in the model) that
nevertheless influences one or more of the flow rates. The outside temperature in our home heating
models was a parameter, and later, in Section 2.4.5, we introduced a climate model that included a
parameter g that governed the strength of the greenhouse effect. We included this in the stock-flow
diagram by adding a parameter box (a.k.a. “flag”) for the parameter along with a control arrow
pointing to the “long-wave radiation” spigot. If we increase the strength of the greenhouse effect,
say by increasing the concentration of atmospheric CO2 , this reduces the outflow of long-wave heat
radiation from the Earth into space. We can indicate this in the diagram by labeling the control arrow
for the GHE strength with a − sign. Thus the complete model with all signs included is as in Figure 26.
Short-wave Planet − Long-wave

radiation Temperature T radiation
+ −
GHE strength, g
Figure 26: Climate model with parameter g that governs the strength of the greenhouse effect (GHE). An increase in g decreases
the outflow of long-wave radiation from the planet.
Let’s look at another example involving a parameter. Water quality in a lake is affected by the
amount of phosphorus in the water. A low level of phosphorus is associated with high water quality.
The lake may be a source of drinking water, provide recreation, and support a healthy fishery. If the
phosphorus content becomes high enough, algal blooms may occur, starving the lake of oxygen and
degrading the water quality of the lake. In the example that follows, we treat the phosphorus content
of a lake as a stock and the rate of inflow of phosphorus from nearby pollution sources as a parameter.
Example 4: Consider the amount of phosphorus in a lake as a stock. The outflow of phosphorus
occurs at a rate proportional to the stock. But the stock also affects the inflow of phosphorus,
because phosphorus in the lake water is a nutrient for plants and algae, which then take up additional
phosphorus that has accumulated in sediments (silt) on the lake bottom. Eventually, these organisms
decay (or are eaten by fishes) and all their phosphorus, including the extra amount they have obtained
from sediments, is released back into the lake water [58]. Additional phosphorus from agricultural run-
off and other pollution sources within the lake’s watershed also flows into the lake; the total inflow
of phosphorus is the sum of the inflows from watershed pollution and from the release of phosphorus
from sediments via algal uptake.
Let’s represent these various effects in a stock-flow model (see Figure 27). The process whereby
organisms take up phosphorus and rerelease additional phosphorus back to the water is indicated by an
amplifying feedback involving the inflow. The watershed pollution rate is shown as a parameter that
affects the total inflow. The outflow is part of a stabilizing feedback loop—like the bathtub model,
when the stock of phosphorus is higher, the outflow of phosphorus is also higher.
Although the pollution parameter is not itself part of a feedback loop, we will see in Sections 4.3
and 4.4 how changes in the pollution parameter can interact with the two feedbacks and lead to large
changes in water quality of the lake.
Inflow of + Stock of Phosphorus − Outflow of

Phosphorus in Lake, S of Phosphorus
+ +
+
Pollution, P
Figure 27: Stock-flow model for phosphorus in a lake, with a pollution parameter.
3.3.3 A Dynamical Zoo

As we already suggested, systems with several stocks, flows, and interacting feedback loops can
exhibit a wide range of more or less complex behaviors. In this section we will—very briefly—look
at some of these different behaviors and the stock-flow models that exhibit them. We start with the
most fundamental kind of population model, which we’ll call the basic demographic model.
Consider a population of animals or human beings in a context in which there is no migration
(an isolated island, for example). The total population is changed by the addition of new individuals
through birth (an inflow) and the loss of existing individuals by death (an outflow). Both processes
tend to increase in speed with the size of the existing population: the larger the population, the more
births and the more deaths. This gives the basic demographic model shown in Figure 28 on the next
page. We can identify two feedback loops in this model: an amplifying one (associated with births)
and a stabilizing one (associated with deaths). The long-term fate of the population will depend on
which feedback is the more powerful.
+ −
Births Population Deaths
+ +
Figure 28: Basic demographic model.
Oscillation
More complicated effects can arise when two or more populations are
connected to one another, as in the classic example of lynx and hare in the
Canadian Arctic. The Canadian lynx (see picture) is a large predatory cat
that ranges over the Canadian Arctic, Alaska, and some northern parts of
the lower 48 states. Lynx feed largely on the snowshoe hare: when hare are
abundant, they can make up over 90 percent of the lynx’s diet. Because of
this, a feedback loop is set up: when hare are abundant, lynx populations
will increase (because there is abundant food for the lynx); when lynx are
abundant, hare populations will decline (because too many of them are being
Figure 29: A Canadian lynx.
eaten). This is indicated in the system diagram (Figure 30), where there is
a positive effect (control) of the hare population on lynx births; and there
is also a positive effect (control) of the lynx population on hare deaths. The
overall feedback around this interaction loop is negative (if we count signs, according to Rule 1 on
page 156, we find three + signs and one − sign). However, this negative feedback does not lead the
whole system to “settle down.” Rather, because of the time delays caused by having two stocks in the
loop rather than one, it results in oscillation of both the lynx and hare populations. Such oscillations
are characteristic of models of this type, with one predator species and one prey species: they are
known as Lotka–Volterra models after the mathematicians who first studied them.
Lynx + Lynx − Lynx

+ +
+ +
Hare + Hare − Hare

+ +
Figure 30: Interaction between lynx and hare populations.
In the case of the lynx and hare, it had been known for many years that their populations oscillate on
a roughly 10-year cycle. In the nineteeth century, the Hudson’s Bay Company held a near monopoly
of the fur trade in Canada, and it maintained careful records of the number of lynx and hare pelts
traded year by year. Estimates of the lynx and hare populations can be derived from these records,
150
hare 150 hare
lynx lynx
Population in thousands
Population in thousands
100
100
50 50
0
0
1840 1860 1880 1900 1920 1940 1840 1860 1880 1900 1920 1940
Year Year
(a) Lotka–Volterra model simulation. (b) Historical data.
Figure 31: Simulation and historical data for hare and lynx populations.
and the oscillations are plain to see (Figure 31(b)). The graphs are a good deal more jagged than those
produced by the simple Lotka–Volterra model, but the overall pattern is quite similar.
Critical Thinking
Does the evidence shown in Figure 31 prove that the predator–prey relationship between
lynx and hare is responsible for the observed oscillation in their populations, according to
the Lotka–Volterra mechanism? Alternatively, does it prove that the Lotka–Volterra model
is wrong for this situation? Think about the discussion of scientific method in Section 2.1.4.
Remark 6: We said earlier that the oscillations in the Lotka–Volterra model can be thought of as
resulting from delays in the system because it has two populations rather than one. Delays can also
appear more explicitly in a system. For example, consider a retail outlet that sells items that take a
long time to manufacture—a furniture store, perhaps, or an automobile showroom. Because of the
delay between placing an order to the manufacturer and the manufacturer filling it, the manager of the
outlet will try to keep some inventory on hand, and will regularly order more if the inventory seems to
be getting low, less if it is getting high. This gives us a stock-flow diagram like that shown in Figure 32
on the next page. The “clock” symbol on the control arrow (which we’ll use only in this one place)
symbolizes the delay before the corresponding control takes effect.
A system like this is prone to oscillation. A sudden “blip” in sales leads to a fall in inventory. The
owner orders more stock to compensate; but because that does not arrive for a few days, she keeps
ordering extra. By the time the new stock starts arriving, sales are back to normal, and all of a sudden
the store is overwhelmed by a flood of new stock, taking inventory far above normal. The owner
compensates by reducing orders, but again it is a while before this takes effect. And so the oscillation
goes on.
It is tempting to say that if the problem results from the delay, reducing the delay (responding more
quickly) should make the problem go away. But it turns out that this is the opposite of the truth. In a
system like this, reducing the delay makes the oscillations faster and more violent—it does not tamp
them down. The way to restore stability is to increase the delay, and to respond less urgently to the
Orders + − Customer
Inventory
Delivered Purchases
−
+
Figure 32: Inventory system with delay.
apparent need. This is an example of the apparently paradoxical and yet correct conclusions that can
result from mathematical modeling.
Overshoot and Collapse

When a population grows according to an amplifying feedback loop, but
that growth depletes a finite resource that the population needs for survival,
the typical behavior exhibited is overshoot and collapse. The textbook
example is the reindeer of Saint Matthew Island, a small (128 square miles)
uninhabited island in the Bering Sea off Alaska.
In 1944, in preparation for an anticipated campaign of war in the Pacific,
the U.S. Coastguard briefly established a base for a navigational radio station
in the inhospitable environment of Saint Matthew Island. The base itself
lasted less than a year. But during that period, the Coastguard introduced
29 reindeer to the island. The idea was that in a last-ditch emergency, the
reindeer might serve as a food source.
Figure 33: St. Matthew Island.
When they evacuated the island, the Coastguard left the reindeer behind.
Untroubled by predators, and finding forage to their liking, the reindeer
population increased rapidly: from 29 individuals at introduction in 1944 to
1350 individuals in 1955 and 6000 individuals in 1962. However, this rapid
growth was sustained by depleting the forage plants which were slow to regenerate in the Arctic
climate. In 1965, a survey found 42 reindeer on the island. One in 1980 found none. The reindeer
population had overshot its sustainable limit and as a result experienced collapse. Notice that the
overshoot was the result of their discovering an energy source (forage plants) that had been stored up
over a long period of time, but which they depleted rapidly—see the stock-flow diagram, Figure 34
on the opposite page. It is hard to avoid the analogy with human use of fossil fuels.
Chaos
Overshoot and collapse might sound like a pretty “chaotic” kind of behavior. Mathematicians,
however, use the word “chaos” in a specific way.
Definition 4
A dynamic model exhibits chaos if small changes in the system conditions lead to large and
unpredictable changes in its long-term behavior.
The weather is the most well-known example here: Edward Lorenz, the mathematician who
discovered the modern concept of “chaos” in the early 1960s, did so while attempting to generate
−
Vegetation Grazing
+ Reindeer −
+ +
Figure 34: Model for the reindeer of St. Matthew Island.
weather forecasts on a then state-of-the-art computer. Lorenz also invented the phrase “the butterfly
effect” to describe the hallmark of chaos, that small causes might lead to large and unpredictable
effects: “Could the flap of a butterfly’s wings in Brazil set off a tornado in Texas?” As we have seen
with our discussion of weather and climate (Section 2.4.1), this unpredictability on the day-to-day
level does not rule out making valid statistical statements about the long-term behavior of the weather
or any other chaotic systems. One way to arrive at such a statistical prediction might be simply to run
your dynamic model thousands of times with slightly different initial data, and then review the pile of
results from these model runs using the tools of descriptive statistics (Section 5.1).
Summary of Ideas: Feedback and Dynamic Networks
• A dynamic stock-flow model is one that explains how stock levels change over time
in a system that is not in equilibrium.
• A feedback loop occurs when the level of some stock directly or indirectly controls
the flow into our out of that stock.
• A feedback loop that moves the system toward some “set point” is called stabilizing
or negative. A feedback loop that moves the system away from a “set point” is called
amplifying or positive.
• We can identify a feedback loop as amplifying or stabilizing using the feedback rule
of signs.
• Some possible behaviors of dynamic models include growth, decay, oscillation,
overshoot-collapse, and chaos.
• A model exhibits chaos if small changes in the system conditions lead to large and
unpredictable changes in long-term behavior.
3.4 The Exponential Model
Objectives
I can identify when the feedbacks in a stock-flow system lead to an exponential

model.
I can give examples of exponential models in physics, biology, economics, and
personal finance.
I can use an exponential model to estimate future population levels.
I can find the half-life or doubling time for an exponential model.
3.4.1 Exponential Models

As we have seen (Definition 1 on page 153), a feedback loop in a stock-flow model occurs when the
flow into or out of a given stock depends in some way or other on the level of that stock. That “in
some way or other” makes the definition of “feedback loop” very general. In this section we will study
an important specific kind of feedback: a feedback for which the flow is directly proportional to the
stock. Such a system is said to be governed by an exponential model.
Definition 1
A stock in a stock-flow system is governed by an exponential model if the net flow

(page 73) into the stock over a given time interval is proportional to the level of the stock at
the beginning of that interval. To understand the dynamics of a stock that is governed by an
exponential model, we therefore need to know two things: the constant of proportionality
(or rate constant) r that connects the stock level and the flow rate, and the length of time
(or time step) T over which that flow is measured.
Example 1: The population of bacteria in Example 3 on page 154 is governed by an exponential

model. Every 20 minutes, there is a net flow of bacteria into the population that is equal to the
previously existing population of bacteria. In other words, there is a 100% increase every 20 minutes.
The time step is T = 20 min and the rate constant is r = 100% = 1. Notice that the rate constant
is a pure number (Definition 2 on page 15), since it is the ratio of two quantities both of which are
expressed in the same units (“bacteria” in this case).
Example 2: The behavior of the thermostat in the example shown in Figure 20 on page 154 is a
feedback loop that is not governed by an exponential model. A standard thermostat has only two
states: “on” (the furnace is producing heat at its rated capacity) and “off” (the furnace is not producing
heat at all). There is a relationship between the heat flow and the temperature stock, yes, but it is not
a proportional relationship.
The most important thing about the rate constant in an exponential model is its sign—positive or
negative. If the rate constant is positive, then each increase in the stock will also increase the net
inflow. That will increase the stock more, which will increase the flow more, which will increase the
3.4. THE EXPONENTIAL MODEL 165
stock more, and so on; we have an amplifying feedback, leading to exponential growth of the stock.
(The behavior of the bacteria in Example 3 on page 154 demonstrates exponential growth.) On the
other hand, if the rate constant is negative, then each increase in the stock will reduce the net inflow
(or increase the net outflow). We have a stabilizing feedback, leading to exponential decay of the
stock.
Example 3: The money in an interest-bearing savings account or CD is governed by an exponential
growth model (if you leave it in the account and don’t spend it!). Every year, the savings account pays
interest that is proportional to the stock of money already in the account. The interest is a flow of
money that is added to the stock. The time step is 1 year, and the rate constant is the rate of interest.
Over time, the stock of money grows exponentially.
Example 4: Consider the cooling of a cup of hot coffee. Over time, the temperature Y of the coffee
in the cup will approach the temperature P of the room that the cup is in. The rate of heat flow out of
the cup is proportional to the temperature difference Y − P. Thus, Y − P is governed by an exponential
decay model (decay because the flow is out of the cup, so considered as a net flow, it has a negative
sign).
As in this example, it often happens that the difference between the level of a stock and some fixed
“set point” is governed by an exponential model. Then we will say that the stock itself is governed
by an offset exponential model—it has the same qualitative behavior as the usual exponential model,
but the equilibrium level has been “offset” to the set point. Offset exponential models are especially
common in problems involving temperatures, but they can occur in other situations too.
Definition 2
A stock in a stock-flow system is governed by an offset exponential model if the net flow
associated with the stock is proportional to the difference between the level of the stock and
some constant level (the set point).
There are three parameters that govern the behavior of an offset exponential model: the rate constant
r, the time step T , and the set point P. If S is governed by an offset exponential model with set point
P, then S − P is governed by an ordinary exponential model with the same rate constant and time step.
Example 5: The bathtub in Example 1 on page 152 is governed by an offset exponential model. The
set point is the equilibrium level, 4.8 in. The time step can be taken to be 1 minute, and the rate
constant then is −0.092 (this follows from the calculations in Example 2 on page 97). The graph in
Figure 18 on page 152 shows the exponential decay of the water level to the set point of P = 4.8 in.
Remark 1: Notice that we still say “decay” in the previous example even though the water level
is increasing toward equilibrium—the word “decay” tells us that the process is slowing down, not
necessarily that the stock level is going down. The same thing would happen in Example 4 if instead
of a hot cup of coffee, I had started with a chilled latte whose temperature was lower than that of the
room.
3.4.2 The Math of Exponentials

In this section we will develop the mathematical formulas that describe the behavior of a stock under
an exponential model. Let’s begin by working out some numbers corresponding to the bacteria model,
Example 3 on page 154.
Imagine that we begin with only 1 cell. This cell is our stock. After 20 minutes, it splits into 2 cells.
So our net inflow is 1 cell, which is equal to the starting population. After 20 minutes, our total stock
is equal to the original population plus the net inflow:
1 cell + 1 cell = 2 cells.
After 20 minutes more, both of those two cells split again. The net inflow is 2 cells, which is again
equal to the previous population:
2 cell + 2 cell = 4 cells.
There is a clear pattern: the stock is multiplied by 2 every 20 minutes.
Problem 1: Suppose we begin with one bacterium of E. coli. How many bacterial cells should we
expect after 2 hours? Assume exponential growth.
Solution: Remember that 2 hours equals 120 minutes, or six “doubling” steps. We can count the
changes for each 20-minute interval successively:
0 mins 1 bacterium
20 mins 2 bacteria
40 mins 4 bacteria
60 mins 8 bacteria
80 mins 16 bacteria
100 mins 32 bacteria
120 mins 64 bacteria
After 2 hours, the original bacterium will have grown to a collection of 64 bacterial cells.
As you can see, the general pattern is that after a number n of 20-minute steps, the original one
bacterium will have grown to
2 × 2 × · · · × 2 bacteria.
| {z }
n times
Mathematicians call the number obtained by multiplying 2 by itself n times “2 raised to the power n,”
and write it 2n (we already used this idea in Section 1.2, Scientific Notation). You can use a calculator
to work out the result (xy ) of raising any positive number x to a given power y.
Problem 2: Starting with one E. coli bacterium, how many does the exponential model predict after
24 hours? Comment on your answer.
Solution: 24 hours is 24 × 3 = 72 twenty-minute doubling periods, so the model predicts a population

of 272 bacteria. A calculator tells us that 272 ≈ 5 × 1021 . The mass of a single E. coli bacterium is
roughly a picogram, that is 10−15 kg, so the total predicted mass of all the bacteria after 24 hours is
5 × 1021 × 10−15 kg = 5 × 106 kg = 5, 000 T.
It’s (fortunately) not possible in real life for a single bacterium to grow to a five thousand tonne ball
of cells in a single day. Long before that stage is reached, the exponential growth model will break
down: negative feedbacks that the simple exponential model does not take into account—shortages of
food or energy, the accumulation of waste products—will have gained a dominant influence.
Remark 2: The answer we obtained to Problem 2 is not realistic, but it does illustrate a very important
aspect of exponential growth: even though it may seem slow to get started, it can produce astonishingly
large numbers in a relatively small number of “doubling periods.” Another way of saying this is that
if there are any limits in a system, an exponentially growing process is likely to hit those limits hard.
Think about this story:
A French riddle for children illustrates another aspect of exponential growth; the
apparent suddenness with which it approaches a fixed limit. Suppose you own a pond
on which a water lily is growing. The lily plant doubles in size each day. If the lily were
allowed to grow unchecked, it would completely cover the pond in 30 days, choking off
the other forms of life in the water. For a long time the lily plant seems small, and so you
decide not to worry about cutting it back until it covers half the pond. On what day will
that be? On the twenty-ninth day, of course. You have one day to save your pond [215,
p. 29].
Many sustainability issues revolve around managing positive feedbacks in the presence of limits,
so this property of exponential growth comes up time and again.
So far we have looked at only one exponential model: Example 3 on page 154. Let’s consider some
other examples.
• From the founding of the nation until the 1970s, U.S. total energy consumption grew
exponentially at a rate of approximately 2.9% per year. (Consumption growth appears to have
leveled off since then; try exploring the “Primary Energy Overview” data series on the U.S.
Energy Information Administration web site [11] for detailed information on U.S. energy
consumption from the 1970s to the present day.)
• A certificate of deposit at my local bank pays 0.2 percent interest per quarter.
• 17% of the atoms in a sample of radon-222 (see Example 3 on page 74) will undergo radioactive
decay in a one-day period.
• The temperature in my bedroom falls by 1 degree per hour for each 10 degrees of temperature
difference between my room and the outside air.
The table below summarizes the rate constants and time steps for these four examples. Notice that in
the first two examples the rate constants have positive signs and in the last two examples they have
negative signs; this reflects the type of feedback (positive or negative) that is involved and the kind
of exponential behavior (growth or decay) that it results in. We can express the pure number r as a
Table 4: Time Steps and Rate Constants
Example Time step (T ) Rate constant (r)

U.S. Energy consumption 1 year 2.9%
Bank CD 0.25 year 0.2%
Radon decay 1 day −0.17
1
Room cooling 1 hour − 10
1
decimal, a fraction, or a percentage—whatever is most convenient. For example, the − 10 on the last
line could have been expressed as −0.1 or −10% if we had preferred.
Now we will give the mathematical formula that will tell us what happens to the stock in any
example of an exponential process (just as the formula 2n , 2 to the power n, told us what should
happen to the stock of bacteria in Problems 1 and 2). We call this the law of exponentials. As well as
the rate constant r and the time step T that we have already defined (Definition 1 on page 164), the
law of exponentials contains one further quantity: the initial value S0 of the stock—its value when the
growth or decay process begins. For example, the initial value in the bacteria model is 1 bacterium;
the initial value in the CD model is the amount you deposit to open the account; the initial value
in the radon-decay model is the number of radon atoms in the sample when it is fresh. The law of
exponentials tells us the level of the stock at any future time in terms of the rate constant, the time
step, and the initial value.
Rule 1: Law of Exponentials
Suppose a stock S is governed by an exponential process with time step T and rate
constant r. If S0 is the initial value of the stock, then the stock Sn after n time steps is
given by
Sn = S0 × (1 + r)n .
Equivalently, the stock S(t) after time t has passed is given by
S(t) = S0 × (1 + r)(t/T ) .
The two forms of the law of exponentials are equivalent, because after n time steps each of which
has length T , the total time t that has elapsed is t = n × T . Dividing both sides by T gives t/T = n,
showing that the exponents in the two forms of the law are in fact equal. Notice that t/T is a pure
number, since it is obtained as the ratio of two quantities that have the same units.
Problem 3: Grandparents decide to deposit $5,000 in a college fund when their grandchild is born.
The fund earns 0.2% interest each quarter. Determine the time step and rate constant, and then use the
law of exponentials to estimate the value of the fund when the child turns eighteen.
Solution: The time step is T = 1 quarter and the rate constant is r = 0.2% = 0.002. The time t is 18
years; we’ll need to express this in the units of the time step: quarters:
4 quarters
18
yr × = 72 quarters.
1
yr

The initial stock is the deposit, S0 = $5, 000. Putting all this into the law of exponentials we get
$5, 000 × (1 + 0.002)72 ≈ $5, 770.
Remark 3: In the previous example, we said that the time step was T = 1 quarter. We could have
used T = 3 months or T = 0.25 yr. These are all equivalent, but we need to be careful that the time
t is also in the same units. Let’s redo the calculation with T = 0.25 yr and t = 18 yr and include the
units:
18 yr
S = $5, 000(1.002) 0.25 yr = $5, 000(1.002)72 .

Notice that the units in the exponent cancel, showing that t/T is a pure number as we said above.
Problem 4: Historically, investment in stocks has yielded annual returns that average 7 percent (but
with considerable variability). Suppose the grandparents instead put their $5,000 in a stock fund that
earned exactly 7% each year for 18 years. Estimate the final value of the fund in this example.
Solution: The time step here is T = 1 yr and the rate constant is r = 7% = 0.07. We have
S = $5,000(1.07)18 ≈ $16,900.
The payoff after 18 years is roughly three times bigger for the stock investment.4
4 But we have assumed that the future performance of this stock portfolio is exactly the same as its average performance in
the past. This is highly unlikely to be the case. In fact, the Stock Exchange Commission requires mutual fund advertisements
to include the warning, “Past performance does not guarantee future results.”
$500
2%
4%
$400 6%
8%
$300
$200
$100
$0
0 5 10 15 20
Time (Years)
Figure 35: Growth of $100 at different annual interest rates over 20 years.
This is a striking example of how apparently small differences in exponential growth rates can
accumulate, over time, into large differences in the final stock. To visualize these differences, take a
look at the chart in Figure 35, which shows the exponential growth of $100 at four different annual
interest rates: 2 percent, 4 percent, 6 percent, and 8 percent. Notice that the curves for the higher
interest rates are not only “ahead” of the lower ones—their advantage increases over time. At higher
interest rates, compounded interest (that is, interest on interest) accumulates much more quickly and
helps the corresponding curves accelerate away from the others.
Remark 4: What makes the law of exponentials true? We can understand it in the same way that
we understood the calculations in the E. coli example beforehand. Definition 1 on page 164 of an
exponential process tells us that after each time step, r times the previous level of the stock is added
to the stock. If the stock after n time steps is Sn , then rSn new stock will be added during the (n + 1)th
time step. So the stock at the end of the (n + 1)th time step will be
Sn+1 = Sn + rSn = Sn × (1 + r).
If we make a table of stock values at each time step, then each one will be obtained by multiplying
the previous one by (1 + r), as shown in Table 5.
Table 5: The Law of Exponentials
Time Step n Stock value Sn

0 S0
1 S1 = S0 × (1 + r)
2 S2 = S1 × (1 + r) = S0 × (1 + r) × (1 + r) = S0 × (1 + r)2
3 S3 = S2 × (1 + r) = S0 × (1 + r)2 × (1 + r) = S0 × (1 + r)3
4 S4 = S3 × (1 + r) = S0 × (1 + r)3 × (1 + r) = S0 × (1 + r)4
You can see that the pattern will continue, giving the law of exponentials
Sn = S0 × (1 + r)n .
Critical Thinking
Exponential growth can produce surprisingly large numbers very quickly, as one sees in
Problem 2 on page 166. That is probably why newspaper writers and others often say
“exponentially” when they want to suggest that something is much bigger than expected. A
Google search will turn up hundreds of examples:
• “Veterans Affairs scandal exponentially bigger than Benghazi for President Obama.”
• “Torchlight II promises to be exponentially bigger than the original.”
• John Krasinski: “Guys have a level of insecurity that’s exponentially bigger than you
think.”
We have seen that the word “exponential” refers to the behavior of a single quantity (a
stock) that changes over time. The quantity behaves exponentially if its net change over a
specified time interval is proportional to its value at the beginning of the interval. It doesn’t
make sense to use “exponentially” in comparing just two values, as is done in the quotations
above (to say nothing of the difficulty of assigning numerical values to government scandals,
video-game awesomeness, or male insecurity). What would be a better way to express the
comparisons made here?
3.4.3 Population Models

One of the standard applications of the exponential model is to various kinds of populations. Recall
the basic demographic model from Figure 28 on page 160, which we redraw as Figure 36.
+ −
+ +
Figure 36: Basic demographic (population) model.
If we assume that both the inflow (births) and the outflow (deaths) in this model are proportional to
the population, then the net flow (inflow minus outflow) will also be proportional to the population. In
other words, we have an exponential model. Here are some examples giving practice in applying this
model to human populations.
Problem 5: The population of the United States was 320 million people in 2015. There were 3.9
million births and about 2.4 million deaths in 2015. Assuming that the population is governed by an
exponential model, and that the birth and death rates remain constant, estimate the population in 2025.
Solution: We assume that both births and deaths are proportional to the total population. The constants
of proportionality are called the birth rate and death rate respectively. Even though there are two flows
involved here, one inflow and one outflow, this is still an exponential model, because each of the flows
is proportional to the population, and therefore so is their difference, the net flow (births minus deaths).
To solve the problem, we begin by using the given information to estimate the birth rate and the
death rate. The time step is 1 year. First, let us find the birth rate:
3.9 million
= 0.01218 ≈ 1.22%.
320 million
Next, we find the death rate:

2.4 million
= 0.0075 ≈ 0.75%.
320 million
The overall rate constant is equal to their difference, r ≈ 1.22% − 0.75% = 0.47% = 0.0047. Now we
apply the law of exponentials to estimate the population after ten years:
Future population = (1 + 0.0047)10 × 320 million ≈ 335 million.
Remark 5: There are two feedbacks in Figure 36. The “birth” feedback is amplifying and the “death”
feedback is stabilizing. We found that overall, the rate constant is positive, r = 0.47%, because the
birth rate is higher than the death rate. Focusing our attention on the feedbacks, we would say that
the amplifying feedback due to births is stronger than the stabilizing feedback due to deaths; the net
effect is that the population is growing. We can indicate the strength of each feedback in the stock-
flow diagram by labeling the control arrows with the sign of the control as before, along with the
appropriate rate, as shown in Figure 37.
+ −
+1.22% +0.75%
Figure 37: Demographic model indicating the strength of each control.
Question 1: Why is the control arrow for the “Deaths” spigot labeled with a +0.75 when this feedback
is negative (stabilizing)?
Answer: Good question! Recall that we labeled this control arrow with a + sign to indicate that an
increase in the population will lead to an increase in the number of deaths. We don’t change this sign
when we include the death rate of 0.75 in the label on the control. This feedback loop includes the blue
outflow arrow, which is labeled with a − sign. Using the feedback rule of signs (Rule 1 on page 156),
the sign on the feedback loop is negative, (−0.75). It’s the sign on the whole feedback loop that tells
us that the rate constant for the death loop is negative, not the sign on the control arrow itself. For now
we just want to introduce the idea that we can measure the strength of a control. We’ll explore this in
more detail in the next chapter. See Section 4.3.
Problem 6: Not all countries are growing in population. The population of Japan has been in decline
for many years. Japan’s population in 2015 was 127 million. That year the country had roughly 1.03
million births, 1.19 million deaths. Use an exponential model to estimate the population of Japan in
10 years.
Solution: As before, we need to calculate the rate constant, r. The birth rate is 1.03/127 ≈ 0.81% per
year, and the death rate is 1.19/127 ≈ 0.94% per year. Overall the net inflow rate is
r = 0.81% − 0.94% = −0.13% = −0.0013.
The negative sign shows that deaths predominate over births. The population would decay exponen-
tially if these birth and death rates remained constant. The law of exponentials gives the population
after 10 years:
Future population = 127 × (1 − 0.0013)10 million = 0.998710 × 127 million ≈ 125 million.
Remark 6: Our stock-flow models for the populations of the United States and Japan are useful
examples for understanding how constant birth and death rates lead to exponential growth or decay
in population models. How well do they do as models of how these two populations are, in reality,
changing over time? Not very well for the U.S., although somewhat better for Japan. We’ve ignored
the effects of migration, a significant factor in population changes in the United States. We’ve also
assumed that birth and death rates remain constant over time, an assumption that is not accurate.
Throughout the world, as infant mortality rates have declined with improved sanitation and access
to health care, birth and death rates have also decreased in a predictable pattern known as the
demographic transition. We will look at more realistic population models in Case Study 7.4.
Think about it. . .
An Indian legend focuses on the properties of exponential growth. According to the story,
an Indian king, delighted by the invention of the game of chess, asks the inventor to name
his own reward. “I have a modest request,” says the inventor. “Place one grain of rice on the
first square of the chessboard, two on the second, four on the third, and so on, multiplying
the number of grains by 2 each time.”
“Done,” says the king, and orders his vizier to make arrangements to pay the inventor his
reward. But long before the 64th square, it becomes clear that the reward can’t be paid—
there is not enough rice in the whole kingdom to satisfy the debt.
If all the rice for the reward were gathered into an enormous cube, how big
would the cube be?
In some versions of the story it is then revealed that the inventor is the god Krishna
himself. That (it is said) is why rice pudding (paal paysam) is traditionally served to visiting
pilgrims—the king is still repaying his debt to Krishna.
3.4.4 Doubling Time and Half-Life

If a stock is undergoing exponential growth, then eventually (if we wait long enough) it will double in
size.
Definition 3
The doubling time of an exponentially growing stock is the length of time it takes to double
in size.
After another doubling time, the stock will have doubled again, becoming four times its original
size. After another doubling time, it will be eight times its original size, and so on. The doubling time
is an important piece of information as we try to understand how rapidly a stock is growing under an
exponential model.
Example 6: In the E. coli model of Example 3 on page 154, the doubling time was 20 minutes. This
short doubling time explains the results we obtained when we tried to calculate the growth over a
24-hour period (Question 2 on page 166).
It was easy to see the doubling time in this example, because we were given this information
directly: “each bacterium divides in two every 20 minutes.” What if we were given a different growth
rate?
Problem 7: Estimate the doubling time of the U.S. population at the growth rate calculated in
Question 5 on page 170.
Solution: From the law of exponentials, the population after N years is
SN = (1 + r)N S0 ,
where r = 0.0047 is the rate constant and S0 is the initial population (320 million). To find the doubling
time we need to find the value of N that makes SN = 2S0 , that is, (1 + r)N = 2. Let’s try a few values
of N. First we try N = 100:
1.0047100 ≈ 1.598.
So 100 years is too few. How about 150 years?
1.0047150 ≈ 2.021.
This is pretty close, but it is a little too many. If we try numbers just less than 150, we soon find that
1.0047147 ≈ 1.992, 1.0047148 ≈ 2.002.
The value of 148 years seems to be the closest.

It wold be very painful if we had to go through this trial and error process whenever we needed to
work out a doubling time. Fortunately, there are mathematical ways to simplify the process.
Rule 2: Doubling Time Approximation
• If a process is growing exponentially with rate constant r, the doubling time is

approximately
70%
time steps
r
Problem 8: Use the doubling time approximation to estimate the doubling time of the U.S. population
(at the growth rate calculated in Question 5).
Solution: The rate constant is 0.47%, so the doubling time approximation gives
70%
≈ 149 time steps (years).
0.47%
This compares well with the figure of 148 years that we arrived at through trial and error.
As decimals, 70% = 0.7 and 0.47% = 0.0047. We could also have done the calculation this way:
0.7
≈ 149 time steps (years).
0.0047
You can use whichever approach you prefer, but be sure to have the same representation in the
numerator as in the denominator—either both percentages, or both decimals.
Remark 7: The doubling time approximation is what it says, an approximation. It is more accurate the
smaller r is. For (positive) values of r less than 10%, the approximation is accurate to within 5% of the
actual doubling time. In Section 4.1 we will see how to use logarithms to make an exact calculation.
Think about it. . .
Al Bartlett (1923–2013) was professor of physics at the University of Colorado. Dr. Bartlett
said
The greatest shortcoming of the human race is our inability to understand the
exponential function.
What do you think he meant? Do you agree? Watch the video of Dr. Bartlett’s lecture [44]
to learn more.
In problems involving exponential decay, the concept that corresponds to the doubling time is the
half-life.
Definition 4
The half-life of an exponentially decaying stock is the length of time it takes to be reduced
to half its original size.
Problem 9: Estimate the half-life of radon-222, given that 17% of it undergoes radioactive decay each
day.
Solution: The rate constant is r = −0.17, and the half-life will be the number N of days such that
(1 + r)N = 0.83N = 0.5.
Let us make a table of possibilities:

N 1 2 3 4 5
(1 + r)N 0.83 0.69 0.57 0.47 0.39
The half-life is between 3 and 4 time steps (days). We’ll see in Section 4.1 that we can use the method
of logarithms to calculate the half-life as 3.7 days.
Corresponding to the doubling time approximation there is a half-life approximation:
Rule 3: Half-Life Approximation
• If a process is decaying exponentially with rate constant r (which will be negative

because this is a decay process), the half-life is approximately
−70%
.
r
This is a positive quantity, obtained by dividing one negative number by another.
For instance, if we apply this approximation to the radon gas example, we estimate the half-life to
be
−70%
≈ 4.1 days.
−17%
(Notice that the minus signs cancel.) This is a bit bigger than the correct value of 3.7 days, but it is
still a reasonable approximation.
Remark 8: (Compare Remark 7 on page 173.) For a (negative) rate constant r that is greater than
−10%, the half-life approximation is accurate to within 7% of the actual half-life. As with doubling
times, we’ll see in Section 4.1 how to carry out an exact calculation, using logarithms.
Summary of Ideas: Exponential Models
• An exponential model is one in which the net flow into a stock is proportional to the
level of that stock.
• The key parameters of an exponential model are the time step T and the rate
constant r. The rate constant is the fraction
Net inflow in one time step
r= .
Stock level
• The law of exponentials says that in an exponential model, the stock after time t is
equal to
(1 + r)(t/T ) × Initial stock.
• An exponential model exhibits growth if r > 0 and decay if r < 0.

• The doubling time in an exponential growth model is the time that it takes for the
stock to double. It is approximately equal to 0.7/r time steps.
• The half-life in an exponential decay model is the time it takes for the stock to be
halved. It is approximately equal to −0.7/r time steps.
3.1. NETWORKS AND CONNECTIONS C
1. Refer to the network depicted in Figure 5 on

page 132. B F H
(a) Find the distance between the top left and bot-
tom right blue nodes.
(b) Find the distance between the bottom left and A E G
top right blue nodes.
D
2. Consider a network having seven nodes, numbered
1–7, and with links connecting two nodes if their Figure 38: Network for Exercises 6, and 7, and Exercise 23
numbers differ by 3. on page 178
(a) Draw a picture of the resulting network.
(b) Is the network connected? If not, how many
components does it have? 7. For the network depicted in Figure 38, calculate
(c) What is the distance between node 1 and node the clustering coefficient for nodes B, E, and G.
7?
8. Refer to the network depicted in Figure 39.
3. Consider a network having seven nodes, numbered (a) Which nodes, if any, exhibit triadic closure?
1–7, and with links connecting two nodes if their (b) Calculate the clustering coefficient for nodes A,
numbers differ by either 3 or 4. B, and G.
(a) Draw a picture of the resulting network.
(b) Is it connected? If not, how many components A B C
does it have?
(c) What is the distance between node 1 and node
3?
D E
4. Does the network of Exercise 3 exhibit triadic

closure? What is the clustering coefficient of a typical
node? F G
Figure 39: Network for Exercise 8 and Exercise 25 on

5. A hub and spoke network (as used by many page 178
airlines) has one central node (the hub) linked to a
number d of peripheral nodes (the spokes), with no
links between one spoke and another. What is the 9. According to the book The Serengeti Rules [59],
clustering coefficient of the hub, and of each spoke, in the elimination of the disease rinderpest by vaccination
such a network? What is the distance from one spoke of domestic cattle affected the giraffe population of
to another? the Serengeti by the following network mechanism:
rinderpest kills wildebeest; wildebeest eat grass; high
6. Refer to the network depicted in Figure 38. densities of grass encourage fires; fires burn down
trees; the upper parts of trees provide food for giraffes.
(a) Which nodes are neighbors of both node B and
Would you expect a reduction in rinderpest to increase
node D?
or decrease the giraffe population? (See Figure 7.3
(b) Find the degree of each node.
in [59].)
(c) Which nodes, if any, exhibit triadic closure?
10. One way of defining the notion of “bridge” 14. A complete network is one in which there is a
in a network is the following: a link between two link between every pair of nodes. Find a formula for
nodes A and B is a bridge if its unavailability would the resilience measure β of a complete network with
significantly increase the distance from A to B. (The n nodes (assume that each link has strength 1 in both
meaning of “significantly” is for you, the user, to directions). How big must n be in order that β ≥ 6?
determine. For example, in Tolkien’s The Lord of the
Rings [323, Book I, Chapter 5], the “unavailability” 15. Consider a directed network having 6 nodes,
of the Bucklebury Ferry forced the Black Riders on a numbered 1–6, and having a (directed) link from node i
20-mile detour, which is surely “significant,” since it to node j if i < j (so, for instance, there are links from 1
lets the next 800 pages happen!) Examine the network directed to 2, 3, 4, 5, and 6, and there are links directed
in Figure 10 on page 136 from this point of view, to 6 from 1, 2, 3, 4, and 5. Find the resilience measure
and identify the bridges. What are the lengths of the β.
“significant” detours one has to make if these are
unavailable? Does this notion of “bridge” correspond
16. A network has five nodes A,B,C,D, and E, and is
well with your intuition? Why or why not?
complete (every pair of nodes is linked). Each link has
a strength, possibly different in one direction from the
11. Imagine that a network, with strong and weak strength in the other. The strengths are listed in the table
links, obeys the following version of the triadic closure below (for instance, the link from A to B has strength 1,
property: if AB and AC are strong links, then there must and the link from B to A has strength 0.2),
be some kind of link (either strong or weak) between B
Node A B C D E
and C. Suppose also that every node in the network has
A 1 0.8 0.7 0.1
at least two strong links to other nodes. Explain why B 0.2 0.7 1 1
every bridge in the network (see Exercise 10) must be C 0.5 0.7 0.4 0
weak. (For the purpose of this exercise, “significantly D 1 0.8 0.6 0.9
increase the distance,” in Exercise 10, can be taken to E 0.2 0.2 1 1
mean “increase the distance to 3 or more.”) This is
one theoretical underpinning of Granovetter’s observa- Compute the resilience measure β from this data. Does
tions about the strength of weak ties. See Section 3.2 the network meet the threshold (approximately β ≥ 6)
of [100]. for resilience?
12. Refer to the network shown in Figure 40. 17. (Class Project) If you are studying this book in
a class, work together to make a drawing of the friend-
(a) Find the clustering coefficients of the red, blue,
and green nodes. ship network of your entire class (ignoring connections
with people outside the class), and apply the concepts
(b) Identify any bridges (in the sense of Exercise 11: of this section to investigate this network. Can you find
links whose removal would leave their end-
examples of triadic closure, bridges, components, or
points at least distance 3 apart).
giant components? What is the greatest distance you
can find between two people who are connected?
3.2. NETWORKS AND BEHAVIOR
18. Identify some examples of the “information

deficit” model of science communication in recent
Figure 40: Network for Exercise 12
news stories. Take one of these stories as an example,
and try to recommend a strategy for communicating the
message that takes greater account of the networked
13. Compute the resilience measure β for the aspect of understanding that we have reviewed in this
networks in Figures 38 and 39. chapter.
19. We introduced this section with the example 23. Apply the same rule used in Exercise 22 to
of vaccination. In fact, network theory provides an Figure 38 on page 176. What happens if C and G speak
important tool for modeling vaccination and disease out? What happens if C and E speak out?
transmission (see [100, Chapter 21] for instance),
though the combination of techniques needed for this 24. Consider Exercise 23 again, but imagine changing
is beyond the scope of this book. Consider, though, the the decision rule: each person will speak out if at least
following simple fact: in an unvaccinated population, a fraction f of their neighbors speak out (0 < f <
each measles patient remains infectious long enough to 1). Investigate what will happen in the examples of
give the disease to 12–18 more individuals. To ensure Exercise 23 for varying values of f .
that epidemics don’t develop and spread, it is necessary
to make sure that the average number of people infected 25. Apply the same rule used in Exercise 22 to
by each patient is reduced below 1. What percentage of Figure 39 on page 176. What happens if B,C, and E
the population needs to be vaccinated to achieve this? speak out? What happens if A, D, and F speak out?
20. In a certain mathematics department, hiring 26. On October 5th, 2017, investigative journalists
decisions are made by a committee of eight people, all Jodi Kantor and Megan Twohey published accounts of
of whom like and respect one another. The committee sexual harassment allegations against Hollywood pro-
has to choose between two candidates, A and B, for a ducer Harvey Weinstein in the New York Times [176].
new professorship. The department head (who chairs There are many articles available about the impact
the committee) is reasonably sure, before the meeting, this story has had on sexual harassment and numerous
that half the members of the committee favor candidate published interviews with Kantor and Twohey about
A and half favor candidate B. However, when the the process of investigating and breaking this story. See
time comes for a decision, the head calls on each for example [210] and [142]. Follow up on this story
committee member in turn to state their preference. “I and its impact. Is it representative of the applications
prefer candidate A,” says the first committee member. of networks as discussed in Exercises 22 and 23? Why
“I prefer candidate A,” says the second member. . . and or why not?
so on through the third, fourth, and on to the eighth
committee member. Candidate A is hired on a vote of 27. Consider the rule of Exercise 22, and imagine
8 to 0. Can you give an explanation for this unexpected that it applies in the network of Figure 10 on page 136,
outcome? Is there a problem with the way different which appeared in our discussion of “the strength of
committee members’ views are represented in this weak ties.” Suppose that A, B, and E speak out. What
process, and can you suggest a way to fix it? happens? Might a result like this lead you to modify
your assessment of the strength of weak ties, perhaps
21. Work out what happens in the two bystander by distinguishing “effectiveness at communicating in-
examples of Figure 14 on page 144 if the intervention formation” and “effectiveness at motivating action”?
thresholds for B1 and B3 are changed in the following
way: B1 ’s threshold is reduced to 0.3 and B3 ’s is 28. A newly settled country is represented by a square
increased to 0.7. portion of the (x, y) plane, with 0 ≤ x ≤ 100 and 0 ≤ y ≤
100, and the sides of the coordinate squares have length
1 mile. At each lattice point (0, 0), (1, 0), and so on
22. The kind of modeling that we have done for up to (100, 100) there is a settler. Each settler at (p, q)
the bystander effect can also be applied to other kinds has a positive relationship with each of the settlers at
of (apparently) “risky” commitments, such as taking the neighboring lattice points (at most four of them,
action to counter some kind of injustice. It may well be namely (p − 1, q − 1), (p − 1, q + 1), (p + 1, q − 1), and
that everyone is aware of the unjust situation, but each (p+1, q+1). There are no unbalanced triangles among
person will only have the courage to speak out once the settlers’ relationships.
they see that at least half of their neighbors are speaking (a) Give an example to show that it is possible for
out. Suppose that this rule applies in the network of some negative relationships to exist under the
Figure 1 on page 127. If D speaks out, what happens? stated conditions.
If B speaks out, what happens? (b) Suppose that each settler has a relationship (pos-
itive or negative) with every other settler within
10 miles. Explain why all these relationships labels, in terms of the model. A generic description
must, in fact, be positive. such as “a + sign indicates that an increase in the
stock increases the outflow” does not provide meaning
specific to the model.
29. A network with positive and negative edges
is called weakly balanced if it lacks the first kind of
red triangle in Figure 17 on page 148 (the “+ + −” 35. The stock-flow diagram in Figure 42 on the next
kind), but it may have red triangles of the “− − −” page depicts relationships between Earth’s tempera-
kind. (If you like, the principle “my enemy’s enemy is ture and stocks of atmospheric methane and methane
my friend” no longer has to apply.) Give an example trapped in frozen soils (permafrost). Label each flow
to show that a weakly balanced complete network arrow and control arrow with the appropriate sign,
does not have to be divided into two factions—in identify any feedback loops, and indicate whether each
other words, the conclusion of Harary’s theorem does one is amplifying or stabilizing.
not apply to weakly balanced networks. There is a
generalization of Harary’s theorem that applies to the 36. Refer to Figure 25 on page 158. Each arrow is
weakly balanced case, though: can you guess what it labeled with a + or − sign. Describe in your own words
is? the meaning of these labels, in terms of the model.
37. Atmospheric carbon dioxide combines with rain-

water to form carbonic acid that can dissolve surface
3.3. FEEDBACK AND DYNAMIC
rocks. This rock-weathering process transports raw
NETWORKS materials for shell-building marine organisms to the
oceans. Through sedimentation these shells eventually
30. Refer to the bacteria stock-flow model in Fig- become limestone. This process of chemical weather-
ure 21 on page 154. Provide appropriate labels for the ing and biosequestration, which takes place on time
flow and control arrows and explain the meaning of scales of hundreds of thousands of years, is depicted
these labels in your own words. in the stock-flow diagram in Figure 43 on the next
page. An increase in temperature tends to accelerate
this process of biosequestration.
31. The stock-flow diagram in Figure 41 on the next
Label the flow arrows and control arrows in the dia-
page shows the relationship between Earth’s tempera-
gram with the appropriate signs and identify a feedback
ture and the stock of ice and snow cover. Label each
loop. Is this feedback amplifying or stabilizing?
flow arrow and each control arrow with the appropriate
sign, identify any feedback loops, and indicate whether
38. The process of biosequestration described in
each one is amplifying or stabilizing.
Exercise 37 is part of the long-term carbon cycle.
(Refer back to Question 1 on page 76 for discussion
32. During the early phase of adoption of a new on the difference between the short, 4-year residence
product or technology, word-of-mouth may be a signif- time of atmospheric carbon and the long-term impacts
icant contributor to the product’s expanding use. Make associated with human-caused carbon dioxide emis-
a sketch of a stock-flow model that depicts this. Use sions from the burning of fossil fuels.) Follow up on
the existing users of the product as a stock. Identify the long-term carbon cycle at [208]. Describe the long-
any feedbacks as amplifying or stabilizing. term carbon cycle in your own words. Can you put the
timescale involved in human terms?
33. Make a sketch of a stock-flow diagram for the
money in a checking account. Include control arrows 39. During a disease epidemic, as the number of
for how the stock affects each flow rate. Label each ar- people infected with the flu increases, it becomes
row with the appropriate sign. Explain how the diagram more likely that uninfected individuals will come into
depends on whether or not the account earns interest. contact with a contagious person. Draw a stock-flow
diagram using the number of individuals who are
34. Refer to the caffeine model in Figure 24 on infected with the disease as a stock. Assume that after
page 157. Each arrow is labeled with a + or − sign. a period of time, a person who is infected recovers, and
Describe in your own words the meaning of these include recovery from the disease as an “outflow.” Use
Glaciers and
Melting
Ice Caps
Absorbed Global
Sunlight Temp
Figure 41: Stock-flow model for Exercise 31 on the previous page; the ice-albedo feedback.
Methane in Melting Atmospheric

Frozen Soils Methane
Short Wave Long Wave

Global Temp
Radiation Radiation
Figure 42: Stock-flow model for Exercise 35 on the previous page; the Arctic methane release feedback.
Atmospheric CO2 Biosequestration Limestone
Short-Wave Long-Wave
Global Temp
Radiation Radiation
Figure 43: Stock-flow model for Exercise 37 on the previous page; the chemical rock-weathering and biosequestration
feedback.
control arrows to show how the stock of individuals What do you think accounts for the difference between
with the disease affects both the inflow and outflow to savings rates and loan rates?
the stock. Identify any feedbacks and indicate whether
they are stabilizing or amplifying. 47. In Example 4 on page 165, what if y − p is
initially negative (the coffee is colder than the room,
40. Revisit the stock-flow diagram from Exercise 39 for instance if we started with a chilled latte, Remark 1
on page 179. Suppose that an effective vaccine for the on page 165)? Is y − p still governed by the same
disease becomes available. Add a stock for vaccinated exponential decay model? Why or why not?
individuals to the model and use control arrows to
indicate the effect this stock has on other flows in the 48. Phosphate rock is the primary source for phos-
model. Suppose that information about the availability phorus fertilizers in agriculture. Mining of phosphate
of the vaccine spreads by word of mouth. Indicate this rock has grown at an average of 3% per year throughout
with an appropriate control arrow in the model. most of the last century, and 160 million metric tons of
phosphate rock were mined in 2010 (globally). If the
41. Refer to the population model shown in Figure 28 mining of phosphate rock were to continue to grow at
on page 160. Each arrow is labeled with a + or − 3% per year, approximately what quantity of phosphate
sign. Describe in your own words the meaning of these rock would be mined in the year 2050?
labels, in terms of the model.
49. Referring to the previous question about the
42. Refer to the lynx–hare model in Figure 30 on mining of phosphate rock, calculate the doubling time
page 160. Each arrow is labeled with a + or − sign. for the amount of phosphate rock mined per year,
Describe in your own words the meaning of these assuming the same growth rate of 3% per year.
labels, in terms of the model.
50. A town with a once popular but now polluted
43. Refer to reindeer model depicted in Figure 34 lake takes steps to halt the flow of phosphates from
on page 163. Each arrow is labeled with a + or − agricultural runoff into its public lake. Once the inflow
sign. Describe in your own words the meaning of these of the pollutant is stopped, the level of pollution in
labels, in terms of the model. the lake declines at a rate of 1.6% per month through
processes of sedimentation and outflow of water from
44. Water vapor in the atmosphere is responsible for the lake. If the amount of phosphate pollution in the
70% of Earth’s greenhouse effect, and a warmer atmo- lake was 50 metric tons before the inflow of pollution
sphere can hold more water vapor. Make a sketch of a was stopped, how much phosphate pollution remains in
stock-flow model depicting the water vapor feedback. the lake after two years?
Identify relationships between the global temperature
and water vapor in the atmosphere. Is this feedback 51. If I place a hot cup of coffee on a table, the
amplifying or stabilizing? temperature of the coffee, Tc , will approach the room
temperature, T0 , over time. The rate of heat flow from
the coffee to the surrounding room is proportional to
temperature difference, Tc − T0 .
3.4. THE EXPONENTIAL MODEL Suppose the temperature of the coffee decreases 1
degree F per minute for every 5 degrees in temperature
45. Research what rates of interest are available on a difference.
balance of $1000 deposited for five years. What are the (a) What quantity is being governed by an exponen-
rate constants and time steps? tial model?
(b) Is this quantity undergoing exponential growth
46. A loan can be thought of as something like or exponential decay?
a savings account but with a negative balance in it
(c) What is the time step?
(representing the fact that you owe the bank money,
rather than the bank owing you). Research what rates (d) What is the rate constant?
of interest are available on a five-year loan of $1000.
52. A short while after drinking a cup of coffee you

have some stock of caffeine in your bloodstream. As 55. The body removes elemental mercury (the form
your body metabolizes the caffeine, it declines at a rate of mercury in older thermometers, dental fillings, or in
proportional to the stock level. Your body metabolizes a compact fluorescent light bulb) from the bloodstream
about 13% of the caffeine in your bloodstream each at a rate of 0.011 milligrams per hour for each mil-
hour. Find the rate constant, r, and time step for the ligram of mercury in the blood. Estimate the half-life
exponential model and determine the half-life for the for elemental mercury in the bloodstream.
level of caffeine in your bloodstream.
56. It was reported recently that China’s rate of eco-
53. Suppose that with no heat source the temperature nomic growth has fallen to 6.9% per year. If China were
of your house will decline 1 degree Fahrenheit per able to continue at this growth rate for the next half-
hour for each 10 degrees Fahrenheit in temperature century (50 years), by what factor would the Chinese
difference with the outside air. The power goes out economy expand over that time?
on a winter night when the outside temperature is at
the freezing point (32 ◦ F), leaving you with no heat 57. Making an assumption about the number of
source until the power is restored, 3 hours later. If the years in a generation, use an exponential model to
house temperature was 72 ◦ F when the power went out,
estimate the number of people from 1000 years ago
what was the temperature of the house when power was
who are in your biological family tree. How does this
restored? compare with the estimated population of the time
(300 million people)? If we make the (unreasonable)
54. U.S. energy consumption has grown at an average assumption that populations 1000 years ago were not
rate of 2.9% per year since the year 1650. Calculate the geographically isolated, about how many times, on
doubling time of U.S. energy consumption assuming average, does each person alive 1000 years ago appear
this growth rate remains constant. in your family tree?
C HAPTER 4
Changing
What does the word growth or growing suggest to you?
• Little kids growing up to be bigger kids, and eventually to be adults?

• The family contracting business that grows from “two guys and a truck” to a company with a
dozen employees and a value of millions of dollars?
• Agriculture, as saplings planted in a field grow over ten years or so to mature peach trees that
yield a harvest each summer (Figure 1)?
• Urban sprawl, as the apparently relentless growth of suburbia swallows up the peach orchard
for the sake of a few dozen big houses on big lots?
• Or that heart-stopping moment when the doctor says, “I’m sorry to tell you we have found a
growth in your liver”?
Growth can be good or bad: it depends what is growing, and

how big it is relative to its surroundings. At the end of the last
chapter we began to look at one of the simplest mathematical
models that can be used to describe growth—the exponential
model. You might remember from the introduction that many
quantities that measure human activity and impact on the world
around us seem to have followed this simple growth model, at
least for the last couple of hundred years. This is a fact that
causes concern to many people, like Gus Speth [305] whom we
quoted before:
The remarkable charts [shown in Figure 1 on
page vi] reveal the story of humanity’s impact on the
natural earth. The pattern is clear: if we could speed Figure 1: A peach.
up time, it would seem as if the global economy
is crashing against the earth—the Great Collision.
And like the crash of an asteroid, the damage is enormous.
But exponential growth models are not the only models of growth that exist; nor does the fact that
some quantity has followed an exponential trajectory in the past mean that it is sure to do so in the
future. In this chapter we will begin (Section 4.1) by taking a closer look at the exponential model,
using a more sophisticated mathematical tool (the theory of logarithms) than we had available before.
Then, in Section 4.2 we’ll think about part of our everyday experience of growth that the exponential

https://doi.org/10.1007/978-3-319-76660-7_4
184 CHAPTER 4. CHANGING
model doesn’t account for at all: growth eventually slows down and stops. (The two-foot toddler
may become a six-foot adolescent, but the six-foot adolescent does not become a thirty-foot adult!)
We’ll develop some simple modeling tools that allow for limits on growth, and test them out in real-
world situations. In Section 4.3 we’ll tie these ideas back into our discussion of feedback loops in
the previous chapter, and then in Section 4.4 we’ll use our new methods to understand the crucial
idea of a tipping point: an abrupt transition from one state to another, brought about by an apparently
small change in some crucial parameter. Many different models can exhibit tipping point behavior,
and we’ll give several examples, including the very real possibility of tipping points in the Earth’s
climate system itself.
To be clear, our world is so complex that it is highly unlikely that any of the models that we’ll
discuss captures all the relevant effects; every model is an oversimplification, as we have already said.
The point is that the models in this chapter give us a repertoire of ways of thinking, or of patterns that
we may recognize. The better stocked our understanding is with such patterns, the better able we will
be to recognize what the world may have in store for us next, and the more prepared we will be to
respond.
4.1. LOGARITHMS AND CHANGE 185
4.1 Logarithms and Change
Objectives
I understand the difference between relative and absolute change.

I know how to use a calculator to find the logarithm of a number.
I know how to use a calculator to find a number with a given logarithm, that is, find
an antilogarithm.
I understand that logarithms convert multiplication to addition.
I can use logarithms to investigate whether a series of data follows an exponential
model.
I can use logarithms for personal finance calculations about saving and borrowing.
I can calculate the exponential reserve index for a resource.
I understand how Arrhenius’s law describes the strength of the greenhouse effect in
terms of greenhouse gas concentrations.
In this unit we are going to learn about the mathematical function called the “logarithm” (or “log”
for short) and its many applications. These include measuring relative change in some quantity,
simplifying calculations involving exponential growth and decay, and expressing the strength of the
greenhouse effect related to a given concentration of atmospheric carbon dioxide.
4.1.1 Relative and Absolute Change
According to data from the United Nations Food and Agriculture Orga- Year Benin Brazil
nization [109], the forested areas (in square miles) in the two countries of 1990 22,640 2,150,000
Benin and Brazil in 1990 and 2015 were as shown in Table 1. Which country 2015 17,150 1,946,000
suffered the greater amount of deforestation in that period? Table 1: Forested areas in two coun-
In one sense, this is an easy question to answer. We can subtract the 2015 tries (square miles).
figure from the 1990 one to find that Benin lost 5,490 square miles of forest
during the period, while Brazil lost 204,000 square miles—over thirty times more. Case closed: the
answer is Brazil, right? Or are you not so sure?
If you are not so sure, you might be thinking like this. That 5,490 square mile loss for Benin
represents almost 25% of its 1990 forest area: three more losses like that, and its forest would be
gone completely. But Brazil’s much larger loss comes from a very much larger initial area (2,150,000
square miles in 1990) and represents less than 10% of it: after nine more similar losses Brazil would
still have some forest left. From this perspective, one could say that Benin’s deforestation, while not
amounting to the same area, was much more impactful than Brazil’s was.
The distinction we’re getting at here is that between absolute and relative change. Let’s define
these notions.
Definition 1
Suppose that Q is some quantity that changes, and we want to measure the size of that
change. We let Q0 stand for the initial value of the quantity (before the change) and Q1
stand for its final value (after the change). Then:
• The absolute change in Q is simply the difference, Q1 − Q0 , which is also sometimes
denoted by ∆Q (the symbol ∆ stands for “difference”).
• The relative change in Q is a measurement that expresses the amount of change in Q
in proportion to the initial value Q0 . Thus
∆Q Q1 − Q0 Q1
Relative Change = = = − 1.
Q0 Q0 Q0
Notice that while the absolute change is a quantity with a unit (hectares in the forestry example
above), the relative change is a pure number (Section 1.1.4), which we may express as a percentage
if we want (25 percent and 9 percent in the forestry example). Later in this section we will briefly
discuss an alternative way of expressing relative changes that involves logarithms.
Which way of measuring change is better? That depends on the purpose of the measurement.
Consider the following examples.
Example 1: In the forest area example above (Table 1 on the previous page), Brazil is simply a much
bigger country than Benin. So, it is reasonable to expect all quantities related to its forestry to be
bigger, including both the absolute areas forested and the changes in those areas over a fixed period
of time. Using the relative change permits a “like for like” comparison between Benin’s forestry
and Brazil’s. When we use absolute numbers, this information is swamped by the difference in size
between the two countries.
Example 2: The yield of soybeans on a certain test plot is 51 bushels per acre. An adjacent plot,
similar in all respects, is planted with a new, genetically modified variety of soybean and yields 57
bushels per acre. The absolute difference is 6 bushels per acre; the relative difference is
57 − 51 6
= ≈ 0.12,
51 51
or about 12 percent. The relative difference is probably the most effective way to describe things here
(“planting the GM crop increased yields by 12 percent”). Many factors influence crop yields (soil
quality, growing season, orientation to sunlight, and so on), and reporting the percentage change gives
a measure of expected improvement that may apply in all these different circumstances. One would
want to carry out experiments with other test plots in different locations, though, before making a
confident claim of a 12 percent improvement across the board.
Think about it. . .
In the 25-year period between 1990 and 2015 the world’s total forested area declined by
500,000 square miles, an area about twice the size of Texas. However, the rate of forest loss
has also declined over the same time period; the annual rate of forest loss is now less than
half what it was in the early 1990s. Do you think this is due to successful efforts in forest
conservation, or is it simply because there’s less forest left to be lost? Look at the Food and
Agriculture Organization’s Global Forest Resource Assessment to learn more [115].
Example 3: Between 2014 and 2015, U.S. natural gas usage increased by 3 percent, coal usage
decreased by 12 percent, oil usage increased by 2 percent, wind increased by 5 percent (and we could
list similar statistics for other energy sources also). By what percentage did America’s overall energy
usage change? This question cannot be answered from the information provided, demonstrating a
limitation of the concept of relative change. By intent, relative changes are scaled (in this case to
the size of the energy sector they represent), and therefore they cannot be added as they stand. In
any situation in which you need to add together a number of changes coming from different sources,
you need to work with the absolute numbers (though once you have finished with your calculations,
you can express the result as a relative change if you like). Just as a matter of interest, U.S. overall
energy consumption decreased very slightly between 2014 and 2015—almost entirely as a result of
the greater efficiency in electricity generation coming from the use of natural gas instead of coal.
Example 4: In a February 2017 article, Vox reported that the availability of avocados in the U.S. had
“surged by 1342 percent.” According to the USDA study that the article was based on, the per capita
availability of avocados was a half pound of avocados per year in 1970. By 2014 this had risen to 6.5
pounds per year [46, 48]. The change in avocado availability over this time was 6 pounds per person
per year, which represents an increase of 1200%:
6lb
= 12 = 1200%.

0.5lb

There’s a trick at work here that we need to be wary of. Quantities that are rather small to begin with
can exhibit large relative changes that aren’t necessarily all that significant. Americans are eating a lot
more avocados now than they were in 1970, but the same USDA report also notes that the per capita
sugar consumption in the U.S. is 130 pounds per year, itself a rise of 10% since 1970. If you are
an avocado grower, the “1342 percent surge” in U.S. avocado consumption, though surely claiming
excessive precision (see Section 1.3.1), will be an important reference point for your business. But
if you are a dietitian, the entire U.S. consumption of avocados is dwarfed by our increase in sugar
consumption, and it is the latter that is more likely to demand your attention, even though in relative
terms it is much smaller.
4.1.2 Logarithms
Your calculator most likely has a button called log (it might also be labeled log10 ). Let’s carry
out some experiments with that.
• Enter 1 and press the log button. You should get the number 0 (zero).
• Enter 10 and press the log button. You should get the number 1.
• Enter 2 and press the log button. The result you get will depend on how many decimal places
your calculator shows, but it should be close to 0.30103.
Using your calculator, fill in the rest of the entries in the table below:
Number 1 2 3 4 5 6 7 8 9 10
Log 0 0.30103 0.47712 0.60206 ? 0.77815 ? 0.90309 ? 1
There are some curious patterns in this table. For example, it is rather clear that log 4 = 2 × log 2 and
that log 8 = 3 × log 2. Moreover, log 6 = log 2 + log 3. What is going on? Once you have filled in the
rest of the table, try to find some more patterns like this.
The patterns that you notice result from the two laws of logarithms, which are these.
Rule 1: Laws of Logarithms
• Law 1: Logarithms turn multiplication into addition: for positive numbers a and b,
log(a × b) = log(a) + log(b).
• Law 2: The logarithm of a power xy (for positive x) is y times the logarithm of x:
log(xy ) = y × log(x).
Example 5: From the calculator, log 12 ≈ 1.07918 and log 15 ≈ 1.17609. Thus the first law of
logarithms tells us that log(12 × 15) ≈ 2.25527. Indeed, 12 × 15 = 180, and you can check that
log 180 ≈ 2.25527, as expected.
Remark 1: Logarithms were invented by John Napier in Scotland in the early

1600s. At that time, calculations were carried out by hand, not by machine.
As you probably remember from grade school, if you have to calculate by
hand, addition and subtraction are much easier and quicker to carry out than
multiplication and division. Napier’s invention, therefore, which allowed one
to multiply numbers by first looking up their logarithms in a table and then
adding the corresponding logarithms, was a practical device for speeding
up calculations—especially the complicated calculations needed to navigate
across the oceans and around the globe.
To carry out Napier’s plan of using logarithms to reduce multiplication to
addition, one needs two processes: to find the logarithm of a given number,
and the reverse—to find what number a given number is the logarithm of.
For instance, in Example 5 above, we found the logarithms of 12 and 15 and
Figure 2: John Napier, 1550–1617. added them to get approximately 2.25527. If we were following Napier’s
procedure, the next question would be to find what number has 2.25527 as its
logarithm. Traditionally, this is called finding the antilogarithm of 2.25527,
and the key for this operation on your calculator may be marked antilog , but it is more likely
to be marked 10x . If you enter 2.25527 in your calculator and press that key, you will get a result
very close to 180—the slight error arises because we rounded off the values of our logarithms to five
decimal places.
Problem 1: Use logarithms (working to five decimal places) to multiply 1.5674 by 17.419 by Napier’s
method. Compare the result with the exact answer given by a calculator.
Solution: The result is shown in the table below.

Number Logarithm
1.5674 0.19518
17.4190 1.24102
27.3025 1.43620
The two numbers that we are originally given are in the left column, their logarithms in the right
column. We add those two logarithms, obtaining 1.43620. Taking the antilogarithm of that (using the
10x key) gives 27.30254 to 5 decimal places. The exact answer is 27.3052406.
Problem 2: Find a number y for which 2y = 128.
Solution: If we take the logarithm of each side of the equation, 2y = 128 becomes
log 128
y × log 2 = log 128, so y = .
log 2
We know that log 2 ≈ 0.30103, and the calculator tells us that log 128 ≈ 2.10721, so y ≈ 2.10721/0.30103 ≈
7.00000. In fact, we can check that
27 = 2 × 2 × 2 × 2 × 2 × 2 × 2 = 128.
| {z }
7 times
250
Problem 3: Find a number s for which 3s = 160.

200
Solution: Again taking logarithms, 3s = 160 becomes 160

150
3x
log 160 100

s × log 3 = log 160, so s = .
log 3 50
The answer we calculate now is 0

0 1 2 3 4 5
2.20412 x
s≈ ≈ 4.62. 4.62
0.47712
Figure 3: Graph of y = 3x .
This example is different from the previous one in that the exponent s
that we get is not a whole number. (And that is to be expected: 34 = 81
and 35 = 243, so the best we can say using whole numbers for 3s = 160 is “s is somewhere between
4 and 5.”) But the value we obtain for s still makes sense: if we join the points (n, 3n ) (that is, (1, 3),
(2, 9), (3, 27), (4, 81), (5, 243), (6, 729), and so on) by a smooth exponential curve, that curve will
pass through the point (4.62, 160) (see Figure 3).
Logarithms and Relative Change

Recall from Section 4.1.1 that the absolute change in some quantity is measured by subtraction,
whereas the relative change is measured by division. Because logarithms convert multiplication into
addition, they convert division into subtraction, and therefore convert relative changes into the more
familiar idea of absolute changes. To make this precise, let’s make another definition.
Definition 2
Suppose that Q is some quantity that changes, and we want to measure the size of that
change. We let Q0 stand for the initial value of the quantity (before the change) and Q1
stand for its final value (after the change). Then the logarithmic change in Q is
Q1

2.3 × log = 2.3 × log(Q1 ) − log(Q0 ) .

Q0
It is expressed in nepers (whatever the units of Q may be).
The word “neper” is intended as a tribute to John Napier. You might also see “centinepers” used: a
1
centineper is 100 of a neper.
Because it depends on the ratio Q1 /Q0 , the logarithmic change is just another way of expressing
the relative change. Its advantage is that, because of the laws of logarithms, the effect of successive
changes can be computed by ordinary addition. Thus, if a quantity increases by 0.38 nepers, and then
further increases by 0.48 nepers, the overall increase is easy to figure:
0.38 + 0.48 = 0.86 nepers.
The corresponding calculation for relative changes (percentages) is more involved. A 0.38 neper
increase corresponds to a relative increase of 46%, that is, multiplication by 1.46. Similarly, a 0.48
neper increase corresponds to a relative increase of 62%, or multiplication by 1.62. Now we need to
multiply to calculate the overall increase factor:
1.46 × 1.62 ≈ 2.37,
corresponding to a 137 percent overall increase. We can check that the two ways of expressing things
match up:
2.3 × log(2.37) ≈ 2.3 × 0.37 ≈ 0.86 nepers,
consistent with the result of our simple addition.
Question 1: What is the point of that pesky numerical factor 2.3?
Answer: That is a good question! To answer it, let’s think about a small relative change in a quantity,
say a change of 1 percent, so that, for example, we might have Q0 = 100 and Q1 = 101. Let’s figure
out the corresponding logarithmic change:
2.3 × log(Q1 ) − log(Q0 ) = 2.3 × 2.00432 − 2 ≈ 0.01.

Thus a relative change of 1 percent corresponds to a logarithmic change of 0.01 nepers or 1 centineper.
This convenient correspondence holds for any small percentage change.1
Rule 2: Logarithmic Approximation
A small relative change in a quantity, of p percent, corresponds approximately to a

logarithmic change of p centinepers.
1 The approximation involved is the same as in the doubling time and half-life approximations, Section 3.4.4
Because logarithms convert multiplication into addition, we can figure the effect of a succession of
relative changes just by adding the corresponding logarithmic changes.
Example 6: The Earth’s current human population is about 7 billion, and its relative increase (growth
rate) is about 1.1% every year. If this trend were to continue, let’s figure what the population would
be a thousand years from now. We could solve this problem using the law of exponentials (Rule 1 on
page 168) but as an example, we will use the logarithmic approximation and the idea of logarithmic
change instead.
From this approximation, the logarithmic increase in the Earth’s population is about 1.1 centinepers,
or 0.011 nepers, per year. In a thousand years at this rate, then, the increase would be 1000 × 0.011 =
11 nepers. This gives us the equation
2.3 × log(Q1 ) − log(Q0 ) = 11.

Here Q0 is the Earth’s current population of 7 billion and Q1 is the future population that we want to
find. We can rearrange the equation using algebra. First divide both sides by 2.3:
11
log(Q1 ) − log(Q0 ) = ≈ 4.78.
2.3
Now add log(Q0 ) to both sides, and find its value using a calculator:
log(Q1 ) ≈ log(Q0 ) + 4.78

≈ 9.85 + 4.78 = 14.63.
Finally, we find the antilogarithm (that is, 1014.63 ):
Q1 ≈ 4.3 × 1014 .
Continuing our present human population growth rate (which, by the way, is much less than the peak
rate of well over 2% seen in the 1960s) yields the inconceivable figure of 430 trillion humans on the
planet a thousand years from now.
4.1.3 Logarithms and the Exponential Model

The law of exponentials (Rule 1 on page 168) says this: Suppose that we have an exponential process
with time step T and rate constant r. Suppose that the initial stock is S0 at time zero. Then, after time
t has elapsed, the stock will have become
S(t) = S0 × (1 + r)(t/T ) .
For an exponential growth process the rate constant r is positive, and the stock grows with increasing
rapidity. The graph of S(t) against t heads upward very rapidly (as you see already in Figure 3 on
page 189), and this can make it difficult to interpret. Logarithms can help here.
Rule 3: Logarithmic Transformation
If S(t) represents the level at time t of a stock that is growing (or decaying) according
to an exponential model, then the graph of log S(t) against t is a straight line.
3.0
2.5
Log of Soybean Production

2.0
1.5
1.0
0.5
0.0
1960 1970 1980 1990 2000 2010 2020 2030
Year
Figure 4: Plot of log of soybean production.
In other words, logarithms convert the steeply growing exponential into the simplest kind of
graph—a straight line. To see this, just apply the laws of logarithms to the expression for S(t) given
above:
log S(t) = log (1 + r)(t/T ) + log S0 = (t/T ) log(1 + r) + log S0 .
The right-hand side here equals mt + c, where m = (1/T ) log(1 + r) and c = log S0 are constants, and
this is the familiar form for the equation of a straight line.
Soybean
Year Logarithm Problem 4: The UN Food and Agriculture Organization estimates
Production
worldwide soybean production for the years 1970, 1980, 1990,
1970 43.7 1.64
1980 81 1.91
2000, and 2010 to have been 43.7, 81, 108.5, 161.3, and 261.6
1990 108.5 2.04 million tonnes respectively [109]. Investigate whether this data is
2000 161.3 2.21 a good fit for an exponential model. If the trend continues, estimate
2010 261.6 2.42 global soybean production in 2020.
Table 2: Soybean production figures and loga- Solution: First we calculate the logarithms of the production
rithms.
amounts, which gives us Table 2. Then we plot a graph of these
logarithms against the year to which they apply. This gives the plot
shown in Figure 4. The five points (shown by small red squares) lie very close to a straight line,
marked on the graph. If we extend that straight line to the right, it cuts the vertical line for the year
2020 at approximately y = 2.6. That is to say, our forecast of world soybean production in 2020 is the
antilogarithm of 2.6, that is 102.6 or approximately 398 million tonnes.
Remark 2: Instead of going through two separate steps—calculating the logarithms and then plotting
them on a graph—it is possible to combine both steps together using semilogarithmic graph paper.
This has a y-axis marked with a special scale for which the distance from the origin measures the
logarithm of y, rather than y itself; the x-axis is marked in the ordinary way. When calculations were
all done by hand, semilogarithmic paper was an important time-saver. With calculators, spreadsheets,
and other electronic tools it is no longer so important. If you want to experiment with this technique,
several online sites allow you to customize and print your own graph paper in various styles, including
semilogarithmic.
Problem 5: When the first draft of this chapter was being written (early October 2014), the Center
for Disease Control was reporting that the viral disease Ebola was “uncontained” in West Africa, with
the number of cases doubling approximately every 20 days. At that time the number of known Ebola
cases was 8,000. If the virus had continued to spread at the same exponential rate, how long before
everyone on Earth would have been infected?
Solution: According to the law of exponentials, using the information provided, the total number of
cases t days from October 2014 is
8000 × 2(t/20) .
We want to know when this figure will equal the current world population of 7 billion (7 × 109 ). That
is,
7 × 109
2(t/20) = ≈ 8.8 × 105 .
8000
Taking logarithms, we get
(t/20) log 2 ≈ log(8.8 × 105 ) = log(8.8) + 5
(notice that since log 10 = 1, log(105 ) = 5). Thus
log(8.8) + 5 20 × 5.94
t ≈ 20 × ≈ ≈ 400 days.
log 2 0.3
Thus, at its then rate of growth, Ebola was on course to infect everyone in the world by the end of
2015.
The fact that we are all still here shows that this did not happen, for the same reason that the growth
in Example 2 on page 166 was not realistic: other limits slow down the exponential spread of the
disease. But the calculation does illustrate the fearsomely rapid rate at which a disease can spread,
especially if adequate medical treatment is not known or not available. For example, in the “Spanish
flu” pandemic of 1918–1920, up to five percent of the world’s entire population may have perished.
See [43].
Here is a problem that requires using an offset exponential model (Definition 2 on page 165).
Problem 6: Harry Potter and Hermione Granger visit the Three Broomsticks in Hogsmeade. Harry
orders a hot Butterbeer. The initial temperature of the Butterbeer is 168 degrees Fahrenheit, and the
ambient temperature of the Three Broomsticks is 56 degrees Fahrenheit. Each minute, the temperature
of the Butterbeer decreases by 0.11(Tb − Ta ), where Tb is the temperature of the Butterbeer and Ta
the ambient temperature of the Three Broomsticks (which remains constant). Harry Potter can drink
the Butterbeer only after its temperature has fallen below 119 degrees Fahrenheit. How long must
Hermione wait for Harry to drink up?
Solution: Let’s begin by identifying the parameters of the model. There are three.
• The rate constant r = −0.11. (We know the sign is negative because this is an exponential decay
model—the temperature of the Butterbeer approaches the “set point” given by the ambient
temperature.)
• The time step is T = 1 min.
• The set point P is the temperature of the Three Broomsticks, P = Ta = 56 ◦ F.
In an offset exponential model, the difference between the stock value and the set point behaves
exponentially. So, consider the difference D = Tb − Ta . The initial value of D at time 0 is
D0 = 168 ◦ F − 56 ◦ F = 112 ◦ F.
D is governed by an exponential model, so that after t minutes the value of D is
Dt = D0 × (1 + r)t/T = 112 × 0.89t (since T = 1 min).
We are interested in when the temperature of Harry’s beer reaches 119 degrees, that is, when
Dt = 119 ◦ F − 56 ◦ F = 63 ◦ F.
Thus we need to solve the equation

63 = 112 × 0.89t .
Using logarithms we get
log 63 − log 112
log 63 = log 112 + t log 0.89, t= ≈ 5 min.
log 0.89
Example 7: In Example 9 on page 174 we tried to figure out the half-life of radon-222, given the
information that 17% of it undergoes radioactive decay each day. We used trial and error to find that
the half-life was somewhere between 3 and 4 days; the half-life approximation rule gave the value 4.1
days, which is too large. We can use logarithms to find a more accurate answer.
According to the law of exponentials, if I start with a certain quantity S0 of radon-222, the amount
remaining after t days will be
S0 × (1 − 0.17)t = S0 × 0.83t .
We want to know for which value of t this expression becomes exactly half the original quantity, that
is S0 × 0.5. In other words we want to solve the equation
0.83t = 0.5.
Taking logarithms, this becomes

log 0.5
t × log 0.83 = log 0.5, or t = .
log 0.83
Both the logarithms appearing here, log 0.5 and log 0.83, are negative numbers (in fact, the logarithm
of 1 is 0, so the logarithm of a positive number less than 1 must be less than 0). But dividing one
negative number by another gives a positive answer:
log 0.5 −0.301
t= ≈ ≈ 3.7 days.
log 0.83 −0.081
This is how we worked out the “more accurate” answer that we gave in Example 9.
Critical Thinking
In the above example, notice that log(0.5) = − log 2. Can you see why this is a consequence
of the first law of logarithms?
We can summarize how to use logarithms to calculate half-lives and doubling times (as in
Example 7) in two rules. (In the second rule, notice that an exponentially decaying process has
negative rate constant, so 1 + r < 1, and thus log(1 + r) is negative also. The minus signs cancel
in the final result.)
Rule 4: Exact Doubling Time and Half-Life Rules
• If a process is growing exponentially with rate constant r, the doubling time is

log 2
time steps.
log(1 + r)
• If a process is decaying exponentially with rate constant r (negative), the half-life is

− log 2
time steps.
log(1 + r)
These rules are more complicated to apply than the approximations from the previous chapter
(because of the logarithms), but they are more accurate. In most situations, though, the approximations
are good enough.
Capital and Exponential Growth

As we saw in the introduction (Figure 1 on page vi), many numbers associated with human activity
appear to be growing exponentially. Partly this results from the growth of the human population—the
sheer number of people on Earth. But population growth is not the only reason for the growth of
human activity. The amount of energy that each individual human being can dispose of has also been
growing, as we have seen in Section 2.2. What has brought this about? Economists would point to the
process of capital formation as crucial. Let’s make a definition.
Definition 3
In economics, the term capital refers to assets that are used to generate income. For
example, a combine that enables a farmer to harvest wheat more quickly; a bridge that
allows goods to flow to a new market; a drug patent that allows a company to make money
from a new treatment. The process by which new capital is created is called investment,
and the process by which capital wears out or becomes obsolete is called depreciation.
Since capital generates income, the amount of income available to invest in new capital depends on
the amount of capital. This is a positive feedback loop, like the one whereby the birth rate depends on
the total population. Similarly, we expect that the rate of depreciation depends on the total amount of
capital. We obtain a stock-flow model like the basic demographic model of Figure 28 on page 160:
+ −
Investment Capital Stock Depreciation
+ +
Figure 5: Simple model for capital formation.
If we assume that income generation (and therefore the amount of wealth available for investment)
and depreciation are both proportional to the existing capital stock, then this is another exponential
Moore’s Law
1011
Transistors Per Processor

109
107
105
103
1970 1980 1990 2000 2010 2020

Date of Introduction
Figure 6: Moore’s law: transistors per microprocessor have doubled about once every two years.
model. To a first approximation, both capital and population grow exponentially. The combined effect
of these forces has led to exponential growth in many measures that quantify human abilities, or
human impacts on the environment.
Critical Thinking
In 1965, and again in 1975, Gordon E. Moore, an electronic engineer working at Fairchild
Semiconductor, was asked to make predictions for the future direction of the semiconductor
industry. In a short article, Moore proposed that semiconductor device complexity (mea-
sured by the number of transistors “crammed onto” a chip) would double roughly every
two years. As you can see in the semilogarithmic plot in Figure 6, this prediction has
held good for roughly forty years or twenty doubling periods, an astonishing stretch of
exponential growth. (Enabling this “Moore’s law” has been an exponential increase in the
capital invested by manufacturers in semiconductor factories.) How long do you imagine
that this process can continue?
The idea of capital, as something you have to invest in and that then enables the production of goods
and services, is a powerful metaphor. Thus, one sees references to intellectual capital (investment in
ideas or learning, like the drug company patent in the example above), to social capital (investment in
the institutions and structures that support the functioning of society), and even to natural capital (the
natural processes in the environment that provide ecosystem services (Definition 3 on page ix)—here
investment often tends to mean spending money to avoid degrading these processes). For some, this
is a powerful explanatory idea that reaches across disparate fields of thought; for others, the idea of
“natural capital” especially seems a step too far, squeezing out other perspectives and ways of valuing
things in favor of the purely economic. See Senator Gaylord Nelson’s remarks quoted on page x.
Example 8: We are all familiar with the idea that money left on deposit at a bank will earn interest (or
to put it the other way around, that you will have to pay interest if you borrow money). This seems a
reasonable expectation to us, but where does it come from? It is closely related to the fact that money
can be used to purchase capital—in other words, that money can buy the ability to make more money.
This is the left-hand (positive) feedback loop in Figure 5 on page 195.
In ancient times, the major income-generating asset was agricultural land. Economists tend to
distinguish land from capital because there is effectively no way to increase the amount of land
on Earth, whereas capital can be increased by human investments; both land and capital, however,
make wealth for the person who owns them. It is intriguing that some ancient societies banned both
the perpetual ownership of land and the charging of interest (see, for example, the Hebrew Bible at
Leviticus 25:23 and 25:36).
Logarithms and Accumulations

Many natural questions involve the way exponential processes interact with accumulations.
Problem 7: A new college graduate in her first job saves $100 per month. The money is deposited, at
the end of each month, in a savings account that earns a rate of interest of 0.2% per month.2 What is
the total in the account after six years, assuming that the monthly savings and the rate of interest stay
the same?
Problem 8: In 1865, the economist William Stanley Jevons (pictured in Figure 7) examined Britain’s
coal supplies. He estimated that Britain’s mineable coal reserves were roughly 1011 tons, and that
annual consumption was about 108 tons (so that at a constant rate of consumption, the reserves would
last a thousand years). But Jevons also noticed that annual consumption was not constant: in fact,
it was increasing exponentially at a rate of approximately 3.5% per year. How long could Britain’s
reserves last if this exponential growth continued?
These problems seem quite different, but in fact they use the same underlying
mathematics. Think about Problem 8. On Jevons’s assumptions, the coal consump-
tion for each successive year is (100 + 3.5)% = 1.035 times the consumption for
the year before. Thus, coal consumption in the year 1865 is 108 tons, in 1866 is
108 × 1.035 tons, in 1867 in 108 × (1.035)2 tons, and so on. After n years, the total
consumption is
Consumption = 108 × 1 + (1.035) + (1.035)2 + · · · + (1.035)n−1 tons, (3)

and the question is when this total will exceed 1011 tons.
Problem 7 involves a similar sum. In this problem it is helpful to think backward
from the end of the accumulation period. The very last $100 that is deposited earns
no interest, the last-but-one earns one month’s interest, the last-but-two earns two Figure 7: William Stanley
months’ interest, and so on until we reach the first $100 deposited, which earns Jevons.
interest for (6 × 12) − 1 = 71 one-month periods. Earning a month’s interest is
equivalent to multiplication by (100 + 0.2)% = 1.002. Thus the total accumulation is
Total = 100 × 1 + (1.002) + (1.002)2 + · · · + (1.002)71 dollars. (4)

To solve problems of this sort it is necessary to know how to find the sum of a series of successive
powers of some quantity x. Such a sum is called a geometric progression.
2 A real financial institution would probably call this “annual interest of 2.4% compounded monthly.” That is a different
way of saying the same thing.
Rule 5: Geometric Progression Formula
Let x be any positive number different from 1. Then
xn − 1 1 − xn
1 + x + x2 + · · · + xn−1 = = .
x−1 1−x
(The two fractions give the same answer. We use whichever form makes the
numerator and denominator positive: the first form if x > 1 and the second form if
x < 1.)
You can check the validity of this formula by multiplying through by (1 − x) and observing how
terms cancel. For instance, if n = 4,
(1 + x + x2 + x3 )(1 − x) = 1 +x +x2 +x3
−x −x2 −x3 −x4 = 1 − x4 .
Solution (to Problem 7): The geometric progression formula tells us that
(1.002)72 − 1
1 + (1.002) + (1.002)2 + · · · + (1.002)71 = ≈ 77.36.
1.002 − 1
From equation (4) on the previous page, we can find the total accumulation by multiplying the
monthly contribution (100 dollars) by this value. Thus the 100-dollar monthly contributions will have
accumulated after six years to approximately $7, 736. The total amount contributed was $7, 200, so
the account has gained a little more than 500 dollars interest during the accumulation period.
We can summarize the idea of this solution in a formula.
Rule 6: Savings Formula
Suppose that a savings account pays an interest rate r per time period, and that an
amount d is deposited in the account at the end of each period. If the accumulation
continues for n periods, the total amount accumulated at the moment the last deposit
has been made will be
(1 + r)n − 1

d .
r
Problem 9: Saving for retirement, an employee deposits $500 per month in a 401(k) account that is
invested in stocks and bonds earning an annual return of 4 percent. The employer matches the worker’s
contributions. After 40 years, what is the total accumulation?
Solution: Because the return is an annual one, we need to work out the total annual contribution to the
account. Because there are 12 months in a year, this is $500 × 12 = $6, 000 employee contributions,
and an equal amount matched by the employer,3 for a total annual contribution of $12, 000. After 40
years, the savings formula gives
3 Employer 401(k) matches are free money (albeit money you can’t spend right now). When you begin regular employment,
be sure to ask about 401(k) matching. Because of the exponential growth of compound interest, the earliest contributions you
make to a retirement savings plan have a disproportionately powerful effect. At the time of writing, 401(k) accounts also enjoy
significant tax advantages.
1.0440 − 1
12, 000 × ≈ 1.14 million dollars.
0.04
A significant nest egg!
One should beware of assuming that a million dollars forty years from now will be worth the same
as a million dollars today, and a prudent retirement plan will take this discounting into account. We
will look at this issue more closely in Section 6.1.3.
Problem 10: To buy a house, you take out a fifteen-year mortgage of

$150, 000. The rate of interest is 0.25% per month (or as your mortgage
company puts it, a “3 percent annualized rate, interest paid monthly”).
What will your monthly payment be?
Solution: This sounds a bit different from the problems we have looked
at so far—it’s a borrowing problem rather than a savings problem. But in
fact, we can use the same ideas to solve it. Think about it this way. First,
imagine that you borrowed the $150, 000 and left the interest to accumulate
without making any repayment. At the end of fifteen years (180 months),
your debt would have increased to
150, 000 × (1 + 0.0025)180 ≈ 235, 100 dollars. Figure 8: Your new residence awaits.
Now imagine that your monthly payments, say of P dollars, are going into
a savings fund whose goal is to accumulate exactly $235, 100 at the end of
fifteen years. From the savings formula we can calculate the rate of savings that will be needed:
1.0025180 − 1

P× = 235, 100.
0.0025
Thus
235, 100 × 0.0025
P= ≈ 1, 035 dollars.
1.0025180 − 1
Over the fifteen years you will have repaid 180 × P ≈ $186, 500 to your mortgage company,
significantly more than the $150, 000 principal of the loan.
The ideas underlying this solution can also be expressed as a rule.
Rule 7: Loan Formula
Suppose that I borrow an amount L for n time periods at an interest rate of r per time
period. Then my periodic payments on the loan will be
r

L .
1 − (1 + r)−n
(Notice the negative power of 1 + r.)
Remark 3: In the loan formula the quantity (1 + r)−n becomes small as n increases, but it is always
greater than zero. Thus, however long the repayment period, the periodic payments cannot fall
below Lr, the amount of interest on the original loan balance. Beyond a certain point, extending the
repayment period does not reduce your payments very much—over time, it merely gives your lender
more of your money in the form of interest.
Now let’s consider the solution to Jevons’s question about coal reserves.
Solution (to Problem 8 on page 197): Remember the setup: coal reserves are supposed to be 1011
tons, with annual consumption of 108 tons increasing at 3.5% per year. After n years, the number of
tons of coal consumed (assuming continued exponential growth) is given by equation (3) on page 197
as
(1.035)n − 1

8 2 n−1 8
10 × 1 + (1.035) + (1.035) + · · · + (1.035) = 10 ×

,
0.035
using the geometric progression formula. We want to know for what value of n this total will equal
the initial reserve of 1011 tons. So we write an equation
(1.035)n − 1

108 × = 1011 .
0.035
Divide both sides by 108 and multiply them by 0.035 to get
(1.035)n − 1 = 0.035 × 103 = 35, so (1.035)n = 36.
Take logarithms on both sides and use the second law of logarithms:
log 36 1.556
n log(1.035) = log 36, so n = ≈ ≈ 104.
log 1.035 1.494 × 10−2
Jevons [173] therefore estimated that if growth continued, Britain’s “thousand-year” coal reserves
would be exhausted in about a hundred years. Thus, he wrote
. . . the check to our progress must become perceptible within a century from the present
time; the cost of fuel must rise, perhaps within a lifetime, to a rate injurious to our
commercial and manufacturing supremacy; and the conclusion is inevitable, that our
present happy progressive4 condition is a thing of limited duration.
Think about it. . .
Was Jevons right? Investigate British coal production. In what year did it reach its peak?
What is British coal production today, as a fraction of the peak? What factors was Jevons
unaware of, and how did they affect his warning that “our present happy condition is a thing
of limited duration”?
Jevons’s book was ahead of its time in many ways. The Jevons paradox on page 94 is just one of
the ideas that it introduced.
Jevons’s work on coal reserves is one of the first known examples of a calculation of an
exponential reserve index. Here are some relevant definitions. Suppose that the total reserve of some
nonrenewable resource is R units, and that the present rate of consumption of that resource is Q units
per year. (In the coal example, R = 1011 tons and Q = 108 tons per year.)
Definition 4
The static reserve index of a nonrenewable resource is the length of time it will last at the
present rate of consumption.
4 By this Jevons means “continually improving.” The political sense of the word “progressive” was not known at that time.
To find the static reserve index, we just divide the reserve by the present rate of consumption (that
is, R/Q). However, it may well be that the consumption rate is not static, but rather increases from
one year to the next. If we assume that consumption increases exponentially with a growth rate of r a
year, the length of time to exhaustion will be shorter—perhaps much shorter.
Definition 5
The exponential reserve index of a nonrenewable resource is the length of time it will last
if consumption increases from year to year at a fixed exponential rate.
Our solution to Question 8 on page 197 gives an example of how to calculate an exponential reserve
index. The underlying mechanism of our calculation can also be condensed into a formula:
Rule 8: Reserve Index Formulas
• For a total reserve of R consumed at an annual rate Q,
Static reserve index = R/Q.
• If the rate of consumption is increasing annually at a rate r, then
log(1 + (R/Q)r)
Exponential reserve index = .
log(1 + r)
Remark 4: The scenario envisaged by these “reserve indexes” is not a realistic one, of course: it
would not happen in practice that one day the coal mines or oil wells would be running at full tilt, and
the next day they would all be mined out or dried up. But the indices are valuable as indicating the
time scale on which something has to change: new resource discoveries, more effective recycling, use
of substitutes, or whatever.
4.1.4 Logarithms and Greenhouse Gas Concentrations

Most of the examples we have studied so far in this section have involved using logarithms in ways
related to exponential growth or decay—to processes that evolve over time. Now we are going to learn
about a different application of logarithms, where time evolution does not play a direct role, and which
is more closely connected to our discussion of relative change in Section 4.1.1. The fundamental
question is this: how does the concentration of the greenhouse gas carbon dioxide (CO2 ) in the
atmosphere (as recorded by the Keeling curve, Figure 18 on page 50) modify the strength of the
greenhouse effect. Let’s recall some of the terminology and ideas from our discussion of the climate
system in Section 2.4.
We explained in that section how Earth’s surface temperature arises from a dynamic equilibrium
between heat energy arriving from the Sun (short-wave radiation) and heat energy reradiated from the
Earth (long-wave radiation). Then we described the greenhouse effect: some gases in the atmosphere,
including carbon dioxide, act like a blanket to the outgoing long-wave radiation from the surface of
the Earth, and so make the planet warmer than the freezing “naked planet” temperature of 255 K.
We presented a very simple model, the glass ball model, to suggest how the greenhouse effect might
operate.
The glass ball model is a crude approximation. It doesn’t include any account of the physics of
actual greenhouse gases, which absorb and emit radiation only at some characteristic wavelengths. It
correctly explains the greenhouse effect as relating to the fact that the upper atmosphere is colder than
the Earth’s surface, but it does not accurately represent the way this comes about. In the glass ball
model the equilibrium between the upper atmosphere and the Earth is established solely by radiation,
whereas in fact, convection (the physical movement of masses of air), adiabatic expansion (gases
become cooler as they expand and warmer as they are compressed), and precipitation (energy transfer
arising from the conversion of water vapor to liquid water) all play an important role. Finally, the glass
ball model doesn’t incorporate any mechanism to understand how the strength of the greenhouse effect
depends on the concentration of greenhouse gases. But of course, this question is of great concern to
us humans! We want to know how much we may expect that the additional greenhouse gases we have
released into the atmosphere are perturbing the climate system. That is, we want to know the radiative
forcing associated with a specified greenhouse gas concentration.
Definition 6
The radiative forcing resulting from a given greenhouse gas concentration is the extra
amount of power (in watts per square meter) that is prevented from leaving the Earth’s
surface by the greenhouse effect. The “extra” is by comparison with some fixed reference
time (sometimes taken to be 1750, roughly the beginning of the Industrial Revolution and
therefore of significant human-caused carbon dioxide emissions).
Figure 9 on the opposite page shows the Environmental Protection Agency’s estimates of the
amount of radiative forcing that several different greenhouse gases5 have contributed over the period
from 1980 to 2015.
As you can see from the figure, carbon dioxide is by far the most significant contributor to radiative
forcing. To calculate the forcing effect of a given concentration of carbon dioxide, scientists use a
basic law named after Svante Arrhenius, whom we already met in our earlier discussion of climate
(see Figure 33 on page 114).
Rule 9: Arrhenius’s Law
The radiative forcing for a concentration Q of carbon dioxide is equal to
A log(Q/Q0 ),
where Q0 is the concentration of carbon dioxide at the initial reference time

(approximately 270 ppmv in 1750) and A is a constant that measures how sensitive
the climate system is to the effects of carbon dioxide.
Because of the first law of logarithms (see Rule 1 on page 188), the factor log(Q/Q0 ) increases by
the same amount each time Q is doubled. Going from a carbon dioxide concentration of 100 parts per
million to 200 parts per million adds the same radiative forcing as going from 200 to 400, or from 400
5 Water vapor, although it is a significant contributor to the greenhouse effect, does not appear in this diagram because there
are no significant “human-caused emissions” of water vapor: its contribution hasn’t changed from the 1750 baseline.
Figure 9: Radiative forcing (relative to the year 1750) over time, according to the U.S. Environmental Protection Agency.
to 800. The crucial constant A (units: W m−2 ) tells us the amount of forcing that would be added if
the carbon dioxide concentration were, hypothetically, multiplied by 10.
The extent to which the Earth warms up in the medium term (say a hundred to two hundred years)
is roughly proportional to the level of radiative forcing. This allows us to define a quantity that plays
a key role in debates about global warming.
Definition 7
The climate sensitivity is the amount by which the Earth would warm (in the medium term)
if the atmospheric carbon dioxide concentration were to double.
Problem 11: Suppose that it is known that the climate sensitivity is 3 degrees Celsius of warming for
a doubling in carbon dioxide concentration. According to Arrhenius’s law, what amount of warming
is associated with a rise in carbon dioxide concentration from its preindustrial level of about 270 parts
per million to its current value of about 400 parts per million?
Solution: Arrhenius’s law (together with the remark that we made above, that temperature rise
is proportional to additional forcing) predicts that warming equals W × log(Q/Q0 ), where W is
a constant. The question tells us that a doubling of carbon dioxide concentration leads to a 3 ◦ C
warming; that is,
3 ◦
3 ◦ C = W × log 2, which tells us that W = C.
log 2
Now to find the warming for an increase from Q0 = 270 ppmv to Q = 400 ppmv, we calculate
400 3 × log(1.48) ◦

W × log ≈ C ≈ 1.7 ◦ C.
270 log 2
The Earth has not yet seen that degree of warming, but it is expected to take some time—up to 100
years—before the effects of any given concentration of carbon dioxide fully make themselves felt.
Meanwhile, of course, the concentration continues to increase!
Figure 10: Likelihood of various estimated values for the climate sensitivity, after [273].
Many of the hardest questions about how to respond to anthropogenic climate change boil down
to improving our understanding of the climate sensitivity. If the climate sensitivity is large, then even
small increases in carbon dioxide concentration may have large effects, and it is worth committing
large resources now to avoid these effects. If the climate sensitivity is small, then expected effects are
smaller and we may have more room to maneuver.
Remark 5: The most difficult part of estimating the climate sensitivity is to quantify the effects of the
various positive and negative feedbacks in the climate system that may amplify (or diminish) the effect
of a given level of radiative forcing. We will learn in Section 4.3 how to do this kind of calculation,
in principle: but to carry it out in practice one needs to measure or model the strengths of all these
various feedbacks, and (especially for the positive feedbacks) this must be done quite accurately. It is
well agreed that without any additional feedbacks, the “bare” climate sensitivity would be of order 1
degree Celsius. The known climate feedbacks are sure to amplify this figure overall, but there remains
some uncertainty about how much. The latest report (AR5) from the Intergovernmental Panel for
Climate Change [253] says that there is “high confidence” that the climate sensitivity lies between
1.5 ◦ C and 4.5 ◦ C. This is still a wide range, though!
One way of expressing the level of our knowledge and uncertainty about the climate sensitivity
is by a graph like that in Figure 10. This graph, which summarizes a range of estimates of climate
sensitivity using various modeling methods and calculations from palaeoclimate records, shows a
probability distribution that summarizes best estimates of the climate sensitivity (at the time the
paper was written). The area under the graph between two estimates of the climate sensitivity (plotted
on the x-axis) gives a measure of how likely the true value is to lie in the range between those estimates.
Thus we can see, for instance, that the climate sensitivity is most likely between 2 and 4 degrees, but
that there is a significant chance that it could be over 5 degrees, and almost no chance that it could
be less than 1 degree. In Chapter 5 we will look more closely at probability ideas like this and at the
significance of the rightward-skewed nature of this particular graph.
Remark 6: Arrhenius’s law is an approximation that relates specifically to concentrations of carbon
dioxide.6 As you can see from Figure 9 on the previous page, the next most significant greenhouse
gas in terms of its radiative forcing effect is methane. Methane is also known as “natural gas”: it
6 You can find a more precise version of the “law” for carbon dioxide, and also the formulas for the corresponding “laws”
for several other greenhouse gases, in Table 1 of [151].
may provide heat for your home, and the recent huge increase in U.S. natural gas production by
fracking has meant that many coal-fueled electricity generating plants have been closed and replaced
by natural-gas-fueled plants. Natural gas is much more convenient in operation, and what’s more,
burning natural gas to generate energy releases less carbon dioxide (for the same amount of energy)
than burning coal or oil (see Table 5 on page 488). All this is good news. However, unburnt methane is
a powerful greenhouse gas, nearly a hundred times more powerful than carbon dioxide. It doesn’t take
much methane leakage in the production and distribution process to outweigh the global warming
benefits of using methane as a fuel for electricity generation instead of coal.7 That’s why the EPA
issued regulations in 2016 limiting the permissible amounts of methane emissions from new natural
gas facilities [108]. These regulations have been repealed by the Trump administration, as part of its
wider-scale withdrawal from action on climate issues.
Summary of Ideas: Logarithms
• The logarithm is a mathematical function that converts multiplication to addition:

for example, log(15) = log(3 × 5) = log(3) + log(5).
• We can also calculate the logarithm of a power: for example, log(156 ) = 6 log 15.
• The antilogarithm function 10x reverses the operation of the logarithm: log(10x ) = x
and 10logt = t.
• If a quantity follows an exponential law of growth or decay, then the graph of its
logarithm against time will be a straight line.
• A geometric progression is the sum of a series of successive powers of some quantity
x. The geometric progression formula is
1 + x + · · · + xn−1 = (xn − 1)/(x − 1) = (1 − xn )/(1 − x).
We can use the geometric progression formula to answer questions about saving and
borrowing.
• The static reserve index of a nonrenewable resource is the length of time that current
reserves would last at the current rate of consumption.
• The exponential reserve index of a nonrenewable resource is the length of time that
current reserves would last at an exponentially increasing rate of consumption.
• The climate sensitivity is the medium-term temperature increase that would be
produced by a doubling of carbon dioxide levels.
7 There are also other possible sources of methane. See Problem 9 on page 28 as well as Exercises 19–22 at the end of the
chapter.
4.2 Logistic Models and the Limits to Growth
Objectives
I can define the growth rate of a data series.

I know how to recognize exponential data in terms of the growth rate.
I know what is meant by the carrying capacity of a system.
I understand the notion of ecological footprint.
I can describe the logistic model for growth in the presence of a finite carrying
capacity.
I know the growth pattern produced by the logistic model.
I know how to use the logistic model to predict the carrying capacity from a data
series.
I understand the concept of peak oil and its relation to the logistic model.
4.2.1 Growth Rates

We’ve said several times that exponential growth—say, in the consumption of
Year Population material resources—cannot persist forever in a finite system. So far, however, we
1900 76,000,000 have not developed any mathematical models that could allow us to discuss what
1910 92,000,000 a slowdown in the growth process might look like. In this section we are going to
1920 105,700,000 learn about the simplest such model, the so-called logistic model. Simple though
1930 122,800,000 it is, this model has been successfully applied, especially to natural resource
1940 131,700,000 problems.
1950 150,700,000 Let’s start with the question of how we might recognize a data series as coming
1960 179,300,000 from an exponential growth model. For example, the Census Bureau [160, 204]
1970 203,200,000 gives the numbers in Table 3 for the population of the U.S. in successive decades.
1980 226,500,000 Is this data a good fit for an exponential growth model? How might we tell?
1990 248,700,000 One good way, as we have seen in the previous section (Problem 4 on page 192),
2000 281,400,000 would be to make a semilogarithmic plot of the data—a graph whose x-axis is
2010 308,700,000 time and whose y-axis is the logarithm of population. If the data come from an
Table 3: U.S. population data. exponential model, then this plot will be a straight line. If you try this, you will
see that the fit to a straight line seems reasonably good, but there is perhaps some
evidence of growth dropping off in more recent decades. (What reasons do you think could account
for that?)
In this section, though, we want to consider another way we could recognize data as coming from
an exponential model. This would be to look at the growth rates of the data series. “Growth rate” is
simply another word for the relative increase, a concept we have already considered in Section 4.1.1
and whose definition we will review below. The difference in terminology reflects that we will be
considering growth rates for a series of data dependent on time, rather than simply for a single change.
4.2. LOGISTIC MODELS AND THE LIMITS TO GROWTH 207
Definition 1
The growth rate of a stock over a time interval; (also known as the relative increase in the
stock over that time interval, see Definition 1 on page 186) is
Final value of stock − Initial value of stock
Growth rate =
Initial value of stock
Final value of stock

= − 1.
Initial value of stock
Since it is obtained by dividing one quantity by another that has the same units, the growth
rate is a pure number.
By definition, an exponential model is one for which the growth rate is a constant—the rate constant,
r, if our interval is one time step. So by looking at the growth rate at different times, we can get an
idea of whether an exponential model is a good fit.
Example 1: From Table 3, the growth rate for the period 1970– Year Population Growth Rate
1980 is
1900 76,000,000 (n/a)
226, 500, 000 − 203, 200, 000 1910 92,000,000 21.1%
≈ 0.115 = 11.5%.
203, 200, 000 1920 105,700,000 14.9%
1930 122,800,000 16.2%
The growth rate may, of course, be different for different periods. 1940 131,700,000 7.3%
For instance, the growth rate for the period 1900–1910 is 1950 150,700,000 14.4%
1960 179,300,000 19.0%
92, 000, 000 − 76, 000, 000 1970 203,200,000 13.3%
≈ 0.211 = 21.1%.
76, 000, 000 1980 226,500,000 11.5%
1990 248,700,000 9.8%
In fact, we can redo our table and include all the growth rates. 2000 281,400,000 13.1%
When we do this, there’s a choice: do we record the growth rate 2010 308,700,000 9.7%
for a given decade in the line referring to the year that starts the
decade, or the year that ends it? We choose to record the growth Table 4: U.S. population data, with growth rates.
rate for each decade in the line of the table referring to the end of
that decade. Thus we get the table in Table 4.
The rates vary quite a bit. For instance, the growth rate over the decade of the 1930s was
exceptionally low—why do you think this might have been? And in the “baby boom” years of the
1950s it seems that the growth rate was unusually high.
Problem 1: The ACME Widget Corporation, a new start-up business, sells 3 million widgets in its
first year of operation, 4 million in its second year, 5 million in its third, and so on. Calculate the
growth rates of ACME’s total sales.
Solution: We make a table for the first seven years of ACME’s operation (Table 5 on the next page).
The first column gives the year, the second column gives the sales for that year, the third the total sales
(obtained by adding up the sales in all the preceding years), and the fourth column is the growth rate
calculated according to Definition 1. Notice that even though ACME’s sales are increasing each year,
their growth rate is actually decreasing (quite rapidly, in fact). ACME’s sales add only a fixed amount
each year, but for true exponential growth we would need to multiply by a fixed amount each year.
That is much harder to achieve.
Annual Sales Total Sales Problem 2: Many business groups present awards to the
Year Growth Rate
(millions) (millions) “fastest growing” company in a certain area, or in a certain
1 3 3 (n/a) business sector. Based on the example above, do you think
2 4 7 133%
3 5 12 71%
that the “fastest growing company” is likely to be a small one
4 6 18 50% or a large one?
5 7 25 38%
6 8 33 32% Solution: The example of ACME Corporation suggests that
7 9 42 27%
as a growing company matures, its growth rate will decrease.
Table 5: ACME Corp sales figures. Remember, the growth rate is a fraction (Definition 1 on the
previous page). For a fraction to be big, it helps for it to have
a small denominator as well as a large numerator; and the denominator in the growth rate is the “total
stock,” in this case, the size of the company. Thus the “fastest growing company” will most likely be
fast-growing because it is small and has not yet encountered the limits of its particular market sector.
100
Many real-world examples of growth are like that of a
80
company expanding into a new market niche. At first, there
are no limits to growth and the process is exponential, with a
60 constant growth rate. As the stock grows, however, constraints
Population
on its further growth become important. These can be of many

40 kinds: for example, a nonrenewable resource might be depleted,
the growth of a tumor might be limited by the size of its host,
20 or a population’s food consumption might approach the total
amount of food available from its environment. Constraints
0
0 20 40 60 80 100 reduce the growth rate, causing growth to tail off. In the
Time simplest case, the growing stock smoothly approaches its upper
Figure 11: Simulation of the constrained growth of a
limit, producing an S-shaped curve like that shown in Figure 11.
population. That is the kind of constraint on growth that we will look at in
this section.
The last example above (of a population limited by its food
supply) introduces an important notion in ecology:
Definition 2
The carrying capacity of an environment for a given biological species is the largest
population of that species that can be sustained in the long term by the resources available
in the environment.
Limits to Growth
The idea of carrying capacity was invented by ecologists studying plant and animal populations. For
example, a fish farm might be said to have a carrying capacity of 100 tons—meaning that its system
can support a healthy population of fish (in various stages of development) whose total biomass is
100 tons. If the fish population grows larger than that, their environment will be degraded, perhaps
severely.
But does the Earth have a carrying capacity for human beings and their activities? If it does, what
is it? What will be the dynamics of our approach to that carrying capacity—a smooth S-shaped curve,
or overshoot, oscillation, or collapse (compare Section 3.3.3)?
Related to this question is the notion of ecological footprint. This idea was invented by Mathis
Wackernagel and William Rees [337] in the early 1990s. Here is the definition.
Definition 3
The ecological footprint of a person or population is the amount of Earth’s surface area that
it would take to sustainably provide all the resources needed for that person’s or population’s
lifestyle.
Of course, not all parts of the Earth’s surface are equally productive—the “footprint” is expressed
in terms of a “global average acre.” One can say, therefore, that “your ecological footprint is 17 global
average acres.” The same idea is sometimes expressed in a different way: “It would take 3.5 Earths to
support everyone on the planet living your current lifestyle.” This gives a graphic sense to the idea of
“overshoot”: we only have one Earth, so if my lifestyle requires more than one Earth to support it for
everyone, then it is overshooting our carrying capacity.
Some, though, would say that the notion of “carrying capacity” is not appropriate when applied
to human beings and their activities. After all, the fish in the fish farm do not have the ability to
improve their environment by choice and creativity. They are limited; we are resourceful—resourceful
enough to keep inventing new technologies that get around apparent limits. An example might be
the replacement of copper wiring by fiber-optic cables for communication circuits in the 1990s and
beyond. Before then, some worried that limited world supplies of copper would slow down the growth
of global connectivity. But we got around the problem—now we make our cables out of sand!
Think about it. . .
Julian Simon, professor of business administration at the University of Maryland and

“cornucopian” economist, wrote [299]:
Because we can expect future generations to be richer than we are, no matter
what we do about resources, asking us to refrain from using resources now so
that future generations can have them later is like asking the poor to make gifts
to the rich.
Others point out, though, that however great our ingenuity, we are still physical beings, dependent
for the foreseeable future on the “life support systems” provided by our planet Earth. Each of us needs
food, water, shelter, and energy; and our systems for supplying the first three have grown increasingly
dependent on an abundant supply of the fourth (that is, energy). If we reach “carrying capacity” for
energy production, what then?
In 1972 these ideas were presented in a highly influential book, The Limits to Growth [191]. The
authors, led by environmental scientist and Dartmouth professor Donella Meadows, built a stock-flow
model of the world involving five basic stocks: human population, natural resources, pollution, food
production, and industrialization. They tried to use their model, called World3, to estimate what the
carrying capacity of the planet might be and in what way human society would approach it.
Think about it. . .
The Donella Meadows Project maintains an online archive of Meadows’s work on systems
thinking and sustainability. They attribute the following quotation to Meadows as her
“guiding message”:
We humans are smart enough to have created complex systems and amazing
productivity; surely we are also smart enough to make sure that everyone
shares our bounty, and surely we are smart enough to sustainably steward the
natural world upon which we all depend.
What do you think she means? Do you agree?
The results of the World3 model were quickly referred to as “predic-

tions,” but that was not how they were intended:
In this first simple world model, we are interested only

in the broad behavior modes of the population-capital
system. By behavior modes we mean the tendencies of
the variables in the system (population or pollution, for
example) to change as time progresses. A variable may
Figure 12: Donella Meadows, environmental increase, decrease, remain constant, oscillate, or combine
scientist and lead author of The Limits to
Growth.
several of these characteristic modes. For example, a pop-
ulation growing in a limited environment can approach the
ultimate carrying capacity of that environment in several
possible ways. It can adjust smoothly to an equilibrium below the environmental limit
by means of a gradual decrease in growth rate. . . . It can overshoot the limit and then
die back again in either a smooth or an oscillatory way. . . . Or it can overshoot the limit
and in the process decrease the ultimate carrying capacity by consuming some necessary
nonrenewable resource. . . . This behavior has been noted in many natural systems. For
instance, deer or goats, when natural enemies are absent, often overgraze their range
and cause erosion or destruction of the vegetation.
A major purpose in constructing the world model has been to determine which, if any,
of these behavior modes will be most characteristic of the world system as it reaches the
limits to growth. This process of determining behavior modes is “prediction” only in the
most limited sense of the word . . . .
The difference between the various degrees of “prediction” might be best illustrated
by a simple example. If you throw a ball straight up into the air, you can predict with
certainty what its general behavior will be. It will rise with decreasing velocity, then
reverse direction and fall down with increasing velocity until it hits the ground. You know
that it will not continue rising forever, nor begin to orbit the earth, nor loop three times
before landing. It is this sort of elemental understanding of behavior modes that we are
seeking with the present world model (quoted from [215], original 1972 edition).
Critical Thinking
It’s a plausible idea that a model, even if it is too crude to provide accurate numerical
predictions, may nevertheless yield an “elemental understanding of behavior modes” in the
way expressed by the Limits authors with their example of a ball thrown into the air. Try to
tease out the assumptions underlying this idea. Can you find reason to believe (or disbelieve)
these assumptions in the case of the Limits model?
The authors of Limits to Growth considered a dozen different policy scenarios for the World3
model. Without policy intervention, the results demonstrated a characteristic “overshoot and collapse”
pattern, which the book attributed to the many delays in the human decision-making process (compare
Section 3.3.3): by the time humanity “notices” the signals that all is not well with the world system
and decides to respond, its environmental footprint has already moved into “overshoot” territory.
“Technological fix” scenarios—for example, imagining that all industrial processes become four times
more efficient and produce four times less pollution—postponed overshoot for a few decades but
did not change the basic pattern. However, the authors found that scenarios in which humanity pre-
emptively took action to slow its approach to ecological limits did manage to avoid the overshoot-and-
collapse pattern. Appearing in 1972, the book recommended that such actions should begin in 1975.
Needless to say, they did not.
The ideas of the book were controversial at the time and remain so forty years later. Critics,
especially from the realm of economics, suggested that the authors of Limits had failed to take into
account the human potential for technological advancement and the power of the market mechanism,
which automatically substitutes other resources for those that are in decline [179], [299]. Others have
pointed out that the world seems to have tracked the Limits modelers’ “business as usual” scenario
fairly accurately over the last forty years [326]. In this unit we are interested in understanding only the
simplest models of “limits” to exponential growth. As we will see, the logistic model that we’ll study
does not allow for an “overshoot and collapse” scenario, but only for gradual approach to a limit.
4.2.2 The Logistic Growth Model

The logistic model is the simplest example of a stock-flow system that exhibits growth behavior with a
finite carrying capacity. It is another feedback loop (like the exponential growth model), but this time
the control function that determines the flow in terms of the stock is not the simple linear one (flow
proportional to stock) that gave rise to the exponential growth model. Here is the definition.
Definition 4
A stock in a stock-flow system is governed by a logistic model if the net flow (Remark 4 on
page 73) into the stock is proportional to the product of two quantities: (a) the level of the
stock and (b) the amount by which the stock level is below a fixed carrying capacity.
To write it in an algebraic formula, if S is the stock level and C is the carrying capacity, then the net
inflow is
kS(C − S),
where k is a constant of proportionality. It is often helpful to express this quantity (the net inflow) in a
different way: put r = kC and write
+
Inflow Stock
Figure 13: The logistic model is a feedback loop, but the control function is nonlinear.
kS(C − S) = kCS(1 − S/C) = rS(1 − S/C).
If the term in parentheses were not there (so that the net inflow was simply rS), we would have an
exponential growth model (Definition 1 on page 164) with rate constant r. When S is small compared
with C, the term (1−S/C) is close to 1, so multiplying by it makes little difference: we still, essentially,
have exponential growth. But as S approaches C, the fraction S/C approaches 1 and so (1 − S/C)
becomes small. Multiplying by this small factor chokes off the exponential growth, giving rise to the
characteristic logistic curve shown in Figure 11 on page 208.
Rule 1: Logistic Model Summary
The logistic model has three parameters, listed below:

• The rate constant r;
• The time step T ;
• The carrying capacity C.
With these parameters set, the net inflow to the stock in each time step is rS(1 − S/C),
where S is the stock level at the beginning of the time step.
Flow Rate
Question 1: Is the feedback in the logistic model amplify-
ing or stabilizing?
Maximum
Answer: That is an excellent question! To answer it,

let us look at the relationship between net inflow and
stock level S. The inflow is given by the control function
rS(1 − S/C). The graph of this function is a parabola, as
Stock Level shown in Figure 14. It starts at zero (when S = 0), rises to
C/2 C
a maximum at S = C/2, and then falls off again to zero at
Figure 14: Control function for the logistic model. S = C.
Now look at the feedback rule of signs, Rule 1 on
page 156. That tells us that the sign we should mark on
the control arrow in Figure 13 should be positive or negative, according to whether an increase in the
stock leads to an increase or a decrease in the corresponding flow.
Because the control function in this example is nonlinear (its graph is not given by a straight line),
the correct sign will depend on what the stock level is. In fact, you can see from Figure 14 that for
100 10
80 8
60 6
Flow rate
Stock
40 4
20 2
0 0
0 20 40 60 80 100
Time
Figure 15: Stock (blue) and flow rate (red) in a logistic model.
S < C/2 the curve is sloping uphill. This means that a small increase in S will increase the flow rate:
the control arrow should be marked with a +. But for S > C/2 the curve is sloping downhill: a small
increase in S will decrease the flow rate, and therefore the control arrow should be marked with a
−. The answer to the question “What kind of feedback do we have?” is therefore “It depends on S.”
When S < C/2 we have a positive (amplifying) feedback, pushing us away from the set point at 0.
When S > C/2 we have a negative (stabilizing) feedback, still pushing us in the same direction, but
now toward a new set point at C.
Figure 15 shows the results from a logistic model run with C = 100 units. The blue curve shows the
stock level. Notice how the stock level appears to be increasing exponentially at first, but the growth
chokes off at the limit set by the carrying capacity. The red curve shows the flow rate (measured on
the right-hand scale). That curve tells us something interesting. Notice that the greatest or “peak” flow
rate comes when there still seems to be plenty of room for further increase in the stock—in fact, as
the green dashed lines indicate, when the stock has reached 50 units, or exactly half of the carrying
capacity. Beyond that point, flow rates decline despite still-increasing stock, as the “limit to growth”
set by the carrying capacity becomes the predominant effect.
Rule 2: Peak Flow Rate Rule
The highest or “peak” flow rate in a logistic model occurs when the stock level is
exactly half the carrying capacity.
You can see this using algebra if you want. The flow rate is
C 2
" #
S r C2

rS 1 − = − S− ,
C C 4 2
by the algebra trick of “completing the square.” The term (S −C/2)2 is always positive (every square
is positive), and it appears with a minus sign. So the maximum value of the flow rate occurs when this
term is zero, that is, when S = C/2. (See Figure 14.)
Definition 5
The S-shaped curve showing the accumulation of the stock in a logistic model is called the
logistic curve.
Example 2: An important example concerns the extraction of a nonrenewable resource. Suppose that
a certain quantity C of a resource (iron ore, coal, oil, or something) is available in the ground. The
process of mining and extracting the resource can be represented by a stock-flow diagram with two
stocks: one for the unmined resource, still in the ground, and one for the total amount of extracted
resource. As the resource is extracted, it is transferred from the “unmined” to the “extracted” stock.
For simplicity, we assume that there is no waste in this process. (See Figure 16.)
Unextracted − + Extracted
Extraction
resource resource
+ +
Figure 16: Logistic model for nonrenewable resource extraction, e.g., mining.
What governs the flow rate (that is, the rate of mining)? There are two influences that are going to
be important.
(a) The more of the resource has been extracted, the more profit the mining companies will
have made. This will encourage more investment in the extraction process—both by way of
the original companies ploughing back their profits into more powerful equipment, and by
encouraging new players to enter the market. This influence is an example of the capital-
investment feedback loop (Definition 3 on page 195). It is represented by the right-hand
(amplifying) feedback loop in Figure 16, showing that the extraction rate depends on the amount
of resource already extracted.
(b) On the other hand, no deposit of a resource is completely uniform. It naturally contains both
“rich” and “poor” sections—in the context of oil, for example, the “rich” areas might be gushers
where the oil naturally spouts from the ground under its own pressure once a well is drilled, and
the “poor” areas those that require complicated and costly techniques like fracking to extract the
oil. The richest parts of the resource deposit will naturally be mined first, and as they become
exhausted, it will become increasingly difficult and expensive to extract more. This influence is
represented by the left-hand (stabilizing) feedback loop in Figure 16, showing that the extraction
rate also depends on the amount of resource remaining in the ground.
If S denotes the total amount of the resource extracted, then C − S is the amount remaining in the
ground. So the model we have sketched above suggests that the rate of extraction will be proportional
to S(C − S). This is a logistic model according to Definition 4 on page 211.
Example 3: In the previous example, the capacity C was set by the amount of the nonrenewable
resource left in the ground. But there are other situations in which a logistic model for growth may be
appropriate. For example, consider the case of a population S growing in an environment with finite
carrying capacity C. When the population is small, one might imagine that it grows exponentially, say
with a growth rate r—in other words, the net inflow per time step is rS. But as the population grows,
Growth Rate
Rate constant, r
Stock
Capacity, C
Figure 17: Graphical representation of growth rates in the logistic model.
food supplies become scarcer and the community has to devote a greater proportion of its resources
to seeking them out. Say that the proportion of resources devoted to seeking food is S/C, reaching
a maximum when S = C and everyone is spending all their time looking for food. Now if S/C of
available resources are devoted to looking for food, only the remainder, a fraction (1 − S/C)), of the
populations’ energies are available for the serious business of reproduction. Thus the net inflow per
time step becomes
rS(1 − S/C),
and once again we have a logistic model. (This model was proposed for population growth with
constrained carrying capacity by Pierre François Verhulst, 1804–1849, who also invented the name
“logistic curve.”)
Growth Rates in the Logistic Model

By our definition of the logistic growth model, a stock S follows the model if the net inflow to S over
a time step is equal to rS(1 − S/C), where r is the rate constant and C is the carrying capacity.
Now recall our definition of the growth rate: the growth rate of S over a time step is the increase in
S over that time step—that is, the net inflow—divided by S itself. So in the case of the logistic growth
model, the growth rate is
rS(1 − S/C)
Growth rate = = r(1 − S/C) = r − (r/C)S.
S
In this formula, r and C are constants, and S is a variable (a quantity that can change). So if we draw
a graph with S on the x-axis and the growth rate on the y-axis, the equation we will have is
y = −(r/C)x + r.
This is the equation of a straight line with x-intercept at (C, 0) and y-intercept at (0, r)—see
Figure 17. Thus, we may formulate the following logistic growth rule.
Rule 3: Logistic Growth Rule
In a logistic growth model, the graph of the growth rate against the stock level is a
straight line heading downhill. The x-intercept of this line is the carrying capacity,
and the y-intercept is the rate constant.
For example, we generated some sample data from a logistic model with carrying capacity 100.
Here it is:
Time 1 2 3 4 5 6 7 8 9 10 11 12
Stock 0.99 1.96 3.85 7.41 13.79 24.24 39.02 56.14 71.91 83.66 91.10 95.34
Growth rate 0.98 0.96 0.93 0.86 0.76 0.61 0.44 0.28 0.16 0.09 0.05
The second row gives the stock levels in our logistic growth model at successive time intervals.
The third row gives the growth rates computed from these stock levels. For example, 0.61 ≈ (39.02 −
24.24)/24.24. According to the logistic growth rule 3 on the previous page, if we graph the stock
levels (second row) on the x-axis, and the growth rates (third row) on the y-axis, we should obtain a
straight line intersecting the x-axis at the carrying capacity, 100. We try the experiment and obtain the
results shown in Figure 18(a).
Examining the figure, it certainly seems as though our expectation is fulfilled. What’s more, we see
that we have invented an important technique for estimating the carrying capacity from the growth
pattern. Suppose that we only had a bit of the data—say the first four values. We could still plot those
four points and draw the best line we could through them. Continuing that line to where it hit the
x-axis, we would be able to estimate the carrying capacity (see Figure 18(b)) long before we actually
reached it. We could then use this information in planning for the future, perhaps even to plan to draw
down the resource in a more orderly way.
1.0 1.0
0.8 0.8
Growth Rate
Growth Rate
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Stock Stock
(a) Scatter plot of growth rate against stock level, for sample (b) Same plot, showing extrapolation from just the first four
data discussed in the text. data points.
Figure 18: Plots of stock against the growth rate for data from a logistic model.
Remark 1: The process that we just described for estimating the carrying capacity from only a few
data points is an example of extrapolation. That means first verifying a relationship within some range
of data values, and then making a prediction by assuming that the same relationship will continue
outside the range of data that we used to establish it. Extrapolation is a common but risky procedure.
You can see by comparing the two sides of Figure 18 that extrapolation from the first four data points
suggests a carrying capacity more like 92 than 100, and this is in an example in which the data have
been specifically generated to fit with the logistic model! Extrapolation can magnify any errors that
exist in the data, and of course it makes no allowance for new phenomena (a more efficient mining
process? an unexpected health risk?) that might arise after the data has been collected. For this reason,
when we make an extrapolation it is best to express it in a conditional form: “if the pattern that we
can see in this data continues, then the maximum level of the stock will be about 92 units.”
Remark 2: The exponential growth model can be considered a special case of the logistic one. In the
exponential growth model, the growth rate is always the same—it is a constant. The graph of growth
rate against stock level will therefore be a horizontal straight line. If we try to find the x-intercept, we
will be in trouble, because a horizontal straight line is parallel to the x-axis and does not meet it. This
reflects the fact that the notion of carrying capacity does not apply in an exponential growth model.
Imagine now that we are studying the mining of some nonrenewable resource, as in Example 2 on
page 214, and that we have reason to believe (or guess) that this process is governed by a logistic
model. Imagine also that we have a table of the total amounts extracted in various years, and that
(either by the graphical process described above (Rule 3 on page 215), or perhaps by some other
method) we have estimated the rate constant r and the total resource stock t. How long, we might
want to know, will it be before we reach the peak rate of mining?
Problem 3: A molybdenum mine is estimated to have the capacity to produce 800 million pounds
of molybdenum in total. Extraction is assumed to follow a logistic model with a rate constant of 3
percent and a time step of 1 year. If 10 million pounds have been mined to date, estimate the number
of years remaining before peak production.
Problems like this can be solved using the time to peak rule.
Rule 4: Time to Peak Rule
In a logistic growth model for resource extraction, with rate constant r and total
extractable resource C, the time left before the peak extraction rate is equal to
log(C/M − 1)
log(1 + r)
time steps, where M is the total amount of resource that has been extracted to date.
Solution (to Problem 3): We apply the time to peak rule. In the notation of Rule 4, we have C =
8 × 108 lb and M = 107 lb. Thus C/M = 80 (a pure number), and the rule gives
log(80 − 1) 1.898
≈ ≈ 150 yr
log(1 + 0.03) 0.0128
(to 2 significant figures) until peak production.
Problem 4: Explain how the time to peak rule confirms that peak production occurs when half the
resource has been extracted.
Solution: When half the resource has been extracted, M = 12 C, so that C/M = 2. Thus C/M − 1 = 1
and log(C/M − 1) = 0, confirming that the time to peak is zero (which is to say, the peak is now!).
Peak Oil
One of the most important applications of these ideas has been to study the likely course of fossil
fuel energy production (especially oil production). In a famous paper [162] published in 1956, Shell
geologist M. King Hubbert applied a logistic analysis—like the one described in Remark 1—to
estimate the future course of U.S. and world production of oil and other fossil fuels. Hubbert caused
a sensation by predicting (accurately, as it turned out) that U.S. oil production would peak sometime
in the early 1970s and then slowly decline. The first energy crisis in 1973 followed shortly after the
8.0
Annual growth rate (percent) 6.0
4.0
2.0
0.0
0 20 40 60 80 100 120 140 160 180 200 220
Total production (billions of barrels)
Figure 19: Fitting a logistic model to U.S. oil production, 1930–1999.
peaking of U.S. oil production, as the country moved decisively from being a net exporter to a net
importer of oil. We can reproduce a Hubbert-like analysis using data available from the U.S. Energy
Information Administration.
In fact, the EIA website [10] gives the amount of oil produced in the U.S. for each year since 1900.
Using that information, one can produce an annual chart of cumulative oil production—that is, the
total amount of oil produced up to that year. (It is this quantity that should follow a logistic model).
From this, we can make a graph of the growth rate versus the cumulative production. Figure 19 shows
the result, based on 70 years of EIA production data (from 1930 to 1999). As you can see, the fit to a
straight line, while far from perfect, is quite good.
Remark 3: The fit suggests an ultimate limit to U.S. oil resources (carrying capacity, where the line
hits the x-axis) of something like 210 billion barrels. However, keep in mind that we stopped plotting
points in 1999! That is, before the recent explosive development of “tight” (fracked) oil in the U.S.
Fracking [132] is a new technology that has dramatically increased production of U.S. oil and gas.
Because of fracking, the most recent few points, if we plotted them in Figure 19, would be far above
the trend line (you can see the most up-to-date raw data on the EIA website we mentioned before [10]).
Is this technological development really a game-changer? Or will it prove to be a temporary “blip”
that does not significantly alter the overall pattern of decline? It is important to understand that
mathematics by itself cannot answer this question for us. The results of mathematical models tell us
what will happen “if present trends continue.” But to understand whether present trends will continue,
or how they might change, we need to look elsewhere.
The same questions that we asked about U.S. oil resources can also be asked about oil resources
worldwide. Here the questions multiply. There are many different oil-producing areas in the world,
all with their own production and depletion rates. Because of all these variables, it is hard to give
an accurate estimate of when global peak oil—the maximum rate of global production—is likely to
arrive. A presentation by Guy Caruso of the EIA at the Tenth Annual Asia Oil and Gas Conference in
2008 summarized 36 different studies, giving a range of dates from 2000 to 2060, with many around
2020. The EIA’s own answer is, “Not soon—but within the present century.”
Definition 6
Global peak oil is the year in which the world’s production of oil is greater than that of any
year either before or after.
Figure 20: One example of a “peak oil” production forecast.
Notice that “peak oil” does not mean “when the oil runs out.” In fact, as its name suggests, it is
the moment when oil extraction and the industry that it supports are at their greatest level ever! It
is more appropriate to think of “peak oil” as the moment when the physical limits of oil production
start to become seriously incompatible with the sociological imperative for endless growth. From that
perspective, it is a significant fact that all the studies cited by the EIA, including their own figures,
place this event within the lifetime of students reading this book. Figure 20 gives one example of such
a prediction.
Think about it. . .
In the conclusion to his famous paper on oil resources, Hubbert presented a graph showing
fossil fuel consumption on a time span from “5000 years ago to 5000 years in the future.”
Then he wrote
On such a time scale the discovery, exploitation, and exhaustion of the fossil
fuels will be seen to be but an ephemeral event in the span of recorded history.
What do you think he means? Do you agree? What are the implications?
Remark 4: Using fossil fuels for energy means extracting them, burning them, and discharging the
waste gases into the atmosphere. Hubbert’s concern was with the input or supply end of this process.
But more recently we have realized that the output or “waste” end of the process generates its own
set of problems. As we have seen in Sections 2.4 and 4.1.4, the billions of tons of carbon dioxide
produced by burning fossil fuels are enough to significantly affect the planet’s climate.
One way in which this idea has recently been expressed is through the concept of a carbon
budget [213]. The Earth’s carbon budget is the amount of additional CO2 that can be added to the
atmosphere in order to have a reasonable chance (about 80%) of keeping the Earth’s temperature
increase (above preindustrial levels) to less than 2 degrees Celsius, a threshold that many scientists
agree represents the “danger line” for the global climate. According to the estimates cited in [213], the
Earth’s carbon budget was about 565 billion tons of CO2 (in 2012—there have of course been more
emissions since then).
The kicker here is that the amount of fossil fuels that oil and gas companies hold as “proven
reserves”—which appear on their balance sheets as assets, which means that the companies are
planning for these assets to be dug up and burned—is equivalent to over 2,500 million tons of CO2 ,
many times higher than the carbon budget. Absent some astonishing new scientific or technological
development, the great majority of these reserves can never be used: they must stay in the ground
if the world’s carbon budget is to be respected. And, remember, this refers only to proven reserves,
not to new ones that exploration may discover. Since fossil fuel companies’ balance sheets and share
prices depend on these “assets,” they will be under strong pressure to blow through the Earth’s carbon
budget as quickly as possible. It may well be that the carbon budget (at the “output” end) rather than
resource depletion (at the “input” end) represents the currently most urgent limit to growth.
Summary of Ideas: Logistic Growth Model
• The growth rate of a stock over a time interval is the increase in the stock over the
time interval divided by the stock level at the beginning of the interval.
• If the stock follows an exponential model, its growth rate is constant.
• In real-life growing processes, growth cannot continue indefinitely at the same rate.
Thus, the growth rate tends to decline as the stock increases.
• The carrying capacity of an ecosystem is the maximum level of a stock that it can
sustainably support.
• The logistic growth model is the simplest model of limited growth. In the logistic
growth model, the growth rate is a linear function of the stock level, declining to zero
when the stock reaches carrying capacity.
• If the stock follows a logistic model, the plot of growth rate against stock level
is a straight line sloping downward. The x-intercept of this line gives the carrying
capacity, and the y-intercept gives the rate constant.
• If a quantity follows a logistic law of growth, its behavior over time is given by an
S-shaped curve. The growth rate reaches its peak at half the carrying capacity.
• These ideas can be applied to the extraction of nonrenewable resources. Peak oil
refers to the time of the global peak in oil extraction.
4.3. MEASURING FEEDBACK STRENGTH 221
4.3 Measuring Feedback Strength
Objectives
I can quantify the strength of a control in a stock-flow system.

I can identify parameters in a stock-flow system.
I can distinguish between linear and nonlinear controls.
I understand that a stock-flow system can have multiple equilibrium states.
I can distinguish between stable and unstable equilibrium states.
I know what is meant by the sensitivity of a stock-flow system to changes in a
parameter.
I can compute the sensitivity in terms of the strengths of the various controls in the
system.
I can compute how sensitivity is amplified by the presence of additional feedbacks.
In our discussion of the logistic model in the previous section, we asked whether a nonlinear
feedback (like the one that occurs in that model) is either amplifying or stabilizing (Question 1 on
page 212). The answer was “it can be either,” depending on the conditions in other parts of the model.
In this section we want to describe that “it can be either” more carefully. A feedback does not “snap”
from amplifying to stabilizing all at once: a feedback may start as “strongly amplifying,” then become
“weakly amplifying,” then “weakly stabilizing,” and then “strongly stabilizing,” as other aspects of
the system change. So to understand changes in feedbacks more carefully, we need to develop some
way of measuring and comparing the strength of feedback loops. That is what we are going to do
now.
4.3.1 The Strength of a Control

Recall that our system diagrams contain two types of arrows. The thick blue flow arrows denote
flows related to the stocks in each model; the thin red control arrows denote relationships by which
a stock level or parameter value controls a flow rate. In our familiar “bathtub” model (see Figure 19
on page 96), the outflow rate of water down the drain is controlled by the water depth, a proxy for
the stock of water in the tub. When the depth increases, so does the flow rate. We indicated this by
labeling the corresponding control arrow with a + sign.
In Section 3.3 we learned the feedback rule of signs: we can determine whether a feedback loop
is amplifying or stabilizing by counting signs around the loop (Rule 1 on page 156). We labeled each
control arrow with a + or − sign according to whether a small increase in the quantity at the “tail”
end of the arrow would tend to increase or decrease the rate of the flow at the “head” end of the
arrow. In Section 4.2, we learned the important fact that the sign that is appropriate can itself depend
on the state of the system: the same feedback can be amplifying in some situations and stabilizing in
others, as we saw for the logistic model on page 212.
Now we would like to go beyond using just a + or − sign and assign to each control arrow a
numerical measure of the “strength” of the effect that it represents. Let’s again revisit our bathtub
model from Section 2.1.1.
+ −
+ 0.46
Figure 21: Bathtub example indicating the strength of the control arrow.
In the bathtub model the outflow rate r is proportional to the depth h, so that r = Ch. The constant
C in Problem 1 on page 95 was 0.46 gallons per minute per inch, so that the flow rate down the drain
was
gal
r = 0.46 × h.
min × in
We were able to use this to calculate a dynamic equilibrium.
Whenever we have a relationship like this, in which a flow is proportional to a stock level, the
constant of proportionality determines the strength of the corresponding control. We can indicate the
strength in the stock-flow diagram as a label on the control, as shown in Figure 21.
Example 1: The strength of the control in the bathtub example is C = 0.46 gallons per minute per
inch. Notice carefully the units of the control strength: flow units (gallons per minute) divided by
stock
units (inches). Previously we had put a + sign on the control arrow to indicate that the flow increases
when the stock increases (Rule 1 on page 156). This + sign is still there, but now it becomes the
sign of the proportionality constant C. A negative proportionality constant would indicate a flow that
decreased as the stock increased.
In a proportional relationship a change in the stock level affects the flow rate, but the strength of the
control itself does not depend on the stock level. We call this a linear control.
Definition 1
A control arrow in a stock-flow diagram is linear if the flow rate at its “head” is equal to
Cs + k, where s is the stock (or parameter) level at its “tail” and C and k are constants. The
strength of the control is the constant of proportionality C.
The controls in an exponential model are linear controls with k = 0. Because we allow the “offset”
parameter k, the controls in an offset exponential model (Definition 2 on page 165) count as linear
also.
Example 2: In Problem 2 on page 156 we noted that your body metabolizes caffeine at a rate
proportional to the stock of caffeine in your bloodstream. This rate depends on a variety of factors
but the average rate is around 15% per hour among healthy adults. Since the rate is proportional to the
stock level, the control is linear, and its strength is
r = 15% hr−1 = 0.15 hr−1 .
Question 1: Why are the units “hours to the negative 1”?
Answer: The units for a control strength are flow units divided by stock units. Caffeine is typically
measured in milligrams (mg) and the corresponding flow units are “milligrams per hour.” Dividing
flow units by stock units gives
mg/ hr mg 1 1
= hr−1 .

= × =
mg hr mg
hr
You can read either 15% hr−1 or 0.15 hr−1 as “15 percent per year.” We’ll see similar units throughout
this section.
Caffeine in −
Metabolism
bloodstream
+0.15
Figure 22: Stock-flow model for caffeine metabolism.
Recall from Section 3.4 that a proportional relationship between a stock and its net inflow is the
hallmark of an exponential model. Let’s revisit our exponential model of the U.S. population for
another example involving linear controls.
As we noted in Example 5 on page 170, this model has two feedback loops: an amplifying “birth”
feedback that is stronger than the stabilizing “death” feedback.
+ −
+1.22% +0.75%
Figure 23: Population model for the U.S.
We calculated the strength of the birth feedback loop to be
+1.22% yr−1 ,
and the strength of the death feedback loop to be
−0.75% yr−1 .
To find the overall strength of the feedback governing the population we just add up the strengths of
the individual feedback loops. This gives
+1.22% yr−1 − 0.75% yr−1 = +0.47% yr−1 .
The fact that this sign is positive tells us that the overall feedback is positive (amplifying) in this
model. Moreover, the overall strength of +0.47% gives us a measure of the expected growth rate of
the overall population. This is exactly how we predicted the U.S. population (neglecting the important
factor of migration!) in Example 5 on page 170.
Remark 1: We need to be careful to distinguish between the sign of a control and the sign of a
feedback loop. Both controls in this example have positive strength: an increase in the population
increases both flows. But the strength of the death feedback is negative, following the feedback rule
of signs, Rule 1 on page 156. The + sign on the death-rate control gets multiplied by the − sign on
the outflow arrow, giving an overall feedback that is negative (stabilizing).
We can summarize the basic ideas of this calculation in a series of bullet points (Rule 1 below). We
are considering only models with a single stock, so that the only possible feedback loops are those
involving two arrows, one stock arrow and one control arrow.8
Rule 1: Rules for Measuring Feedbacks
• A feedback loop in a single-stock system consists of a control arrow and a flow arrow.
• The control arrow has a strength, which is measured in flow units divided by stock
units.
• Use the feedback rule of signs to determine the sign of the strength of the feedback
loop.
• To find the overall feedback in the system, add the strengths of all the feedback
loops.
• As in the feedback rule of signs, a negative overall feedback is stabilizing; a positive
overall feedback is amplifying.
Problem 1: In a hypothetical fish farm, there is a stock of S adult tilapia. Every month, 1,000 young
tilapia are added to the stock, 3% of the stock die from natural causes, and 25% of the stock are
harvested. Make a stock-flow diagram and compute the strength of all the feedback loops. What is the
overall feedback? Is it positive or negative? What is the equilibrium level of S?
Solution: In the stock-flow diagram, there is one inflow (at a constant rate) and two outflows
controlled by feedback loops. The diagram is shown in Figure 24 on the opposite page. We see that
there are two negative feedback loops, one of strength −3% per month and one of strength −25% per
month. The overall feedback is therefore stabilizing, with a total strength of −28% per month:
1 1 1
−3% − 25% = −28% .
month month month
We can find the (dynamic) equilibrium by setting the inflow equal to the outflow and solving for
the stock level, S. The inflow is 1,000 tilapia per month. The outflow is 28% of stock per month, or
0.28S month−1 :
1 tilapia
0.28S = 1, 000
month month
tilapia 1 month

S = 1, 000 × ≈ 3, 600 tilapia.
month
0.28
Notice that this is in fact an offset exponential model with stabilizing overall feedback: the “set point”
or equilibrium is 3,600 tilapia and the feedbacks will push the stock back toward that level if it goes
above or below the equilibrium.
8 More complicated feedback loops with multiple stocks, such as those in Example 3 on page 157, can also be analyzed,
but the methods involved are beyond the scope of this book.
+ Natural
death
−
Inflow of
Tilapia stock, S
juveniles
−
+
Harvesting
Figure 24: Stock-flow model for fish farm.
4.3.2 Stable and Unstable Equilibrium Points

When we introduced feedbacks in Section 3.3, we said that if a stock is near an equilibrium, a
stabilizing feedback will tend to push that stock back toward the equilibrium, while an amplifying
feedback will tend to push the stock further away. Now we’re going to turn that idea around and
characterize equilibrium points themselves as being stable or unstable according to whether the total
feedback in the system tends to stabilize (push back toward that equilibrium point) or amplify (push
away from that equilibrium). As we’ll see, nonlinear systems (those whose controls are not linear
in the sense of Definition 1 on page 222) can have many equilibria, some stable and some unstable.
Our guiding idea in the next two subsections is to find a workable definition of “feedback strength”
in nonlinear systems, which will allow us to say that a stable equilibrium is one at which the overall
feedback is negative (stabilizing) and an unstable equilibrium is one at which the overall feedback is
positive (amplifying). Let’s start with some examples.
Definition 2
An equilibrium (of a stock-flow system) is stable if after any small disturbance of the
equilibrium, the system tends to return to the equilibrium state.
Example 3: In Example 1 on page 222, we saw that the water depth at equilibrium was 4.8 inches.
The feedback is stabilizing; if the water is a little above or below this point, it will return to this
equilibrium. Therefore this equilibrium is stable.
Definition 3
An equilibrium (of a stock-flow system) is unstable if after any small disturbance of the
equilibrium, the system tends to move further away from the equilibrium state.
Example 4: Imagine a sharpened pencil balanced on its point. Theoretically, the system is in
equilibrium. But any small disturbance—even a puff of wind—will set a feedback process going that
will result in the pencil falling over and landing in a stable equilibrium position flat on the table. This
is an example of an unstable equilibrium. It also helps explain why we don’t in real life see systems
resting in unstable equilibrium—small disturbances are a fact of life, and any such disturbance will
destroy the unstable equilibrium and move the system toward a completely different state.
Critical Thinking
The definitions of stable and unstable equilibrium allow for the possibility of an equilibrium
that is neither stable nor unstable. Find the logical loophole that permits this. In systems
with more than one stock, such equilibria are very common. Think about the Lotka–Volterra
model (Figure 30 on page 160), which is in equilibrium at 0 lynx, 0 hare. What happens if
2 lynx are added (but no hare)? What happens if 2 hare are added (but no lynx)?
Outflow Example 5: Next we revisit Example 4 on

150 Inflow page 159, where we discussed phosphorus
levels in a lake. Recall that the stock of
Flow rate (tons/yr)
phosphorus, S, in the lake affects the water

100 quality. When the phosphorus content is low,
C the water quality is high; if the phosphorus
level becomes too high, the quality of the
B water is degraded. The outflow of phosphorus
50 from the lake is proportional to the stock. The
inflow comes from two sources. One source is
P
the amount of phosphorus that arrives in the
A lake from local pollution sources such as agri-
0
0 50 100 150 200 250 300 cultural runoff; we treat this as a parameter P.
Phosphorus (tons) The other source comes from phosphorus re-
leased from lake-bed sediments by algae; and
Figure 25: Graph of inflow and outflow of phosphorus when the pollution because algae are fertilized by phosphorus,
level is P = 15 tons per year. their level—and therefore the rate of inflow
of phosphorus from this source—depends on
the already existing stock of phosphorus in the
lake. Because of this dependence, the stock of phosphorus in the lake controls the inflow of new
phosphorus. A stock-flow diagram for this model is given at Figure 27 on page 159.
The key to the model is the way in which the phosphorus stock controls the inflow of new
phosphorus, which was studied in [58, 205]. To get a handle on the behavior that the authors of those
studies found, and its results for our model, it will be helpful to look at it graphically. In Figure 25
we’ve plotted both the inflow and outflow of phosphorus assuming that the pollution parameter, P, is
15 tons per year. Notice that the dependence of the inflow on the phosphorus stock is nonlinear.
Problem 2: How many equilibrium points are there in the phosphorus example depicted in Figure 25?
Estimate the phosphorus level at each equilibrium.
Solution: A dynamic equilibrium occurs whenever the inflow and outflow are equal. Graphically,
these are places where the plots of the inflow and outflow intersect. There are three such points; we’ve
marked them as A, B, and C in Figure 25. At point A the phosphorus level is about 30 tons, and
the flow rate is approximately 15 tons per year. At equilibrium B, the stock level is 130 tons, and at
equilibrium C, the stock of phosphorus is 200 tons. Note that at all three points, the pollution rate is
the same: 15 tons/yr. The high level of phosphorus at point C can occur due to heavy algal growth
“recycling” phosphorus from sediments back into the water (called a eutrophic state). We’ll see in
Section 4.4 how it’s possible for the system to shift from a “low” phosphorus to a “high” phosphorus
equilibrium, and vice versa.
Problem 3: Characterize each equilibrium point from the previous problem as either stable or
unstable.
Solution: Look again at the graph in Figure 25 and imagine that the phosphorus level is a little higher
than it would be at equilibrium point B, say 150 tons. Which flow is larger in this region of the graph,
the inflow or the outflow? We can see that it’s the inflow that is larger. When the inflow exceeds the
outflow, the phosphorus level should move up, away from the equilibrium at B and toward C.
If instead the stock level were a little below the equilibrium level at B, around 100 tons, then the
outflow would exceed the inflow, pushing the stock level down toward the equilibrium at point A. This
means that the equilibrium at B is unstable. By considering similar changes around points A and C we
can conclude that each of these equilibrium points are stable. If there is a small change from A, the
stock level will be pushed back toward A. The same holds for point C.
Outflow
150 Inflow
Flow rate (tons/yr)
100
C
B
50
P
A
0
0 50 100 150 200 250 300
Phosphorus (tons)
Figure 26: Plots of the inflow and outflow of phosphorus, indicating stable and unstable equilibrium points when the pollution
level is P = 15 tons/yr. The gray arrows indicate the direction in which the phosphorus stock level will move.
Remark 2: In Figure 26, the stock of phosphorus is indicated on the horizontal axis. The vertical
axis indicates the flow rates. An increase in phosphorus stock corresponds to rightward movement
in the graph; a decrease in phosphorus corresponds to leftward movement. The gray arrows indicate
the direction the phosphorus stock would move at each nonequilibrium point. For example, when the
stock of phosphorus is at 75 tons, the outflow is higher than the inflow; the stock will move down (that
is to say, to the left), toward the equilibrium at A. These gray arrows allow one to visualize directly
that the equilibria at A and C are stable and that the equilibrium at B is unstable.
4.3.3 Nonlinear Controls

The manner in which phosphorus is “recycled” from sediments in a lake is an example of a nonlinear
control. Many interesting stock-flow systems have nonlinear controls. Other examples that we have
considered so far include the logistic control from Definition 4 on page 211 (where the inflow to the
stock S is proportional to S(1 − S/C)) and the control given by Stefan’s law for outgoing thermal
radiation from a planet (Rule 1 on page 109; outflow from the stock is proportional to T 4 , where T is
the temperature). The graphs of these control functions are curves, not straight lines; that is what is
meant by saying that they are nonlinear.
In Problem 1 on page 212 we asked whether the feedback in the logistic model should be considered
amplifying or stabilizing. The surprising answer was that it depends on the value of S. An amplifying
feedback means that a (small) increase in the stock would yield an increase in the corresponding
flow, or in other words that the graph of the control function slopes “uphill.” This is true when S <
C/2. A stabilizing feedback means that a (small) increase in the stock would yield a decrease in the
corresponding flow, or in other words that the graph of the control function slopes “downhill.” This
will be true when S > C/2. Thus, as the evolution of the logistic model progresses, we see a gradual
change from an amplifying feedback regime (pushing the stock away from the equilibrium value
S = 0) to a stabilizing feedback regime (pushing the stock toward the equilibrium value S = C).
If we think about it, this idea tells us what we should mean by the strength of a nonlinear control.
Suppose that the control function expresses the flow, F, in terms of the stock S. Imagine that the stock
level is changed by a small amount, say from an original level S0 to a new level S1 . Then the flow level
will also change by a small amount, from its original level F0 to its new level F1 . The ratio
Small change in Flow F1 − F0
=
Small change in Stock S1 − S0
is what we will mean by the strength of the control.
Definition 4
The strength of a nonlinear control arrow is measured by making a small change in the
stock or parameter level at its tail and noting the corresponding small change in the flow
rate at its head. The ratio
Small change in Flow
Small change in Stock or Parameter
is the strength of the control. In general this strength will depend on the stock (or parameter)
value.
Example 6: Consider the control function for a logistic model with C = 100 and r = 0.3, that is,
F = 0.3S(100 − S).
(This is the control function we graphed in Figure 14 on page 212.) When S0 = 30, we obtain F0 =
0.3 × 30 × 70 = 630. Making a “small change” of 1 unit in the stock, we take S1 = 31 and then
F1 = 0.3 × 31 × 69 ≈ 642. Thus the strength of the control at the stock value S0 = 30 is approximately
642 − 630
= 12.
31 − 30
Problem 4: Compute the strength of the control in the logistic model of Example 6 when S0 = 75.
Solution: We calculate in the same way as before. Let’s set out our work in a table:
S0 = 75 F0 = 0.3 × 75 × 25 ≈ 562
S1 = 76 F1 = 0.3 × 76 × 24 ≈ 547
The strength of the control at the stock value S0 = 75 is approximately

562 − 547
= −15.
76 − 75
Notice that we obtained a negative number, reflecting the fact that for this stock value (greater than
C/2) the overall feedback is stabilizing.
Question 2: You said, “a small change in the stock.” Exactly how small do you mean by “small”?
Flow Rate
Answer: Great question! To answer it, think about what
we are doing in terms of the graph of the control function
F1
(Figure 27). The quantity (F1 − F0 )/(S1 − S0 ) that we have
computed is actually the slope (“rise over run”) of the red line F0
segment in this figure. What “small” actually means is “small
enough that the red line segment from (S0 , F0 ) to (S1 , F1 ) is
indistinguishable from the green curve joining those points.”
In the figure we have taken S0 = 20 and S1 = 40, and we can S0 S1
Stock Level
see a tiny amount of daylight between red and green. Taking
S0 = 30 and S1 = 31 will definitely be “small enough” for our Figure 27: The strength of the control is the slope of the
red line.
purposes.
Remark 3: If you have studied some economics, you may be

familiar with the idea of the marginal cost of producing an item. Suppose a factory can make 10,000
widgets for $511.17, but that it can make 10,001 widgets for $511.18. In that circumstance, one says
that the marginal cost of the 10,001st widget (that is, the extra cost added by producing it) is one cent.
The marginal cost is the small change added to the total cost by making a small change in the number
of widgets produced. That is like our control strength, which is the small change added to the flow by
making a small change in the stock or parameter that is controlling it.
Now that we have the notion of strength even for a nonlinear control, we can give the rule that
we hinted at at the beginning of Section 4.3.2, relating overall feedback strength to the stability of
an equilibrium point. The overall feedback strength is, of course, the sum of the strengths of all the
individual feedback loops (if there happens to be more than one).
Rule 2: Equilibrium Types Rule
Imagine a system with a single stock and nonlinear controls, at an equilibrium value.
Compute the strength of each control (at the equilibrium value) and from these obtain
the overall feedback strength by addition, as in Rule 1 on page 224. Then:
• If the overall feedback strength is negative, then that equilibrium is stable.
• If the overall feedback strength is positive, then that equilibrium is unstable.
Example 7: Let’s compute the strength of the overall feedback for the phosphorus model at an
equilibrium point, say the point A in Figure 26 on page 227. The equilibrium level of the stock S
is approximately 30 tons. In the stock-flow diagram below, we have calculated the strengths of the
various controls at this equilibrium point and marked them on the appropriate arrows. The inflow

Phosphorus in Lake, S Phosphorus
+0.04 +0.5
+1
Pollution, P
Figure 28: Stock-flow diagram for lake phosphorus model, showing control strengths at equilibrium A.
feedback is amplifying with strength +0.04 yr−1 and the outflow feedback is stabilizing with strength
−0.5 yr−1 . The strength of the overall feedback is
0.04 yr−1 − 0.5 yr−1 = −0.46 yr−1 .
Since the strength of the overall feedback is negative, this feedback is stabilizing. By Rule 2 on the
previous page this means that point A is a stable equilibrium, which confirms what we had already
found graphically (Figure 26 on page 227).
Notice that the mere presence of one amplifying feedback loop in a system does not by itself make
the equilibrium unstable; that amplifying feedback might by canceled out by a greater amount of
stabilizing feedback somewhere else (as occurred in Example 7). What matters is the overall amount
of feedback from all the loops combined.
Question 3: How did you come up with those two numbers, 0.04 and 0.5, for the strengths of the
controls?
Answer: Let’s talk about the easy one first! The output control, the red line in Figure 25 on page 226,
is a straight line through the origin. That means that the corresponding control is linear, and its strength
is just the slope of the line, which is 0.5 = 12 , since the line passes through the origin and the point
(200,100).
For the other control, you would need to know the mathematical expression for the blue curve in
Figure 25. We haven’t been telling you what that is, because it’s a bit messy, but if you really want to
know, it is P + 100S4 /(1304 + S4 ), where P is the pollution parameter (here equal to 15). If you call
this expression f (S), what one has to do is to evaluate
f (31) − f (30)
,
31 − 30
assuming that a change of 1 in the stock level is “small enough” (see Question 2 on the previous
page). Aren’t you glad we did this calculation for you? In most of the problems in this book that
involve nonlinear controls, we will already have calculated the control strengths, as we did in this
example.
Figure 29: Kelp forest at Monterey Bay Aquarium.
Here is another example.

Problem 5: Kelp forests are marine ecosystems that provide shelter and food for many different
species, supporting diverse plant and animal communities [238]. Sea urchins graze on the kelp; if
the urchin population is high enough, the kelp forest ecosystem can be overgrazed, collapsing into
an “urchin barren.” Here we treat the amount of kelp biomass as the stock in a stock-flow model.
The biomass is measured in kilograms per square meter. The inflow represents growth of existing
kelp fronds as well as the establishment of new fronds. The outflow of kelp occurs through the
shedding of kelp detritus and through grazing of kelp by sea urchins. The inflow and outflow units are
kg week−1 m−2 , or kilograms per urchin per square meter.
We treat the urchin population as a parameter affecting the outflow. The outflow due to grazing
depends on the urchin population. The stock-flow diagram in Figure 30 shows the strengths of each
control at a particular equilibrium, where u = 15 urchin m−2 (urchins per square meter), corresponding
to a healthy kelp forest ecosystem. Find out whether the equilibrium is stable or unstable.
new kelp + Kelp − kelp detritus

growth Biomass Q and grazing
+.019 +0.034 +.01
Urchin
population, u
Figure 30: Kelp forest stock-flow model.
Solution: There are two feedback loops here, one amplifying with strength +0.019 week−1 and one
stabilizing with strength −0.034 week−1 . The overall feedback strength is
+0.019 week−1 − 0.034 week−1 = −0.015 week−1 .
Thus by Rule 2 on page 229, the equilibrium is stable.

Notice that the control arrow from the urchin population to the outflow spigot, of strength
0.01 kg week−1 urchin−1 , does not appear in any feedback loop and thus does not affect the
calculation above. These kinds of arrows will be relevant in the next section, when we think about
the effects of a change in a parameter on the equilibrium state.
4.3.4 Changing the Equilibrium
Consider a stock-flow system that is resting at a stable equilibrium (overall stabilizing feedback) that
depends on some parameter. For example, we might consider the phosphorus model in the dynamic
equilibrium described in Example 7 on page 230, where the pollution level was P = 15 t/ yr and the
equilibrium level of phosphorus was S = 30 tons. Imagine now that the pollution level increases to
P = 16 t/ yr. The equilibrium will also change, rising from 30 tons to a little over 32 tons. We could say
2t
that the sensitivity of the equilibrium to the parameter P is a little more than = 2 yr, meaning
1 t/ yr
that a 1 t/ yr change in the parameter produced a slightly more than 2 t change in the equilibrium. Or
we might consider the bathtub system (Example 1 on page 70), which is in dynamic equilibrium with
inflow equal to 2 gallons per minute and equilibrium level 4.8 inches. If the inflow is decreased to 1.5
gallons per minute, the equilibrium level will decrease to 3.6 inches: we could say that the sensitivity
(4.8 − 3.6) in in min
of the equilibrium level to change in the input flow rate is = 2.4 . Or for a
(2 − 1.5) gal min −1 gal
third example (related to our work in Section 2.4.5), consider the climate model with a “greenhouse”
parameter g. With g = 0.9 we see an equilibrium temperature of T = 298 K; changing the parameter to
g = 0.8 changes the corresponding equilibrium value to T = 294 K. We could say that the sensitivity
of the equilibrium temperature to a change in the parameter g is roughly 4 K per 0.1 change in g, or
4/0.1 = 40 K per unit. Let’s define exactly what we mean by “sensitivity.”
Definition 5
Suppose a stock-flow model has a parameter P. When the parameter value P is P0 , the model
rests in a stable equilibrium where the value of a certain stock is S0 . If P is changed to a
new value P1 (close to P0 ), then the equilibrium stock level will also change to a new value,
S1 . The sensitivity of the stock S to the parameter P is
i S −S
1 0
h
Sens(S; P) or just Sens = .
P1 − P0
You may also see this written Sens = ∆S/∆P, where the symbol ∆ means “change in”: “the
sensitivity is the change in the stock divided by the change in the parameter.”
In this section we’ll see how to calculate these sensitivities in terms of the strengths of the controls
and feedbacks in our stock-flow model. Here is the rule.
Rule 3: Sensitivity Rule
• Suppose that a stock in a stock-flow system S is maintained at a stable equilibrium

value S0 by an overall negative feedback of strength q. (Note that q will be a negative
number.)
• And suppose also that a parameter P affects the net flow into S by a control of strength
p. (This p could be positive—if increasing the parameter P causes an increase in
inflow or decrease in outflow—or negative—if increasing the parameter P causes a
decrease in inflow or increase in outflow.)
• Then the sensitivity of S on P is Sens(S; P) = −p/q. In other words, a small change
∆P in P will produce a change
p
∆S = − ∆P
q
in the equilibrium value of the stock.
Let’s see how this works out in the phosphorus model.

Problem 6: Compute the sensitivity of the phosphorus level S to a change in the pollution level P for
the equilibrium described in Example 7 on page 230 and depicted as point A in Figure 26 on page 227.
Solution: The pollution parameter P affects the total inflow. The strength of the pollution control is
p = 1 (a pure number in this case). In Example 7 we computed the strength of the overall feedback as
q = −0.46 yr−1 . Using the sensitivity rule, we get
p 1
sensitivity = − =− ≈ 2.2 yr.
q −0.46 yr−1
This indicates that if the pollution rate increases from 15 to 16 tons per year, the equilibrium stock of
phosphorus will increase from 30 to a little more than 32 tons:
p
∆S = − ∆P
q
ton
= 2.2 yr × 1 = 2.2 tons.
yr
It is important to understand that the results of these sensitivity calculations are valid only for small
changes in the parameter. (Refer back to Question 2 on page 229.) For example, suppose that the
pollution rate rises from 15 to 25 tons/yr. This is not a small change. But we can still use our graphical
approach to find the new equilibrium value. The change in P has shifted the entire inflow curve up
by 10 tons/yr, and as a result the equilibrium has moved to a new point, indicated by A on the graph
shown in Figure 31 on the next page.
20
Problem 7: The strengths for the feedbacks at the new equilibrium (point A in Figure 31 on the next
page) are shown in the stock-flow diagram Figure 32 on the next page. Calculate the sensitivity of the
stock S to a change in the pollution level P. Is the system more or less sensitive to a change in the
pollution level then it was at the equilibrium in Problem 6?
Outflow
150 Inflow
Flow rate (tons/yr)

100
50
P A
0
0 50 100 150 200 250 300
Phosphorus (tons)
Figure 31: Plots of the inflow and outflow of phosphorus when the pollution rate is P = 25 tons/yr.
Solution: The strength of the overall feedback is now q = 0.28 − 0.5 = −0.22 yr−1 . The strength of
the pollution control is still p = 1. Calculating the sensitivity, we get
p 1
sensitivity = − =− ≈ 4.5 yr.
q −.22 yr−1
An increase in the pollution rate of one ton per year is now expected to increase the equilibrium stock
by 4.5 tons:
p
∆S = − ∆P
q
ton
= 4.5 yr × 1 = 4.5 tons.
yr
The system is more than twice as sensitive to a change in P as it was when in equilibrium with
P = 15 t/ yr (Problem 6 on the previous page).

+0.28 +0.5
+1
Pollution, P
Figure 32: Stock-flow diagram for phosphorus at new equilibrium (P = 25 t/ yr).
Question 4: Consider the kelp forest at equilibrium as described in Problem 5 on page 231. If the
sea urchin population increases, do we expect the equilibrium level of kelp biomass to increase,
decrease, or remain the same? Calculate the sensitivity of the kelp biomass Q to a change in the
urchin population u.
Solution: If the urchin population increases, this increases the outflow of kelp biomass. Thus we
expect a decrease in the equilibrium level of the stock.
The strength of the overall feedback in this system can be found from the data in Figure 30 on
page 231 as
q = +0.019 week−1 − 0.034 week−1 = −0.015 week−1 .
The strength of the control of the outflow is +0.01 kg week−1 urchin−1 . Considered as ap-
plying to the net flow (inflow minus outflow), the control should have the opposite sign, p =
−0.01 kg week−1 urchin−1 . By the sensitivity rule,
p 1 −0.01 kg week−1 urchin−1
Sens(Q; u) = − = −p × = − ≈ −0.67 kg urchin−1 .
q q −0.015 week−1
Be careful to note that both p and q are negative in this example. According to the sensitivity rule, that
yields a negative sensitivity; that is, an increase in urchin population will lead to a decrease in kelp
biomass, in accordance with our expectations.
Remark 4: When a parameter is changed, the system does not pass instantaneously to its new
equilibrium. There is some period during which the stock needs to adjust itself to the newly changed
flows; this is called a transient effect. We are assuming that the parameters change slowly enough
that transient effects can be neglected. This assumption is known as a quasi-static approximation.
The length of time the transient effects last is governed by how long it takes the stabilizing feedback
loops to restore the stock to equilibrium. Thus, the less the overall strength of the stabilizing feedback,
the longer the transients will last. We will return to this idea, called “critical slowing down,” in the
section on tipping points (Section 4.4).
Amplification Through Additional Feedback

There is another way to present these same ideas that is often considered in climate modeling (see
Remark 5 on page 204). This is to start by computing the sensitivity of a “bare” model, stabilized by
only the most basic kind of negative feedback. For example, in a climate model, one might consider
only the negative feedback given by Stefan’s law, which predicts that the net outflow of heat from the
Earth will increase if its temperature increases.9 Having calculated the sensitivity of our equilibrium
value to a parameter change in this “bare” model, we may later want to incorporate the effect of
additional feedbacks and ask how they change the “bare” sensitivity. Attempting to model climate
accurately involves many of these feedbacks: global temperatures influence ice cover, vegetation
levels, water vapor levels in the atmosphere and therefore cloud cover, release of methane trapped in
permafrost, and a variety of other things, all of which in turn influence the inflow of heat to the Earth
via short-wave radiation or its outflow via long-wave radiation or both. Scientifically, quantifying
these feedbacks is a bear of a problem. But the mathematical question—how does one incorporate the
effect of additional feedbacks, assuming one knows how big they are?—has an answer that is much
less hard to explain.
Suppose that the stabilizing feedbacks in a “bare” model have total strength qbare (qbare is negative
because this is a stabilizing feedback). According to the sensitivity rule, the sensitivity of the “bare”
system to a control with strength p is
p
Sensbare = − .
qbare
Now suppose that the model is augmented by considering additional feedbacks, of overall strength
qadditional (this can be of either sign, since the additional feedbacks could be positive or negative). The
feedback for the augmented system (bare system together with the additional feedbacks) is then
9 Reminder: This is a negative feedback because it is a control of positive strength applied to an outflow: a greater outflow
of heat from the Earth, produced by higher temperatures, will tend to reduce those temperatures.
qaugmented = qbare + qadditional .
Using the sensitivity rule again, the sensitivity of the augmented system is
p
Sensaugmented = − .
qaugmented
Therefore
Sensaugmented qbare qbare 1
= = = q . (5)
Sensbare qaugmented qbare + qadditional 1+
additional
qbare
Notice that this ratio does not depend on the control strength p. All control sensitivities get multiplied
by this same amount when the additional feedbacks are included. The amount by which we multiply
is called the amplification factor:
Rule 4: Law of Amplification
Imagine a stock-flow model stabilized by a certain (“bare”) level of negative

feedback. When additional feedbacks (positive or negative) are included in the model,
all sensitivities are multiplied by the same factor A, called the amplification factor.
The amplification factor A associated with additional feedback qadditional augmenting
a system with bare feedback qbare is
1
A= ,
1−k
where k = −qadditional /qbare is called the feedback quotient.
The formula 1/(1 − k) is just the same as the formula in equation (5). The various minus signs are set
up so that k will have the same sign as the additional feedback qadditional .
Critical Thinking
Have you ever used a microphone or PA system? If so, you are probably familiar with
the implications of the feedback quotient k. In this situation it represents the fraction of an
initial sound at the microphone that is received back at the microphone via the loudspeakers.
If this fraction is greater than 1, the system will “howl” uncontrollably. But if k is less than
one, but still close to it, the system will not “howl,” but it will “boom”—every slight sound
at the microphone will be turned into a series of echoes that pass many times around the
feedback loop before they die away. Our analogy is not exact, because of the delay in sound
transmission from speaker to microphone: think about what difference that may make.
The graph of the function 1/(1 − k) that relates the amplification A

factor to the feedback quotient is shown in Figure 33. Notice how big the
amplification factor gets (and how steep the graph is) as k approaches
1, that is, as the additional positive feedback qadditional almost cancels
out the bare negative feedback qbare . What this tells us is that in these
circumstances, the system is going to be hugely sensitive to small
disturbances. It also tells us that even a tiny bit of extra positive feedback
will make the system even more hugely sensitive. In fact, it may be
enough to push it past a tipping point (see Section 4.4).
Remark 5: So what is the amplification factor in climate models? Most
of the uncertainties in estimating the climate sensitivity (Definition 7 k
1
on page 203) arise from uncertainties in estimating feedbacks; there
is reasonable agreement that the “bare” climate sensitivity is approxi- Figure 33: Amplification factor.
mately 1 K for a doubling of carbon dioxide. But feedbacks are difficult
to understand and quantify. One older estimate for the overall positive
feedbacks in the climate system gave a range of values for k, the feedback ratio, between 0.4 and
0.78 [157, page 156]. Using the law of amplification, this corresponds to a range of values for A
between 1.7 and 4.5. Multiplying by the 1 K “bare” sensitivity gives the range 1.7 K to 4.5 K for the
“augmented” climate sensitivity of the Earth—very close to the range suggested in the IPCC’s AR5
report [253] over a decade later. For the future of humanity, it matters a great deal in which end of this
range the climate sensitivity falls.
Think about it. . .
Elon Musk, founder of Tesla, is quoted as saying [353]:

We’re running the most dangerous experiment in history right now, which is
to see how much carbon dioxide the atmosphere can handle before there is an
environmental catastrophe.
Do you agree? Why or why not?
Critical Thinking
Algal blooms are not unusual in southern Florida’s Lake Okeechobee during the summer
months, but in 2016 an algal bloom began in the early spring and grew to encompass over
300 square miles. By the end of June, Florida governor Rick Scott had declared a state of
emergency in surrounding communities affected by the bloom.
What is the role of phosphorus in Lake Okeechobee’s algal blooms? Are there other con-
tributing factors? What effects do these blooms have? See NASA’s Earth Observatory [311]
and the additional references provided there.
Summary of Ideas: Measuring Feedback Strength
• The strength of a control arrow is the constant of proportionality between small

changes in the flow rate at its head and small changes in the stock or parameter at its
tail.
• If a control is nonlinear, the strength of the control may depend on the state of the
system.
• The strength of a feedback loop is the product of the strength of its control arrow and
the sign of its flow arrow.
• The overall strength of feedback is the sum of the strengths of all the individual
feedback loops.
• The sensitivity of the system to a change in a parameter is the constant of
proportionality between a small change in the equilibrium level of the system and
a small change in the parameter that gave rise to it.
• The sensitivity rule (Rule 3 on page 233) allows you to compute the sensitivity of a
system in terms of the strengths of the various controls that it contains.
• The sensitivity of the system may be changed if the system is augmented by
additional feedbacks. The factor by which the sensitivity is changed is called the
amplification factor.
• The amplification factor can be computed in terms of feedback strengths by the law
of amplification (Rule 4 on page 236).
• Computing the amplification factor associated with feedbacks in the climate system
is one of the most uncertain parts of climate modeling.
4.4. TIPPING POINTS 239
4.4 Tipping Points
Objectives
I understand how a tipping point may occur in a model that includes a nonlinear
feedback and a parameter.
From the model equations, I can estimate the parameter value at which a tipping
point occurs.
I can relate the appearance of a tipping point to the increasing sensitivity of the
system as described in Section 4.3.4.
I understand the notion of critical slowing down and how it may provide empirical
evidence that a tipping point is approaching.
I can apply these concepts to the Earth’s climate, and specifically to the Budyko model
for the ice-albedo feedback.
I understand that many environmental systems exhibit tipping point behavior.
I am aware of some other possible tipping points in the global climate system.
A press conference on December 3, 2013, at the National

Academy of Sciences in Washington, DC, was the public launch of
a new expert report, “Abrupt impacts of climate change: Anticipat-
ing surprises.” The report is available free online from the National
Academies Press [349].
“Report warns of climate change tipping points within our
lifetime,” headlined a press release related to the report from UC
Berkeley [285]. The press release quotes Berkeley professor Tony
Barnosky, one of the report’s authors, as saying
Our report focuses on abrupt change, that is, things

that happen within a few years to decades: basically,
over short enough time scales that young people living
today would see the societal impacts brought on by Figure 34: Cover of NAS report.
faster-than-normal planetary changes.
Barnosky is professor of integrative biology and a member of the

Berkeley Initiative in Global Change Biology (BIGCB).
But what exactly is a “tipping point”? In this section we will examine the mathematics of tipping
points, using our work on simple climate models (Section 2.4) and the model for phosphorus in a lake
(Problem 2 on page 226) as examples. In the previous section we saw that some models (for example,
the phosphorus model) may have multiple equilibria: there is more than one point at which the system
can be “in balance.” We also saw how a change in a parameter could change the configuration of those
equilibria (Section 4.3.4). Now we’ll see how the combination of these two key ideas gives rise to the
possibility of sudden jumps in the equilibrium state. These are the “tipping points.”

Figure 35: The naked planet climate model, from Section 2.4.
4.4.1 Solving the Climate Model Graphically

Let’s review our work on climate models from Section 2.4. We started with the simplest possible
model, which we referred to as the “naked planet” model because it ignored the effect of the
atmosphere. In this model, Earth’s temperature is modeled by a stock-flow system. The “stock” here
is a stock of heat, which is measured by its proxy, the Earth’s surface temperature T . On the left
of Figure 35, heat flows in from the Sun via short-wave radiation, at a constant rate of 14 L(1 − α)
watts per square meter of Earth’s surface; here L is the solar constant, approximately 1350 watts per
square meter, and α is the albedo, currently around 0.3. On the right-hand side, heat leaves Earth
via long-wave radiation, at a rate dependent on the temperature T . In the “naked planet” model, we
assume that the rate of this heat flow is given by Stefan’s law as σ T 4 watts per square meter, where
σ ≈ 5.7 × 10−8 W K−4 m−2 is Stefan’s constant and T is the temperature in kelvins. The temperature
dependence here gives rise to a stabilizing feedback loop that keeps Earth’s temperature stable; the
equilibrium temperature for the model is the value of T for which the incoming and outgoing heat
flows are equal.
Inflow In Section 2.4 we rearranged the equations to get a

600 Outflow formula for the equilibrium temperature,
Energy flow (W/m2 )
r
4
L(1 − α)
400 T= .
4σ
But we could also have solved the model by drawing
200 a graph; that is the way will we think about it in this
section.
Specifically, we could plot two curves showing energy
0 flow rate against temperature: one for the incoming
220 240 260 280 300 320
energy from the Sun, and one for the outgoing energy
Temperature, T (kelvins)
radiated by the Earth. At the point where the two curves
Figure 36: Graphical representation of equilibrium in the intersect, the incoming and outgoing energy flow rates
“naked planet” model. are equal; this will be the equilibrium temperature. This
is what we have done in Figure 36.
Question 1: Why is the graph of the incoming energy
flow rate a horizontal line?
Answer: In the naked planet model, the incoming flow rate is a constant, 14 L(1 − α) ≈ 236 W/ m2 .
Because this quantity does not depend on T , its graph is a horizontal line.
The graph of the outgoing energy flow rate is curved upward according to Stefan’s law. As you
can see in Figure 36, the incoming and outgoing curves intersect at roughly 255 K, the temperature
predicted for the naked planet model by our calculations in Section 2.4.
You’ll recall from Section 2.4 that this prediction was too small because we had not taken the
greenhouse effect of the atmosphere into account. Our next model attempted to do so by considering
the atmosphere as a “glass ball,” transparent to the incoming short-wave solar radiation, but opaque
to the long-wave infrared heat radiated from Earth’s surface. This gave us the “glass ball model.” In
deriving this model in Section 2.4.4 we considered the temperature of the “atmosphere layer” and
“Earth surface layer” separately, but the result as far as the energy balance of the Earth itself was
concerned was that we saw the same (constant) rate of energy inflow but a different rate of energy
outflow, namely 12 σ T 4 rather than σ T 4 as in the naked planet model.
A graphical representation of equilibrium in the glass
ball model is shown in Figure 37. The green curve shows Inflow
600 Outflow (naked planet)
the energy outflow in the glass ball model as a function
Outflow (glass ball)
of temperature: for comparison we’ve also shown the
Energy flow (W/m2 )

energy outflow curve in the naked planet model (red)
and the energy inflow “curve” (actually a straight line, 400
the same for both models, blue). In our graph, we find
that the “outgoing energy” curve for the single layer
greenhouse model has moved lower, and therefore its 200
point of intersection with the “incoming energy” line
has moved further to the right (see Figure 37). As we
computed in Section 2.4.4, the equilibrium temperature 0
220 240 260 280 300 320
is now a little above 300 K. Temperature, T (kelvins)
At the end of Section 2.4 we introduced a parameter
to model the varying strength of the greenhouse effect. Figure 37: Graphical representation of equilibrium taking the
Using this parameter g (the greenhouse effect strength, greenhouse effect into account.
or
GHEstrength for short), the energy outflow rate is
1
× σ T 4.
1+g
When g = 0 we have no greenhouse effect and we get the naked planet model. When g = 1 we
have “full” greenhouse effect, and we get the glass ball model. Intermediate values of g model the
greenhouse effect at different “strengths.” A corresponding stock-flow diagram is shown in Figure 38.

+ −
GHE strength, g
Figure 38: Climate model with greenhouse parameter.
Question 2: What does the − sign on the control arrow from the parameter to the outgoing flow rate
tell us?
Answer: It tells us that as the parameter g increases (that is, the greenhouse effect gets stronger), the
outgoing energy flow rate decreases. We can see this from the formula: the bigger g is, the smaller is
the factor 1/(1 + g).
As g varies, we obtain a family of energy outflow curves, of the same general shape, between the
green and red curves in Figure 39 on the next page. As g increases (that is, the greenhouse effect gets
stronger), the curves will move down toward the green curve and the equilibrium point will move
further to the right. Conversely, as g decreases (that is, the greenhouse effect gets weaker), the curves
will move up toward the red curve and the equilibrium point will move further to the left.
The Ice-Albedo Feedback
In our picture so far, there is only one equilibrium

Inflow temperature point for each value of the greenhouse pa-
600 Outflow (g = 0) rameter g. But we have neglected an important issue: the
Outflow (g = 0.3)
effect of Earth’s temperature on its albedo and therefore
Energy flow (W/m2 )
Outflow (g = 0.6)
Outflow (g = 1) on the level of incoming energy. Recall that the albedo,
400 denoted by α, is a measurement of how “shiny” the
Earth is—how much of the energy arriving from the Sun
gets reflected immediately back into space.
200 We have been working with the approximation α ≈
0.3. That is close to correct for the Earth as it is at
present. But what if the average temperature of the Earth
0 were to get to freezing or below?
220 240 260 280 300 320
Temperature, T (kelvins)
Problem 1: Suppose that instead of the balmy global
average temperature 288 K (15 ◦ C) we currently enjoy,
Figure 39: Equilibrium at various values of g.
Earth’s global average temperature was 255 K (−18 ◦ C).
Describe how you would expect this to affect the Earth’s albedo.
Solution: With Earth’s average temperature well below freezing, we would expect most of Earth to be
covered in ice and snow. These are highly reflective, so more of the Sun’s energy would be reflected
back into space, decreasing the solar radiation that is absorbed by the planet. This corresponds to a
higher albedo.
Remark 1: In the previous problem we imagined the Earth having a surface temperature well below
freezing. Let’s take a moment to imagine the process of the planet cooling down. As the planet cools,
the snow and ice cover increases; this leads to a higher albedo, reflecting more of the Sun’s energy
back into space. The inflow of solar radiation the Earth absorbs is decreased, which cools the planet
further, increasing the extent of the snow and ice cover, which in turn increases the albedo even
more. This amplifying feedback is the ice-albedo feedback, one of many amplifying and stabilizing
feedbacks that appear when the Earth’s climate system is analyzed more carefully.
Problem 2: Modify the stock-flow diagram in Figure 35 on page 240 to incorporate the effect of the
ice-albedo feedback.
Solution: The ice-albedo feedback involves the planet temperature and the inflow of short-wave
radiation from the Sun. The stock of heat (represented by the temperature) affects the inflow,
indirectly, through the amount of snow and ice. We can add a control arrow from the temperature
stock to the inflow of short-wave radiation. If the temperature increases, the amount of ice and snow
decreases, decreasing the albedo and increasing the inflow of solar radiation. We indicate this by
labeling the control arrow with a + sign. The resulting feedback loop is amplifying.
Short-wave + Planet − Long-wave

+ + −
GHE strength, g
Figure 40: Climate model with ice-albedo feedback.
Critical Thinking
In the previous problem and the discussion to follow, we represent the ice-albedo feedback
by a control between the planet temperature and the inflow of solar radiation. This is a
single-stock model, and it therefore allows us to use the ideas about single-stock models and
feedback strength developed in Section 4.3. But there are a lot of details about ice formation
and so on tucked away in that one control arrow! A more detailed way of including the ice-
albedo feedback in our stock-flow diagram would be to treat the Earth’s snow and ice cover
as a second stock (see Exercise 31 on page 179 in Chapter 3), but then we would be unable
to use the ideas of Section 4.3, which are crucial to the discussion that follows. Think about
the tradeoff here, which is one that applies to all kinds of modeling: is my model too simple
to be useful, or too complicated to be understandable? Compare the quotation from [156]
on page 74.
To incorporate the ice-albedo feedback in the graphical representation of our model we need to
be more specific about this relationship between the temperature T and the albedo α. The albedo
should have the current value, 0.3, when the Earth is relatively warm (when T > 280 K or so) and
a much higher value, about 0.7, when the Earth is frozen (for T < 260), with a smooth transition in
between. This is the blue curve in Figures 41(a),(b) below. The exact mathematical details of how
this is arranged are not actually very important. The important thing about the resulting model is its
qualitative behavior; specifically, what happens to the equilibrium when the model is modified in this
way.
The equilibrium for the naked planet model (g = 0) has now moved so far to the left that we cannot
see it on the scale of our graph (it’s at about 200 K). More interesting, though, is what happens to the
model in the “glass ball” case, g = 1. Just as we saw with the model for phosphorus concentrations
in a lake (Problem 2 on page 226), there are now three possible equilibria, represented by points of
intersection of the incoming and outgoing energy flow curves (Figure 41). One is just above 300 K
(this is the same equilibrium that we had found before for g = 1), one at about 270 K, and one at about
245 K. In Figure 41(b) we have magnified the blue and green curves and explicitly labeled the three
equilibrium points A, B, and C. We can describe them as follows.
• At point A, the planet is more or less frozen solid. It stays in this low-temperature equilibrium
because the reflective ice and snow sends a lot of the Sun’s heat back into space before the
planet has a chance to warm up. This is “Snowball Earth.” This equilibrium was proposed as a
theoretical possibility by a Russian climatologist, Mikhail Budyko, in the early 1960s, on the
basis of a model like the one we have just discussed. Budyko at first did not believe that Earth
had ever actually been in a “snowball” state. Later, however, geological evidence was discovered
that suggested that several “snowball” events had in fact occurred in geological history, most
recently in the Neoproterozoic period, approximately 650 million years ago; see [177] for more
on the evidence for this and the background to the model that we are using here.
• Point B is the “middle equilibrium.” We’ll see in a moment that this is an unstable equilibrium
point (Section 4.3.2), like the middle equilibrium in the lake-phosphorus model: one therefore
expects that this is not a state in which Earth would remain for long.
• Point C is (roughly)10 our present climate: not too hot and not too cold. We’ll call this
“Goldilocks Earth.”
400
Inflow Inflow
600 Outflow (g = 0) Outflow (g = 1)
Outflow (g = 1) 300
Energy flow (W/m2 )
Energy flow (W/m2 )

400 C
200
B
200
100
A
0 0
220 240 260 280 300 320 220 240 260 280 300 320
T temperature (kelvins) T temperature (kelvins)
(a) Graphical representation of climate model with ice-albedo (b) Magnified version of previous figure, showing three
feedback. equilibria when g = 1.
Figure 41: The ice-albedo feedback.
Problem 3: Which equilibrium points in Figure 41(b) are stable? Which are unstable?
Solution: We follow the same line of reasoning we used in determining the stability of the equilibrium
points in the phosphorus model (Problem 3 on page 227). To check whether a particular equilibrium
is stable or unstable, consider a nearby temperature and compare the inflow and outflow rates at that
point. Let’s look at the equilibrium point A, where T = 245 K. Imagine now that the temperature is
a bit warmer than this, say 250 K. Here, the green curve is above the blue curve: the rate of energy
inflow is less than the outflow energy rate. The Earth will cool and the temperature will move to
the left, down toward T = 245 K. On the other side of this equilibrium (at 230 K for example), the
inflow rate exceeds the outflow rate. When the temperature is in this region, it will move up, toward
the equilibrium at A. Whenever the system is near the point A, the dynamics of the Earth system will
drive the temperature toward A. The point A “attracts” nearby points; this is a stable equilibrium.
Repeating this analysis for the other two points, we see that C is also a stable equilibrium and B is an
unstable equilibrium.
Remark 2: In any physical system, small changes around an equilibrium are natural. For this reason
we don’t expect a real-life system to settle into an unstable equilibrium like the one at point B in
the Earth system above: even if the system arrives at this equilibrium by chance, the slightest change
would drive it away again toward one of the stable states, A or C. See Example 4 on page 225.
10 Actually, it’s a little warmer than our present climate, which you may remember corresponds approximately to g = 0.65.
In this illustration we’ve used g = 1 for simplicity, but the results with g = 0.65 would be qualitatively similar.
400 400
Inflow Inflow
Outflow g = 1 Outflow g = 1
300 Outflow g = 0.7 300 Outflow g = 0.5
Energy flow (W/m2 )
Energy flow (W/m2 )

C C
200 B 200
A A
100 100
0 0
220 240 260 280 300 320 220 240 260 280 300 320
(a) When the greenhouse strength decreases from g = 1.0 to (b) When g decreases to g = 0.5, points C and B meet as a
g = 0.7 the equilibrium points B and C move toward each single equilibrium. This is a tipping point.
other.
Figure 42: Graphical representations of the climate model with ice-albedo feedback as it approaches a tipping point.
4.4.2 Sudden Changes in Equilibrium

Now let’s imagine that the Earth is in its “Goldilocks state” at point C in Figure 41, when the
greenhouse parameter g slowly11 begins to decrease, representing a reduction in greenhouse strength.
What will happen?
Reducing g will cause the rate of energy outflow to increase. In other words, the green curve will
rise a bit. As it does so, points A and C will move to the left, and point B will move to the right. Points
B and C will therefore approach each other, as we see in Figure 42(a).
The Earth’s temperature follows point C, moving steadily to the left (decreasing) as g decreases,
until round about g = 0.5 and a temperature of 276 K, something strange happens (Figure 42(b)).
At this value of g, the “Goldilocks point” C and the unstable equilibrium point B, which have been
moving toward each other since we started decreasing g from its initial value of 1, finally meet. The
outgoing energy curve just grazes the incoming energy curve.
Problem 4: Describe the behavior of the climate system if the greenhouse strength g decreases a small
amount more, to below g = 0.5.
Solution: When g = 0.5, there are two equilibrium points, and the Earth has arrived at the higher of
the two, at about 276 K. If g continues to decrease, even by a tiny amount more, the outflow curve will
“lift off” the inflow curve completely at this point, and the only equilibrium left will be the “Snowball
Earth” state, near T = 225 K. The Earth’s temperature will therefore make an abrupt shift from near
T = 276 K to a stable equilibrium at T = 225 K, as shown in Figure 44(a) on the opposite page for
g = 0.45.
What this means is that the Earth’s temperature, having decreased steadily with g, from the initial
value of something like 300 K at g = 1 to 276 K at g = 0.5, will suddenly “flip over” to the Snowball
Earth value of about 225 K. Think of steadily increasing your finger’s pressure on the toggle of a
standard light switch. At first, the toggle moves steadily too: but, at a certain point, the switch “flips
over” into a completely different state. That is like the behavior that we are describing for the Earth
system (in fact, the mathematical modeling is quite similar).
11 How slowly? Slowly enough that the climate system can come to equilibrium before g changes very much. See Remark 4
on page 235.
This is what we mean by a tipping point.
Definition 1
A tipping point of a system with a parameter is a point at which one or more possible
equilibrium states of the system make discontinuous jumps in response to a small change in
the parameter.
In our imagination, we have allowed a steady decrease in the greenhouse effect

(as measured by the parameter g) to take the Earth’s climate system through a
tipping point to the “Snowball Earth” state (point C in Figure 44(a)). Now imagine
that the greenhouse effect starts strengthening again.
Problem 5: Suppose the greenhouse strength now gradually rises from g = 0.45
back up to g = 1.0. What state will the climate be in?
Figure 43: An ordinary light
Solution: We can refer back to Figure 44(a), where the outflow is plotted for both
switch is an example of “tip-
ping point” behavior. values of g. The single equilibrium for g = 0.45 (point C) at T = 225 K will move
steadily up to T = 245 K as g increases to 1.0. There are three equilibria again, but
the system remains in the “Snowball Earth” state. In fact, in order to switch back to a warm climate,
the greenhouse strength will have to rise all the way beyond g = 1.5, where the system passes through
a second tipping point, shown in Figure 44(b).
400 400
Inflow Inflow
Outflow g = 1 Outflow g = 1.5
300 Outflow g = 0.45 300 Outflow g = 0.45
Energy flow (W/m2 )
Energy flow (W/m2 )
200 200
C C
100 100
0 0
220 240 260 280 300 320 220 240 260 280 300 320
(a) When the greenhouse strength decreases to g = 0.45, the (b) Once the system is in the “Snowball Earth” state, the
equilibrium at T = 276 K abruptly switches to the single greenhouse strength needs to increase to g = 1.5 before the
equilibrium at T = 225 K, the “Snowball Earth.” climate will tip back toward the “Goldilocks” state again.
Figure 44: Representations of climate model with Earth in a “Snowball Earth” state.
Think about it. . .
Once a system has passed through a tipping point, the change usually cannot be reversed
simply by returning the parameter value to where it was before the “tipping” occurred.
In our Snowball Earth example, once the Earth has entered the “snowball” state, it will stay there
unless and until the system passes through another tipping point (shown in Figure 44(b)). That
requires increasing g to approximately 1.5—a “super” greenhouse effect. Current understanding of
the “Snowball Earth” episodes in the Earth’s ancient history is that they were ended by exactly this
mechanism—volcanoes released sufficient CO2 over millions of years to initiate a super greenhouse
effect strong enough to pass through this tipping point and return the Earth to the “Goldilocks”
equilibrium C.
Example 1: For another tipping point example, let’s revisit our model of phosphorus pollution in a
lake developed in Problem 2 on page 226 and the following discussion. In Figure 45, we’ve plotted
the inflow and outflow rates of phosphorus to the lake when the parameter P (the rate of inflow of
external pollution, such as agricultural runoff) is (a) P = 15 tons/yr and (b) P = 25 tons per year.
Outflow Outflow
150 Inflow 150 Inflow
Flow rate (tons/yr)
Flow rate (tons/yr)

100 100
C
B
50 50
P A
P
A
0 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
Phosphorus (tons) Phosphorus (tons)
(a) P = 15 tons/yr. (b) P = 25 tons/yr.

Figure 45: Graphical representation of model for phosphorus pollution as it approaches a tipping point.
The first thing you surely notice is that the graph of the inflow rate of phosphorus in Figure 45 has
a very similar shape to the graph of the inflow rate of energy we saw in the ice-albedo feedback
models like those in Figure 41 on page 244. A sigmoidal curve, like the ones we see here, is
characteristic of systems in which tipping points occur. Indeed, this is what makes the mathematical
analysis so powerful. The mathematics we developed for phosphorus pollution helped prepare us for
understanding the ice-albedo feedback, and our work on tipping points for the ice-albedo feedback
now informs our understanding of phosphorus pollution! The mathematical picture is similar for the
dynamics of kelp forests and sea urchin populations (Problem 5 on page 231), the dynamics of a
savanna controlled by wildfires, and many other systems that may not appear at all similar at first
glance.
Problem 6: In Problems 6 and 7 (page 233) we computed the sensitivity of the stock of phosphorus to
a change in the pollution rate P for each of the cases shown in Figure 45. Using those results, describe
a possible relationship between the sensitivity to a change in a parameter and a tipping point.
Solution: We can see that in the system shown in Figure 45(b) the equilibrium point A is nearing
a tipping point. We computed the sensitivities as 2.2 yr (when P = 15 t yr−1 ) and 4.5 yr (when P =
25 t yr−1 ). As the system approaches a tipping point, the stabilizing feedback weakens, and therefore
the system becomes more sensitive to small disturbances. We conjecture, then, that the appearance of
a tipping point may be signaled by the sensitivity becoming very large.
As a check on this idea, let’s increase the pollution rate a bit more to P = 27 tons/yr, bringing the
equilibrium even closer to its tipping point, and then recompute the sensitivity. The relevant control
strengths are shown in Figure 46 on the next page.

+0.38 +0.5
+1
Pollution, P
Figure 46: Stock-flow diagram for phosphorus in a lake showing the strength of each control when P = 27 t yr−1 .
Problem 7: The stock-flow diagram in Figure 46 shows the strength of each control when P = 27
tons/yr and the equilibrium point A has moved very close to the tipping point (the inflow is shifted up
little higher from what is shown in Figure 45 on the previous page(b)). Calculate the sensitivity of the
phosphorus stock to a change in the pollution rate P at this equilibrium.
Solution: The strength of the overall feedback has now weakened to q = −0.12 yr−1 . The strength of
the pollution control remains p = +1. Using the sensitivity rule (Rule 3 on page 233), we obtain
p 1
Sens(S; P) = − =− ≈ 8.3 yr.
q −0.12 yr−1
Notice how dramatically the sensitivity is increasing as we approach the tipping point. When we
changed the pollution rate from 15 tons/yr to 25 tons/yr, the sensitivity doubled. A small additional
increase to 27 tons/yr doubled the sensitivity again. As we approach the tipping point, the overall
stabilizing feedback is approaching zero. Once we pass through the tipping point, the low phosphorus
equilibrium (A) will disappear and the system will “flip” to the high-phosphorus, eutrophic state (C).
We summarize our observations in a rule.
Rule 1: Tipping Point Sensitivity Rule
As a stock-flow system (with a single stock) at stable equilibrium approaches a

tipping point, the total strength of the stabilizing feedback in the system approaches
zero. Equivalently, the sensitivity of the system to changes in a parameter becomes
arbitrarily large.
Critical Slowing Down

So how can we tell whether we are approaching a tipping point?
In our simple models for climate or for lake pollution, we could vary the parameters (g and P
respectively) and look for points where the incoming and outgoing curves just graze each other, as in
Figure 42(b). But that does not really get at the observational question: is there some aspect of the
system itself that we could notice “from within,” so to speak, without having to create a model, that
could warn us that a tipping point was approaching?
In fact, there are several such possibilities. One of them is called critical slowing down; see [318].
The authors of this paper explain it as follows:
Figure 47: Critical slowing down in an ecosystem, from van Nes and Scheffer [332].
An important question in climate forecasting is whether any early warning of an

approaching threshold change or “tipping point” in the climate system can be provided,
before it is reached. . .
In general, for a system approaching a threshold where its current state becomes
unstable, and it makes a transition to some other state, one can expect to see it become
more sluggish in its response to small perturbations. . . This is referred to as “critical
slowing down” in dynamical systems theory. This phenomenon has long been known
about, but it has only recently been applied to climate dynamics.
The key words here are more sluggish. As our system approaches its tipping point, the stabilizing
feedback that holds it in equilibrium becomes weaker, according to Rule 1 . But this stabilizing
feedback is simply the “restoring force” that the system applies in trying to get back to equilibrium
in response to some short-term disturbance. With less restoring force available, the system will take
longer to respond to the same temporary disturbance—that is, it will be more sluggish—than it would
have done if it were in equilibrium further away from the tipping point. This is critical slowing down.
Definition 2
The tendency of a system in equilibrium to respond more sluggishly to shocks as its

parameters approach a tipping point is called critical slowing down.
Example 2: Van Nes and Sheffer [332] investigated a model for a tropical savanna ecosystem
that could pass through an “overgrazing” tipping point. In their simulations they could clearly
observe critical slowing down in the system’s response to a “shock” (a sudden change in the animal
population)—see Figure 47. It’s important to understand, though, that these are simulations, not
experiments carried out on actual ecosystems. So far as we (the authors) are aware, there are as yet
no real-world examples in which critical slowing down has generated an unambiguous signal that has
caused people to take preventive action. And one has to ask, if scientists discerned an unambiguous
“critical slowing down” signal of an impending climate transition, would humanity be willing to listen
to this unwelcome news, and take possibly difficult action? Such questions take us beyond the world
of mathematics.
Figure 48: Risks of various climate tipping points at various levels of global warming, from Lenton et al. [192].
Climate System Tipping Points

We looked at the “Snowball Earth” tipping point because we can discuss it reasonably accurately using
our very simple climate-modeling ideas. However, no responsible scientist believes that this particular
tipping point is likely to be reached any time soon. Our present concerns are related to increasing the
strength of the greenhouse effect, not decreasing it. Climate scientists have examined a number of
potential tipping points in the climate system that are much more likely to pose significant near-term
risks. These include:
• irreversible collapse of a major ice sheet or ice cap (the Greenland ice cap is the most likely
candidate here; see Exercise 44 on page 62);
• shutdown of the Atlantic thermohaline circulation, which warms the waters of the Gulf Stream
(the “scientific” premise of the 2004 movie The Day After Tomorrow);
• dieback of Amazonian rainforest and its replacement by savanna.
These are among nine potential tipping points whose probabilities are considered in the paper [192].
See also [318] and [42].
Remark 3: The Tipping Point is also the title of a book by science journalist Malcolm Gladwell [131].
Gladwell’s thesis in the book is that “The tipping point is that magic moment when an idea, trend, or
social behavior crosses a threshold, tips, and spreads like wildfire.” Though both share the underlying
idea of “small changes producing large effects,” Gladwell’s sociocultural application of the “tipping
point” metaphor is not exactly the same as the physical one considered here.
Think about it. . .
Dr. Gavin Schmidt, climatologist at NASA’s Goddard Institute for Space Studies, says
We are a society that has inadvertently chosen the double-black diamond run
without having learned to ski first. It will be a bumpy ride.
Figure 49: Visual representation of Earth at a tipping point. Illustration by Cheng “Lily” Li. Used with permission.
Summary of Ideas: Tipping Points
• It is possible (even normal) for a stock-flow system to have more than one equilibrium
state.
• As the parameters of a system are varied, the equilibrium states will change.
• A tipping point is a parameter value at which the configuration of equilibrium states
undergoes a discontinuous change. The example that we studied involves an unstable
and stable equilibrium moving toward each other and then canceling out.
• If a parameter passes through a tipping point, the system may “jump” to a new
equilibrium state. Such jumps are not usually reversible by simply undoing the change
in the parameter.
• Critical slowing down is the effect whereby a system’s response to shocks becomes
more sluggish as a tipping point is approached.
• Scientists have identified several potential tipping points in the Earth’s climate-
environmental system.
Year Population
4.1. LOGARITHMS AND CHANGE 1 4,201
2 5,101
3 6,195
1. Find a number x such that 5x = 16, 000; express 4 7,524
your answer to 2 decimal places.
Using an exponential growth model, estimate the pop-
ulation of stoats in year 5.
2. Find the solution x to the equation 6.03x = 2, 718.
8. The table below gives the population of Japan (in
3. In 1990, 7.6 million children worldwide died millions) at 20-year intervals for the past 100 years.
before reaching the age of five. In 2015, the corre- Year Population
sponding figure was 3.7 million [166]. Express this
change in absolute terms (as a difference), in relative 1910 51
terms (as a percentage), and in logarithmic terms (in 1930 64.5
centinepers). Which way of expressing the change do 1950 83.2
you think is most effective and useful? 1970 103.7
Would you have expected so large a reduction in 1990 123.6
child mortality? If the figure was a surprise to you, 2010 128.1
why do you think that might have been? What other Investigate the appropriateness of an exponential
information might be useful to provide context for growth model for these data. (You should find that the
these figures? (For example, it may be relevant to know model of exponential growth is not appropriate beyond
that more children overall were born in 2015 than in 1970 or so. Japan’s population has leveled off, and
1990.) indeed, more recently is beginning to decline.)
4. One of the most rapidly growing populations in 9. The UN Food and Agriculture Organization gives
the world is that of South Sudan. According to the CIA the following figures for annual worldwide beef pro-
World Factbook, this population is growing at the rate duction (in millions of tons):
of 4.1% per year. Use logarithms to find the doubling Year Beef Production
time for the population of South Sudan at this rate
1969 41.8
of exponential growth. Compare your answer with the
1979 48.1
doubling time approximation (Rule 2 on page 173). 1989 54.2
1999 58.7
5. The population of Niger is growing at a rate 2009 67.5
of approximately 3.5 percent per year. If this trend Investigate the appropriateness of an exponential
continues, approximately how long will it take for growth model for these data by drawing a graph of the
Niger’s population to double? logarithm of annual beef production against time. You
should find that an exponential growth model is a good
6. In a positron-electron tomography (PET) scan, the fit. Use the graph to estimate annual beef production in
patient is injected with radioactive glucose. The glu- 2019, assuming the trend in the data continues.
cose loses 31.5 percent of its radioactivity each hour.
Find the half-life, using the exact half-life formula. 10. U.S. energy use is currently about 100 Quads
per year (1 Quad = 1 quadrillion BTU). Convert this
to watts (joules per second).
7. A breeding pair of stoats finds its way to a remote
island in the South Pacific. Some time later, the number
of stoats on the island is counted for four successive 11. The power output of the Sun is 3.85 × 1026 watts.
years, with the following results: If U.S. power consumption continued to grow at the
rate of 2.9% per year, about how long would it be until of unobtanium is 6.3 × 108 kilograms per year, and this
the U.S. required the total power output of the Sun (see is increasing at an annual rate of 7 percent, calculate
Exercise 10)? the exponential reserve index.
12. An engineering company owns capital equipment 18. Suppose that the climate sensitivity is 6 degrees
valued at $14 million. The worth of the capital equip- Celsius of warming for a doubling in carbon dioxide
ment depreciates by 13% each year. In a year in which concentration (this is at the high end of the range of
the company has the use of capital valued at C dollars, possibilities, but not out of the question). According to
they expect to generate 1.4×C dollars of net profit. The Arrhenius’s law, what amount of warming is associated
company’s board has approved reinvesting 8 percent of with a rise in carbon dioxide concentration from its
net profits in new capital equipment each year. What late-1950s level of about 315 parts per million to its
will the value of the company’s capital be in ten years’ predicted 2030 value of about 415 parts per million?
time? Comment on your answer.
19. Estimate the total weight of the Earth’s atmo-
13. An extravagant mathematics professor has credit sphere, as follows: The atmospheric pressure (at the
card debt of $12, 755. The monthly interest rate on the Earth’s surface) is about 15 pounds per square inch.
professor’s credit card is 1.75%. If he wants to pay off This figure represents the weight of the column of
the debt in 18 months, what monthly payment does he atmosphere lying over every square inch of the Earth’s
need to make? surface. Multiply by the surface area of the earth
(4πR2 , where R is the radius of about 4000 miles) to
get the weight of the whole atmosphere. Express your
14. Global proved oil reserves in 2011 were ap-
answer in tons. Be sure to convert units appropriately.
proximately 1.5 × 1012 barrels. Global consumption
was about 3 × 1010 barrels per day. Estimate the static
reserve index, and the exponential reserve index assum- 20. It has been estimated that 500 billion tons of
ing a 3.5% annual increase in consumption.12 methane may be locked under the Arctic permafrost
in the form of compounds called clathrates, which are
stable only at low temperatures. As the Earth warms,
15. Estimate the exponential reserve index for oil on
these clathrates may become unstable and methane
the assumption that total global reserves are (a) twice,
would then bubble out through the ground: this is
(b) five times, and (c) twenty-five times the proved
already happening to a small extent in Siberia. Suppose
reserves figure quoted in the previous exercise (using
that 5 percent of the total 500 billion tons of methane
the same exponential growth rate of 3.5% annually).
are released. What fraction of the total weight of the
atmosphere does that represent? Express your answer
16. The rare metal indium (atomic number 49) is in parts per billion (ppb).
used in engineering and electronic applications (such as
LCD display screens). Suppose that the global reserve 21. The previous question estimated the effect of a
of indium is 10,000 tons and that current extraction release of methane from Arctic clathrates in terms of
rates are 620 tons per year. Compute the exponential parts per billion by weight. However, such concentra-
reserve index for indium, on the assumption of 3 tion figures are usually expressed in terms of parts per
percent annual growth. billion by volume. The laws of physical chemistry tell
us that to convert one to the other, one must multiply by
17. Planet Pandora has a reserve of 5.9 × 1011 kilo- a conversion factor, the average molecular mass of the
grams of unobtanium. If the current rate of extraction atmosphere divided by the molecular mass of methane.
12 The term “proved reserves” in this exercise has a specific technical meaning. They are, by definition, the “quantity of
energy sources (oil) estimated with reasonable certainty, from the analysis of geologic and engineering data, to be recoverable
from well established or known reservoirs with the existing equipment and under the existing operating conditions” (Business
Dictionary [267]). It is therefore possible for “proved reserves” to increase even as oil is extracted, because new sources are
established, new technical means become available (e.g., fracking), or new operating conditions apply (e.g., a price increase
rendering formerly uneconomic wells viable). The proved reserves represent only a fraction of the total amount of oil that is
actually available. That total, however, whatever it may be, is being steadily diminished by human activity. No more oil is being
created.
This conversion factor is approximately 1.8. Multiply do you think of the methodology used to make this
your answer to the previous question by 1.8 to obtain a calculation?
figure in parts per billion by volume (ppbv).
26. Suppose that a population of bacteria growing in
22. The current concentration of atmospheric a culture is described by a logistic model. At successive
methane is approximately 1800 ppbv. According time intervals, the population is given in the table
to [151], the radiative forcing produced by an increase below. Compute the growth rate for each time interval.
in methane concentration from a background level of Minutes Population Growth Rate
M0 ppbv to a new total of M ppbv is approximately
√ 0 25000 (n/a)
20 35400 38%
p
0.0406 M − M0
40 46600
watts per square meter. Calculate the radiative forcing 60 61200
resulting from the methane release event hypothesized 80 77700
in Exercise 20 on the previous page. Compare to 100 94800
120 110700
the levels of radiative forcing shown in Figure 9 on
page 203, and comment on your answer.
27. For the previous exercise, plot the growth rates
against time and confirm that the logistic model is a
23. The term geoengineering refers to a spectrum of
good fit. Estimate the carrying capacity.
proposed “technological fixes” to the climate change
problem. For instance, it has been proposed to increase
28. The total sales by month of a new product are
the effective albedo of the Earth (reflecting more heat
given in the table below (in thousands of units sold).
into space) by injecting sulfate aerosols into the upper
Initially sales are increasing rapidly, but then they
atmosphere. By what percentage would one need to slow as the market becomes saturated. Assuming sales
change the Earth’s albedo to counteract all the forcing follow a logistic model, estimate the total sales of the
shown in Figure 9 on page 203? new product.
Month Cumulative Sales (thousands)
24. Sulfate aerosols (see previous exercise) settle out
of the atmosphere eventually, and therefore to maintain 1 484
2 1028
a fixed change in effective albedo one would need
3 2140
to inject these aerosols continuously; as the radiative
4 4243
forcing increased, the rate of aerosol injection would 5 7666
have to increase to keep pace. Suppose that this geo-
engineering process continues for a hundred years and 29. Zinc ore is extracted from a mine. The following
then, in year 101, suddenly stops (perhaps because data describe the total amount of ore extracted (in
of worldwide political changes). What do you expect thousands of tons) after the mine has been in operation
would happen? for various numbers of years:
Years Extracted
4.2. LOGISTIC MODELS AND THE 1 411
LIMITS TO GROWTH 2 752
3 1,333
25. You can find many “footprint calculators” online, 4 2,231
all of which embody assumptions about the goods and 5 3,394
services we consume, and about the land areas needed 6 4,492
to support them.13 Try using one of these calculators to Use the logistic model to estimate the final amount of
estimate your personal “footprint.” Does your lifestyle ore that can be extracted from the mine.
overshoot? If so, read the FAQs to find out why. What
13 The one that is closest to Wackernagel’s original methodology [337] is probably the Global Footprint Network calculator
at [237].
30. The table below (from [30]) shows the weight (in account. Suppose that each week you deposit a $250
grams) of a human fetus at different gestational ages paycheck into this account. Using the stock level as the
(number of weeks after conception). horizontal axis (the “x-axis”) and the flow rate as the
vertical axis (the “y-axis”), make a plot of the inflow
Time (weeks) Weight (grams) rate as a function of the stock level, under different
8 1 assumptions as follows:
10 4 (a) Assume that the checking account pays no inter-
12 14 est.
14 43 (b) Draw a possible inflow for a checking account
16 100 that pays interest.
18 190
20 300 (c) Draw an inflow rate for a checking account that
22 430 pays interest only when the account contains at
24 600 least $2000.
Compute the growth rates for each two-week period.

37. For the previous exercise, make a plot for a
Suppose that a baby kept growing throughout the 40
possible outflow rate (on the same axes as before). It
weeks of pregnancy at the same rate as between weeks
may be easiest to first decide what the outflow rate
8 and 10. At birth, how much would the newborn
should be when the account has $0, then decide how
weigh?
that will change as the stock level increases. Describe
your reasoning.
31. Investigate whether a logistic model is a good fit
to the data in the previous exercise.
38. Carefully work through the analysis for the
stability of the two remaining equilibrium points from
32. In 1950, around the time Hubbert wrote, U.S. Problem 3 on page 227
cumulative oil production was about 40 billion barrels.
If we use a logistic model with r = 7% and C = 200
39. In 2016 southern Florida experienced multiple
billion barrels, when would the time to peak formula
harmful algal blooms (HABs), many related to water
predict that U.S. peak production would occur?
discharged from Lake Okeechobee (see discussion in
the Critical Thinking box on page 237). The Indian
33. According to the United Nations, the total popu- River Lagoon provides a particularly striking example
lation of the Dominican Republic was 2.38 million in of the potentially devastating effects of algal blooms;
1950, 3.31 million in 1960, 4.52 million in 1970, and see [139]. Why is it necessary for officials to release
5.83 million in 1980. What can you conclude about the water from Lake Okeechobee, knowing it could make
growth rate of the population based on this data? harmful algal blooms more likely elsewhere? Are the
impacts of harmful algal blooms in the U.S. limited to
34. Describe the meaning of the term carrying capac- Florida?
ity in your own words.
40. A stock-flow system models the level of water
in a lake. Near a certain equilibrium point, there are
4.3. MEASURING FEEDBACK STRENGTH three feedback loops: a positive feedback loop with
strength 0.1 month−1 , a negative feedback loop with
strength −0.2 month−1 , and a positive feedback loop
35. Describe the units in which the strength of a with strength 0.15 month−1 . Find the overall feedback
control arrow (running from a stock to a flow in a in the system. Is the equilibrium point stable or unsta-
stock-flow system) should be measured if the stock ble?
is measured in kilograms and the flow is measured in
cubic meters per second. 41. A stock-flow model for the population of bees in a
hive includes a control arrow linking the concentration
36. Consider a stock-flow system for a checking of a certain environmental pollutant to the rate of bee
account. The stock is the amount of money in the deaths. Near a certain equilibrium, when the pollutant
level is 0.1 micrograms per liter, the rate of bee deaths

is 2,000 per year; when the pollutant level is 0.12 mi- 4.4. TIPPING POINTS
crograms per liter, other factors being constant, the rate
of bee deaths is 2,100 per year. What (approximately) 44. Carefully work through the analysis of the sta-
is the strength of the control that links pollutant levels bility of equilibrium points B and C in the ice-albedo
to bee deaths? feedback model shown in Figure 41(b) on page 244.
42. Imagine that the climate system is in equilibrium, 45. Find the equilibrium temperature of the Earth
stabilized by negative feedback. Scientists discover a in our simple climate model when g (the greenhouse
previously unrecognized positive feedback related to effect strength) is equal to 0.8. Assume that the albedo
the slowing of tree growth caused by global warming. is α = 0.3.
Their best estimate of the feedback quotient k related
to this new feedback, that is, the ratio of the strength of 46. On the website for this book, at http://
the new positive feedback to (minus) the net strength of math-for-sustainability.com/budyko, you will
the already known negative feedback, is 0.65. By what find a simulator that allows you to work with the
factor does this new feedback amplify the sensitivity of Budyko climate model (incorporating ice-albedo feed-
the climate system? back) described in the text (see page 245). Use the
simulator to find how big the downward “jump” in
43. This exercise gives a different way of approach- temperature is as the climate passes through the tipping
ing the law of amplification (Rule 4 on page 236). point at g = 0.5.
Imagine a system with stock S stabilized at an
equilibrium point by a negative feedback of strength 47. Use the Budyko model simulator described in
qbare . We introduce a new feedback of strength
the previous exercise to answer the following question:
qadditional . Think of this in the following way: there
Suppose that the Earth has entered the “snowball” state.
is a parameter P that controls S through a control of
strength qadditional , but also we are going to force P and The greenhouse parameter g returns to its previous
S to be equal. value of about 0.7. What will the Earth’s surface
temperature be now?
(a) Show that the sensitivity of S to changes in P is
exactly the feedback quotient k.
48. Use the Budyko model simulator described in
(b) Imagine S changes by one unit. Then P must the previous exercise to answer the following question:
change by one unit (because they are the same). Suppose that the Earth has entered the “snowball” state.
Now, because of the control of sensitivity k, S How big must g become to tip the Earth back into the
must change by an additional k units. “Goldilocks” state again?
(c) But now P must change by an additional k units,

49. On the website for this book, at http:
so because of the control, S must change by an
additional k × k = k2 units. But then so must P, //math-for-sustainability.com/phosphorus,
so S must change by an additional k × k2 = k3 you will find a simulator that allows you to work
units, and so on. Following out these ideas, show with the phosphorus pollution model described in
that the initial 1-unit change in S gets amplified Example 1 on page 247. Using this simulator, figure
to out the rate of phosphorus pollution (given by the
1 + k + k2 + k3 + · · · parameter P) when the system passes through a
units. tipping point from the “low phosphorus” to the “high
phosphorus” state.
(d) Use the geometric progression formula (Rule 5
on page 198) to show that the sum above equals 50. Using the phosphorus model simulator described
1/(1 − k), as predicted by the law of amplifica- in the preceding exercise, figure out the rate of phos-
tion.
phorus pollution (given by the parameter P) when the
system passes through a tipping point from the “high
phosphorus” to the “low phosphorus” state.
Flow rate (tons/yr) Outflow

150 Inflow
100
50
Phosphorus (tons)
50 100 150 200 250 300
Figure 50: A phosphorus model with an unrealistic inflow, used in Exercise 51.
51. Suppose, in the context of Example 1 on

page 247, that the inflow and outflow rates of phospho-
54. In the context of the kelp forest ecosystem model
rus in a lake are given in Figure 50 (these are unrealistic
in the previous problem, explain what is meant by
flows that have been manufactured for this problem).
saying that the sensitivity of the kelp forest ecosystem
Determine any equilibrium points and decide whether
is high (in a particular state)?
each one is stable or unstable. Suppose that the inflow
is increased by adding P tons per year of agricultural
runoff, in a manner similar to Example 1 on page 247. 55. Based on Problems 6 and 7 on page 233, discuss
Do you expect this model to exhibit any tipping points how the sensitivity of the kelp forest ecosystem may be
as P increases? Explain. changing as the system approaches a tipping point.
52. On the website for this book, at http:// 56. Scientists study the response of a river ecosystem
math-for-sustainability.com/kelp-forest, to a perturbation (given by a one-time “surge” of
you will find a simulator that allows you to work with nutrients). In the absence of any pollution, the time the
the kelp-sea urchin model described in Problem 5 on ecosystem takes to recover from the surge (the response
time) is 3.6 weeks. Response times in weeks for other
page 231. Using this simulator, determine the density
levels of pollution (measured by a parameter p) are
of the urchin population (given by the parameter u)
when the system passes through a tipping point from
the “kelp forest” state to the “urchin barren” state. Value of p 0 0.05 0.1 0.15 0.2
Response time 3.6 4.4 5.6 7.7 12.5
53. Using the kelp forest model simulator described
in the preceding exercise, determine the density of the Based on the theory of critical slowing down, the
urchin population (given by the parameter u) when the scientists suspect that the system may be approaching
system passes through a tipping point from the “urchin a tipping point. Can you estimate for what p-value the
barren” state to the “kelp forest” state. tipping point will occur?
C HAPTER 5
Risking
In the models studied in earlier chapters, we used past and

present information to make predictions about the future.
While these predictions are useful, we know that they are
not going to be exactly correct! Why might that be? Nothing
in life happens in a completely predictable way. In any
modeling, there is always some uncertainty—some aspect
that we cannot perfectly predict. That is partly because all
models are oversimplifications, and partly because even the
best model may be disturbed by some small random event
(the proverbial flap of a butterfly’s wings in Brazil that sets
off a tornado in Texas; see Definition 4 on page 162).
Fortunately, we have the ability to assess and even to Figure 1: Is this a risky activity?
measure uncertainty, and to make decisions accordingly.
Here are some examples of the ways we do that. Do these
sound familiar? Can you think of more?
• “I would love to go skydiving, but my boyfriend thinks it is too risky.”

• “There is a 40 percent chance of rain tomorrow.”
• “You’re more likely to be hit by lightning than to catch Ebola in the U.S.”
• “Sometimes the wind blows, and sometimes it doesn’t. It’s too erratic to be a reliable source of
energy.”
• “If we don’t take action to reduce human emissions of heat-trapping gases, there is a fifty-fifty
chance of warming the planet by at least 5 degrees Fahrenheit.”
• “Regular mammograms can reduce a woman’s long-term risk of dying from breast cancer by
22 percent.”
• “I’ll buy the lottery ticket. I could be a millionaire tomorrow. There’s always a chance!”
We don’t know in advance where the lightning will strike, or which ticket will win the lottery, or
who will be unlucky enough to develop breast cancer. But that doesn’t mean that we can’t say anything
about these various possibilities. We all understand that some are more likely than others.
Example 1: Try it for yourself. Rank the following from most to least likely (for you).
(a) Rolling a five with a single roll of a fair die.

https://doi.org/10.1007/978-3-319-76660-7_5
260 CHAPTER 5. RISKING
(b) Getting an “A” in your Math for Sustainability class.

(c) Snow falling in your town on the next 4th of July.
(d) Snow falling in your town on the next 4th of January.
(e) Being attacked by a shark in the next year.
(f) Dying in an automobile accident in the next year.
(g) Shaking hands with a (current or former) U.S. president in the next year.
What’s more, we can (at least sometimes) assign numbers to these likelihoods. The chance of rolling
a five is one in six, or about 0.17. There were 53 shark attacks in the U.S. in 2012, so we could divide
by the U.S. population of about 3 × 108 and say that the chance of a random U.S. person suffering
a shark attack in a year is about 1.7 × 10−7 , a million times smaller than the chance of rolling a five
with a fair die. (Later, we will have to think carefully about the assumptions we make in doing this
sort of calculation.)
Any kind of realistic thinking about sustainability involves likelihoods and risks. Climate change is
real, but how bad will it get? The answer is uncertain: not just because we can’t tell what humans will
do in the future, but also because the limits of our knowledge about the climate system (especially
about its various positive and negative feedbacks) mean that we can only give a range of possibilities
for the key number, the climate sensitivity (see Figure 10 on page 204). Nuclear power does not
produce any greenhouse gas emissions in normal operation, and it is constant and reliable, but what
about the possibility of a catastrophic accident? This question asks us to balance a steady benefit that
is nearly certain against a small risk of disaster. It is hard to do so without numbers. How likely are
you to die from a nuclear reactor meltdown? (Answer: Less likely than you are to die from a shark
attack, and much less likely than you are to die in an automobile accident, even if you live next door
to the reactor. See Case Study 7.6.)
The math involved in working with these sorts of
numbers is called probability and statistics. In this chapter,
we are going to learn about these techniques and how they
can help us make good decisions when faced with limited
knowledge and uncertain outcomes.
Of course, numbers cannot tell us the whole story.
Many people are much more concerned about dramatic,
horrifying risks, especially in contexts that evoke feelings
of powerlessness, than they are about apparently routine
risks, especially where it seems that one has some control.
For example, people tend to be much more concerned
Figure 2: Is this a risky activity? about airplane accidents than automobile accidents, even
though it is statistically clear that airplane travel involves a
considerably lower risk (per passenger-mile traveled) than
automobile travel does: see the data in Section 8.2.6. The
mathematician has the right to point this out, but he or she does not have the right to tell people that
their risk perceptions are therefore “irrational” and should be ignored. How human beings respond to
risk is a fact of (human) nature, hard-wired into us over millions of years; we will need to work with
it, not against it, to deal with the new risks of a transition to a sustainable world.
5.1. UNDERSTANDING DATA 261
5.1 Understanding Data
Objectives
I understand what is meant by a data set.

I can distinguish categorical and numerical variables in a data set.
I understand what is meant by the distribution of a numerical variable.
I can use a box plot and a histogram to give a visual summary of a distribution.
I can use descriptive statistics such as the median, the mean, and the interquartile
range to summarize the “center” and “spread” of a distribution.
I can identify outliers and extreme values.
I can identify a skewed distribution and know why skewness can be important.
I understand that real-world data sets often, but not always, follow the normal
distribution.
5.1.1 Data and Distributions

Knowledge has been a human pursuit for at least as long as recorded history. We, as a species, appear
to have a strong desire to understand what has been, what is, and what can be—Carl Sagan went
as far as to say, We are a way for the cosmos to know itself [283]. But how do we generate that
understanding? How do we know whether what we believe is correct? One way of answering those
questions begins with observation.
It is through observation that we collect information about reality. To secure a robust understanding,
we want precise information. That requires taking many observations. Once we have recorded the
results of many careful observations, we have to be able to build understanding from them. We need
ways to summarize many observations to help us understand the bigger picture.
Definition 1
A data set is a collection of observations.
• The objects described by the data set are the individuals. Individuals might be people,
but they might also be animals or things (like individual tubes of toothpaste) or even
occurrences (like individual tosses of a coin).
• The data set records several kinds of information (for example, height or weight)
about each individual. These kinds of information are the variables of the data set.
The information of a specific kind (height, for example) belonging to a specific
individual is called the value of that variable that applies to that individual.
Let’s look at an example.

Example 1: Table 1 is a snapshot of part of a data set recording information from an online survey of
students in an imaginary class.
Number Gender GPA cents height

1 Female 2.0 40 66.3 • The individuals in the data set are the students. The indi-
2 Male 2.9 41 71.8 viduals in this snapshot are represented by identification
3 Male 4.0 96 62.9 numbers 1–10. (There are 25 students in the whole class.
4 Female 2.9 112 62.1
You can download the full data set from this book’s
5 Female 2.3 22 64.8
6 Female 3.8 60 63.3 website.)1
7 Female 3.7 14 71.2
8 Other 2.4 97 66.8
• There are four variables in the data set, recording each
9 Male 2.0 71 63.1 individual’s preferred gender identification, GPA, height
10 Male 3.6 132 69.9 (in inches), and the total amount of change (in cents) that
they could find in their pockets when the survey was taken.
Table 1: Example student data.
• The first variable (gender identification) is called a categorical variable, because it describes
the individuals by categories or characteristics. The other three variables take on numerical
values and are therefore called numerical variables. In this book we will focus our attention
on numerical variables.
Here’s another example with a more environmental theme.
Lake Phosphorus
Site ID Land Use Area Concentration Example 2: Table 2 gives data (from a USGS study [130])
(km2 ) (µg/L) on phosphorus concentrations at lake study sites in the Puget
1 Residential 0.41 16 Sound area of Washington State. Each study site is a different
2 Residential 0.18 19 lake and is categorized by the primary use of the adjacent
3 Mixed 0.05 20
0.14 13
land: Residential, Forest, or Mixed (both forest and residen-
4 Residential
5 Forest 0.13 12 tial).
6 Forest 0.18 23 The individuals in the table are the study sites, identified
7 Mixed 0.09 26 by a site ID, numbered 1–12. The three variables in the data
8 Forest 0.45 4
9 Forest 0.17 11
set record the primary use of land adjacent to the study site,
10 Forest 0.20 17 the surface area of the lake in square kilometers (km2 ), and
11 Mixed 0.65 16 the concentration of phosphorus in the water in micrograms
12 Forest 0.28 19 per liter (µg/L). The first of these is a categorical variable
Table 2: Measured phosphorus concentrations and land- and the other two are numerical variables.
use data for lakes used to evaluate phosphorus loading by
If we want to understand each study site individually,
residential area run-off.
Table 2 provides a great way to organize the data. We can
see that that Study Site 4 was residential, that its lake area
was 0.14 km2 , and that the corresponding phosphorus concentration was 13 µg/L. Often, though,
we want to focus on one variable, like the phosphorus concentration, for example, and get a general
picture of the whole group from the perspective of that one variable. To put it differently, we want to
focus on the distribution of the variable of interest. We’ll define that in a quite general way:
Definition 2
The distribution of a numerical variable is any type of organization that summarizes the
values the variable takes and how often those values are observed.
1 It is available at http://math-for-sustainability.com/sample-data in spreadsheet format for further analysis.
Let’s consider the concentration of phosphorus at each study site as given in Table 2. The simplest
way of organizing the data set is to sort it as a list in ascending order:
4, 11, 12, 13, 16, 16, 17, 19, 19, 20, 23, 26.
This sorted list is a distribution; we can see the minimum and maximum values in our data set, what
values the variable takes and how often it takes each value. There are also other more convenient and
informative ways to describe a distribution. We will look at several of them in this section.
Visuals for Distributions: Box Plots

Numerical lists and tables often don’t convey a picture of a distribution that is easy to grasp, especially
when a large number of individuals are involved. To summarize the information, we may use a visual
representation instead. The simplest of these visual representations is called a box plot. We shall see
that a box plot is simply a visual summary of the most important information contained in an ordered
list of values.
First, we need to discuss some terminology. Namely, we need to define the median, and the upper
and lower quartiles.
Example 3: The student data set in Table 1 is part of a larger one that has 25 individuals in total.
Let’s consider the heights of these 25 individuals. We’ve organized the values of the height variable
in ascending order.
59.8, 60.4, 62.1, 62.9, 63.1, 63.3, 63.9, 64.2, 64.8,

64.8, 65.1, 66.3, 66.4, 66.8, 67.2, 68, 68.3, 68.3,
69.5, 69.6, 69.7, 69.9, 71.2, 71.8, 72.8.
The simplest question we could ask about the distribution is, “Where is the middle?” Where is the
center of the distribution? One way to answer that question is to find the middle item in the ordered
list above. Since there are 25 items in the list, the middle one is the 13th, marked in red, which turns
out to be 66.4.
Definition 3
The median of a variable in a data set is a number m for which 50% of the values of the
variable are above m and 50% of the values of the variable are below m. Informally, the
median is the value that “splits the data set in half.”
In our example of heights, the median is 66.4. There are 12 students shorter than the median height
and 12 taller. (The one remaining student is exactly of median height.)
Problem 1: Consider the set of numbers below. Find its median:
29, 31, 42, 47, 45, 23, 35, 26, 44.
Solution: The first step is to put them in order:
23, 26, 29, 31, 35, 42, 44, 45, 47.
Now we locate the middle number. Since there are 9 numbers, the middle number is the 5th one, or
35. Therefore the median is 35.
Problem 2: Consider the set of numbers below. Find its median:
7, 12, 1, 10, 4, 15.

Solution: The first step is to put them in order:
1, 4, 7, 10, 12, 15.
Notice that there is an even number (six) of items in this data set. The “middle” is between the third
and fourth value—that is, between 7 and 10. In this situation, the convention is that we take the
midpoint of the interval from 7 to 10 to be the median. That is, we add the two middle values (7 and
10) and divide by two. Hence, the median is taken to be
7 + 10 17
= = 8.5
2 2
for this data set.
The median tells us the center of the distribution. Another question we might want to ask, though,
is: how wide is the distribution? The simplest measure of width or “spread” is the range.
Definition 4
The range of a numerical variable in a data set is the difference between the greatest and
least values (called the extreme values) of the variable.
Although the idea of “range” is easy to understand, it has the disadvantage that it can be greatly
affected by a few extreme values, or “outliers.” For example, consider household incomes in the U.S.
The median U.S. household income in 2015 was $56,100 [35]. But according to Forbes magazine,
the richest person in the U.S., Bill Gates, had a net worth of approximately $75 billion in March,
2015, corresponding (if safely invested) to an annual income of roughly 3 billion dollars. The poorest
household was not noted in Forbes, but we would be safe to assume that its annual income was close
to 0 dollars. Would it be useful to say, “Household incomes in the U.S. range from zero to three billion
dollars per year”? No, because the great majority of households are not at either extreme. To obtain a
more representative notion of “spread” we introduce the idea of quartiles.
We have seen that a median splits the values of a variable into two parts, the “lower half” and
the “upper half.” To give us an idea how spread out the distribution is, we can find the medians for
the “lower half” and the “upper half” separately. We call these values the lower quartile and upper
quartile.
Definition 5
The lower quartile for a variable in a data set is the median of the lower half of the values
of that variable. The upper quartile is the median of the upper half.
Thus 25 percent (or one-fourth) of individuals are below the lower quartile and 75 percent (or
three-fourths) are above. For the upper quartile, the percentages are reversed, with 75 percent (or
three-fourths) below and 25 percent (or one-fourth) above.
Think about it. . .
For U.S. annual household income in 2015, the lower quartile is approximately $27,300 and
the upper quartile is approximately $102,500 (this is based on the work of the University of
Minnesota Population Survey Center, as presented at [35]). What do you think about this?
Is this a wider or a narrower spread than you would have expected?
Let’s consider the height values from the data set mentioned earlier. Highlighted again in red is the
median:
59.8, 60.4, 62.1, 62.9, 63.1, 63.3, 63.9, 64.2, 64.8,
64.8, 65.1, 66.3, 66.4, 66.8, 67.2, 68, 68.3, 68.3,
69.5, 69.6, 69.7, 69.9, 71.2, 71.8, 72.8.
First, let us find the upper quartile. For that, we take the upper half of the data set—the values above
the median:2
66.8, 67.2, 68, 68.3, 68.3, 69.5, 69.6, 69.7, 69.9, 71.2, 71.8, 72.8.
Notice that we have 12 values; therefore, two share the middle: 69.5 and 69.6. We take the midpoint
of these two values to get the upper quartile:
69.5 + 69.6 139.1
= = 69.55.
2 2
Next, let us find the lower quartile. For that, we take the values below the median:
59.8, 60.4, 62.1, 62.9, 63.1, 63.3, 63.9, 64.2, 64.8, 64.8, 65.1, 66.3.
Again, we have 12. Two share the middle: 63.3 and 63.9. We take the midpoint of these two values to
get the lower quartile:
63.3 + 63.9 127.2
= = 63.6.
2 2
Now that we have computed the median, the lower quartile, and the upper quartile, we can visualize
the distribution of the height variable using a box plot, as shown in Figure 3. The box encloses all the
values between the lower and upper quartiles. A line is drawn inside the box to indicate the median
value. Thin lines, the whiskers, are drawn to indicate the location of the minimum and maximum
values.
60 62 64 66 68 70 72 74
Figure 3: Box plot for heights in example data set.
2 You might wonder whether to include the median itself in this group. Different textbooks give different answers to this
question: some include the median in both the “upper half” and the“lower half”; some include it in neither. In this book we
will not include the middle in either “half.” That is, if we have an odd number of individuals (like 25 here), the “lower half”
is points 1 through 12 in order, and the “upper half” is points 14 through 25. If we had an even number of individuals, say 24,
the middle would fall between points 12 and 13; the “lower half” would be points 1 through 12 and the “upper half” points 13
through 24. This is the same convention that is followed by the built-in statistical functions in the TI-83 series of calculators. If
you use other statistical programs or packages, be aware that the alternative definition may produce slightly different answers.
Notice that:
• The vertical line in the center of the box is drawn at the median, 66.4.
• The two edges of the box are at the lower and upper quartiles, 63.6 and 69.55.
• The “whiskers” extend to the least value, 59.8, and the greatest value, 72.8.
A box plot can give us an idea of how widely distributed the values of a variable are. Each half of
the box represents 25% of the data, so the whole box represents the “middle 50%.” The width of the
middle 50%, that is the width of the box, is a measure of spread called the interquartile range.
Definition 6
The interquartile range (IQR) for a variable is the difference between the lower and upper
quartiles. In other words, it is the width of the “box” in a box plot for the variable.
A value that is unusually far “outside the box” is called an outlier. Bill Gates’s net worth is an
outlier. André the Giant, who died in 1993, is an outlier for the height variable (7 feet 7 inches). So
is Jyoti Amge of India (currently the world’s shortest mobile adult person at just about 2 feet). An
outlier can be “outside the box” in either direction.
How far outside the box is “unusually far”? People give different definitions, and to some extent it
depends on the type of problem being analyzed, but one common definition is that an outlier must be
“outside the box” by more than 150% of the width of the box (that is, the interquartile range). That is
the definition we shall use in this book.
Definition 7
An outlier is a variable value that is more than 1.5 times the IQR (interquartile range) below
the lower quartile, or more than 1.5 times the IQR above the upper quartile.
If there are outliers, we draw them as separate points in the box plot. Instead of extending right to
the end of the range, the “whiskers” extend only as far as the least and greatest values that are not
outliers. Here is an example.
Problem 3: The phosphorus concentration data in Table 2 on page 262 are taken from a larger study
on phosphorus concentrations in 28 study sites in the Puget Sound region. The full sorted list of values
for phosphorus concentrations, in µg/L is
4, 7, 10, 11, 11, 12, 13, 15, 15, 15, 15, 15, 16, 16, 16, 16, 16, 17, 19, 19, 19, 20, 20, 20, 23, 23, 26, 28.
Calculate the median, the lower quartile, and the upper quartile for this data set, and make a box plot
including outliers.
Solution: Here, the middle two values are both 16, so the median is also 16.
The lower half of the data set, which contains 14 values, is
4, 7, 10, 11, 11, 12, 13, 15, 15, 15, 15, 15, 16, 16.
The middle two values are marked in red. The lower quartile is the midpoint between these two values:
13 + 15
lower quartile = = 14.
2
The upper half of the data set, also containing 14 values, is
16, 16, 16, 17, 19, 19, 19, 20, 20, 20, 23, 23, 26, 28;
19 + 20
upper quartile = = 19.5.
2
Now that we know that the lower quartile is 14 and the upper quartile is 19.5, we can compute the
IQR. It is the difference between the quartiles:
IQR = upper quartile − lower quartile = 19.5 − 14 = 5.5.
An outlier is a value that is more than 1.5 times the IQR above or below the upper or lower quartiles,
respectively:
1.5 × 5.5 = 8.25.
We are looking for values that are more than 8.25 below the lower quartile or more than 8.25 above
the upper quartile:
14 − 8.25 = 5.75, 19.5 + 8.25 = 27.25.
The variable takes the value 4, which is below 5.75, and the value 28, which is above 27.25, so these
values are outliers. There are no others.
In a box plot with outliers, the whisker on the left extends to the lowest value that is not an outlier,
which is 7 in our example. Similarly, the whisker on the right extends to the highest value that is not
an outlier, which is 26. The final box plot is as shown in Figure 4.
5 10 15 20 25 30
Figure 4: Box plot of phosphorus data set, with outliers. Data from [130].
Remark 1: Free online calculators to generate box plots can be found in many places. Unfortunately,
they do not all follow the same conventions about how to define quartiles (see footnote on page 265)
and outliers as we do, so check before using them!
Visuals for Distributions: Histograms

A box plot provides a simple way of summarizing the distribution of a numerical variable. It does this
by showing where “most” of the values of the variable fall. If we want to get a more complete picture,
the next step might be to use a histogram, which is a chart showing how often the values of a variable
fall into various intervals.
The first step in drawing a histogram for a numerical variable from a data set is to divide the range
of values of the variable into a number of equal intervals, called class intervals. Then we draw up
a frequency table, showing how many individual values of the variable belong to each of the class
intervals.
Definition 8
The number of times a variable (in a data set) takes on a value in a certain class interval is
called the frequency of the variable for that interval.
Let us once again consider the heights (in inches) from our sample data set in Example 3 on
page 263. Sorted in order, these are
59.8, 60.4, 62.1, 62.9, 63.1, 63.3, 63.9, 64.2, 64.8,
64.8, 65.1, 66.3, 66.4, 66.8, 67.2, 68, 68.3, 68.3,
69.5, 69.6, 69.7, 69.9, 71.2, 71.8, 72.8.
We decide to use five class intervals. The range is 72.8 − 59.8 = 13 inches. In order to divide this
into five equal intervals, the width of each interval must be 13 ÷ 5 = 2.6 inches. Our intervals will
therefore be 59.8 to 62.4, 62.4 to 65.0, 65.0 to 67.6, 67.6 to 70.2, and 70.2 to 72.8.
Now we make a frequency table showing how many data values lie in each class interval (Table 3).
Interval 59.8–62.4 62.4–65.0 65.0–67.6 67.6–70.2 70.2–72.8

Frequency 3 7 5 7 3
Table 3: Frequency table for height data from Example 3 on page 263.
For example, the value 5 in Table 3 shows that 5 students had heights in the central interval 65.0–
67.6. You can verify this from the listing above. (It is helpful to have arranged the data in order before
starting to count frequencies!)
Question 1: What should I do if one of the data values is on the boundary between two class intervals?
For example, if one of the heights is exactly 65.0, should I count it in the second interval (62.4–65.0)
or the third (65.0–67.6)?
Answer: We will use the convention that if there is a choice, you always take the rightmost interval
(the one with the larger endpoint values). In the example above, we would assign the data value 65.0
to the third interval.
Question 2: How do I know how many class intervals to use?
Answer: It depends on what aspects of the data we want to represent! A helpful rule of √ thumb,
however, is that if the total frequency (the number of values) √ is N, we should take about N class
intervals. (In the example, N = 25, so this suggests taking 5 = 25 intervals, as we did.) This approach
is called the square root heuristic.
Once we have made a frequency table, we can use it to draw a histogram.
Definition 9
A histogram is a graphical representation of a frequency table. It is a chart with adjacent

vertical bars. Each bar represents one entry from the frequency table: its base (on the x-
axis) is the class interval that it corresponds to, and its height (on the y-axis) represents the
frequency.
0
60 62 64 66 68 70 72 74
Figure 5: Histogram of class heights corresponding to frequency table given in Table 3.
Figure 5 shows a histogram corresponding to the frequency table given above (Table 3 ). The
leftmost vertical bar corresponds to the first class interval, 59.8–62.4. Its left- and right-hand sides
have x coordinates 59.8 and 62.4. Its height, 3 (the y coordinate of its top), corresponds to the
frequency 3 associated with this class interval. Similarly for the other 4 vertical bars. (Just as a matter
of appearance, then, a histogram looks like a column bar chart. It is its use to represent frequencies
that makes it specifically a histogram.)
Question 3: Now that I have drawn the histogram, I can see clearly that the height distribution seems
to have two separate “peaks.” What could account for that?
Answer: You have just discovered one of the key reasons that we draw histograms—they make
important properties of a distribution visual and intuitive. In this example, the two-peaked or bimodal
property of the distribution probably comes from having a population that contains people of different
genders: male heights are typically 6 inches or so greater than female heights, so a sample containing
many males and females is likely to show a bimodal height distribution.
Example 4: Make a histogram from the phosphorus concentration data set in Problem 3 on page 266.
Use 6 class intervals.
Solution: We recall that the data are
4, 7, 10, 11, 11, 12, 13, 15, 15, 15, 15, 15, 16, 16, 16, 16, 16, 17, 19, 19, 19, 20, 20, 20, 23, 23, 26, 28.
√
There are 28 individuals, and 28 is between 5 and 6, so the square root heuristic would suggest
that we should use either 5 or 6 class intervals. In fact, the question tells us to use 6. The range is
28 − 4 = 24, so the width of each interval is 24/6 = 6 micrograms per liter. Thus the class intervals
are 4–8, 8–12, 12–16, 16–20, 20–24, 24–28.
Now we can make the frequency table. We need to remember the rule that if there is a choice, a
data value should be counted in the rightmost class interval. For instance, the “16” values should all
be counted in the 16–24 interval, not the 12–16 interval. See Table 4
Interval 4–8 8–12 12–16 16–20 20–24 24–28
Frequency 2 3 7 9 5 2
Table 4: Frequency table for the phosphorus concentration data given in Problem 3 on page 266
0
5 10 15 20 25 30
Figure 6: Histogram of phosphorus concentrations corresponding to Table 4 on the previous page [130].
We can use this frequency table (Table 4 on the previous page) to draw the histogram shown in
Figure 6. In contrast to the previous example (distribution of student heights), this one has a single
peak. The class interval 16–20 stands above the rest, and the frequencies fall away smoothly on either
side.
Remark 2: You can generate histograms with a spreadsheet program, such as Excel. As with box
plots, you can find also free online software to do this.
To summarize, we can use these visual representations (box plots and histograms) to communicate
a large amount of information in a simple way. They tell us, among other things,
(a) where the center of the distribution appears to be,
(b) how widely dispersed the values are around that center (spread),
(c) and whether that dispersion is symmetric.
5.1.2 Descriptive Statistics

Histograms and box plots provide a graphical summary of the key features of a distribution. By
contrast, a descriptive statistic is a number used to summarize the center, spread, or skewness of a
distribution. In this section, we will look at some of these descriptive statistics.
Measures of Center
We have already explored one measure of center: the median. Another measure of center—one that
you most likely are already familiar with—is the mean, also sometimes called average.
Definition 10
The mean of a set of numerical values is the sum of all the values in the set divided by the
number of values in the set. In a formula,
Sum of all members of V
Mean of set V = .
Number of members of V
In media reports, phrases like “average height” or “average temperature” usually refer to
the mean.
Problem 4: Find the mean of the heights of the students in the “mock” data set3 that we considered
in Example 3 on page 263.
Solution: There are 25 height values, and their sum is 1660.25. Dividing by 25, we obtain
1660.2
≈ 66.4
25
for the mean of the distribution. In this case, the mean and the median are the same. This will not
usually be the case.
Problem 5: At a certain university, the annual salaries of 11 employees are
$36,072, $41,976, $44,892, $48,744, $51,948, $53,964,

$71,712, $76,968, $98,532, $107,640, $189,756.
Find the median and the mean.
Solution: The data are already sorted for us. The median salary is just the middle number, $53,964.
But the mean is the sum of all the salaries divided by 11; this turns out to be much larger: $74,746.
The difference reflects the fact that our distribution is skewed to the right, with the one outlier salary
at the high end having a large influence on the mean. Figure 7 is a box plot summarizing this data set,
including the outlier. Notice how we added a “diamond” to the plot to indicate the mean.
50 100 150 200

Figure 7: Box plot of employee salaries (in thousands of dollars). Note the diamond indicating the mean.
We will say more about the idea of “skew” on page 274.
Critical Thinking
In news reports and other places you may well have seen arguments like this:
The average household income in the U.S. is $56,000. The average cost of a
family health insurance plan (on the individual market) is $10,000. Therefore,
the average family paying for health insurance on the individual market must
$10, 000
spend ≈ 18% of their income paying for health insurance.
$56, 000
Analyze this argument critically. Even supposing that the figures are accurate, what
additional assumptions does it involve?
Problem 6: Suppose a class of 20 students takes a midterm exam. A frequency table for their scores is
below. Find the mean, median, range, and IQR. Draw a box plot reflecting the distribution of scores.
3 Available from http://math-for-sustainability.com/sample-data/.
Score 65 70 75 80 85 90 95 100
Number of students
3 4 3 1 4 2 2 1
with score
Solution: There are

3 + 4 + 3 + 1 + 4 + 2 + 2 + 1 = 20
students in the class.
The mean requires we add up all the scores and divide by 20. There were 3 scores of 65, 4 scores
of 70, and so on up to 1 score of 100. So the mean is
(3)(65) + (4)(70) + (3)(75) + (1)(80) + (4)(85) + (2)(90) + (2)(95) + (1)(100)

= 79.5.
20
Because 20 is an even number, to find the median we will need to take the midpoint of the 10th and
11th values (when written in ascending order). We can find these values using the frequency table.
Values 1 to 3 are 65, 4 to 7 are 70, 7 to 10 are 75, and 11 is 80. So we need to take the midpoint of 75
and 80 to find the median.
75 + 80
median = = 77.5.
2
The range is the difference of the highest and lowest scores: 100 − 65 = 35.
To find the IQR, we first need to know the upper and lower quartiles. The upper quartile is the
median of the top 10 values, and the lower quartile is the median of the bottom 10 values. Because 10
is even, the median of the bottom 10 will be the midpoint of the 5th and 6th values. The bottom 10 are
Score 65 70 75
Number of students
3 4 3
with score
The 5th and 6th values are both 70. Therefore the lower quartile is 70. The top 10 values are
Score 80 85 90 95 100
Number of students
1 4 2 2 1
with score
The 5th value is 85 and the 6th value is 90. So the upper quartile is their midpoint, 12 (90 + 85) = 87.5.
Now we can compute that the IQR is the difference of the upper and lower quartiles:
Interquartile range (IQR) = 87.5 − 70 = 17.5.
With the information we’ve found, we can draw a box plot (Figure 8). As in the previous solution, we
add a “diamond” to indicate the location of the mean. There are no outliers in this plot.
70 80 90 100
Figure 8: Box plot of student scores.
Measures of Spread
A measure of spread is a descriptive statistic that tells us how “widely spread out” a distribution is
about its center. The most important measure of spread we have talked about so far is the interquartile
range (IQR)—the difference between the upper and lower quartiles, which we could also call the
width of the “middle half” of the distribution. (The range—the difference between the highest and
lowest values—can also be considered a measure of spread, but it is not a useful descriptive statistic
because of its extreme sensitivity to outliers.)
In reading books and scientific literature you are likely to come across another measure of spread
called the standard deviation. The standard deviation is related to the interquartile range in roughly
the same way as the mean is related to the median. To compute it, rather than simply arranging the
data in order and choosing appropriate values from the list, one must take all the data and perform
a calculation involving sums, squares, and square roots. The exact formula for standard deviation is
not part of this book. It is worth knowing, however, the notation4 σ for the standard deviation, and
you should know that it (like the IQR) measures the extent to which data are “spread out” around the
center of the distribution. It is useful also to be aware of the ways in which standard deviation and
IQR are related:
Rule 1: Standard Deviation Rules
• For any distribution, the interquartile range can be at most 3 standard deviations: that
is, IQR ≤ 3σ .
• For many data sets (specifically those that follow a normal distribution, which we
will be discussing in Section 5.1.3), the relationship is much closer: the interquartile
range is approximately 1.35 times the standard deviation:
IQR ≈ 1.35 × σ , σ ≈ 0.74 × IQR.
Let’s give a table to summarize the descriptive statistics we’ve learned about so far.
Table 5: Descriptive Statistics
Name Tells Us What to Know About It

About
Median Center 50% of values are above it and 50% are below it.
Mean Center The usual “average”: it is the sum of all values divided by the number
of values.
Interquartile Spread It’s the difference between the upper quartile and the lower quartile.
Range (IQR) It captures 50% of values around the median.
Standard Spread A more complicated and mathematically powerful measure of
Deviation spread, about 74% of the IQR for the most common kind of
distribution.
Range Spread The difference between the highest value and the lowest one. Too
sensitive to outliers to be a reliable statistic.
4 This is the Greek letter “sigma,” the same letter that we already used in Section 2.4 for Stefan’s constant.
mean = median
median median
mean mean
Leftward Skew No Skew Rightward

No SkewSkew
Figure 9: A distribution can be skewed to the left or the right.
Skewness
Look again at the histogram of class heights in Figure 5 on page 269. It is clear that this histogram
is symmetric about its middle point: if you were to place a mirror at the middle value (x = 66.3),
the left-hand half would be the exact reflection of the right-hand half. This tells us that a randomly
selected student from the class is just as likely to fall 3 or more inches, say, below the middle height
of 66.3 inches as they are to fall 3 or more inches above it.
By contrast, consider the distribution of household incomes, which we discussed on page 264.
The median annual household income in the U.S. is about $56,000. But no household income is less
than zero, whereas the incomes of the richest households (the notorious 1 percent) extend from about
$300,000 into the tens or hundreds of millions. The distribution is lopsided—much more stretched out
to the right than to the left. (You see the same effect on a smaller scale in the salary data in Problem 5
on page 271.) The technical term for this lopsidedness is that the distribution is skewed.
Definition 11
A distribution is skewed if it stretches out further to one side of its center than to the other.
We refer to rightward skew or leftward skew according to which side of the center the
distribution is more stretched out: rightward skew means that the distribution has a longer
“tail” on the right, and leftward skew means that it has a longer “tail” on the left.
These definitions are illustrated in Figure 9 (this figure shows distributions as smooth curves, which
we can imagine as idealized histograms in which the class intervals become very narrow).
One indicator than a distribution is skewed is the difference between the mean and the median. The
mean is sensitive to long tails and thus tends to be closer to the tail than the median. We saw this in
the box plot for the employee salary example, Figure 7 on page 271. Figure 10 on the opposite page
shows three box plots corresponding to the three histograms in Figure 9.
Why is skewness important? A skewed distribution alerts us to the presence of asymmetrical risk
in a given situation. For example, think about auto insurance. If we took 10,000 car owners of roughly
your age and followed them for a year, we could make a histogram to show the distribution of the
amount of money that they lost owing to auto accidents in that year. It doesn’t take much thought to
see that this distribution is going to be strongly skewed to the right. The majority will drive safely all
year and have zero losses. Quite a number will have minor “fender bender” accidents whose repair
cost is a couple of hundred dollars. But a few will completely wreck their cars, racking up losses in
the thousands or tens of thousands of dollars. Even though the median loss may be quite small—a
hundred dollars or so, maybe—the strong rightward skew alerts us to a small chance of catastrophic
outcomes. If wrecking your car would ruin your life (maybe you can’t get to work anymore, you
lose your job, now you can’t pay the rent, your significant relationships unravel. . . ) you may want to
purchase insurance to guard against this risk. (If you are rich enough to just walk out and buy another
car, that is another story. Insurance may not be a worthwhile investment for you. But few people are
in so fortunate a position.)
All of these points are relevant when we
look at the rightward-skewed distribution of
climate risks represented by a graph such as
Leftward Skew
that shown in Figure 10 on page 204. Re-
call that the histogram in that figure repre-
sents (our best estimate of) the distribution of
potential outcomes resulting from repeatedly No Skew
“rolling the climate dice”—a “roll” in this
case being a hypothetical doubling of carbon
dioxide concentrations from preindustrial lev-
Rightward Skew
els. The median outcome, about a 2 ◦ C tem-
perature rise, is very scary but, in the opinion
of many scientists, something that humanity
might succeed in adapting to. But the long
Figure 10: Illustrating skewness via box plots.
rightward tail does include some outcomes
that would ruin your life—in fact, that would
ruin the life of nearly everyone on Earth. To continue the analogy, no one is rich enough to go out
and buy a new planet if we total this one (nor, as far as we know, are there any for sale). There is a
strong case, then, for humanity taking out an insurance policy, in the form of a collective commitment
to reduce emissions of carbon dioxide and other greenhouse gases. That doesn’t mean that we know
for sure that a catastrophic outcome is coming (nobody takes out auto insurance because they know
for sure they are going to total their vehicle). All that is needed is an understanding that the strongly
rightward-skewed risk distribution includes outcomes whose costs we could not possibly meet.
5.1.3 The Normal Distribution

If you study many large data sets taken from natural populations,
you will find that they often take on the symmetric “hill” or “bell”
shape that we already noticed in the phosphorus concentration data of
Figure 6 on page 270. Another example is shown in Figure 11. This
distribution was taken from a National Institutes of Health study on
body mass index (BMI) that was conducted in 1990 (see [263]). (The
BMI is a measure of body size that takes into account both one’s weight
and one’s height. For adults, the normal range of BMI is considered
to be 19–25, with a BMI above 30 defined as “obese.”) Again, we
see that the histogram follows a bell-shaped pattern. In fact, in this
example the authors have drawn a smooth bell-shaped curve as a fit to
the data provided by the histogram. The curve that they have drawn
has a special shape. It is called a normal distribution curve.
We will not give a mathematical definition of the normal distribution
in this book (that belongs, rather, to a course in statistics). What you
need to know is that the normal distribution is smooth, bell-shaped, and Figure 11: Body mass index dataset,
symmetric, and that it frequently occurs in natural data. Why? Because from [263].
mathematicians have shown5 that whenever a numerical variable is
obtained as the aggregate (sum) of many small, independent random quantities, it will tend to follow
5 This is called the central limit theorem.
a normal distribution. If we think about body weight for instance, there are a huge number of life
influences—genetic, environmental, behavioral—that together produce someone’s weight as it is
today. If we imagine that these influences are independent and cumulative, then the mathematics will
lead us to expect that body weights will follow a normal distribution. This is, indeed, approximately
true.
A normal distribution is completely specified by its center and spread, usually given in terms of
its mean6 and standard deviation. The interquartile range, for a normal distribution, is about 1.35 × σ
(see Rule 1 on page 273); to put that another way, for a variable that follows a normal distribution,
about half the values of the variable lie within 0.67 ≈ 1.35/2 standard deviations of the mean. It is
also helpful to have some information about where “most” values lie:
Rule 2: 95 Percent Rule
For a variable that follows a normal distribution, about 95% of the values lie
within two standard deviations of the mean, and almost all lie within three standard
deviations.
Problem 7: Consider the BMI data set above, and assume it follows a normal distribution. If the mean
BMI is 25.44 and the standard deviation is 4.9, what range of BMI should include 95 percent of the
individuals sampled?
Solution: 95 percent of individuals should have a BMI between the lower limit of 25.44 − (2)(4.9) =
15.64 and the upper limit of 25.44 + (2)(4.9) = 35.24.
Remark 3: Not all data sets follow a normal distribution. If you review our discussion above, the key
word you should focus on is independent: a variable will follow a normal distribution if it is made up
of the aggregate of many independent small quantities. What does this mean? We’ll give a definition
later (Definition 10 on page 291) but the idea is natural enough: independent quantities don’t have any
“influence” connecting them. When this assumption fails, the comforting assurance provided by the 95
percent rule breaks down, and extreme events become much more likely. According to some authors
like Cathy O’Neil [256], baseless assumptions of independence fueled the misplaced confidence in
risk modeling and quantification that in turn led to the 2008 financial meltdown.
Think about it. . .
Cathy O’Neil’s book Weapons of Math Destruction [256] includes the story of her journey
from Harvard PhD to math professor to hedge fund analyst (with an inside view of
the financial crisis) to writer and blogger about the dangers of overreliance on opaque
mathematical models. The rest of the book contains many specific examples reinforcing
her argument. In the concluding chapter she writes:
Big Data processes codify the past. They do not invent the future. Doing that
requires moral imagination, and that’s something only humans can provide.
Try to express this in your own words. Do you agree? We’ll return to the theme of moral
imagination at the end of this book, Section 6.4.
6 The symbol for the mean is another Greek letter, “mu” (µ). We already mentioned that the symbol for standard deviation
is “sigma” (σ ).
Summary of Ideas: Data and Distributions
• A data set is a collection of information. Objects in a data set are called individuals.
Characteristics of each individual are called variables.
• Variables can be categorical, meaning they categorize individuals by some trait, or
numerical, meaning they describe individuals by some numerical measurement.
• A distribution of a numerical variable is an efficient way of telling us what values
the variable takes and how often those values are observed.
• The median of a numerical variable splits the distribution so that 50% of values are
above, and 50% are below. The upper quartile is the median of the top 50% of values
and the lower quartile is the median of the bottom 50% of values. The interquartile
range (IQR) is the difference between the upper and lower quartiles.
• The mean of a numerical variable is the sum of all its values divided by the number
of values. The mean and the median are two different ways of measuring the “center”
of a distribution.
• An outlier is a value of a numerical variable that is significantly far from the rest of
the values. We define an outlier to be any value that exceeds the upper quartile, or is
below the lower quartile, by more than 150% of the IQR.
• A box plot visually represents the distribution of a numerical variable by showing the
quartiles, the median, the upper and lower values, outliers (if any), and possibly also
the mean.
• A histogram visually represents the distribution of a numerical variable by a column
bar chart in which each bar corresponds to a range of values for the variable and the
height of the bar reflects the number of times a value in that range is observed.
• A distribution for a numerical variable is skew if it stretches out further to one side
of its median than to the other.
• A normal distribution is a bell-shaped curve that gives a good approximation to the
distribution of many naturally observed numerical variables.
5.2 Probabilities and Predictions
Objectives
I understand the ideas of a random process and an outcome.

I know what is meant by a probability model.
I can define the terms sample space, outcome, and event in the context of a
probability model.
I can find the probability of an event.
I know what it means for two events to be independent, and what it means for them
to be mutually exclusive.
I can find the union and intersection of two events and determine their probabilities
under appropriate conditions.
I understand the law of large numbers and can avoid fallacies in applying it.
I can use probability language to describe levels of uncertainty.
5.2.1 Random Processes

In the previous section, we talked about various ways to summarize the distribution of numerical
data—variables—from a data set. These included box plots and histograms, which summarize the
data visually, as well as descriptive statistics that measure the center (the median and the mean) and
the spread (the IQR and the standard deviation) of the distribution. Summarizing data in these and
other more sophisticated ways is a crucial skill in science.
But where does all this numerical data come from? It could come from a complete listing of some
population, like all the students in a class or all the households in the nation (this is the origin of the
term “statistics,” literally “state-istics”). But there is another common way that data can be generated:
from the successive outcomes of a random process. Let’s begin with a very broad definition.
Definition 1
Any activity whose result cannot be determined with certainty is a random process. An
outcome is the result of a single instance of the random process.
Example 1: The weather is an example of a random process that is familiar to all of us. We cannot
perfectly predict the weather from day to day because the dynamics of the whole atmosphere is too
complex a process to simulate even on our most powerful computers (see Section 2.4.1). That’s why
weather forecasts consist of ranges of temperatures (highs and lows, for instance) and “percentage
chances” of events like rain, hail, and snowfall. An outcome of the weather process might be “the
temperature and precipitation in my home town at 2 p.m. tomorrow.” Even though the forecast does
not tell us this outcome with certainty, it still allows us to make decisions while taking potential risks
into account—like deciding to go ahead with a picnic despite a 20% chance of rain or deciding to halt
roadwork when there is a 60% chance of thunderstorms.
5.2. PROBABILITIES AND PREDICTIONS 279
Example 2: Genetic mutation is a random process that is important to all of us. A mutation is an
unpredictable change in the genetic code that governs the growth and development of each living
thing. Each of us human beings, as well as each plant and animal, is continually exposed to influences
that have a tiny chance of causing a mutation: radiation (both natural and human-caused), mutation-
inducing chemicals, and transcription errors in the copying process that makes DNA for new cells.
Most mutations are probably never noticed; some are harmful or lethal (such as a mutation that causes
cancer); but others may be beneficial. For instance, plant and animal breeders create new varieties
by selectively reinforcing the effect of beneficial random mutations; this is how our food crops and
domestic animals have evolved over centuries or millennia from their wild predecessors. (See Case
Study 7.5 for more about this.)
Example 3: Random processes make themselves known in a slightly different way in the working
of the scientific method (Section 2.1.4). Imagine a scientist carrying out an experiment to measure
some quantity—perhaps the average concentration of arsenic in the water of a lake (see Problem 5
on page 38). The good scientist will not simply take one water sample, analyze it once, report the
resulting number, and go home. Why not? That would not be following the scientific method, which
requires “stress testing” your model (here, your estimate of the arsenic concentration in the lake). No,
the conscientious scientist will carefully repeat the experiment several times, and she will most likely
find that the results are slightly different each time. Not every aspect of the measurement process
can be precisely controlled, which is to say that the measurement is a random process. The various
outcomes of this process (that is to say, the different measured values) form a data set in the sense of
Definition 1 on page 261. The process of obtaining a single “best” value for the quantity of interest,
together with an appropriate precision (Section 1.3), from this spectrum of observations is called
(statistical) inference, and we’ll discuss it further in Section 5.4 once we have developed the necessary
ideas about probability.
Example 4: Games of chance were the first random processes to be studied seriously by mathemati-
cians, in the seventeenth century. Flipping coins (one or several), rolling dice, and so on, give examples
of random processes for which we can write down the complete list of all possible outcomes, called
the sample space. What’s more, if the coins or dice are assumed to be “fair,” then the outcomes in
the sample space are all equally likely, and this gives a natural way to assign probabilities—measures
of likelihood—to more complex events like “rolling a total of 5 with two dice.” We are going to use
probability theory, which was derived from studying games of chance, to help us understand all kinds
of random processes. That is why you will see a lot of dice rolls and coin flips among the examples
for the next few pages, before we get back to applying the theory to sustainability questions.
5.2.2 Probability Models

Let’s think about the basic features of a game of chance. First of all, there are the outcomes of the
game. As we have already defined them, these are the possible results of playing the game once.
If, for instance, our game is “flip a single coin,” then there are two possible outcomes, “heads” and
“tails.” If our random process is “roll a single die,” the possible outcomes are 1, 2, 3, 4, 5, and 6. (We
could also denote them by , , , , , and , but it is more convenient to deal with numbers!) If
our game is “roll two dice and record the score from each one,” then there are 36 possible outcomes
ranging from (1, 1) to (6, 6). But if our game is “roll two dice and record the total score,” there are
only 11 possible outcomes, ranging from 2 to 12.
The list of all the possible outcomes is called the sample space.
Definition 2
The sample space is an exhaustive list of all the possible outcomes. It is usually denoted
by the letter S and is written using set notation (the list of possible outcomes is enclosed by
curly brackets).
In the example of rolling a single die, the sample space would be expressed by
S = {1, 2, 3, 4, 5, 6}.
The sample space might be very large in some examples—too large for us to explicitly write out all
the possibilities. But the concept of the sample space is still an important one.
Remark 1: Though the sample space may be large, we’ll assume that it isn’t infinitely large. Infinitely
large sample spaces come up often enough that any serious treatment of probability has to address
them. But doing so properly involves a lot of complicated mathematics that would distract us from
the main points of this chapter. We’ve chosen to take an end run around these issues by assuming that
sample spaces are always finite.
The central idea of probability theory is that of an event. Imagine a game of chance, like rolling a
single die. An event is “a thing that may or may not happen” each time we play the game. A good
example might be “rolling an even number,” which we could denote by the letter E. Sometimes we
play the game and the die comes up 2, 4, or 6. We say that the event E happens or occurs on that roll.
Other times we play the game and the die comes up 1, 3, or 5. Then we say that the event E does not
happen on that roll.
It turns out that there is an equivalent way to express this same idea. Remember that we are
assuming that there is a finite sample space—a finite list of outcomes, one and only one of which
must occur each time we play. Some of those outcomes match up with the event E, and some don’t.
In the case of the example in which E is “rolling an even number” with a single die, the outcomes that
match up to E are 2, 4, and 6. Instead of speaking of the event E as something that happens or not,
we could simply speak of this collection of outcomes—mathematically speaking, this subset {2, 4, 6}
of the sample space. And this would convey the same information: “I rolled an even number” means
exactly the same thing as, “The outcome I obtained belongs to the subset {2, 4, 6}.” This provides us
two alternative languages for speaking of events: one in terms of “things that may or may not happen”
on each play of the game and one in terms of “collections of outcomes,” that is, subsets of the sample
space. Both of these languages are important, and we will try to give key definitions in both versions.
Definition 3
An event is any collection of outcomes (that is to say, any subset E of the sample space S).
To say that “E happens” on some play of the game is the same as to say that the outcome of
that particular play belongs to the subset E.
Here are some examples, for the random process of rolling a die:
• Event A = {1, 2, 3, 4}, or rolling a number less than or equal to 4.

• Event B = {2, 4, 6}, or rolling an even number.
• Event C = {1, 2, 3, 5, 6}, or rolling anything but 4.
• Event D = {1}, or rolling a 1.
Notice that we can think of a particular outcome, like rolling a 1, as part of many different events—
in this example, the events A, C, and D. To put it another way, if we roll the die and a 1 comes up, then
events A, C, and D all happen on that roll (as do many other possible events that we didn’t mention).
There are also two extreme examples that are always allowed to count as events:
• Event 0/ = { }, which contains no outcomes at all7 and thus never occurs, and
• Event S = {1, 2, 3, 4, 5, 6}, the entire sample space (containing every outcome), which always
occurs whatever the outcome.
Probabilities
A game of chance is not completely specified when we have defined the sample space and listed the
corresponding outcomes. For example, the sample space for a flip of a biased coin and the sample
space for a flip of an unbiased coin are the same, namely {Heads, Tails}. What’s still needed is
information about the likelihood or probability of different events.
Definition 4
The probability of an event is a real number between 0 and 1 that measures how likely that
event is to happen. If E is an event, its probability is denoted by P(E), so that 0 ≤ P(E) ≤ 1.
(It is also common to specify probabilities in terms of percentages, so one might talk of a
“25% chance” rather than a probability of 0.25.)
There are some reference points on the probability scale: P(E) = 0 means that E is impossible,
P(E) = 1 means that E is certain, and P(E) = 12 means that E is as likely as not. Probabilities
cannot be assigned in a completely arbitrary way: they have to obey some rules that will be set out in
Section 5.2.3. First, though, let us gather up the concepts we have introduced into a single definition.
Definition 5
A probability model is a mathematical representation of a game of chance. It is defined by

• the sample space, made up of all the possible outcomes,
• the events, which are collections of outcomes (or, to put it another way, subsets of the
sample space), and
• the probability associated with each event.
If we make a table listing all possible outcomes together with the probability of each one, we obtain
a data set (Definition 1 on page 261). The individuals of this data set are the outcomes; there is one
(numerical) variable that gives the probability of each specific outcome. This data set is called the
distribution of the probability model. There is one condition that it must obey:
7 The symbol 0/ is called the null set or empty set symbol.
Rule 1: Law of Total Probability
In any probability model with (finite) sample space S, the sum of the probabilities of
all the individual outcomes in the sample space must equal 1.
In simple cases, like flipping a fair coin or rolling a fair die, all the outcomes are equally likely.
(This is because these objects are symmetric, so interchanging “heads” and “tails” on the coin, or
interchanging the way two faces of the die are labeled, will not affect any of the probabilities—which
is another way to say that these probabilities are all equal.) Probability models of this type were the
first to be studied.
Definition 6
Outcomes in a probability model are said to be equally likely if they have the same
probability. If all outcomes in the sample space are equally likely, then we say the
probability model is uniform.
For a uniform probability model, there is a direct way to calculate the probability of any event E.
We simply divide the number of outcomes in the event E by the number of outcomes in the sample
space S.
Rule 2: Equally Likely Probability Rule
When all outcomes in the sample space S are equally likely (that is, when the
probability model is uniform), the probability of an event is given by the formula
Number of Outcomes in Event E
Probability of Event E = P(E) =
Number of Outcomes in Sample Space S
Problem 1: Assuming a uniform probability model, find the probabilities of each of the possible
events associated with a single flip of a coin. (Recall that an “event” can be any collection of
outcomes—including none or all of them, the “extreme examples” from page 281.)
Solution: The sample space associated with flipping a coin is S = {H, T }, where H means heads, and
T means tails.
The events are the subsets of the sample space. So, let’s list all the possible events. We’ll give them
descriptive names, but we could also denote them by letters like A, B,C, D:
• Head = {H}.
• Tail = {T }.
• Either = {H, T }. (This is the event that we get either heads or tails.)
• Neither = { } = 0.
/
Finally, we need to determine the probability of each event. Since we assume that outcomes are equally
likely, we can use Rule 2 to determine the probabilities.
Table 6: Solution to Problem 1

Number of Outcomes
Event Probability
in Event
1
Head 1 = 0.5
2
1
Tail 1 = 0.5
2
2
Either 2 = 1.0
2
0
Neither 0 = 0.0
2
Question 1: I thought you said that for a uniform model all the probabilities were equal! So how come
we now have one event with probability 1 and one with probability 0?
Answer: A uniform model means that each outcome has the same probability. But an event can include
a number of different outcomes; the more outcomes in the event, the more probable it will be.
Problem 2: A uniform probability model is given by rolling a fair (six-sided) die. Calculate the
probability of each event listed below
• Event A = {1, 2, 3, 4} (rolling a number less than or equal to 4).

• Event B = {2, 4, 6} (rolling an even number).
• Event C = {1, 2, 3, 5, 6} (rolling anything but 4).
• Event D = {1} (rolling a 1).
Solution: For each event, we use Rule 2 to calculate the probability.
• Event A = {1, 2, 3, 4}. Thus Event A contains exactly four distinct outcomes. Using Rule 2, we
see that
P(A) = 46 = 23 ≈ 0.67.
We can use the same process to compute the probabilities of the other events.
3 1
• Event B = {2, 4, 6}. Thus P(B) = 6 = 2 ≈ 0.5.
5
• Event C = {1, 2, 3, 5, 6}. Thus P(C) = 6 ≈ 0.83.
1
• Event D = {1}. Thus P(D) = 6 ≈ 0.17.
Problem 3: Two fair coins are flipped. Find the probability of exactly one head, using the equally
likely probability rule.
Solution: Beware! It’s tempting to think that the equally likely outcomes here are “0 heads,” “1 head,”
and “2 heads,” which would make the answer 13 . But this is incorrect.8 Let’s look at the situation more
carefully.
8 You would be in good company if you made this mistake, though. The eighteenth-century French mathematician Jean le
Rond d’Alembert, one of the founders of probability theory, made it too.
It’s helpful to imagine that the two coins are of different sorts—say, a penny and a nickel. This helps
us to see that the sample space actually contains four outcomes, not three: HH (heads on both coins),
HT (heads on penny, tails on nickel), T H (tails on penny, heads on nickel) and T T (tails on both).
The event E =“exactly one head” is made up of two outcomes, E = {T H, HT }. Thus, by Rule 2,
Number of Outcomes in Event E 2 1

P(E) = = = .
Number of Outcomes in Sample Space S 4 2
1
The correct answer is P(E) = 2 = 0.5.
Problem 4: Suppose that in Problem 3, instead of flipping two separate fair coins, we flip the same
fair coin twice in succession, and ask (again) for the chance of exactly 1 head. Will we get the same
answer, or a different one?
Solution: We will obtain the same answer. Though it may be helpful at first to think of two different
sorts of coin, like a penny and a nickel, nothing changes in the solution if instead we think about the
first and second flips of the same coin.
Problem 5: We roll a fair die twice in succession. What is the probability of the event E that the total
score (the sum of both rolls) is at most 3?
Solution: The previous problem has taught us that it amounts to the same thing to imagine rolling
two separate dice, say white and black. The sample space then contains 6 × 6 = 36 equally likely
outcomes, running from ( , ) through ( , ). Exactly three of these 36 outcomes have total score
less than or equal to 3, namely
E = {( , ), ( , ), ( , )}.
Thus, by Rule 2,
Number of Outcomes in Event E 3 1

P(E) = = = .
Number of Outcomes in Sample Space S 36 12
1
Thus the answer to our problem is P(E) = 12 ≈ 0.083.
Not all probability models are uniform. For instance, imagine a die in which more of the total
weight is concentrated toward one face.9 Then the symmetry of a fair die is violated—there will be
a tendency for your loaded die to land with the heavier face downward, so that the face opposite the
heavier one will be more likely to come up than the equally likely rule would suggest. Therefore, the
probability model for a loaded die is nonuniform.
Rule 2 on page 282, the equally likely probability rule, applies only to uniform probability models,
where it tells us the probability of each outcome. In the nonuniform case, we need to know specifically
what probability is associated with each individual outcome (that is, we need to know the distribution
of the probability model, as explained after Definition 5). Then to find the probability of an event (a
set of outcomes), we add together the probabilities of all the outcomes that make up the event. This
gives us a generalized version of Rule 2, which is displayed below.
9 It is not hard to create a loaded die, but we won’t give instructions here!
Rule 3: Nonuniform Probability Rule
For a nonuniform probability model (in which the outcomes need not be equally
likely),
Probability of Event E = P(E) = Sum of the Probabilities of Outcomes in E.
When the distribution actually is uniform, the probability of each one of N equally likely outcomes
must be 1/N. In this case, then, Rule 3 gives the same result as Rule 2.
Problem 6: Suppose you have a die that is loaded so that the probability of the outcome 2 is 25%,
whereas each of the other outcomes (1, 3, 4, 5 or 6) has probability 15%. (Notice that the total,
15% + 25% + 15% + 15% + 15% + 15%, equals 100% or 1, as required by the law of total probability,
Rule 1 on page 282.) What is the probability of
1. rolling an even number?
2. rolling a number less than or equal to 2?
3. rolling a number other than 2 or 4?
Solution: For each case, we want to list the outcomes in the event and add their probabilities.
1. We call rolling an even number “Event A.” As a set of outcomes, this is the event A = {2, 4, 6}.
The probability is then
P(A) = 25% + 15% + 15% = 55%.
2. Similarly, rolling a number less than or equal to 2 can be defined as Event B = {1, 2}. The
probability is then
P(B) = 15% + 25% = 40%.
3. Finally, rolling a number other than 2 or 4 can be defined as Event C = {1, 3, 5, 6}. Then
P(C) = 15% + 15% + 15% + 15% = 60%.
Problem 7: In 2009, scientists at the MIT Joint Program on the Science and Policy of Global Change
produced a probability model for the Earth’s overall temperature increase over the present century,
based on the best available models of global climate. They investigated two scenarios: a “no policy”
scenario whereby current trends, including increasing fossil fuel consumption and carbon dioxide
emissions, continue unchecked, and a “policy” scenario that includes aggressive efforts to limit
greenhouse gas emissions. Under the “no policy” scenario, they estimated the following probabilities
for various ranges of the “temperature anomaly” (that is, the increase in global temperatures over
preindustrial levels). In reading this table, bear in mind that many scientists agree that a 2 ◦ C
temperature anomaly represents a dangerous threshold (and some think that even that may be too
high).
Temperature anomaly under 3 ◦ C 3–4 ◦ C 4–5 ◦ C 5–6 ◦ C 6–7 ◦ C 7 ◦ C+

Probability (no policy) 0.01 0.12 0.29 0.34 0.15 0.09
Calculate the probability that the anomaly remains below 5 degrees under the “no policy” scenario.
Figure 12: The MIT “Greenhouse Gamble” spinners. “No policy” at the left; “with policy” at the right. From [219].
Solution: The event “anomaly remains below 5 degrees” is made up of three outcomes: “below
3 degrees,” “between 3 and 4 degrees,” and “between 4 and 5 degrees.” These outcomes have
probabilities 0.01, 0.12, and 0.29. Thus, the overall probability of “anomaly remains below 5 degrees”
is
0.01 + 0.12 + 0.29 = 0.42.
This is less than an even chance!
Example 5: Under the “policy” scenario (whereby the world community works together to limit
greenhouse gas emissions), the probabilities calculated by the MIT researchers look a bit more
favorable:
Temperature anomaly under 2 ◦ C 2–2.5 ◦ C 2.5–3 ◦ C 3 ◦ C+

Probability (with policy) 0.20 0.43 0.27 0.10
There is still only a 20 percent chance of keeping warming below 2 degrees, even under the policy
scenario. But the more extreme outcomes become much more unlikely. Temperatures are predicted to
rise regardless of policy implementation. It is simply a question of how much.
In order to help people visualize these probabilities, the MIT researchers developed a human terms
analogy (compare Section 1.1.5) between the random process represented by the climate system and
an actual game of chance. They represented the two distributions, with and without policy, as arising
from the spins of two “roulette wheels” (Figure 12). Continuing with “business as usual” (without
implementing a greenhouse gas reduction policy) corresponds to a spin of the wheel on the left;
implementing a policy corresponds to a spin of the wheel on the right. Which gamble do you prefer
to take? Visit Greenhouse Gamble [219] and “Spin the Greenhouse Gamble wheels.” You can also
read about the history of the model and its “inaugural spin” by the president of MIT (who fortuitously
landed on a low-warming outcome and “saved the world.”)
Example 6: Let’s check that the numbers in the MIT Greenhouse Gamble example (Problem 7) obey
the law of total probability, Rule 1 on page 282. In the “no policy” case,
0.01 + 0.12 + 0.29 + 0.34 + 0.15 + 0.09 = 1.00;
and in the “with policy” case,
0.20 + 0.43 + 0.27 + 0.10 = 1.00.
In both cases the probabilities of all the individual outcomes add up to 1, as required.
Remark 2: People sometimes also express probability information in terms of odds. By definition,
the odds are expressed as a ratio:
Odds of an Event E
=Probability that E Happens : Probability that E Does Not Happen
=P(E) : (1 − P(E)).
Odds of an Event E =
Probability that E Happens : Probability that E Does Not Happen
= P(E) : (1 − P(E)).
Thus, if an event has a probability of 23 , the odds of the event are 23 : 13 , or 2 : 1. If the event has a
probability of 13 , its odds are 13 : 23 = 1 : 2, or, as it is also expressed, “2 : 1 against.”
We will not use the language of odds elsewhere in this book.
5.2.3 Laws of Probability

Once we know the sample space for a uniform probability model, the probabilities of all events are
assigned automatically by the equally likely probability rule. But for a nonuniform model, this is not
true. We need to know the distribution of the model—the probabilities of individual outcomes—as
a starting point (see Rule 3 on page 285). These probabilities cannot be assigned in a completely
arbitrary way, because they must satisfy the law of total probability (Rule 1 on page 282). In this
section we will look at some other rules that any assignment of probabilities must satisfy. For the
finite probability spaces that are our main examples, these laws are necessary consequences of the
definitions that we’ve presented so far. When mathematicians work with infinitely many possible
outcomes, these laws remain true, but some must be given independent logical status. As we said
before, we won’t let these issues bother us.
Definition 7
The complement of an event E, denoted by E c , is the set of all outcomes in S that are not
in event E. That is to say, E c happens exactly when E does not happen.
For example, suppose we are rolling a standard six-sided die and the event we are interested in is
rolling an even number: E = {2, 4, 6}. The complement of this event, denoted by E c , is rolling an odd
number: E c = {1, 3, 5}.
Problem 8: Suppose you are rolling a twelve-sided (dodecahedral) die, with sides numbered from 1
to 12. What is the sample space? If event A = {2, 5, 6, 7, 8, 9, 10, 11, 12}, what is its complement?
Solution: The sample space of the 12-sided die is
S = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}.
If Event A = {2, 5, 6, 7, 8, 9, 10, 11, 12}, then the complement is
Ac = {1, 3, 4},
or all the numbers in S that are not in A.
Rule 4: Law of Complements
The probability of an event plus the probability of its complement is equal to 1:
P(E) + P(E c ) = 1.
We can sometimes use this law as a shortcut. It allows us to calculate the probability of an event by
considering its complement:
P(E) = 1 − P(E c ).
When we know the probability for an event’s complement, we can determine the probability of that
event. Let’s do an example to clarify this point.
Problem 9: Suppose you are rolling a twelve-sided fair die (where each outcome from 1 to 12 is
equally likely). What is the probability of event A = {2, 5, 6, 7, 8, 9, 10, 11, 12}?
Solution: We could calculate P(A) directly using equation 2 on page 282. A slicker way, however, is
to use the complement rule. Let’s first find the complement of A:
Ac = {1, 3, 4}.
Now we can use equation 2 to calculate the probability of Ac :

3 1
P(Ac ) = = = 0.25.
12 4
From the complement rule, P(A) + P(Ac ) = 1. Thus
P(A) = 1 − P(Ac ) = 1 − 0.25 = 0.75.
Before we discuss the next rules, we need two definitions. Both definitions discuss how we can
consider two events together. The first talks about the union of two events, which we can think of as
all the outcomes captured by both events. The second talks about the intersection of two events, which
we can think of as the outcomes they have in common. Let’s first define them precisely and then look
at some examples.
Definition 8
The union of two events A and B is the set of all the outcomes that are in either A or B
or both. We write the union of A and B as A ∪ B. To put this another way, the event A ∪ B
happens if either A happens, or B happens, or they both happen.
Definition 9
The intersection of two events A and B is the set of all the outcomes that are in both A and
B. We write the intersection of A and B as A ∩ B. To put this another way, the event A ∩ B
happens if both of the events A and B happen.
Here are some sample calculations.

Problem 10: Suppose you are rolling a 6-sided die, so the sample space is S = {1, 2, 3, 4, 5, 6}.
Determine the union and intersection of each of the following pairs of events:
• A = {1, 2, 3, 4}, B = {2, 4, 6}.

• A = {1, 3, 5}, B = {2, 4, 6}.
• A = {1, 2, 5}, B = {2, 5}.
• A = {2}, B = {2}.
Solution: For each bullet, we write the union and intersection below.
• A = {1, 2, 3, 4}, B = {2, 4, 6} :
◦ The union is all outcomes in A and/or B: A ∪ B = {1, 2, 3, 4, 6}.

◦ The intersection is all outcomes in both A and B: A ∩ B = {2, 4}.
• A = {1, 3, 5}, B = {2, 4, 6}:
◦ A ∪ B = {1, 2, 3, 4, 5, 6}.
◦ A ∩ B = 0.
/
(Notice that there is nothing that A and B share, so their intersection is the empty set 0.
/ Events
whose intersection is empty are said to be mutually exclusive or disjoint.)
• A = {1, 2, 5}, B = {2, 5}:
◦ A ∪ B = {1, 2, 5}.
◦ A ∩ B = {2, 5}.
Here, we have a special case in which B is completely included in A (the mathematical language
for this is “B is a subset of A”). In this case, A ∪ B = A and A ∩ B = B.
• A = {2}, B = {2}:
◦ A ∪ B = {2}.
◦ A ∩ B = {2}.
In this example A and B are both the same event, {2}. Their union and intersection, then, are
also both equal to the same event.
Our next rule relates to the probability of the union of two events. Remember, the union A ∪ B is
the event “either A occurs or B occurs or both.” At first, we might suppose that the probability of
this event would be the sum of the probabilities of events A and B. This is correct if A and B are
mutually exclusive (that is, if they have no outcomes in common). But if they are not, simply adding
the probabilities of A and B will “double count” the common outcomes in A ∩ B. To obtain a rule that
is valid for all circumstances, we have to correct for this double counting.
Rule 5: Law of Addition
The probability of the union A ∪ B of two events A and B is given by the equation
P(A ∪ B) = P(A) + P(B) − P(A ∩ B).
• If A and B are mutually exclusive, which is to say that they have no outcomes in
common, that is, A ∩ B = 0,
/ then this rule takes the simpler form
P(A ∪ B) = P(A) + P(B).
Problem 11: Suppose the trial is the roll of a 6-sided fair die. Let Event A be rolling an even number
and Event B be rolling a number greater than or equal to 4. What is the probability of rolling an even
number or rolling a number greater than or equal to 4 (that is, P(A ∪ B))?
Solution: Let us write out the sets first. The sample space is S = {1, 2, 3, 4, 5, 6}. Event A = {2, 4, 6}
and event B = {4, 5, 6}. By looking at events A and B, we see that A ∩ B = {4, 6}.
Therefore,
P(A ∪ B) = P(A) + P(B) − P(A ∩ B)
= 36 + 36 − 26
4 2
= 6 = 3 ≈ 0.67
We can check our answer by looking at A ∪ B = {2, 4, 5, 6}. Since A ∪ B contains 4 out of the 6
outcomes in the sample space, we know that our work is correct. But why does this formula work?
We can draw a Venn diagram (Figure 13) to see why.
S
A B
4
2 5
6
1 3
Figure 13: Venn diagram for Problem 11.
When we calculate P(A), we are considering the probability of getting 2, 4, or 6. When we calculate
the probability of getting P(B), we are considering the probability of 4, 5, or 6. So, adding P(A)+P(B)
is considering 2, 4, 6, 4, 5, or 6. But notice that we are counting 4 and 6 twice! When we subtract
P(A ∩ B), we are taking away the duplicate 4 and 6.
5.2.4 Independence and Repeated Trials

Games of chance can be repeated—an idea that we have already used in Problems 4 and 5 on page 284.
The coin can be flipped again, or the die rolled again; once a card is drawn from the deck it can be
replaced, and the deck reshuffled, and another card drawn. Moreover, the results of the second coin
flip, or dice roll, or card draw are independent of those of the first one, in the sense of the definition
below.
Definition 10
Events A and B are independent if knowing whether or not A occurred does not give any
new information about how likely B is to occur.
For example, suppose I flip a fair coin and it comes up heads. That does not tell me anything about
how likely it is to come up heads the next time I flip it (the coin has no “memory”). The events
A = {head on first trial}, B = {head on second trial}
are independent.
Question 2: In the last example above (the card draw) why were you careful to specify that the first
card drawn should be replaced and the pack reshuffled?
Answer: The simple answer is that if we didn’t replace the first card, we would not be repeating the
exact same game—one would be a draw from a deck of 52 cards and the next from a pack of 51. More
formally, if we made our two draws without replacement, the results would not be independent. If my
first draw gave me the ace of spades, I then would know for certain that my second draw would not. I
gain information about the second draw from the result of the first draw, contrary to the definition of
independence.
Events do not have to belong to separate repetitions to be independent, though.
Example 7: Suppose I roll a fair die, for which each of the outcomes {1, 2, 3, 4, 5, 6} is equally likely.
Define events A and B, relating to the same roll, by saying that event A is “I roll a number less than
or equal to 4,” and event B is “I roll an even number.” The probability of event B is 36 = 0.5. If I
know that event A occurred, and want to work out the probability of B, I now have four equally likely
outcomes instead of six, namely {1, 2, 3, 4}, and two of them are even. With this extra knowledge I
find the probability of B to be 24 = 0.5 again. This is the same as before, which means that A and B are
independent.
Problem 12: Consider rolling a fair die, as above, but now let A be “I roll a number less than or equal
to 3” and let B be “I roll an even number.” Are these events independent?
Solution: These events A and B are not independent. If I know that A occurred, then I have three
equally likely outcomes, {1, 2, 3}, of which only one is even. So I now calculate the probability of B
to be 13 ≈ 0.33. This is different from my initial value P(B) = 0.5 in the absence of any information
about A, and this difference tells me that A and B are not independent.
The law of multiplication tells us about the probability that two independent events will both occur.
Rule 6: Law of Multiplication
If event A and event B are independent, then the probability of their intersection,
A ∩ B, is equal to the product of their individual probabilities, that is,
P(A ∩ B) = P(A) × P(B).
(Remember that the intersection A ∩ B is the event “A and B both occur.”)
Example 8: Let’s apply the law of multiplication to calculate P(A ∩ B) in Example 7 on the previous
page. The probability of event A, “My roll is ≤ 4,” is 23 , and the probability of event B, “My roll is
even,” is 12 . Since the events are independent, the law of multiplication gives
2 1 1
P(A ∩ B) = × = .
3 2 3
To check that this answer is correct, we observe that the event A ∩ B, “My roll is even and ≤ 4,”
contains the two outcomes 2 and 4 only. So by the equally likely rule,
2 1
P(A ∩ B) = = ,
6 3
in agreement with our calculation above.
Problem 13: A fair die is rolled twice. Calculate the probability that the first roll is ≤ 4 and the second
roll is ≥ 4.
Solution: That word “and” clues us in that we are dealing with the intersection of two events: Event A,
“the first roll is ≤ 4,” and Event B, “the second roll is ≥ 4.” Because they refer to separate repetitions
of a game of chance, Events A and B are independent.
Event A covers the four outcomes {1, 2, 3, 4} out of 6, and Event B covers the 3 outcomes {4, 5, 6}
out of 6. So by the equally likely rule (Rule 2),
4 3
P(A) = , P(B) = .
6 6
By the multiplication rule,
4 3 12 1
P(A ∩ B) = × = = .
6 6 36 3
This is the answer that we want.
It is a special feature of games of chance that they can be repeated many times, yielding independent
results. In real-life situations, one has to take great care with assumptions of independence. It is not
enough to know that one event cannot directly influence the other. Consider the following example.
Problem 14: Eddie and Freddie are adult brothers. Let E be the event “Eddie is over 6 feet tall” and
let F be the event “Freddie is over 6 feet tall.” Are E and F independent?
Solution: These are not independent events. If Eddie is exceptionally tall, it is somewhat likely that
he inherited this trait from one or both of his parents. And if one or both parents are exceptionally tall,
it is somewhat likely that they passed this trait on to Freddie also. Thus Freddie is somewhat more
likely than a randomly selected adult to be exceptionally tall also. In this way, event E (“Eddie is over
6 feet tall”) gives us information about the probability of event F (“Freddie is over 6 feet tall”).
Remark 3: The principle underlying the previous example could be expressed as follows: “if two
events A and B are both influenced by a third event C, then A and B may not be independent, even
though they don’t directly influence each other.”
Here is another similar example.
Problem 15: A mortgage broker arranges two loans on the same day: one to the Bernheimer family
in Idaho and one to the Qian family in New York. Based on past experience, the broker estimates that
there is a 4% chance that the Bernheimer loan will go into default and that there is a 2% chance that
the Qian loan will go into default. The broker calculates the probability that both loans will default,
based on the assumption that the potential defaults are independent events. What answer does the
broker obtain? Comment on the assumption of independence.
Solution: Let B be the event “the Bernheimer loan defaults” and let Q be the event “the Qian loan
defaults.” We are given that P(B) = 0.04 and P(Q) = 0.02 (these are just the percentages in the
question re-expressed as decimals). Thus, assuming independence, the multiplication rule gives the
probability of both loans defaulting as
P(B ∩ Q) = P(B) × P(Q) = 0.04 × 0.02 = 0.0008,
or 0.08% if we want to express it as a percentage, as this broker seems to like to do.

What about the assumption of independence? At first glance, it seems clear that B and Q are
obviously independent: there is no way that the Bernheimers in Idaho can be influencing the Qians
in New York, whom they don’t know and have never met. But independence is not just about the
presence or absence of direct causal influence. Think about the possibility of a nationwide economic
crisis. Such a crisis would cause many mortgages to fail, all across the nation; and conversely, if
brokers reported a lot of mortgage failures, that would be evidence for a crisis. But that means that
even one mortgage failure is a tiny piece of evidence for a crisis, and is therefore a tiny piece of
evidence for the likelihood of other mortgage failures also. The events are therefore, strictly speaking,
not independent.
Of course with just two mortgages like the Bernheimer’s and the Qian’s, the deviation from
independence is very, very slight. But in 2008, thousands upon thousands of mortgages were bundled
together into securities that were then resold on a secondary market. The valuation of these securities
was based on a rating of the intrinsic risks of the underlying bundle of mortgages, and that rating
was based on assumptions of independence.10 These assumptions turned out to be grossly flawed, for
reasons very like those in this example. In a way, the whole gigantic 2008 financial crisis is a painful
reminder to take seriously the fact that mathematical models (like the multiplication rule) come with
conditions attached (like independence). If you don’t check that the conditions apply, don’t expect
your answers to be worth much.
The Law of Large Numbers

As we mentioned above, there are some random processes that it is difficult to imagine repeating. I
may say today that “there is a 40 percent chance of rain tomorrow,” but once tomorrow has arrived
I cannot rewind to today and repeat the process! However, many random processes are repeatable.
These include games of chance (by definition), but also sampling processes in science such as
Example 3 on page 279 (gathering separate samples of lake water in order to analyze them for arsenic).
Here, the measurements that the scientist gathers from successive samples can be treated as successive
independent outcomes from a process analogous to a game of chance—a process whose distribution
is centered at the correct value of the quantity being measured (the arsenic concentration), and has
10 The story is told in Cathy O’Neil’s book [256], which we already introduced on page 276.
spread determined by the precision of the measuring equipment and the intrinsic variability of arsenic
concentrations in different parts of the lake.
Imagine we have a random process that is repeatable in this way, like measuring the arsenic
concentration in a sample of lake water. We may refer to each individual instance of the random
process (each individual measurement, in our example) as a trial. Suppose we carry out a large
number, N, of trials; then we will obtain a list of N independent outcomes, and this list of outcomes
will be a data set (Definition 1 on page 261).
Let E be an event arising from the outcomes—for example, E could be the event “the measured
arsenic concentration is greater than 3 µg/ L.” Some of the outcomes will belong to E and some of
them will not. The (absolute) frequency of E in the data set11 is the number of outcomes in the list
that belong to E—in other words, the number of times that E happens. It is a whole number between
0 and N. The relative frequency is obtained by dividing the absolute frequency by N, so that we now
get a number between 0 and 1.
Definition 11
The relative frequency of an event E in a given set of trials is the number of times an
outcome in event E happens, divided by the total number of trials.
Number of Trials in which E Happens
Relative Frequency of Event E = .
Total Number of Trials
For example, suppose that our random process is to toss a fair coin, and that we are interested in
how often the outcome is Heads. Then our sample space is S = {Heads, Tails}. Let’s define event E
to be {Heads}.
You could pull out a penny from your pocket and start flipping. But we can also use a coin toss
simulator, for instance the one available at random.org [144], which will flip virtual coins, roll virtual
dice, and simulate numerous other random processes. Let’s try 10 trials. When we ran our simulation,
we obtained the outcomes shown in Figure 14:
Figure 14: Result of flipping ten virtual coins at random.org [144].
So the relative frequency of heads, in these 10 trials, is 0.6 (or 60%, if you prefer to express it as a
percentage):
6 (number of Heads)
Relative Frequency of Heads = = 0.6.
10 (number of Trials)
We suggest that you try this out also. Did your experiment produce the same relative frequency as
ours? Maybe, but more likely not. Different experiments will produce different results. However, it
11 Definition 8 on page 268 (frequencies for histograms) is an example of this. The relevant event E is “the variable takes
on a value in a certain class interval.”
is an observed fact that as we increase the number of trials, the relative frequency of a given event
tends to settle down and approach the probability P(E) of E. For example, we ran another 200 trials
using [144]. This time, we got 96 heads. Therefore, the relative frequency is
96 (number of Heads)
Relative Frequency of Heads = = 0.48.
200 (number of Trials)
As you can see, it looks as though the relative frequencies are approaching 0.5 (the probability of
Heads for a fair coin). Let’s try to understand this.
Suppose we flip a fair coin 20 times and ask, as in Problem 4 on page 284, what are the probabilities
of various numbers of heads. There are now 220 (approximately a million) equally likely possible
outcomes, each of which is a sequence of heads and tails like HT HHT T T T HT HHHT T HT HT . If
we followed the process we used to solve Problem 4 we would need to list all of these equally likely
outcomes and count how many contain 0 heads, how many contain 1 head, and so on. This would
undoubtedly be a tedious task if we had to do it by hand, but fortunately mathematics12 can help.
The left-hand histogram in Figure 15 shows the results of such a computer calculation, presented as a
histogram. We can see that the outcomes tend to cluster around the “expected” value, 10. In fact, the
probability of obtaining between 8 and 12 heads (that is, a relative frequency between 0.4 and 0.6) is
nearly 75 percent.
0.06
0.15
0.04
0.10
0.02
0.05
0.00 0.00
0 5 10 15 20 0 50 100 150 200

(a) Probabilities of obtaining various numbers of heads in 20 (b) Probabilities of obtaining various numbers of heads in 200
flips of a fair coin. The probability of obtaining between 8 flips of a fair coin. The probability of obtaining between 80
and 12 heads is ≈ 0.737. and 120 heads is ≈ 0.996.
Figure 15: Probability distributions for obtaining heads in (a) 20 and (b) 200 flips of a fair coin.
If we increase the number of trials more, we find that this “clustering” will get stronger. The right-
hand histogram in Figure 15 is the result of the same calculation, but carried out for 200 trials instead
of 20. Now, the probability of getting the relative frequency in the same range (between 0.4 and 0.6,
that is, between 80 and 120 heads out of 200) has risen to over 99 percent.
These are examples of the law of large numbers.
12 The specific mathematical objects we need are called the binomial coefficients. We will not study these in detail in this
book.
Rule 7: Law of Large Numbers
Imagine a random process (such as a game of chance) that can be repeated

independently many times. Let E be an event associated with this random process
and suppose that P(E) = p. Then the relative frequency of E after sufficiently many
trials is almost certain to approach p.
Remark 4: The words “sufficiently many” and “almost certain” are deliberately left quite vague.
Many mathematicians, beginning with Jacob Bernoulli (1713), have worked on making them more
precise. Here is one example of what they have come up with. Let ε be any small positive number.
For each N (number of trials), consider
EN = “the relative frequency of event E, in N independent trials, will be between p − ε and p + ε.”
EN is itself an event, and one can ask about its probability. The statement of the law of large numbers
is that however small ε is, the probability of EN approaches 1 as N becomes very large.
This is a complicated statement (and for the purposes of this book, the simpler one given in Rule 7 is
adequate). However, we saw an example of how the complicated statement works in the coin-tossing
example above. Here the probability of “Heads” is p = 0.5, and we were considering the small “error”
ε = 0.1. With two trials, the chance that the relative frequency is in the range 0.4 = p−ε to 0.6 = p+ε
is13 50 percent; with 20 trials, about 73 percent; with 200 trials, over 99 percent. It is clear that these
probabilities are approaching 1 as N becomes larger.
We have used the word “distribution” in two ways in this chapter: in Definition 2 on page 262 to
refer to the pattern of a data set, and in Definition 5 on page 281 to refer to the list of probabilities of
the outcomes in a probability model. The law of large numbers ties these definitions together. If a data
set is obtained from many independent trials using a specific probability model, then the distribution
of the data will reflect the distribution of the underlying probabilities.
Remark 5: The law of large numbers does not say that if the relative frequency (of heads, say) is
unusually large in the first hundred or so trials, then it will have to be unusually small in the next
hundred or so trials so as to “balance out.” It claims only that in the very long term—perhaps after
tens of thousands of trials—whatever unusual behavior happened in the first hundred trials will almost
certainly be swamped by the more frequent occurrence of “average” behavior.
Example 9: Consider flooding, like that caused by Hurricane Harvey in 2017. The hurricane was
described as causing a “five hundred year flood” in many areas. What does this mean? The National
Weather Service maintains a network of flood gauges across the U.S., and the “five hundred year
flood” level for a particular gauge is the level ` such that
1
P(gauge level exceeds ` in any given year) = .
500
Thus the law of large numbers tells us that in the (very) long term, about one in every five hundred
years will see a gauge level exceeding `. It does not say, however, that if you have endured one “five
hundred year flood” you are now safe for the following 499 years. Sadly, the random processes that
generate extreme weather, like successive flips of a coin, have no “memory.”
Section 5.3.5 sketches how these “five hundred year” levels are calculated from observed data.
13 See Problem 3 on page 283.
Think about it. . .
In a game of roulette at the Monte Carlo casino on August 18, 1913, the roulette ball
had fallen into the black twenty times in succession (“red” and “black” are equally likely
outcomes in roulette). Many gamblers bet heavily that the next spins of the wheel would
come up red, reasoning that according to the law of large numbers, the wheel would have
to “restore the balance” between black and red. What do you think of their reasoning?
(In fact, gamblers who reasoned in this way lost millions of francs. The ball fell into the
black another six times before the record-breaking streak came to an end.)
Remark 6: You may have noticed something else about the histograms in Figure 15 on page 295.
Not only do they concentrate about the “expected” value, as the law of large numbers predicts—their
shape is also familiar. In fact, they seem to approximate the bell curve, or “normal distribution,” that
we discussed at the end of the preceding unit—see Section 5.1.3. This is not a coincidence. It is an
expression of the central limit theorem, which says that we may always expect the relative frequencies
in repeated trials of this sort to approach a normal distribution. As mentioned in Section 5.1.3, this
partly explains why the normal distribution is so common in actual data.
5.2.5 Understanding Uncertainty

What is probability? We defined the probability P(E), in Definition 4 on page 281, as “a real number
between 0 and 1 that measures how likely E is to happen.” If E belongs to a random process that
allows for many independent repetitions (such as a game of chance, our fundamental example), then
the law of large numbers allows us to tighten this up: the probability of E is the “limit” that the relative
frequency of E will approach if we make a very large number of trials. This is sometimes called the
frequentist understanding of probability.
However, not every random process does allow for such repetition. As already mentioned, we
understand a statement such as “there is a 40 percent chance of rain tomorrow” even though there
is no way we can make many independent trials of the way tomorrow’s weather evolves from today’s.
Examples like this are better understood as reflecting “degrees of belief” or assurance that we should
give to a statement: the 40 percent chance of rain means that we believe (or should believe) that rain is
quite possible, but that it is a little more likely not to rain than it is to rain. This kind of interpretation
of probability is called a subjectivist understanding.
In many cases both interpretations make sense: we can say that the probability that a fair coin lands
Heads is 12 either to mean that we rationally believe that Heads and Tails are equally likely or to
mean that after many flips, the coin will land Heads about half the time. Notice that the frequentist
and subjectivist interpretations give the same answer, 12 . It can be shown mathematically that under
certain assumptions, if both ideas apply, then they are bound to agree.
Look back at Section 2.1.4, where we described the scientific method. We explained there
that science works by continually “stress testing” its models—trying to find evidence that would
disconfirm them, or prove them wrong. The more such “stress tests” a model passes, the greater
assurance we can have about it. As you would expect from the discussion above, we can describe
our level of assurance by using the language of probability theory (especially in its subjectivist
interpretation).
Example 10: How many African elephants are alive in the wild today? This is not a population that
one can count directly: ecologists must use various sampling methods to estimate it. Consider two
possible ways of expressing the results of one such sampling:
(a) “There are 511, 752 African elephants alive in the wild today.”
(b) “There is a 95% probability that the number of African elephants alive in the wild today is
between 450, 000 and 650, 000.”
Hopefully, you recognize from Section 1.3.1 that (a) makes a measurement claim with inappropriate
precision. The alternative (b) is an example of how probability language can be used in specifying
precision. It gives a range of values (one of the ways that we learned to express the precision of a
measurement claim), together with a level of assurance that we can attach to the claim that the true
number of elephants lies in the specified range.
Because scientific models are continually being improved, we can say that uncertainty is a key
element of the scientific method. The point of science is to reduce uncertainty (by gathering more
or better data, or by more accurate or insightful modeling, or by better understanding fundamental
processes)—but in order even to talk about reducing uncertainty, you have to have a language that
allows uncertainty to exist. That is why a statement with no uncertainty—that is, one that is not open
to any challenge whatever—is not a scientific statement.
This is a perfectly reasonable way to use the word “uncertainty,” and is well understood among
scientists. A problem arises, though, because the general public consistently understands things in a
different way [55]. To many members of the public, “uncertain” often sounds like “unsure,” so the
scientist’s statement “The probability is 95%” (intended to convey a high level of certainty) provokes
the response “So, you’re not actually sure?” (After all, every dollar-store marketer knows that there
is a huge psychological difference between 95 cents and one dollar.) The situation is not helped by
paid shills who capitalize on this confusion (“Doubt is our product,” as was stated in a memo by the
tobacco company Brown and Williamson as early as 1969) to buy time for polluting industries to stay
in operation without the “burden” of appropriate regulation [257].
This is a communication gap, and it’s hard to know the best way to address it. One proposal
that could perhaps be helpful [346] is to match up different probability levels with different kinds
of language used in legal cases, so that for instance, the standard of proof in a criminal case
(“beyond reasonable doubt”) corresponds to a probability of at least 0.99, and that in a civil case
(“preponderance of the evidence”) corresponds to a probability of at least 0.5. The Intergovernmental
Panel on Climate Change has adopted a version of this idea also, so that, for instance, an IPCC report
will say that a conclusion is “virtually certain” if its probability is at least 0.99, “very likely” if its
probability is at least 0.9, and so on. In [346] you will find a detailed table listing a proposed 11-
point scale of certainty, with corresponding probabilities, informal descriptions of scientific and legal
certainty, and IPCC language (where applicable) for each point on the scale.
Think about it. . .
What do you think about the proposal to replace specific numerical probabilities with
quasilegal language that we have cited from [346]? Do you believe that it would improve
communication between scientists and the public, or not? Why?
It’s worth noting that public estimates of the comparative precision of different sciences often vary
significantly from reality. For instance, the public judges forensic science [55] to yield the most
precise scientific results possible, despite the fact that the real-life accuracy of forensic science is
comparable to that of psychology, judged by the public—in those same surveys—as one of the least
precise sciences. Because this belief is thought to originate from television portrayals of the wonders
of forensic science in crime scene investigations, it is sometimes called the CSI effect.
Summary of Ideas: Probabilities and Predictions
• A random process is an activity, such as a game of chance, whose result cannot

be determined with certainty. The result of a single instance of a random process is
called an outcome.
• The sample space is an exhaustive list of all the possible outcomes.
• An event is any set of outcomes coming from the sample space.
• A probability model is a mathematical representation of a random process. It is
defined by its sample space (that is, its list of possible outcomes), together with the
probability of each outcome.
• A probability model is uniform if each outcome has the same probability, which is
1
then equal to .
Number of Outcomes
• The probability of an event is the sum of the probabilities of the individual outcomes
that comprise the event.
• Events are independent if knowledge about the occurrence of one does not influence
our understanding of the probability of the other.
• The law of large numbers states that after many independent repetitions of some
random process, the relative frequency of an event E (that is, the fraction of the trials
in which E occurs) will almost certainly approach the probability of E.
• Probabilities can be used to model levels of assurance and therefore to quantify the
uncertainty that is intrinsic to the scientific method.
5.3 Expectations and Payoffs
Objectives
I can define a random variable in a probability model.

I can calculate the expected value of a random variable.
I can compare policy options in terms of their expected utility.
I understand the concept of risk aversion.
I understand the importance of extreme values when considering policy options.
In this section, we will ask how we use probabilistic information—

uncertain information—to make decisions. These could be range from indi-
vidual, personal decisions like “should I buy the lottery ticket?” or “should I
take out an insurance policy?” all the way up to huge policy decisions, like
“should my company invest in this energy-saving technology?” or “should we
institute a carbon tax to reduce the impacts of climate change?” What these
decisions have in common is a range of outcomes about which we have only
limited knowledge—knowledge of their probabilities—together with some
understanding of the payoff of each outcome, the amount we expect to gain
or lose.
In these circumstances, mathematical economics provides a model called
expected utility theory for decision-making.14 We will learn how this kind of
Figure 16: A lottery ticket (it lost). analysis works and how it can be applied. We will also talk about its limitations.
Does it, in fact, accurately model our attitudes to risky decisions—either
what they should be, or what they actually are? Especially, how relevant is
probability theory, with its emphasis on what is to be expected in the long term
after multiple repetitions, to one-time, high-consequence events like climate
change?
5.3.1 Random Variables

How much is a lottery ticket worth?
It depends which ticket it is, of course. If you scoop the PowerBall grand prize today, your ticket
may be worth 40 million dollars. Or it may be worth a million, ten thousand, a hundred, or a smaller
prize. Most likely, it is worth nothing at all. But you don’t know (until after the drawing). The value of
your ticket depends on a random process. We call a numerical variable that depends on the outcome
of a random process like this a random variable.
Recall that the sample space for a random process is the list of all the possible outcomes. A random
variable, as we just explained, gives a numerical value that depends on the outcome of the random
process. So a random variable will be completely defined if we list its value for each possible outcome,
that is, for each element of the sample space. (Reminder: In this chapter, we’re formally considering
only random processes that have finite sample spaces. That implies that any random variable will have
only a finite number of values.)
14 Cost-benefit analysis is a version of expected utility theory.
5.3. EXPECTATIONS AND PAYOFFS 301
Example 1: The payoff from a lottery ticket is a rather complicated random variable. Let’s illustrate
the idea by drawing a diagram for a simpler random variable—a bet on a coin flip. Suppose that a
coin is flipped and I stand to win W dollars, where W is $10 if the coin comes up tails, but nothing if
the coin comes up heads. Then W is a random variable based on the coin flip (a random process). We
could illustrate it as shown in Figure 17.
W
S
heads 0
tails 10
Figure 17: Illustration of the random variable W , the payoff for a certain bet on a coin flip.
As you can see from the picture, the random variable W can be envisaged as a rule that assigns a
real number to each of the elements in the sample space S. Mathematicians call this kind of rule a
“function,” and this idea is the basis of the formal definition of random variable.
Definition 1
Consider a random process with sample space S. A random variable, X, associated to this
random process is a function from S to the set R of real numbers. Put differently, a random
variable is a numerical value that depends on the outcome of the random process.
Example 2: Here’s another example, a random variable for which the underlying random process is
the roll of a die. We decide to name this random variable X.
X
S
0
10
10
−20
10
−20
Figure 18: A random variable X based on the roll of a die.
If X dollars is the payoff for a bet, then I pay you $10 if the number rolled is 2, 3 or 5, and you pay
me $20 if the number rolled is 4 or 6. If the number rolled is 1, there is no payment either way. Would
you take this bet? (Think about it. We’ll talk about the answer later.)
We can use the values of a random variable to specify events (collections of outcomes; see
Definition 3 on page 280). For example, if X is the random variable shown above, then “X = 10”
specifies the event { , , }.
Problem 1: In Example 2 on the previous page, what is the probability of the event X = 10?
Solution: The event X = 10 comprises three { , , } of the six equally likely outcomes { , , , , , }
of the dice roll. So, by the equally likely probability rule (Rule 2 on page 282), we may write
3 1
P(X = 10) = 6 = 2 = 0.5.
The probability that X = 10 is one-half (or “50 percent,” if you like to express it that way).
Problem 2: Suppose a fair coin is tossed 2 times and we wish to keep track of how many heads occur.
What might be an appropriate random variable in this situation? Draw a diagram (like those we drew
before) showing how the values of the random variable relate to the sample space.
Solution: As we saw in our discussion of Problem 4 on page 284, the sample space for this problem
contains four equally likely outcomes: HH, HT , T H, and T T . The random variable that we are
interested in (let’s call it Y ) is the number of heads we obtain from those 2 flips. It can take only
the values 0, 1, or 2. The diagram in Figure 19 shows the random variable as a function matching each
outcome with a number.
Y
HH
2
HT
1
TH
0
TT
Figure 19: Illustrating the random variable that gives the total number of heads in two coin flips.
We can rephrase some of the questions from Problem 3 on page 283 in terms of this random
variable. For example, what is P(Y = 2)? What is P(Y = 1)? Our solution to that problem tells us
that the answers are 14 and 12 respectively. You can see this directly from the diagram (Figure 19):
Y = 2 includes just one of the four equally likely outcomes, and Y = 1 includes two of the four
equally likely outcomes.
5.3.2 Expected Values

You probably figured that the bet in Example 2 on the previous page would not be a wise one for you
to take. The reason is that in the long run, each outcome is equally likely. In 600 dice rolls, let’s say,
we might expect that each face would come up about 100 times. If each face came up exactly 100
times, we could calculate your gains and losses:
Outcome Number of times Gain or loss each time Total gain or loss
100 $0 $0
100 $10 $1,000
100 $10 $1,000
100 –$20 –$2,000
100 –$20 –$2,000
100 $10 $1,000
If we total the right-hand column we see an overall loss of 1,000 dollars in playing the game 600
times. True, this assumes that the six outcomes all occur exactly in proportion to their probabilities.
But the law of large numbers tells us that in many trials, we should expect the outcomes to occur at
least approximately in proportion to their probabilities, Thus, in the long run, you should expect to
steadily lose money if you keep playing this game.
This idea of “what we should expect in the long run” gives its name to the key concept of this
section, the expected value (also known simply as the expectation) of a random variable.
Definition 2
The expected value or expectation of a random variable X, written E(X), is found using
the following process:
(a) List all the values that the random variable can take.
(b) Multiply each value by the probability that the variable takes that value. The result of
this multiplication is the contribution to the expectation coming from that particular
value of X.
(c) Add up all the contributions (results of the multiplications) coming from step (b). The
sum that you obtain is the expectation E(X).
Putting it in the form of an equation,
E(X) = a × P(X = a) + b × P(X = b) + c × P(X = c) + · · · ,

where a, b, c, . . . are the possible values of X.
Problem 3: A fair die is rolled (once). Calculate the expected value of the score.
Solution: There are six possible outcomes, { , , , , , }, each of which has probability 16 . The
score is a random variable (say Z) that assigns to each dice roll its numerical value (so that Z( ) =
1, Z( ) = 2, and so on). Thus there are six values of Z, from 1 to 6, and their contributions are
1 1
respectively 1 × , 2 × , and so on. The total of the contributions, which is the expected value of Z,
6 6
is
1 2 3 4 5 6 1 + 2 + 3 + 4 + 5 + 6 15
E(Z) = + + + + + = = = 3.5.
6 6 6 6 6 6 6 6
Problem 4: Calculate the expected value of the random variable X in the dice game of Example 2 on
page 301.
Solution: As the definition of expected value tells us, we make a table listing all the values that X
can take and the corresponding probabilities. To make it clear how we calculated the probabilities,
we have also listed the outcomes (of the random process, that is, the roll of the die) that make up the
various events (values of the random variable) that we consider. The expected value is the sum of the
contributions in the right-hand column.
Value of X Event Probability Contribution

0 { } 0.17 $0.00
10 { , , } 0.50 $5.00
–20 { , } 0.33 −$6.66
The expected value is the sum of the contributions, which is −$1.66.

What does the “expected value” of X tell us? It is certainly not the value we “expect” on any
individual trial—there is no outcome for which X = −1.66 (in the same way that there is no U.S.
family with the “average” number of children, which is not a whole number). Rather, it means that
the average of the X-values in a large number of independent trials is expected to be close to E(S).
Equivalently, in a large number N of independent trials, the total of the X-values is expected to be
close to N × E(X). For example, if you play the game 600 times in succession you should expect that
your net winnings will be approximately
600 × E(X) = 600 × (−$1.66) ≈ −$1, 000.
This is the same result that we calculated at the beginning of this section, a loss of a thousand dollars.
Rule 1: Law of Large Numbers for Random Variables
Imagine a random process (such as a game of chance) that can be repeated many
times independently. Let X be a random variable associated with this random process
and suppose that E(X) = m. Then the average value of X after sufficiently many trials
is almost certain to approach m.
Compare this with the law of large numbers for probabilities, which we formulated in Rule 7 on
page 296. As in that case, there is a mathematician’s version of this law too, avoiding the “sufficiently
many” and “almost certain” language in terms of limits and ε’s. But we will not need it.
Problem 5: I invite you to play the following game: First, you pay me a dollar. Then, we flip a fair
coin. If the coin comes up heads on the first flip, I pay you 50 cents and the game is over. If not, we
flip again. If the coin comes up heads on the second flip, I pay you $1.25 and the game is over. If not,
we flip a third time. On this last flip, if the coin comes up heads I pay you 3 dollars, and if it comes
up tails you get nothing. Do you accept my invitation?
Solution: Let’s calculate your expected gain or loss from this game. Define a random variable W (for
“winnings”) to be the net amount, in dollars, that you win (or lose if negative). There are four possible
values for W :
• W = −$1.00 + $0.50 = −$0.50 if the first flip comes up heads; the probability of this event is
1
2.
• W = −$1.00 + $1.25 = +$0.25 if the first flip comes up tails and the second comes up heads;
the probability of this event is 12 × 12 = 14 .
• W = −$1.00 + $3.00 = +$2.00 if the first and second flips come up tails and the third comes
up heads; the probability of this event is 12 × 12 × 12 = 18 .
• W = −$1.00 if the first, second, and third flips come up tails; the probability of this event is
1 1 1 1
2 × 2 × 2 = 8.
We make a table of possibilities for W
Value of W Probability Contribution

−$0.50 0.5 −$0.2500
+$0.25 0.25 +$0.0625
+$2.00 0.125 +$0.2500
−$1.00 0.125 −$0.1250
The expected value is E(W ) = −$0.2500 + $0.0625 + $0.2500 − $0.1250 = −$0.0625. So you stand
to lose, on average, a little over 6 cents each time you play this game. Unless each play brings you at
least 7 cents worth of thrills, you probably should not accept my invitation.
Critical Thinking
The reference to “7 cents worth of thrills” may seem frivolous (and indeed it is). But it is
also a reminder that decision-making processes need to take into account all the factors
relevant to a situation, some of which may not be so easily quantified as the expected
financial payoff. In what ways might focusing attention on expected financial payoffs bias
the outcome of a decision-making process? What factors might that focus tend to neglect,
or at least have difficulty taking into account?
Problem 6: Your car is worth $7, 500. You pay $40 per month for a one-year insurance policy that
will reimburse the value of the car if it is totaled (to keep things simple, let’s assume that for a lesser
accident the policy will pay nothing). The probability that you will total your car in any given year is
2 percent. Find your expected gain (or loss) from buying the policy. Comment.
Solution: The total annual cost of the policy is 12 × $40 = $480. Let
X be the random variable expressing your overall gain or loss for the
year. There are two possible values for X: either you do not total your
car (probability 98 percent), in which case X = −$480, or you do
(probability 2 percent) in which case X = $7, 500 − $480 = $7, 020.
We can compute the expectation directly:
E(X) = 0.98 × (−$480) + 0.02 × $7, 020 ≈ −$470 + $140 = −$330.
It appears that buying the insurance policy is a bad deal for you (after Figure 20: Total loss.
all, your expected loss is $330). But “expected (financial) loss” may not
capture all the information that is relevant to your decision-making (we
discussed some ways in which that could happen on page 274). Perhaps the monetary values do not
fully capture your profile of risks and rewards. Moreover, this is not a “trial” that you will get to repeat
many times—the law of large numbers is therefore arguably less relevant to your situation. Perhaps
buying the insurance is worth your while despite the “expected” financial loss.
Things look different, of course, from the perspective of the insurance company. Now we are in the
world of the law of large numbers: the company probably insures thousands of drivers just like you.
From past experience, it has a good idea of what fraction (2 percent) will total their cars in any given
year. Expected value is a good guide for the insurance company, and it sets the policy cost so that its
expected profit is positive.
An example like this shows that expected monetary values are not always the best guides to
decision-making, particularly in a situation in which we are dealing with small risks of potentially
catastrophic events that will not often be repeated. All three of the phrases in italics apply when we
consider large-scale environmental risks such as climate change.
Problem 7: Two dice are rolled, and a random variable M is defined to be the product of the numbers
that they display. (For instance, if the roll is { , }, the value of M is 15.) Find the expectation of M.
Before we start working, let’s mention that a quicker method to solve this problem will be explained
later (Example 3 on the opposite page). Right now the point is to show how we can calculate directly
from the definitions even in a complicated-looking problem like this one.
Value of M Probability Contribution

Solution: There are 36 = 6 × 6 possible outcomes to this trial,
1 1/36 0.0278
2 2/36 0.1111
each of which is a pair of dice values like { , }. Since the
3 2/36 0.1667 dice rolls are independent, each of these 36 possible outcomes
4 3/36 0.3333 has probability 16 × 16 = 36
1
. Let’s list all the possible values
5 2/36 0.2778 of M in a chart, as shown below. The top row represents the
6 4/36 0.6667 possible outcomes for the roll of the first (white) die, and the
8 2/36 0.4444 first column represents the possible outcomes for the roll of the
9 1/36 0.2500
second (black) die. For example, if I roll a on my first die and
10 2/36 0.5556
12 4/36 1.3333 a on my second, then my roll is represented by the red 6 cell
15 2/36 0.8333 in the chart (showing an outcome with M = 6).
16 1/36 0.4444
18 2/36 1.0000
20 2/36 1.1111 1 2 3 4 5 6
24 2/36 1.3333 2 4 6 8 10 12
25 1/36 0.6944 3 6 9 12 15 18
30 2/36 1.6667
36 1/36 1.0000
4 8 12 16 20 24
5 10 15 20 25 30
Table 7: Calculating expectation in the two-dice multi- 6 12 18 24 30 36
plication problem.
Now we need to make a table (Table 7) of all the possible
M-values, their probabilities, and their contributions to the
expectation. Each cell of the 6 × 6 chart above has the same
1
probability, 36 , so the probability of each M-value is just the
number of times that value appears in the chart, divided by 36.
That is how the probabilities in Table 7 are obtained. The “contribution” of each M-value, as explained
in Definition 2 on page 303, is the product of that M-value and its probability. Finally, the expected
value of M is obtained by taking the sum of all the contributions. The result is 12.25.
Suppose that X and Y are two random variables. We can combine them in various mathematical
ways, and the result is another random variable. For example, we could add them, or multiply them. In
Problem 7, if we suppose that X and Y are the outcomes of the first and second dice rolls respectively,
the problem asks us to find E(X ×Y ).
You might wonder whether there is a shortcut here. If the random variable Z is gotten by combining
X and Y in some arithmetic way (recall that random variables are real numbers by definition, so
they can be added and multiplied and so on), can you find the expected value of Z by combining the
expected values of X and Y in the same way? Indeed there are rules that allow you to do this, but some
of them come with conditions. Let’s take a look at them.
Rule 2: Linearity Rules for Random Variables
Let X and Y be random variables and let c be a constant. Then
E(X +Y ) = E(X) + E(Y ), E(c × X) = c × E(X).
For the second rule we need to know what it is for two random variables to be independent. You
can probably guess what the definition must be by thinking about the definition of independent events
(Definition 10 on page 291).
Definition 3
Random variables X and Y are independent if knowing the value of X does not give any
new information about how likely any given value of Y is to occur.
Rule 3: Product Rule for Independent Random Variables
Let X and Y be independent random variables. Then
E(X ×Y ) = E(X) × E(Y ).
Example 3: Think about Problem 7 again. Let X be the outcome of the first roll and Y the outcome of
the second roll. These are independent random variables. We know that
1+2+3+4+5+6
E(X) = E(Y ) = = 3.5,
6
by our solution to Problem 3. Thus the answer to the problem, E(X × Y ), equals E(X) × E(Y ) =
3.5 × 3.5 = 12.25. A lot less painful than our calculation by hand when we first solved Problem 7!
The product rule is definitely not true without the independence assumption. See Exercise 42 on
page 340 at the end of the chapter for an example.
5.3.3 Expected Utility as a Guide to Decisions

Our discussions about bets and insurance purchases suggest a way to make decisions in the presence
of uncertainty. This is the expected utility rule—not a mathematical rule like the linearity and product
rules that we just discussed, but a proposed guideline for decision-making. It’s easy to state:
Rule 4: Expected Utility Rule
When making a decision between uncertain alternatives, choose the strategy that has
the greatest expected payoff.
We use the word “payoff” here in a general sense to cover not only financial payoffs but any other
kind of reward or incentive that can be measured numerically. Economists often speak of such rewards
as “utilities” (in this context “utility” does not mean the power company; it is just a fancy word for
“usefulness”), and that is why this decision-making rule is called the “expected utility rule.”
Example 4: It may or may not rain today, and I am wondering whether I should bring my umbrella
to work. Some economists suppose that, consciously or unconsciously, my decision process involves
quantifying the pleasures and inconveniences involved in my various decision possibilities, something
like this: Carrying my umbrella on a sunny day is an inconvenience that costs me 2 units of utility;
the joy of a sunny day without that inconvenience is a gain of 3 units of utility. On the other hand, if
I don’t have an umbrella and it rains, I get wet at a cost of 15 units of utility. Finally, if it rains and I
have my umbrella, I don’t get wet but I don’t enjoy the day either: zero units either way.
This information can be codified into a payoff matrix for the random variable U that measures the
amount of utility I gain from my “trial” under different conditions:
U Rain Shine
Umbrella 0 –2
No Umbrella –15 +3
Suppose now that I hear from the weather forecast that the probability of rain is 40% (and the
complementary probability of “shine” is 60%). What should I do? Using Definition 2 on page 303, I
can calculate the expected values of the utility U under two options: I take an umbrella (let’s call this
expected value E(U|umbrella)) or I don’t (call this E(U|no umbrella)). We obtain
E(U|umbrella) = 0 × 0.4 + (−2) × 0.6 = −1.2,

E(U|no umbrella) = (−15) × 0.4 + 3 × 0.6 = −4.2.
My expected utility if I take the umbrella is −1.2 units, which, even though negative, is greater than
my expected utility of −4.2 units if I don’t take it. So, if I follow the expected utility rule, I should
take my umbrella.
Things come out differently if the probability of rain is 10% and of shine is 90%. Now
E(U|umbrella) = 0 × 0.1 + (−2) × 0.9 = −1.8,

E(U|no umbrella) = (−15) × 0.1 + 3 × 0.9 = +1.2.
In this case, the expected utility is higher if I don’t take my umbrella. The expected utility rule tells
me to leave it at home and risk a shower.
Problem 8: I expect to live in my house for the next 20 years. I consider whether to install a set of
rooftop solar panels, which will cost me $17,500 up front and will subsequently save me $1,200 per
year on my electricity bill. There is a catch, though. The savings from solar panels depend on “net
metering tariffs,” which are mandated by law: these require the utility company to buy back from me
any excess power that my solar panels generate and that I can’t use at that exact moment. Without net
metering, my annual savings would be only $200. I estimate that each year there is a 5 percent chance
(independent of previous years) that the government, under the influence of power-industry lobbyists,
will repeal net metering.15 On the basis of the expected utility rule (and considering only the financial
aspects of the situation), is it advisable for me to install the solar panels?
Solution: We need to find the expected financial payoff from the solar installation. To do this, we first
need to understand what is the probability that net metering will still be in effect in year n (where n = 0
represents the present year and n = 19 is the last year of the 20-year life span we are considering).
Think of it this way: in any given year, the chance that net metering is not repealed that year is 0.95.
Thus, the chance that it has not been repealed (is still in effect) in year n is equal to the chance it has
not been repealed in that year or any previous year, that is,
0.95 × · · · × 0.95 = 0.95n ,
| {z }
n times
by the multiplication rule (Rule 6 on page 292). The chance that net metering has been repealed in
year n (the complementary event) is then 1 − (0.95)n by the complement rule (Rule 4 on page 288).
Now we can figure the expected savings on my electric bill from solar panels in any given year
n. To to this, we add together the contribution to the expected value of these savings if net metering
has not been repleaded by year n and the contribution to the expected value of these savings if net
metering has been repleaded by year n:
$1, 200 × (0.95)n + $200 × (1 − (0.95)n ) = $200 + $1, 000 × (0.95)n .

We find the total expected savings over the 20 years by summing the above expression over all the
values of n from 0 to 19:
$200 × 20 + $1.000 × (1 + (0.95) + (0.95)2 + · · · + (0.95)19 )

1 − (0.95)20

= $4, 000 + $1, 000 ×
1 − 0.95
≈ $4, 000 + $1, 000 × 12.83 = $16, 830;

using the geometric progression formula (Rule 5 on page 198). This total gives my expected savings
from solar power. But I also need to account for the cost of the solar panels themselves, yielding
Expected Financial Utility of installation = −$17, 500 + $16, 830 = −$670.
Applied in the way the problem requires, the calculation gives a negative expected utility from my
solar installation: I should not install the panels.
There are several interesting points we could make here. First, the negative value is entirely due
to regulatory uncertainty: if I knew for sure that the net metering tariff would be maintained, I’d
see a positive expected payoff of $6,500. Second, even with that uncertainty, the negative expected
financial utility is quite small. I might well decide that it is outweighed by the good feelings I get
about reducing my carbon dioxide emissions, or even the good feelings I get from knowing that
my neighbors see me as caring about reducing my emissions. Third, you might remember from an
economics or personal finance class that a dollar in twenty year’s time should not be valued the same
as a dollar right now. That’s correct, and we will study how such time discounting affects this sort of
calculation in Section 6.1.
15 As this book was going to press a 30% tariff was put in place on imported solar panels (effectively increasing the price
of these panels in the U.S.). Such panels accounted for 80% of the U.S. market in 2017. While this tariff doesn’t impact those
who already own solar panels, it will impact the businesses that install them.
Think about it. . .
What is the logical status of the expected utility rule? It is certainly not a mathematical
theorem, since it relates to the behavior of human beings, not just to mathematical objects.
There are at least three different ways to understand the rule.
• Descriptive: as an analysis of mental processes. “This is what is going on inside
our heads when we make decisions. Our brains are trying to find the best possible
outcome as measured by expected value.”
• Predictive: as a model of human behavior. “This may not be how our brains actually
do work, but it is a useful way to model the decisions we end up taking.” Many
classical economists would take this point of view.
• Normative: as a guide or an ethical principle. “This is how you ought to make
decisions (especially important ones).” Or, “This is how a rational person makes
decisions” (subtext: “and you don’t want to be irrational, do you?”).
How would you express the significance of the expected utility rule? Or do you see it as an
inappropriate way to make choices at all—if so, why?
Problem 9: Your community is subject to repeated flooding from rainfall in the nearby mountains.16
In any given year the rainfall (also known as precipitation) can be classified as “low,” “medium,”
“high,” or “exceptional,” with probabilities respectively of 27 percent, 48 percent, 24 percent,
and 1 percent. The corresponding annual amounts of damage due to flooding are $21,000 for
“low” precipitation, $260,000 for “medium,” $870,000 for “high,” and $4.6 million for “extreme”
precipitation.
The U.S. Army Corps of Engineers proposes to build a dam to protect your community. The
construction of the dam will be financed by a bond issue. Repayments on the bond will cost $100,000
annually. The dam will reduce the flood damage to zero for “low,” “medium,” and “high” precipitation
amounts. However, if an “extreme” precipitation event occurs—a “hundred-year flood”—the dam will
burst and the total damage will amount to $25 million.
Should your community build the dam?
Solution: We can make a payoff matrix for this problem, with random variable C the annual cost to
your community, in thousands of dollars, of various scenarios:
Low Precip Medium Precip High Precip Extreme Precip

Probability 27% 48% 24% 1%
Cost, No Dam, $000s 21 260 870 4,600
Cost, With Dam, $000s 100 100 100 25,100
Now let’s use the probabilities to compute the expected values of C:
E(C|no dam) = 21 × 0.27 + 260 × 0.48 + 870 × 0.24 + 4600 × 0.01 ≈ 385,
E(C|dam) = 100 × 0.27 + 100 × 0.48 + 100 × 0.24 + 25100 × 0.01 ≈ 350.
Recall that C represents the cost, we see that the expected cost to our community is lower if the dam
is built. The recommendation of the expected utility rule is clear: Build the dam!
16 Adapted from an example in [155].
But if you lived downstream, you might well be a bit queasy about this recommendation. Those
expected values of C are quite close, after all. And whereas your expected losses without the dam are
made up of many small and moderate-sized events, the expected loss with the dam is dominated by the
chance (albeit tiny) of a huge catastrophe. Is that a gamble you think you want to take? If you don’t
want to take that chance, you might be exhibiting risk aversion. We’ll discuss this important idea in
the next section.
5.3.4 Risk Aversion
Question 1: Which would you prefer to have: a guarantee of receiving $50, or a commitment that you
will receive $100 if a fair coin flip lands Heads (and nothing if it lands Tails)?
Many people would say that they would prefer the guarantee of money in their pocket. The expected
payoffs in both situations are the same, namely 50 dollars, but one situation offers greater certainty in
that payoff than the other. This preference for greater certainty is what social psychologists call risk
aversion.
Definition 4
A person making a choice displays risk aversion if they prefer greater certainty in the
outcome, even at some cost to their expected payoff.
To explain the phrase “even at some cost,” think about the dam example again. The downstream
dweller who, we imagined, might be doubtful about the recommendation to build the dam is willing
to trade an additional expected cost of $35,000 annually for the greater certainty of not being
overwhelmed by a massively damaging, dam-busting flood. In the coin-flip example, supposing that
you said that you would prefer a guaranteed 50 dollars to the coin flip, we could start asking further
questions: would you prefer a guaranteed 45 dollars to the coin flip? What about 40 dollars, and so
on? In this way your level of risk aversion could be quantified.
Utility Treatment of Risk Aversion

Some social scientists try to incorporate the notion of risk aversion within the framework of expected
utility theory. This is done by supposing that the actual “utility” or benefit that individuals receive
from a certain financial payoff is not equal to that payoff, but instead is related to that payoff by a law
of diminishing returns: an additional dollar delivers more utility if it is the first dollar you receive, and
less utility if it comes on top of a thousand dollars that you have received already. One ends up with
a “utility function” (relating financial gains to their corresponding utilities) something like the graph
in Figure 21 on the next page. Notice how the graph bends over to the right (mathematically, we say
that it is concave), reflecting the law of diminishing returns.
Question 2: How does a concave “utility function” like Figure 21 on the next page explain risk
aversion?
Answer: Glad you asked that! To understand it, let’s think about our coin flip example again. The
option of a guaranteed 50 dollars corresponds, as we’ve marked on the graph (see the middle red
dashed line), to a “utility” of 8.3 units. If we think about the option of 100 dollars dependent on a coin
flip, though, we see that it gives a 50% chance of obtaining zero units of utility and a 50% chance of
obtaining 13.5 units (see the other red dashed line). The expected utility is therefore
20
15
13.5
Utility
10
8.3
0
0 20 40 50 60 80 100
Financial gain (dollars)
Figure 21: A risk-averse utility function. Notice how the graph “bends over to the right.” As explained in the text, this “bending”
is a sign of risk aversion.
(0.5 × 0) + (0.5 × 13.5) ≈ 6.8 units of utility.
Since 6.8 < 8.3, the person with this utility function will prefer the certainty of 50 dollars in hand,
and will walk away from the risk.
In fact, we can say exactly what the dollar value of this certainty is (to this particular person).
Looking at the graph again, we can figure out the dollar value that corresponds to 6.8 units of utility.
To do this, we look at 6.8 on the y-axis, read across to the utility curve, and then down to the x-axis
(the light blue dotted lines in Figure 21). The perhaps surprising conclusion is that for this individual,
the “for-certain” sum would have to be reduced to $38.54 before they would consider taking a chance
on the coin flip.
Question 3: So does this assumption of a concave utility function enable us to understand all kinds of
attitude to risk?
Answer: Not so fast! In fact, there are several puzzles about our attitudes to risk that expected utility
theory, even with this concave-utility-function tweak, does not seem able to account for. One of these
is that risk-averse behavior in the domain of possible gains (what we have been discussing so far)
often turns into risk-seeking behavior in the domain of possible losses. Suppose that you are stopped
for a minor traffic offense. The cop offers you a choice17 between a $50 fine or a coin flip that would
lead to no penalty if the coin lands Heads and a doubled fine of $100 if it lands Tails. Now which
option would you take? Many people, even those who showed themselves risk-averse in the previous
thought experiment, are now likely to “take a chance” on walking away with no penalty—in other
words, to seek out risk rather than avoid it. The puzzling feature here is that from the perspective of
your total wealth the two situations are the same. What the cop is offering you is, essentially, a certain
loss of $100 together with an opportunity to play the coin-flipping game with which we introduced
this section (Question 1 on the previous page). So why might you be risk-averse in one situation
and risk-seeking in the other? It seems that what you see as your baseline or starting-point—what is
sometimes called your framing of the situation—greatly affects your risk attitude.
17 Set aside, for a moment, the fact that this would be illegal. We are just doing a thought experiment here.
In a famous paper [327] in 1979, behav- 20

ioral economists Daniel Kahneman and Amos
Tversky introduced a more complex model
called prospect theory to provide a more
10
realistic account of human decision-making
under risk. Prospect theory employs a utility
function like that graphed in Figure 22. The
Utility
positive side of this function (that is, for x > 0) 0
is the same as the one we already graphed
in Figure 21, but the function takes a sharp
turn at x = 0—we have emphasized that by −10
drawing this part of the curve in green instead
of in blue. Notice that the curve “bends”
the other way (convex instead of concave) −20
−100 −50 0 50 100
when x < 0, reflecting potential risk-seeking
behavior in the loss domain. And what is x = Financial gain (dollars)
0? According to prospect theory, this is our
Figure 22: A utility function that might be used in prospect theory.
“frame” or reference point for the decision,
which we arrive at by a prerational, heuristic
process.
There is more to prospect theory than has been described here, especially a description of how
we underweight high probabilities (relative to certainty of an event) and overweight low probabilities
(relative to certainty of the complementary event). The point is, though, that these are not simply
arcane descriptions of strange patterns of human behavior that are unlikely to occur in practice. Almost
every sustainability decision confronts us with a spectrum of risks and rewards including outcomes
that have low probability but could be hugely significant in terms of their payoffs (financial or
otherwise). And in almost every sustainability decision it makes a huge difference what our reference
point or “frame” may be. Prospect theory suggests that if our reference point is the “American
lifestyle” as known to middle-class people in the twenty-first century, we may be willing to incur
risk in order to protect it (the green part of the curve in Figure 22). But from the reference point of
median global incomes, that same behavior might appear as wanton risk-seeking (on the blue part of
the curve). (According to data from Pew Research [184], 88% of Americans are in the “upper-middle”
or “high” income category when considered by global standards.)
Think about it. . .
A regulatory agency (such as the U.S. Environmental Protection Agency) is often required
by law to be highly risk-averse. For example, under the Safe Drinking Water Act the
EPA establishes maximum contaminant level goals (MCLGs) for various contaminants in
drinking water, which are the maximum concentrations of contaminants that can be present
while ensuring no adverse health effects whatsoever. The legally mandated maximum
contaminant levels (MCLs) are then set as close to the MCLGs as is practicable. Reflect
on the consequences, both positive and negative, of such a risk-averse regulatory policy. Do
you believe that this policy stance is the correct one? Why or why not?
5.3.5 Extreme Values

Let’s think back to the flood example, Problem 9 on page 310. An important lesson from this and
similar problems is that in the real world, we need to be able to quantify the magnitude of extreme
events, like the “hundred year flood” (the flood gauge level so large that it has an annual probability of
1%) or even the “five hundred year earthquake” (the earthquake so big that it has an annual probability
of 0.2%). Even if they are very unlikely, such events can have such large consequences that they need
to be taken into account in planning. In this section we want to describe very briefly how numerical
values for these extreme events are arrived at.
In the language that we have been developing, we have a random variable X (such as the
daily rainfall or temperature or flood gauge value) and we are interested in the probability of an
exceedance—an event like
“X ≥ ` for at least one day in a given year,”
where ` is some fixed level. As we explained in Example 9 on page 296, the “one hundred year” level
for X is that value of ` for which the probability of an exceedance is one in one hundred (0.01); the
“five hundred year” level is that value of ` for which the probability of an exceedance is one in five
hundred (0.002); and so on.
How, though, can one estimate the probabilities of these rare exceedances? After all, we most
likely have never witnessed such an event; we certainly have not witnessed enough hundred-year or
five-hundred-year periods to estimate the probabilities directly using the law of large numbers. It turns
out that there is a way to make use of the distribution of smaller “extreme” events, like “rainiest days,”
to estimate the probabilities of more extreme events that we have never seen directly. How can that be
possible?
Consider for a moment what we have learned about the central limit theorem (Section 5.1.3). We
can start with more or less any kind of probability model, it says,18 and any random variable X
associated with it. Then, if we can make many independent repetitions and take the average value
of X over all these repetitions, we’ll get a new random variable, Y . According to the central limit
theorem, the probability model for the resulting average variable Y will be of a very restricted type,
in fact, a normal distribution. What’s more, to find which normal distribution Y has, we just have to
estimate the mean (“center”) and the standard deviation (“spread”). The point to hold on to is that
averaging many independent repetitions reduces the wide spectrum of available probability models to
a single family of models. All of these have a clear family resemblance (the bell-shaped normal curve)
and differ only as to where that curve is centered and how wide it is spread—two quantities that we
call the parameters for the normal distribution family.
It turns out that there is a result similar to the central limit theorem if we start with any kind of
probability model and associated random variable X, keep taking the values of X from independent
repetitions, and then form not the average but the maximum of these X-values—the extreme value.
Once again, the extreme value is a random variable (let’s call it Z this time) whose distribution belongs
to a restricted family, called simply the extreme value distribution family. It is true that the EVD family
is less elegant than the normal family, in particular it has three parameters instead of two, but the basic
idea is the same: By finding the best-fitting distribution from this restricted family, we can get hold of
the probabilities of extreme events, provided that we have a data set covering sufficiently many years
even of quite ordinary events.
Example 5: In [194], Tom diLiberto writes for NOAA’s online ClimateWatch Magazine, “How can
we call something a thousand-year storm if we don’t have a thousand years of climate observations?.”
He takes as his example the extreme rainfall near Beckley, West Virginia, in June 2016, which was
said to be a “thousand-year event,” and he steps the reader through the process of using the available
observations together with the extreme value distribution to reach this conclusion. The key ideas are
18 There is some mathematical fine print, of course.
(a) Histogram of extreme daily rainfall values in 82 separate (b) Extreme value distribution fitted to the given data.
years.
Figure 23: Extreme rainfall in Beckley, West Virginia, from [194].
found in the two graphics in Figure 5.3.5. In the left graphic are the data from 82 years for which
complete observations are available. The random variable Z here is the greatest one-day rainfall over
an entire year, and the distribution of the observed values of Z has been represented on a histogram
with class intervals of 0.25 in. On the right, you can see the best fit to these data from the 3-parameter
family of extreme value distributions. This is a curve (like the normal distribution curve), but it is not
symmetric—it is skewed to the right. Using this curve one can now find, for any specified rainfall level
`, the probability of an exceedance of ` in a one-year period. In particular, one can find that for ` =
7.25 in, the probability of an exceedance is 0.001, or one in a thousand. The June 2016 storm dumped
about 8 inches of rain on Beckley in a day, which explains why it was classified as a “thousand-year
event.”
Think about it. . .
You can find several easy-to-use extreme value distribution calculators online, including
through links provided on this book’s website.19 Try one of these out, either on data that you
make up, or (if possible) on real data for rain- or snowfall, flood depth, earthquake strength,
or other examples. Are you surprised by the result?
The very brief sketch of extreme value theory in this section is intended only to explain that
there is sound mathematics behind often-used phrases like “hundred-year drought.” But as with any
mathematical theory, correctly applying it depends on certain assumptions; in this case, one very
important assumption is that the future will be like the past. By upending this assumption, climate
change adds a further level of complexity to extreme-value modeling in the environmental context.
The Intergovernmental Panel on Climate Change has issued a “Special Report on Extreme Events” to
accompany its Fifth Assessment Report. In some cases, they find, climate change will make extreme
events (especially high-precipitation events) much more likely. In other cases, extreme events may
have the same or even a reduced probability. The picture is uncertain—and that very uncertainty is
another vulnerability.
19 Visit http://math-for-sustainability.com/extreme-value-calculator.
Think about it. . .
Some of the above treatment of extreme value theory is adapted from Hadlock’s book [146],
which has the cheery title Six Sources of Collapse. Just after discussing extreme value
theory, Hadlock goes on to comment:
Having spent many years in the world of industrial disasters and risk analysis,
I’ve become convinced of two principles: (1) we often don’t understand the
magnitude and diversity of the risks, and (2) we are too often afflicted with an
optimistic bias when assessing the facts. Thus, when our calculations tell us
what we would like to hear, we’re more likely to accept them than when they
shake our fundamental assumptions or threaten to cost us money.
Do you agree with him? Why or why not?
Summary of Ideas: Expectations and Payoffs
• A random variable is a numerical variable that depends on the outcome of a random

process. For example, the payoff from a bet, or the temperature in a certain location
at midday tomorrow.
• The values of a random variable can be used to define events.
• The expected value E(X) of a random variable X is the sum of contributions from
each of its values. The contribution from the value a is the product of a and the
probability of the event “X = a.”
• If X and Y are random variables, then E(X +Y ) = E(X) + E(Y ). The corresponding
law for products, E(X ×Y ) = E(X) × E(Y ), holds if X and Y are independent.
• The law of large numbers for a random variable X says that the average value of X,
after many independent trials, will be close to E(X).
• The expected utility rule is a guide for decision-making under uncertainty: choose
the option with the greatest expected utility.
• Risk aversion is the tendency to prefer a policy that offers greater certainty in the
outcome, even if that comes at some cost.
• Extreme value theory is a branch of probability theory that allows statisticians to
quantify the risks of “extreme” events, even ones that have never yet been observed.
5.4. ASSIMILATING NEW INFORMATION 317
5.4 Assimilating New Information
Objectives
I can define and use the notion of conditional probability.

I can distinguish false positive from false negative errors.
I can define the sensitivity and selectivity of a test.
I can construct a two-way frequency table and use it to solve problems related to
conditional probability.
I know what the prosecutor’s fallacy is and why it is important.
I understand how statistical inferences are used to update model parameters based
on new evidence.
I know what a credible interval is.
In the year 2012, there were 53 “unprovoked” shark attacks on swimmers in U.S. waters. The U.S.
population that year was about 312 million, so we could estimate that the probability of a random
person in the U.S. being attacked by a shark is
53
≈ 1.7 × 10−7 .
3.12 × 108
It is easy, but wrong, to say that this is the probability that you will be attacked by a shark. Why is this
wrong? Because you know much more about yourself than you do about a “random person” in the
U.S. Maybe you are a big wave surfer and spend your time in the water off the beaches of Maui and
Kauai. Then your chances of being attacked by a shark will be greater than those of a “random person”
in the U.S. Maybe you are an alpine climber who never hits the beach and your idea of summer fun
involves scrambling among the jagged spires of California’s Sierra Nevada. Then your chances of
being attacked by a shark will be less than those of a “random person” in the U.S. Your chances of
being struck by lightning will be greater, but we’re not discussing that right now.
Like our judgment of the probability that you will be attacked by a shark, risk judgments in
sustainability science are regularly updated by new information. The language that mathematicians
use to describe this updating process is called conditional probability. That is what we will study in
this section.
5.4.1 Conditional Probabilities

In Section 5.2 we defined independent events (Definition 10 on page 291). Recall that events A and
B are independent if knowing whether or not A occurred does not give any new information about
how likely B is to occur. Implicit in this definition, then, is the idea that under other circumstances we
might need to revise our estimate of the probability of A if we knew that event B had already occurred
(or was certain to occur). This is the basic idea of conditional probability.
Definition 1
The conditional probability of an event is its probability after some new piece of
information is taken into account. A conditional probability is written like this: P(A|B),
which is read as “the probability of A given B.” This means that the probability that event A
occurs if we already know that event B occurs.
Problem 1: Consider the rolls of a fair die. Let A be the event “we roll an even number” and let B be
the event “we roll a 1, 2, 3, or 5.” Find P(A|B).
Solution: The sample space consists of six equally likely out-

A B comes, {1, 2, 3, 4, 5, 6}. In terms of these outcomes, we have the
events A = {2, 4, 6} and B = {1, 2, 3, 5}. We can represent the way
4,6 2 1,3,5
that events A and B overlap by a Venn diagram as in Figure 24.
Without prior information about event B, the event A is made
up of 3 equally likely outcomes out of 6. Therefore, by Rule 2 on
page 282, the probability of A is
Figure 24: Venn diagram for problems 1 and 2 3
P(A) = 6 = 0.5.
If we know that B occurred, though, things change. Now there are
only 4 possible equally likely outcomes, 1, 2, 3, and 5, and of these only one belongs to event A.
Therefore, by Rule 2 on page 282 again, the conditional probability of A given B is
1
P(A|B) = 4 = 0.25.
Problem 2: With the same setup as in Example 2, calculate the conditional probability P(B|A). Is this
the same as P(A|B) or different? Can you explain why?
Solution: Looking at the Venn diagram again, if we know that A occurred, we now have only 3
possible equally likely outcomes, 2, 4, and 6. Only one of these (namely 2) belongs to event B.
Therefore, by Rule 2 on page 282 again, the conditional probability of B given A is the different
number
P(B|A) = 13 ≈ 0.33.
We can understand the reasoning better if we make a two-way table of the outcomes, like Table 8:
Table 8: Two-Way Frequency Table
Number of outcomes A occurs A does not occur Total

B occurs 1 3 4
B does not occur 2 0 2
Total 3 3 6
The entries in the table show the number of equally likely outcomes that fit with each of the possible
combinations of events. For instance, the red 3 in the upper right of the table20 shows that there are 3
outcomes (namely 1, 3, 5) for which event B occurs but event A does not.
20 When we refer to the “upper right,” “lower left,” and so on of these two-by-two tables, we’re referring only to the main,
central part, not to the “totals” row and column. Thus the “upper right” in Table 8 is 3, not 4.
When we are computing P(A|B) we know that B occurred. We are therefore looking only at the
upper row of the table. The probability is
Number of outcomes in upper row for which A occurs 1 1

P(A|B) = = = .
Total number of outcomes in upper row 1+3 4
Similarly, when we are computing P(B|A) we know that A occurred. We are looking only at the left-
hand column of the table. The probability is
Number of outcomes in left column for which B occurs 1 1

P(B|A) = = = .
Total number of outcomes in left column 1+2 3
The numerators of these fractions are the same (namely, the contents of the top-left cell of the table)
but the denominators are different.21 In working with conditional probabilities, the most important
question to ask is: what is the denominator?
Example 1: Let’s look more closely at shark attacks. You can download detailed information about
these from the database at [296]. We find that of the 53 shark attacks recorded in 2012, 11 were on
females and 42 on males. The U.S. population skews 50.8% female (according to the Census Bureau),
so we could make the following table for 2012 (assuming a total U.S. population of approximately
312 million):
Table 9: Shark Attacks

U.S. Persons Male Female
Not Attacked by Shark 153,500,000 158,500,000
Attacked by Shark 42 11
Now from the table we can read off some conditional probabilities. Suppose that you are a U.S.
male (but otherwise we know nothing about you, i.e., you are a “random” male in the U.S.). Then the
probability that you are a shark attack victim is
Men Attacked by Sharks in the U.S. 42

P(attacked|male) = = ≈ 2.7 × 10−7 .
Men in the U.S. 42 + 153, 500, 000
If you are a “random female U.S. person,” the corresponding probability is
Women Attacked by Sharks in the U.S. 11

P(attacked|female) = = ≈ 0.7 × 10−7 .
Women in the U.S. 11 + 158, 500, 000
These numbers answer the questions, “How likely is a randomly chosen person (of one or other
gender) to be attacked by a shark in a given year.” As expected, they are extremely small. A shark
attack is a remote risk for most people, whatever their gender identification.
If we ask a different question, however, we will get a different answer. Suppose we ask “How likely
is a random shark attack victim to be male?” Now we want the probability
Men Attacked by Sharks in the U.S. 42

P(male|attacked) = = ≈ 0.79.
People Attacked by Sharks in the U.S. 42 + 11
Similarly, if we ask “How likely is a random shark attach victim to be female?” we will get the answer
21 Reminder: The numerator is the figure on top; the denominator is the one on the bottom. In the fraction 2 , the numerator
7
is 2 and the denominator is 7.
Women Attacked by Sharks in the U.S. 11

P(female|attacked) = = ≈ 0.21.
People Attacked by Sharks in the U.S. 42 + 11
The reason that these two numbers are so different from the previous set is that they are answering
different questions and therefore have different denominators. If you are interested in evaluating your
own risks of becoming a shark attack victim, you probably want to be asking the first set of questions.
If you are a physician who regularly works with shark attack victims, it might be significant to ask the
second set of questions instead.
These examples help explain the general formula for conditional probabilities.
Rule 1: Conditional Probability Rule
Let A and B be events related to the same random process. The conditional probability
of A given B is found from the formula
P(A ∩ B)
P(A|B) = .
P(B)
To explain this, imagine22 that we make a large number, N, of independent repetitions of the
process. Then according to the law of large numbers, the total number of outcomes for which event B
occurs will be approximately N × P(B), and of those outcomes approximately N × P(A ∩ B) will be
ones for which both A and B occur. In the two-way frequency table (Table 10),
Table 10: Two-Way Frequency Table
Number of outcomes A occurs A does not occur

B occurs N × P(A ∩ B) N × P(Ac ∩ B)
B does not occur N × P(A ∩ Bc ) N × P(Ac ∩ Bc )
the top left entry is N × P(A ∩ B), and the sum of the first row is N × P(B). Dividing one by the other,
as in our examples, gives
N × P(A ∩ B) P(A ∩ B)
P(A|B) = = , (6)
N × P(B) P(B)
as the conditional probability rule states.
Remark 1: We defined A and B to be independent if knowledge about A does not affect our assessment
of the probability of B, i.e., if P(B|A) = P(B). Knowing the conditional probability rule, we can rewrite
this as
P(A ∩ B)
= P(B) or P(A ∩ B) = P(A) × P(B).
P(A)
This is just the multiplication rule (Rule 6 on page 292) from the section where we defined
independence. Notice, by the way, that because in this formula it makes no difference when we
interchange A and B, we find that
P(A|B) = P(A) if and only if P(B|A) = P(B);
that is, A is independent of B if and only if B is independent of A. You probably believed this already,
but it is nice to have a mathematical proof.
22 It is not necessary that the process be repeatable in real life. This is just a “thought experiment.”
Critical Thinking
A newspaper headline claims, Sitting All Day Increases Your Risk of Dying by 50%. Since
you know that everybody is certain to die at some time or other, this headline is hard to
understand. Can you think of a way of interpreting the headline that makes some sense of
it? Express your interpretation in the language of conditional probability. What question(s)
would you like to ask the reporter in order to clarify whether your interpretation is the
correct one?
5.4.2 Inference
Question 1: In a murder mystery, the detective suspects that someone has slipped arsenic into
Professor Plum’s nightly tot of Scotch. How can we determine how much arsenic is in the whisky?
Answer: This question can be answered by a direct measurement. Chemical techniques can be used
to separate out the arsenic, purify it, and weigh the result to determine the quantity contained.
Now for a similar question on a larger scale.
Question 2: A lake contains 6 × 1011 gallons of water. It is suspected that the lake has been polluted
by arsenic. How can we find the total amount of arsenic in the lake?
Answer: Unlike the previous one, this question cannot be answered by a direct measurement. There
is no way to drain the entire lake, separate out the arsenic from the nearly one trillion gallons of water,
and then weigh it. Instead of direct measurement we must use inference to find the total amount of
arsenic. For instance, we may take a number of samples, analyze each of them, and then estimate the
overall amount of arsenic from the sample data. This estimate should not be just a bare number—it
should also come with a specified precision (Section 1.3.1) so that we know a credible range within
which the true value is likely to be found. Compare the discussion in Section 5.2.5.
Definition 2
Inference is the process by which we learn more about some variable that we can’t access
directly (the subject of inference) by taking into account the information given by some
evidence.
In Question 2, the subject of inference is the amount of arsenic in the lake, and the evidence is
the set of results obtained by analyzing the individual samples. Before we consider examples of this
sort (where the subject of inference is a numerical variable), though, we’ll look at an easier sort of
example, where the subject of inference is an event H that may or may not happen.23 Does the patient
have cancer? Is the well water contaminated? Did the defendant commit the crime?
The need for inference arises because we cannot see directly whether H occurs. What we can do is
carry out a test. A test is another event E that we can observe and which, we have reason to believe,
will yield some evidence about the unobservable H. The test can return a positive result (E occurs,
23 The event H is sometimes called the hypothesis, and another name for this kind of inference process is hypothesis
testing.
giving evidence that the hypothesis H may be true) or a negative result (E does not occur, giving
evidence that the hypothesis H may be false).
There are four possible scenarios for the occurrence of events E and H, as shown in Table 11.
Table 11: Inference Table

H occurs H does not occur
E occurs P Q
E does not occur R S
Events in class P are called true positives: the test E returns a positive result, and the true answer
H is also positive. Events in class Q are false positives: the test returns a positive result, but the
true answer is negative. Similarly, events in class R are false negatives, and events in class S are
true negatives. In classes P and R the inference process is functioning correctly: the test accurately
tells us about the unobserved condition H. Events in classes Q and R are different kinds of error in
the inference process: the test has not returned accurate information. We would like to design our
inference process to minimize the chances of these two types of error.
Definition 3
A false positive error is one for which a test returns a positive (“yes”) answer when the real
answer is negative (“no”). A false negative error is one for which a test returns a negative
(“no”) answer when the real answer is positive (“yes”).
Example 2: The following are false positives.

• Your computer rejects a file for being malware when it is actually safe.
• A person tests positive for hepatitis C when he actually doesn’t have it.
• The fire alarm goes off when there is no fire.
Example 3: The following are false negatives:

• Your computer determines it is safe to download a file that is actually a virus.
• A person tests negative for breast cancer when she actually does have it.
• A carbon monoxide detector does not go off when there are dangerous levels of carbon
monoxide in the air.
False positive and false negative errors are very different, but both can be bad. For instance, consider
the example of cancer screening. A false negative error means that someone has a cancer that we
don’t catch through screening. That is a terrible thing. But false positive errors also have their cost.
For instance, a healthy person is screened as having cancer. The best possible outcome from this
is a lot of anxiety. Worse outcomes might include unnecessary and invasive medical interventions,
including surgeries, which carry their own risks. And what if the “false positive” was for an untreatable
condition? What would you do if you received a positive AIDS test (in the time before effective
treatments existed), only to learn years later that it was a mistake? Such cases have occurred.
Imagine that we can somehow carry out a large number of independent repetitions of the test, and
that we fill in the cells P, Q, R, and S in Table 11 with the frequencies of the corresponding kinds
of events: P the frequency of true positives, Q the frequency of false positives, R the frequency of
false negatives, and S the frequency of true negatives. Filled out in this way, the table is called a two-
way frequency table, like the one in Table 8 on page 318. In terms of this table we can define two
important properties of our test:
Definition 4
Suppose that H is an unobservable condition and that E is a test for H. Then

(i) The sensitivity of the test is the conditional probability P(E|H), that is, the proportion
of outcomes satisfying H for which the test E returns a positive result. In terms of
Table 11, the sensitivity is P/(P + R).
(ii) The specificity of the test is the conditional probability P(E c |H c ), that is, the
proportion of outcomes not satisfying H for which the test E returns a negative
answer. In terms of Table 11, the specificity is S/(Q + S).
Thus the more sensitive a test is, the fewer false negatives it will produce; the more specific it is, the
fewer false positives it will produce. An ideal test would be 100% sensitive and 100% specific. But in
the real world, there is a trade-off between false positives and false negatives: the more sensitive you
make your test (the more you want to make sure that you avoid a false negative), the less specific it
will become (and thus the more false positives you must be prepared to accept).
Critical Thinking: It’s Not All About Causation
The way we usually think about medical or other tests is that the unobserved hypothesis
H causes the evidence E (with some possibility of error): the cancer causes a positive
screening result, the burglar trips the burglar alarm, and so on. The language of conditional
probability, though, does not have causality assumptions built in: we can (and will) talk
about both P(E|H) and P(H|E). The flip side of this is that we can’t draw conclusions about
causality from conditional probability data alone: even the ideal case P(E|H) = P(E c |H c ) =
1 (in other words, 100% sensitivity and 100% specificity) tells us nothing by itself about
whether one event causes the other. Can you devise an example that illustrates this?
You may have heard people say “correlation does not imply causation.” This is a catchy
slogan to remind us that we can’t draw conclusions about causality only from data about
conditional probabilities and similar measures of “association.”
Problem 3: In a certain region of the country, asymptomatic women between the ages of 40 and 50
are invited to participate in breast cancer screening (mammography). The probability that a randomly
selected woman from this group has breast cancer is 0.8 percent. If a woman has breast cancer, the
probability is 90 percent that she will have a positive mammogram (that is, the sensitivity of the test
is 90 percent). If a woman does not have breast cancer, the probability is 7 percent that she will still
have a positive mammogram (that is, the selectivity of the test is 100 − 7 = 93 percent). Imagine that
you are the medical adviser to a woman who has just learned that she has a positive mammogram.
What would you tell her is the probability that she actually has breast cancer?
This question is taken from the book [128]. The author tells a story about it:
Dr. Konrad Standing [a pseudonym] is chief of a department in a university teaching

hospital, a prominent figure. . . with more than three decades of professional experience.
A few years ago, I asked him whether the physicians at his institution would participate
in a study of diagnostic intuitions being conducted by my colleagues and me at the Max
Planck Institute for Human Development. . . . To set an example, he volunteered himself.
The first diagnostic task he worked on concerned the routine breast cancer screening of
women, as described [in Problem 3 on the previous page].
....
Department chiefs are not used to having their reasoning tested. Dr. Standing was visibly
nervous while trying to figure out what he would tell the woman. After mulling the
numbers over he finally estimated the woman’s probability of having breast cancer, given
that she has a positive mammogram, to be 90 percent. Nervously, he added, “Oh, what
nonsense. I can’t do this. You should test my daughter; she is studying medicine.”
Solution: Let’s see if we can do better than Dr. Standing. The first step is to make a two-way frequency
table. Imagine a sample of 10, 000 potential patients.24 The hypothesis H is the event “the subject has
breast cancer” and the test (or evidence) E is the event “the subject has a positive mammogram.” We
can fill in our table as follows.
First, we know that the (unconditional) probability of event H is 0.8 percent. That means 80 =
10, 000 × 0.8% out of our 10, 000 patients are expected to have cancer, and 10, 000 − 80 = 9, 920 to
be cancer-free. These numbers are the totals of the left-hand and right-hand columns respectively.
We are also told that “if a woman has breast cancer, the chances are 90 percent that she will have
a positive mammogram.” That means that of the 80 in the left-hand column of the table, 80 × 90% =
72 are expected to reside in the top-left cell (cancer and positive mammogram) and the remaining
80 × 10% = 8 in the bottom-left (cancer, negative mammogram). Similarly, we are told that “If a
woman does not have breast cancer, the probability is 7 percent that she will still have a positive
mammogram.” Thus, of the 9, 920 in the right-hand column, 9, 920 × 7% ≈ 694 reside in the top right
cell and the remaining 9, 920 × 93% ≈ 9, 226 in the bottom right. This allows us to completely fill in
the two-way frequency table, Table 12:
Table 12: Breast Cancer Screening
Patients Screened Cancer No Cancer Total

Positive Mammogram 72 694 766
Negative Mammogram 8 9,226 9,234
Total 80 9,920 10,000
Now to answer the question that was asked, “given that a woman has a positive mammogram, what
is the chance that she has cancer?” We answer this by looking in the first row of the table, which
represents the 72 + 694 = 766 patients in our sample who had a positive mammogram. Of these, 72
actually had cancer. The desired probability is therefore
72
P(cancer|positive test) = ≈ 9%.
766
That was not so hard, was it, Dr. Standing?
24 The exact number N = 10, 000 in the sample does not matter; it will cancel out in the end, like the N in equation (6) on
page 320. It is helpful, though, to think of a large N.
Think about it. . .
We saw in the breast cancer example that given a positive screening test, the chances that
a patient in that sample actually has breast cancer are less than 10 percent—that is, the
chance that a positive screening result is a false positive is over 90 percent. (Guys, similar
or even less impressive figures also apply to positive results for prostate cancer (PSA)
screening [28].) Did you know this? Does this knowledge affect how you think you might
respond if you received a positive screening at some time in the future? Does it affect your
willingness even to take a screening test? Why or why not?
Remark 2: Constructing a two-way frequency table, as we have done here, is mathematically

equivalent to a technique called Bayes’ theorem that is commonly taught for solving problems of
this kind. But though the two techniques are mathematically the same, research by Gerd Gigerenzer
(the author of [128]) has shown that they are not psychologically the same. Gigerenzer presented the
case study above to 48 physicians. For half of them, the case study was presented in a way that would
lead them to try to use Bayes’ theorem; for the other half, in a way that would lead them to construct
a two-way frequency table. Of the 24 who tried to use Bayes’ theorem, only 8 got close to the correct
answer; of the 24 who constructed the frequency table, 19 were close to the correct answer. And
Dr. Standing?
With frequencies, he “saw” the answer, remarking with relief, “But that’s so easy”. . . .
His daughter’s help was no longer needed.
Problem 4: In 1990 a 55-acre landfill in Tonawanda, New York, was found to be contaminated with
radioactive waste [330]. The waste is related to the development of atomic weapons during World War
II. In the early 1960s a new neighborhood was constructed adjacent to the landfill. Remediation of the
site is underway and is expected to be complete in 2024.
Tonawanda is just one of hundreds of sites around the country that were involved in the early
development of atomic weapons and atomic energy following World War II [172]. Suppose that
a randomly selected site has a 4% chance of being contaminated and that a soil sample from a
contaminated site has a 60% chance of testing positive for radiation. A soil sample from a non-
contaminated site has an 18% chance of testing positive. If a soil sample from a randomly chosen
site tests positive, what is the probability that the site is contaminated?25
Solution: Let’s make a two-way frequency table that summarizes the information we have. Suppose
we had 1000 soil samples from randomly chosen sites. A randomly chosen site has a 4% chance of
being contaminated, so 4% of our samples are from contaminated sites:
0.04 × 1000 = 40.
Forty of our samples come from contaminated sites and 960 samples come from uncontaminated sites.
Of the 40 samples from contaminated sites, 60% test positive:
0.60 × 40 = 24,
and the remaining 16 test negative.

Of the 960 samples coming from uncontaminated sites, 18% test positive:
0.18 × 960 = 173,

25 This problem is motivated by an example case study in [106].
and the remaining 787 test negative. We construct a two-way frequency table (Table 13 on the next
page).
Table 13: Radioactivity Testing
Samples Contaminated Not Contaminated Total

Positive Test 24 173 197
Negative Test 16 787 803
Total 40 960 1000
Now, if a soil sample tests positive, what is the probability the site is contaminated? We use the first
row of the table. There are 24 + 173 = 197 samples that tested positive. What percentage of these are
from contaminated sites?
24
P(contaminated|positive test) = ≈ 12%.
24 + 173
We conclude that if a sample tests positive, there is a probability of approximately 12%, or 0.12, that
it comes from a contaminated site.
Problem 5: Groundwater contamination by methane is sometimes a result of developing natural

gas wells via hydraulic fracturing (fracking) [259]. However, groundwater can also be contaminated
by naturally occurring methane unrelated to fracking [222]. Suppose that in a particular region of
Northern Pennsylvania methane was known to occur naturally (before fracking) on 7% of well sites,
and that 14% of water wells on such sites showed methane contamination. Some years after fracking
in the region began, 3.5% of all water wells throughout the region showed methane contamination.
Suppose a recent home owner in the region puts in a new well and soon discovers methane
contamination in the water. What is the probability that the methane is naturally occurring, and not a
consequence of the fracking activity?
Solution: This is once again an inference problem, with the hypothesis H (now unobservable) being
“Natural methane at site” and the evidence E being “Methane contamination found in well water.” We
use a two-way frequency table to determine the conditional probability
P(H|E) = P(natural methane|methane contamination found in water).
We imagine 10, 000 water wells, of which 7%, or 700, are on sites where methane may be naturally
present and 14% of these 700, or 98, have dissolved methane in the water. This allows us to fill in the
first column of the frequency table, Table 14 on the opposite page.
We also know that 3.5% of all wells, or 350 of the 10, 000, show methane contamination. Since 98
are from naturally occurring sources of methane, 350 − 98 = 252 are not, allowing us to fill in the first
row. That in turn allows us to fill in the only missing entry, in the bottom right corner. Since the total
of all entries in the table must be 10, 000, the bottom-right number must be
10, 000 − 98 − 602 − 252 = 9, 048.
This completes the construction of the two-way frequency table, Table 14.
To answer our question, we use the top row of the table. There are 350 wells with dissolved methane,
98 of which are contaminated from naturally occurring sources:
98
P(natural methane|methane contamination found in water) = ≈ 0.28.
350
Table 14: Methane Contamination

Samples Natural Methane No Natural Methane Total
Methane Contamination Found 98 252 350
No Methane Contamination Found 602 9048 9,650
Total 700 9,300 10,000
The probability that the observed methane contamination arises from natural sources is about 0.28, or
28%.
We note that research on fracking impacts to groundwater has not presently reached a definitive
conclusion. Some studies show increased impacts near fracking developments, while other studies
suggest that naturally occurring methane is widespread in the region.
5.4.3 The Prosecutor’s Fallacy

Suppose that A and B are events. We’ve seen in our examples that the probabilities P(A|B) and P(B|A)
may be different (perhaps hugely different). They express the answers to two different questions (“how
likely is A, if we know that B occurred?” and “how likely is B, if we know that A occurred?”) and they
therefore have different denominators. Nevertheless, they are easily confused, sometimes with tragic
results.
Unfortunately, this mistake has happened in courtrooms often enough to receive a name—the
prosecutor’s fallacy. Imagine a courtroom situation in which a defendant is on trial for a major
crime. The hypothesis H is the event “the defendant is innocent.” The prosecutor, wanting to show
that the defendant is guilty (which is to say, that H is false) presents some piece of evidence (for
example, a DNA match). Call this evidence E. The prosecutor then calls expert witnesses who testify
that “this match is extremely unlikely to have come about by chance”—that is, P(E|H) is very small.
A prosecutor is within their rights to bring such evidence, of course. The danger is that the jurors
may hear the expert witness as saying “the defendant is extremely unlikely to be innocent”—that is,
P(H|E) is very small. Especially if the expert witness quotes a numerical value for P(E|H), the jurors
may be intimidated by mathematics and persuaded to convict when they should not do so.
Suppose for example that the numbers happen to be exactly the same as those in the cancer
screening example on page 323. We’ve rewritten the frequency table in Table 15—only the labeling
of rows and columns has been changed: “Your Honor,” says the prosecutor, “there is only a 7 percent
Table 15: Prosecutor’s Fallacy
Subjects Guilty Not Guilty Total

Positive Test 72 694 766
Negative Test 8 9,226 9,234
Total 80 9,920 10,000
chance (694/9, 920) that a defendant who is not guilty would test positive. I submit that this evidence
should lead the jury to convict!”
This is the prosecutor’s fallacy. The jury should be asking not, “what is the chance that a defendant
who is not guilty would test positive?” but, “what is the chance that a defendant who tests positive
is not guilty?” In this example, the answer is 694/766 ≈ 91%. The test is not very convincing as
evidence of guilt!
Example 4: Sally Clark was a British “solicitor” (attorney). Her son Christopher
was born in September 1996. In December, Christopher died in his sleep. A
second son, Harry, was born the next year and also died in infancy. Clark was
arrested and later convicted of double murder. A significant piece of evidence in
her trial was provided by pediatrician Professor Sir Roy Meadow, who testified
that the probability of two instances of sudden infant death in the same family
was “about 1 in 73 million.” Many people—including, perhaps, members of the
jury—believed that this number gave the chance that Sally Clark was innocent.
She was convicted in 1999 and sentenced to life imprisonment.
Would you have been persuaded by Roy Meadow’s evidence? Hopefully not.
It is now a notorious example of the prosecutor’s fallacy. Meadow’s testimony
gave the figure
P(two infant deaths|Clark is innocent) ≈ 1.4 × 10−8 .
In other words, he gave the probability of two instances of sudden infant death
Figure 25: Cover of [289], which under the assumption that Clark was innocent. But even if this is the right answer
describes the Sally Clark case.
Reprinted with permission.
(which there is reason to doubt), it is not the answer to the right question. What
the jurors needed to know is
P(Clark is innocent|two infant deaths).
That is, they needed to know the likelihood that Clark was innocent given two instances of sudden
infant death. As we saw in Example 1 on page 319, these two conditional probabilities (P(H|E) and
P(E|H)) can differ by many orders of magnitude.
After several appeals—and protests from eminent statisticians—Sally Clark was released from
prison in 2003, with the appeal judge stating that Meadow’s evidence “should never have been before
the jury in the way that it was.” But it was too late. As her family stated, “Sally was unable to come to
terms with the false accusations, based on flawed medical evidence and the failures of the legal system,
which debased everything she had been brought up to believe in and which she herself practised.” A
few years after her release, Sally Clark was dead. Following a comprehensive review of Meadow’s
evidence in other cases, several other women were acquitted or had their convictions overturned.
Meadow himself was struck off the medical register in 2005 for professional misconduct, but he was
reinstated after an appeal. The whole story is a terribly sad one, which might have been avoided had
the jurors been instructed on how to use a very simple tool, the two-way frequency table.
5.4.4 Inference and Numerical Variables

The examples that we have considered so far apply the ideas of testing and inference only to events,
which you can think of as being like yes-or-no questions (“Does the patient have cancer?”). But similar
ideas can be applied to scientific research of many different kinds, including those where the subject
of inference is a numerical variable, not a categorical “yes or no.” The aim of this section is to show
by some simple examples how the same ideas (evidence, inference, and conditional probability) that
we have already used for hypothesis testing can also be applied to inferences related to numerical
variables. We’ll also emphasize that they provide scientists with a realistic way not only to estimate
the value of a numerical parameter (for example, the level of some pollutant in a lake, which we
discussed in Section 5.2.1) but also to set credible bounds for the potential error in such an estimate
(so that any resulting calculations can be made with appropriate precision; see Section 1.3.1).
Many readers and instructors may want to treat this as an “extra credit” section: while the ideas
that it uses are conceptually no different from those earlier in the chapter, they are much denser
computationally. At the same time, you should realize that this is only the briefest of introductions.
If you want to understand more, you’ll need to read a book (or take a course) on statistics,26 which
is the art of using probabilistic methods to make well-justified inferences from data. An example of
such a book is the work [134] of Alan Grafen and Rosie Hails.
Before we begin, let’s set up some terminology.
• A while ago, we introduced the notion of a parameter (Definition 2 on page 102), and we have
made use of it regularly since then. We originally defined parameter in the context of stock-flow
models, but the general idea is broader than that:27 in any kind of scientific model (including
one that has an unpredictable element, that is, which is a random process by our definition), a
parameter is some external quantity that controls the behavior of the model in some way.
• If we have a model for some system (large or small), we will want to test it. Recall
(Section 2.1.4) that science is a process whereby we continually put our models to the “reality
test”; when applied to a real situation, the model must describe it with sufficient accuracy. In
the context of our present discussion, the scientific method proceeds as follows.
◦ First, we use observation to collect a large set of data about the process we are trying to
model.
◦ Then, we use this observed data to learn more about the parameter values in the model—
that is, to estimate the values that “fit” best with our observations.
◦ Finally, we use the model, complete with estimated parameter values, to make predictions
in a new situation. If the predictions don’t match reality, the model fails; we modify or
reject it and seek a better one. If they do agree well enough with reality, the model passes
the test (so far).
• In terms of the picture of inference we’ve given in Definition 2 on page 321, we are trying
to understand in a systematic way how new evidence E (in the form of observations or
experimental data) can be used to improve our pre-existing understanding of the parameter
values in our model. In the examples discussed in Section 5.4, the “parameter value” was
simply the occurrence or not of a single event, the hypothesis H, and our “pre-existing
understanding” was expressed by our knowledge of P(H). In problem 3 on page 323, for
example, the hypothesis H was “the subject has breast cancer,” and our pre-understanding—
before screening—was that the probability of this event was 0.8%. We can think of this pre-
understanding as a probability model with only two outcomes: “H is true” and “H is false.”
For more complex questions, such as estimating the value of some parameter p, our pre-
understanding is expressed by a more complex probability model (Definition 5 on page 281)
for the values of p. This probability model is called the prior distribution for p.
• After we have collected our data, we have some new evidence E. We use the evidence to update
our understanding of the probability model for p. The new model, taking the evidence into
account, is called the posterior distribution for p.
• The posterior distribution is not a fixed and final thing. As yet more evidence comes in, that
posterior distribution can be “recycled” and used as the prior distribution for another updating
process. In this way, our understanding of the correct value of p, and of the precision with which
it can be claimed, is continually being refined by taking into account all the evidence we have
accumulated so far.
26 If you do so, you will learn that statisticians are divided into two philosophical camps: frequentist and Bayesian. Both
groups obtain similar conclusions in practical problems, but differ about the justification for their methods. In keeping with the
rest of the chapter, the approach presented here is fundamentally Bayesian.
27 We already used this more general notion of “parameter” in Section 5.3.5.
Prior distribution for p

(State of our knowledge before experiment)
Evidence
Model that uses
(results of experiment
parameter p
or observations)
Update
Posterior distribution for p

(State of our knowledge after experiment)
Figure 26: The process of (Bayesian) statistical inference.
The process is summarized in Figure 26. Using our pre-existing knowledge (the prior distribution)
together with our model, we arrive at a prediction, which is then compared with the evidence (the
results of experiment or observation). The result is a new distribution for p, the posterior distribution,
which combines both items—our prior knowledge and the new evidence—in a mathematically
rigorous way. The blue dashed arrow at the left shows how our knowledge is continually improved:
when additional evidence arrives, the posterior distribution that we have previously calculated
becomes the prior for another cycle of updating using the additional evidence.28
Of course a diagram like this is all very well, but the key question is what happens inside the red
circle labeled “Update”? How are we supposed to put the evidence and the predictions of our model
together? But in fact, that is not so hard. Our discussions in the earlier part of the chapter have already
shown us the way.
Example 5: First let’s review an example from Section 5.4.2 from this perspective. Consider, for
instance, Problem 4 on page 325, where we want to test a site for radioactive contamination. The
parameter p that we want to investigate is in this case a “yes or no” one: is the test site contaminated?
In the language of Section 5.4.2 we called this our hypothesis. The prior distribution was our
knowledge that a randomly chosen site has a 4% chance of being contaminated (P(H) in the notation
of Section 5.4.2), and our model was that our test site was a randomly chosen one. Our evidence E was
a test result indicating contamination. We used a two-way frequency table to combine this evidence
with the prior information to get an updated estimate, P(H|E) ≈ 12%, that our site was contaminated.
This updated estimate is what we are calling the posterior distribution.
Let’s try applying similar ideas to a slightly more complex problem.
Problem 6: Your friend hands you a bag containing three pennies. They tell you that one of the three
pennies is fair, one of them is biased (coming up heads 65% of the time), and one is a trick coin with
28 There is an important condition that must be satisfied here: the newly obtained evidence should be genuinely new
information, that is, it should be independent of the evidence used in the previous updating cycle or cycles. When this condition
is not satisfied a more complicated updating process needs to be used,
two heads (so that it comes up heads every time). You draw a coin at random from the bag and flip
it five times. Each time, it comes up heads. What is the probability that the coin you drew is the trick
one?
Solution: We can solve this problem using the same technique as we used before, a frequency table.
Now, however, the frequency table will have 3 columns instead of 2, corresponding to the 3 possible
“parameter values” that we are trying to choose among: the fair coin, the biased coin, and the double-
headed (trick) coin. We’ll call these three possibilities hypotheses H1 (fair), H2 (biased), and H3
(double-header). Our evidence E is the occurrence of five heads in a row.
The prior distribution is provided by the fact that we drew a coin randomly from the bag. So, before
carrying out the experiment, the hypotheses H1 , H2 , H3 were equally likely. Each had probability
1
3 ≈ 0.33.
Now we must figure the probabilities P(E|Hi )—how likely the evidence that we observed (five
heads in a row) would be on each of the given hypotheses. One calculation is easy: given hypothesis
H3 , the five heads would be certain to occur, that is, P(E|H3 ) = 1. The other probabilities can be
figured using the multiplication rule (Rule 6 on page 292), keeping in mind that the 5 flips will be
independent events:
P(E|H2 ) = 0.65 × 0.65 × 0.65 × 0.65 × 0.65 = 0.655 ≈ 0.116,
P(E|H2 ) = 0.5 × 0.5 × 0.5 × 0.5 × 0.5 = 0.55 ≈ 0.031.
Now let’s build our frequency table. Imagine N = 3,000 repetitions29 of the whole process. Because
of the prior distribution, in 1, 000 of them we draw the fair coin, in 1, 000 we draw the biased coin,
and in 1, 000 we draw the double-header. Using the probabilities we have just calculated we can fill in
the frequency table (Table 17 on the next page).
Table 16: Biased Coin Problem

Coin Chosen Fair (H1 ) Biased (H2 ) Trick (H3 ) Total
Five Heads (E) 31 116 1,000 1,147
Not Five Heads (E c ) 969 884 0 1,853
Total 1,000 1,000 1,000 3,000
Each column of the table must contain 1, 000 outcomes. In the first column, for example, we know
that 31 = 1, 000 × 0.031 of those 1, 000 outcomes should be five heads, leaving 1, 000 − 31 = 969 for
the complementary outcome. We calculated the second and third columns in the same way.
Now to answer the question. We know that five heads—the evidence E—actually occurred, so we
are restricted to the first row of the table, which contains 1, 000 + 116 + 31 = 1, 147 equally likely
outcomes. Of these, 1, 000 are consistent with hypothesis H3 (you drew the trick coin). Thus
1, 000
P(H3 |E) = ≈ 0.872.
1, 000 + 116 + 31
The chance that you drew the trick coin (on the basis of the evidence) is about 87%. In the same way
we can calculate
31 116
P(H1 |E) = ≈ 0.027, P(H2 |E) = ≈ 0.101.
1, 000 + 116 + 31 1, 000 + 116 + 31
These three figures together give us the posterior distribution, as shown in Table 17 on the next page.
29 As in Section 5.4.2, the specific value of N does not matter—it will cancel out in the end—but it is helpful to think of a
large number.
Table 17: Biased Coin Problem: Posterior Distribution

Coin Chosen Fair (H1 ) Biased (H2 ) Trick (H3 )
Probability 0.027 0.101 0.872
Notice that the sum of the three probabilities in the posterior distribution is 0.027 + 0.101 + 0.872 = 1,
as we expect from any probability model (Rule 1 on page 282).
It looks as though the evidence is pretty convincing in favor of Hypothesis H3 (you drew the trick
coin). But what if some additional evidence turns up? Our model can handle that too.
Problem 7: You perform the experiment of the previous problem 6 on page 330 and obtain five heads
in a row, as stated. Solving the problem in the way explained above, you deduce that there is an
87% chance that you drew the trick coin. Just to be sure, though, you flip the coin that you drew one
more time. This time it comes up tails! What is your new estimate for the posterior distribution (the
probabilities of H1 , H2 , and H3 )?
Solution: One thing is for sure: you didn’t draw the double-headed coin! So, from leading the race,
Hypothesis H3 is going to drop back to last place—in fact, to zero probability. To compute in more
detail we just need to use the same technique as before—a frequency table.
Our evidence E now is a single flip of Tails. Our prior distribution is our best knowledge before that
new, independent evidence came in—that is, the same as the posterior distribution from the previous
problem (remember the blue dashed arrow in Figure 26?). Thus, if we imagine N = 10, 000 repetitions,
0.027 × 10, 000 = 270 of them will belong to H1 , and similarly 1, 010 will belong to H2 , and 8, 720
will belong to H3 . Now, on the basis of Hypothesis H1 , half (0.5 × 270 = 135) of the 270 outcomes
in its column will be tails (consistent with the evidence E) and half will be heads (not consistent with
E). Similarly, in the column for H2 , 0.65 × 1, 010 ≈ 657 will be heads (not consistent with E) and
1, 010 − 657 = 353 will be tails (consistent with E). The resulting frequency table is shown below
(Table 18).
Table 18: Biased Coin Problem After New Evidence

Coin Chosen Fair (H1 ) Biased (H2 ) Trick (H3 ) Total
Flip Tail (E) 135 353 0 488
Flip Head(E c ) 135 657 8,720 9,512
Total 270 1,010 8,720 10,000
Since we know that the evidence E (a single flip of Tails) did in fact occur, our posterior distribution
is obtained by selecting among the equally likely outcomes in the first row only:
135 353
P(H1 |E) = ≈ 0.277, P(H2 |E) = ≈ 0.723, P(H3 |E) = 0.
135 + 353 135 + 353
This updated posterior distribution is shown in Table 19 on the opposite page. Based on the new and
old evidence—all the evidence accumulated so far—Hypothesis H3 is definitely rejected, and of the
other two, H2 (you drew the biased coin) has an approximately 72% chance and H1 (you drew the fair
coin) has an approximately 28% chance. This is a dramatic example of how new evidence can change
our understanding!
Table 19: Biased Coin Problem: New Posterior Distribution

Coin Chosen Fair (H1 ) Biased (H2 ) Trick (H3 )
Probability 0.277 0.723 0.0
Example 6: Here is an example that is specifically about the measurement process in science. Before
we dive in, let’s say clearly that a student in this course is not expected to be able to solve problems of
this sort “from scratch.” The purpose of the example is to show how the model illustrated in Figure 26
on page 330 can be applied to everyday scientific questions, such as estimating the true value of some
quantity based on a series of measurements that may involve random errors or “noise.”
We want to estimate the concentration of arsenic in a reservoir that supplies drinking water to the
local community (compare Problem 5 on page 38). By “estimate,” we mean not only to give a single
number for the arsenic concentration, but also a good idea of the precision of the measurement. We
take nine samples and analyze them, obtaining the following results (expressed in micrograms per
liter):
4.2, 5.9, 5.7, 5.2, 2.6, 6.1, 5.8, 5.3, 4.5.
This list of results is our evidence, E. We assume that the measuring technique used is known to have
an accuracy of about ±1.3. More technically, our assumption is that the result of a measurement is
a random variable having a normal distribution (Section 5.1.3) whose mean is the true value of the
concentration and whose standard deviation is 1.3 micrograms per liter. Given this information, we
want to estimate the true value of the concentration and give reasonable limits on the precision of our
estimate.
Here’s how we could go about solving this problem. (In practice, these steps would be carried
out automatically by computer software for data analysis.) We observe that the parameter p that we
want to estimate is the concentration of arsenic in the reservoir water. Our model is also described
explicitly: the various observations will be independent samples from a normal probability model
with mean equal to p and standard deviation 1.3.
(a) We start by selecting a prior distribution for p. Remember, the prior distribution is supposed to
reflect the state of our knowledge before any measurements have been taken. In this case, we
don’t know anything at all before the first measurement, and it seems reasonable to reflect our
ignorance by choosing as our prior distribution one that makes all the values of p between 0
and, say, 10, equally likely.30
(b) Now we subdivide the range from 0 to 10 into a number of class intervals. In our example
we will use 100 class intervals,31 each of width 0.1, so that the hypotheses we are testing
have the form “H1 : p lies between 0 and 0.1” through “H100 : p lies between 9.9 and 10.” Our
prior distribution is now a uniform probability model with outcomes H1 , . . . , H100 , each having
probability 0.01.
(c) For each of these 100 class intervals—or 100 hypotheses, if you like—we calculate the
probability P(E|Hn ), that is, the probability that that particular value of p would give rise to
the evidence E that we see. Fortunately, this is something the computer will do for us.
30 There are, in fact, excellent theoretical reasons for choosing this so-called uniform prior in the present situation. This
theory also makes it clear that the upper limit of 10 does not matter—we could choose any upper limit L as long as the range
from 0 to L was large enough to accommodate all the measurements with several standard deviations left over.
31 The more we use, the greater the accuracy. In fact, cutting the range of values for p into intervals like this is forced on us
by our decision to work only with finite sample spaces. With fancier mathematics, we could cope with “the whole infinite range
of possible values from 0 to 10” as a sample space. But our work-around, using a large but finite number of class intervals, is
fine for an example.
15 15
10 10
Probability (%)
Probability (%)
5 5
0 0
3 4 5 6 7 3 4 5 6 7
(a) Posterior distribution for arsenic concentration after nine (b) Posterior distribution after updating the distribution of part
initial observations (Example 6 on the previous page). (a) (now used as the prior) by nine further observations
(Example 7 on the opposite page).
Figure 27: Posterior distributions for arsenic concentrations.
(d) Using these probabilities, we compute a very long and thin frequency table: 2 rows and 100
columns. The columns are labeled by our 100 hypotheses; the total number of outcomes in each
column is the same (because we assumed a uniform prior distribution, as in (a) above); the
outcomes in the nth column are divided between the E and E c rows according to the probability
P(E|Hn ) that we calculated in step (c) above.
(e) Finally, since we know that the evidence E was actually observed, we can look across the
equally likely outcomes in the first row of the giant frequency table to find the posterior
distribution of p: the probability, given the observed evidence, that p falls into each of the
class intervals.
We’d like you to notice two things from this brief description of the estimation process. First, it is
clear that what we are describing is a complicated calculation, which we are glad to have software to
carry out! Second, however, the overall structure of the calculation is the same as that in the biased
coin problems, Problems 6 on page 330 and 7 on page 332, which were simple enough to analyze by
hand.
The result of the calculation is shown in Figure 27(a). For each class interval, this shows the
posterior probability that the arsenic concentration p falls into that class interval, given the evidence
E that we have accumulated. In other words, the distribution given by this histogram describes our
best available information about the concentration p on the basis of the evidence E.
If we want to reduce this information to one or two useful numbers, we might apply the descriptive
statistics of Section 5.1 to summarize this distribution. For example, the mean of the distribution is
approximately 5.0. It would be reasonable, then, to say that this is our best estimate of the arsenic
concentration. The probability (according to the posterior distribution) that the concentration is less
than 4.2 is approximately 2.5%, as is the probability that the concentration is greater than 5.8. Thus
we can say that “there is a 95% probability that the true value of the arsenic concentration lies between
4.2 and 5.8.” This statement is analogous to (b) in Example 10 on page 297; it specifies the precision
of a measurement claim and attaches a level of assurance to that precision.
Remark 3: We could, if we wished, ask for a greater level of assurance—say 99% rather than 95%.
Then we would have to accept a lesser level of precision. In fact, we could say that “there is a 99%
probability that the true value of the concentration lies between 3.9 and 6.1.” Ranges of precision with
associated assurance levels, like this, are technically called credible intervals, but we won’t make use
of this technical terminology.
Example 7: In the situation of Example 6, we carry out nine more measurements and obtain the
results
5.5, 5.1, 5.3, 5.6, 3.6, 6.2, 4.4, 4.8, 6.6.
What now should be the posterior distribution?
Recalling the solution to Problem 7 on page 332, what we should do now is to use the posterior
distribution shown in Figure 27(a) as our new prior distribution when evaluating this new evidence.
If we do that and let the computer grind away again, we obtain the results shown graphically in
Figure 27(b).
What changes do you notice? The mean of the posterior distribution (our best estimate of the true
concentration) has shifted a little as a result of the new evidence—it is now about 5.1 rather than 5.0.
More obvious, though, is the fact that this new posterior distribution is significantly narrower than
the old one. In fact, the 95% credible interval now runs approximately from 4.5 to 5.7, a width of 1.2
units (micrograms per liter), in contrast to the previous credible interval (based only on the initial nine
measurements), which had a width of 1.6 units. This is an example of a general principle.
Rule 2: Sampling Rule
When taking a series of measurements to estimate some parameter, the more

independent measurements that are taken, the narrower the posterior distribution will
be. In many cases, the “spread” of the posterior distribution will be approximately
1
proportional to √ , where N is the number of observations taken.
N
We can see this rule in action in the examples above. Doubling the√ number of observations reduced
the “spread” from 1.6 to 1.2. The ratio, 1.6/1.2, is pretty close to 2 ≈ 1.4.
Example 8: The process that we have outlined is one for obtaining a data-based estimate—with a high
level of assurance—of the range of values within which some parameter lies. Suppose, for example,
that the result of our sampling is that with an assurance level of 99 percent, the arsenic concentration
in the lake is between 0.1 and 0.3 micrograms per liter. Since the value 0 does not lie in this range, it
would be correct to say that with a high level of assurance, arsenic has been found in the lake.
Now imagine how this information may be reported in the local newspaper. “Scientists ‘sure’
that arsenic is present in local lake!” Scandal ensues. But recall that (according to the EPA, whose
estimates are usually quite conservative) a safe level of arsenic in drinking water is 10 micrograms
per liter. The level of arsenic that the scientists have detected in the lake “with high assurance” is fifty
times smaller than that. The point is that a “high assurance” of a nonzero effect is not any kind of
assurance that it is an effect we should actually be worrying about. For that, one needs to know more
about the numbers themselves.
Example 9: Here is a real-life example of this sort of confusion. In October 2015, the World Health
Organization announced that it was listing processed meats like bacon and hot dogs among the “Class
I carcinogens,” those substances for which a highly assured link has been established between them
and cancer in humans. This led to some alarming headlines like this one from a British newspaper:
UN health body says bacon, sausages and ham among most carcinogenic substances
along with cigarettes, alcohol, asbestos and arsenic.
No doubt this headline helped sell newspapers, but it is sensationalist and irresponsible. The effect of
regular smoking on cancer incidence is about a thousand times stronger than the effect of regularly
eating processed meats; the fact that the two effects are both nonzero does not make them of equal
magnitude. The New York Times did a much better job:
The report placed processed meat into its Group 1 category, which means the panel found
“sufficient evidence” that it could cause cancer. While other substances in this group
include alcohol, asbestos and tobacco smoke, they do not all share the same level of
hazard.
Critical Thinking
In your reading you may encounter the phrase “statistically significant.” We’ve avoided
using this language because its technical interpretation belongs to the frequentist model of
statistical inference rather than the Bayesian one that we have been sketching. However,
its meaning in practice is something like the phrase “nonzero with high assurance” that
we were looking at above. In particular, when you hear that something is “statistically
significant,” do not jump to the conclusion that it is therefore practically significant, in
the sense that it will make a difference in day-to-day life. It may do so, but that is a separate
question. Try to find out the actual size of the reported measurement or effect and compare
it with the sizes of other measurements in the same class of things, or (if possible) with
warning levels set by reputable authorities, like the EPA in Example 8 on the previous page.
Summary of Ideas: Assimilating New Information
• The conditional probability P(A|B) is the probability of some event A when we

know for certain that some other event B has occurred or will occur. It is given by the
formula
P(A ∩ B)
P(A|B) = .
P(B)
• The conditional probabilities P(A|B) and P(B|A) are different. In the formula above,
they have the same numerator but different denominators.
• In statistical inference, we try to understand a variable that is not directly observable
(the subject of inference) based on evidence such as test results. When the subject
of inference is an event (that may or may not occur) it is often called the hypothesis.
• A false positive occurs when a yes/no test returns a yes answer when the correct
answer is no. A highly specific test will produce few false positives.
• A false negative occurs when a yes/no test returns a no answer when the correct
answer is yes. A highly sensitive test will produce few false negatives.
• Statistical inference can be applied to parameters that are real numbers (such as
the concentration of a pollutant), as well as to yes/no hypotheses. The result of such
inference may be expressed with appropriate precision as a range that, according to
the evidence, has a high probability of including the correct value of the parameter.
6. The following numbers represent data on con-

5.1. UNDERSTANDING DATA centrations of particulates (a form of air pollution) in
the air on a winter day in various locations around
1. Consider the set of numbers below. Calculate the Fairbanks, Alaska. The units are micrograms per cubic
median and the quartiles, and determine whether there meter:
are any outliers:
6, 14, 15, 25, 12, 20, 3, 18, 17, 45, 15.
26, 5, 27, 12, 22, 25, 26, 23, 29, 23, 30, 40, 33, 26.
Find the median and the upper and lower quartiles.
Draw a box plot summarizing the distribution. Calculate the IQR and identify any outliers.
2. The daily high temperature records, in degrees 7. Draw a box plot for the data in the previous
Fahrenheit, for a location in Central Pennsylvania dur- problem.
ing the month of May are given below.
8. The following data show the number of claims paid
67, 63, 69, 72, 79, 80, 74, 68, 73, 75, 72, 69, by the National Flood Insurance Program by year (in
59, 61, 72, 73, 69, 66, 82, 73, 66, 70, 71, 76, thousands) in the years 2004 through 2014:
75, 71, 68, 71, 76, 77, 77.
56, 214, 25, 23, 75, 31, 29, 78, 150, 18, 14.
Find the mean and the interquartile range, to 1 decimal
place. Sort the data on number of claims, find the median,
calculate the interquartile range, and find any outliers.
3. Construct a box plot for the previous problem.
9. Make a histogram for the data of the previous
4. Consider the following data set of per capita problem.
emissions of carbon dioxide by country for 2013 (in
metric tons per year). Calculate the interquartile range 10. Refer to the “mock” data set available at http://
and find any outliers.32 math-for-sustainability.com/sample-data,
which gives some information about the members of
Nation CO2 an imaginary class.
Austria 7
(a) Find the mean GPA of the class, to 1 decimal
Bangladesh 0.4
place.
China 8
Iceland 6 (b) Find the median GPA of the class, to 1 decimal
Philippines 1 place.
Poland 8 (c) Find the range of the height variable, to 1
Slovenia 7 decimal place.
South Africa 9
(d) Consider the “number of cents” variable. How
Ukraine 6
many outliers for this variable does the data set
United Kingdom 7
contain?
United States 16
Venezuela 6
11. Suppose the data set below represents temper-
ature readings from sites throughout New Hampshire
5. The bar chart in Figure 28 on the next page shows
during the afternoon of an October day:
the death rate (per 100,000 people) grouped by age for
the year 2004 in the United States. Describe the skew 32, 38, 45, 28, 37, 45, 37, 31, 47, 40
of the distribution and explain why you might expect 32, 41, 37, 41, 41, 39, 38, 39, 39, 42
this distribution to be skewed in this sort of way. 35, 40, 40, 34, 37, 40, 39, 35, 40, 36
39, 34, 35, 37, 38, 43, 36, 35, 38, 41.
32 You can find the complete data set on our website at http://math-for-sustainability.com/emissions-data.
Deaths per 100,000 in group

15,000
10,000
5,000
<1 1-4 5-14 15-24 25-34 35-44 45-54 55-64 65-74 75-84 85+
Age at death (in years)
Figure 28: Bar chart for Exercise 5 on the previous page. U.S. rates of death by age group, 2004. Data obtained from [90].
Original source: [117].
Represent this data in the form of a histogram, using an 17. If two fair 6-sided dice are rolled, what is the
appropriate number of class intervals. probability that at least one die shows a 6?
12. Show that there are no outliers in the data set of 18. Referring to the the previous problem, describe
Example 6 on page 271. Plot a histogram for it. the complement of the event “At least one die shows a
6.” Calculate the probability of this event.
13. The data set below gives the mean extent of Arctic
sea ice, in millions of square kilometers, for successive 19. A fair coin is to be tossed 3 times. Write out the
Novembers from 1981 until 2015: sample space for this random process.
11.21, 11.96, 11.68, 11.32, 11.42, 11.81,
11.59, 11.73, 11.57, 11.37, 11.16, 11.91, 20. If a fair coin is tossed 3 times, what is the
11.78, 11.35, 11.05, 10.61, 10.94, 10.79, probability that at least one of the tosses is Heads?
11.03, 10.58, 10.94, 10.8, 10.32, 10.66,
10.51, 9.88, 10.08, 10.64, 10.26, 9.89,
10.03, 9.94, 10.26, 10.28, 10.11. 21. Referring to the previous exercise, describe the
complement of the event “At least one of the tosses is
Find the median, mean, and interquartile range, and Heads.” Calculate the probability of this event.
draw a box plot for this data set.
22. If a fair coin is tossed 3 times, what is the
14. The Arctic sea ice extent in November 2016 probability that at least two of the tosses are Heads?
was 9.08 million square kilometers. Compared with
the data from the previous 35 years (Exercise 13), is
23. Referring to the previous exercise, describe the
this an outlier? Is it an outlier compared with the data
complement of the event “At least two of the tosses are
from the 20-year period 1981–2000? Comment on any
Heads.” Calculate the probability of this event.
inferences you might draw.33
24. What is the probability that a randomly chosen

5.2. PROBABILITIES AND PREDICTIONS human being lives in the United States?
15. Two fair 6-sided dice are to be rolled. Write out 25. The board game Discworld: Ankh-Morpork uses
the sample space for this random process. a 12-sided (dodecahedral) die. Unfortunately, I lose the
special die, but I still want to play the game. Someone
16. If two fair 6-side dice are rolled, what is the suggests just using a pair of ordinary (6-sided) dice
probability that the dice show the same number? instead, and then adding their scores. What do you
think of this suggestion? Can you think of a better way
33 Links to Arctic sea ice data and further discussion can be found on our website at
http://math-for-sustainability.com/sea-ice-data.
to replicate the behavior of a fair 12-sided die using a is a face card.” Are the events A and B independent or
pair of ordinary dice? not? Explain.
26. You read in a book on poker that the probability 33. Revisit Exercise 32 and suppose that after the first
of being dealt three of a kind in a five-card poker hand draw the card is not replaced. Are the events A and B
is 1/50. Explain the meaning of this statement. independent or not? Explain. Is your answer different
from the answer given in the previous exercise. If so,
27. A survey of students at a certain college found why?
that 61 percent had not exercised at all in the four
days leading up to the survey, and that 17 percent had 34. Revisit Exercises 32 and 33. In each case calcu-
exercised exactly once in those four days. Find the late P(A ∩ B).
probability that a randomly chosen student exercised
twice or more during the four-day period. 35. According to the Centers for Disease Control,
about 128,000 people are hospitalized each year for
28. With the data from the previous exercise, find the foodborne illnesses; 3,000 of these cases are fatal.
probability that two randomly chosen students totaled Consider the events “Marge in Montana is hospitalized
two or more workouts between them during the four- for a foodborne illness” and “Nick in New Jersey is
day period. hospitalized for a foodborne illness.” Give reasons why
these events may not be independent.
29. You roll two fair dice and write down the two
scores that you obtain. Then you toss a fair coin to 36. The United Nations Environment Program sur-
determine how to combine the scores: if the coin lands veyed 63 developing countries regarding environmen-
heads, you add the two scores, and if it lands tails, you tal risks. A list of major environmental risks was
multiply them. What is the probability that the result of presented and governments were asked to identify
this process is the number 9? which of those risks they considered to be significant
problems in their own country. For example, out of
the 63 countries surveyed, 55 (87 percent) listed “soil
30. Abby, Boaz, Carmen, Dani, and Eduardo work
degradation - erosion and loss of fertility” as a prob-
in the same office. Their employer must choose two
lem that their country experienced; 22 countries (35
of them to attend a conference in Paris. To avoid
percent) listed “inadequate domestic water supply” as
unfairness, the choice will be made by drawing two
a problem. What is the probability that a randomly
names from a hat. What is the probability that neither
selected country experiences both problems? Describe
of the two men (Boaz and Eduardo) is chosen?
any assumptions that you make and comment on their
plausibility.
31. Marcus is a student who has an 8 a.m. class. There
is a 70 percent chance that Marcus will remember to set 37. Suppose that 10,000 barrels of nuclear waste are
an alarm before he goes to bed the previous evening. If housed in a storage facility. Each barrel has a 1 in
he does not set an alarm, there is a 15 percent chance 100,000 chance of developing a leak in a given year. If
that he will wake up and go to class anyway. If he does leaks in different barrels are independent events, what
set an alarm, Marcus wakes when the alarm goes off is the probability that at least 1 barrel develops a leak
and tosses two fair coins. If they both land heads, he over a 20 year period?
goes back to bed; otherwise he goes to class. What is
the probability that Marcus makes it to his class in the
morning?
5.3. EXPECTATIONS AND PAYOFFS
32. Consider a standard deck of 52 playing cards. A 38. A random variable S is defined based on the result
card is randomly drawn. The card is replaced, the cards of two independent coin tosses, as follows: S = 100 if
are re-shuffled and a second card is randomly drawn. both land Heads, S = 25 if only one lands Heads, and
Event A is: “The first card drawn is a face card (Jack, S = −140 if neither lands Heads. Find E(S).
Queen or King).” Event B is: “The second card drawn
39. A biased coin lands Heads with probability 0.45 people will show up and buy tickets. On the day of the
and Tails with probability 0.55. Let X be the number fundraiser there is a 28% chance of rain. What are the
of heads obtained in 100 flips of the coin. Find E(X). expected ticket receipts?
(Hint: Let Y1 , Y2 , and so on equal 1 if the coin lands
Heads on the first flip, second flip, and so on, and 0 if it 46. I want to travel from my home town to San
lands Tails. Then X = Y1 + · · · +Y100 . Use the linearity Francisco. Three air carriers compete for my business:
rule (Rule 2 on page 307). Epsilon Airways, JetGreen Airways, and Untied Air-
ways. On the way, one or more of the following events
may happen to me, which will increase or decrease my
40. A gambling game works as follows. You pay $5
satisfaction with the journey as follows:
to play. Then, a die is rolled and a coin is tossed. If the
• Flight canceled: −15 units of satisfaction.
coin lands Heads, your payoff is the number of dollars
rolled on the die (from 1 to 6). If the coin lands Tails, • Baggage lost: −5 units of satisfaction.
you receive a double payoff ($2 if the die rolled a 1, • Cramped seat: −2 units of satisfaction.
and so on). Would you play this game? • Free upgrade: +4 units of satisfaction.
From experience I know that the probabilities of these
41. Consider the following gambling game. You pay events on the different airlines are
$1. Two fair coins are flipped. If they both come up Epsilon JetGreen Untied
heads, you win $4. If they both come up tails you win Canceled 0.1 0.08 0.05
your dollar back. If one is heads and the other is tails, Lost bag 0.05 0.08 0
you lose. Calculate the expected value of playing this Cramped 0.1 0.2 0.3
game. Upgrade 0.2 0 0.1
Which airline should I choose to maximize my ex-
42. Let X be a random variable denoting the number pected satisfaction?
(between 1 and 6) obtained by rolling a fair die.
Calculate E(X 2 ) and E(X)2 . Are they the same or 47. Suppose you are considering purchasing a home
different? Comment. in the Miami area; the current market value of the home
is $300,000. You plan to keep the home for 20 years.
43. Ten fair coins are flipped, and a random variable Assume we have the following possible scenarios with
estimated probabilities (these estimates are fictional):
Y is defined to be 1,000 if all ten come up heads and
0 otherwise. What, approximately, is the expectation of • Optimistic outcome (20% chance): Home val-
ues will rise 3% per year (80% increase in value
Y?
over 20 years).
• Middle outcome (25% chance): Home value
44. In 2012, Hurricane Sandy inflicted an estimated will rise 1% per year (22% increase in value
$32 billion of damage on New York City and sur- over 20 years).
rounding areas. The probability of a hurricane of this
• Pessimistic outcome: (30% chance): Due to in-
or greater intensity striking the New York area in creasing risks from sea level rise home values
any given year is estimated to be 1.6%. Let D be a decline by 40%.
random variable representing the dollar value of the • Natural Disaster: (20% chance): Home is de-
total hurricane damage in the New York area in any stroyed by natural disaster; insurance pays 75%
given year. What (if anything) can be said about E(D)? of original market value.
(The question does not provide sufficient information • Market Collapse (5% chance): Home value
to evaluate E(D) exactly. It asks whether you can say plummets to 30% of original market value after
anything at all about E(D) on the basis of the given natural disaster strikes nearby neighborhoods.
information. For example, is it greater than or less than What is the expected value of the home 20 years from
a certain value?) now?
45. The Student Sustainability Circle plans to host 48. The following table gives hypothetical estimates
an outdoor fundraiser. The tickets for the event cost for the annual damage to the U.S. economy from
$1.60 each. If the weather is not rainy, then 500 people climate change, depending on the rise in mean global
will show up and buy tickets; if the weather is rainy, 80 temperature, and also the estimated probability that the
temperature rise will be in the given range by the end Redo your calculations. Would you now recommend
of the century. (These numbers are partly based on approval of the dam proposal?
estimates from published research, and depend on the (b) You have recalculated as in (a) and still decide
policy goals of various nations; see, for example, [154]. to reject the proposal. The Corps of Engineers comes
Note that some losses—such as species loss—are not up with a new design, which will resist up to a five
considered in most estimates because of the difficulty hundred year flood (a “super extreme” event). Assume
of putting an economic value on them. We also note that the loss rates and probabilities without the dam are
that these costs are usually reported as a percentage of the same as in (a), but that the new dam will cost four
gross domestic product (GDP) and the numbers below times as much as the previous proposal and will lead
are based on the current GDP.) to a $50 million loss if it fails (with annual probability
Using this information, compute the expected value 0.2%). What is your recommendation now?
of the annual damages from climate change to the U.S. While there is one right answer to each calculation in
economy at the end of the century. this question, there are no right or wrong answers to the
“what would you do?” questions, which are intended to
rise in global Annual Probability
help guide you in exploring your own thoughts about
mean temp cost to U.S.
risk aversion or risk tolerance.
< 1.5 ◦ C $90 billion 50%
1.5 - 2 ◦ C $180 billion 15%
2-3 ◦ C $320 billion 25% 52. Home values have been rising in South Florida
> 3 ◦C $750 billion 10% since 2010. A recent Bloomberg article [114] high-
lights homeowners who are selling their homes out of
concern that home values will eventually decline due
49. It has been estimated that the probability that
to the risks of sea level rise and climate change. One
a newly opened restaurant will fail during its first
seller featured in the article characterized the current
three years of business is 0.6. If you are an investor
market as “pessimists selling to optimists.” Discuss this
proposing to put $1 million into a new restaurant, what
in terms of risk aversion.
annual profit share do you need to receive from the
restaurant to convince you to invest in it rather than in
government bonds yielding an interest rate of 2% with 53. Is risk aversion rational? Discuss.
no risk? Assume that you want your expected return
over three years to be equal to what you would get from
the bonds (no risk aversion), and that if the restaurant 5.4. ASSIMILATING NEW INFORMATION
fails, you will recover only 50% of your capital.
54. A fair (six-sided) die is rolled. Let A be the event

50. The behavior of a highly risk-averse individual is
modeled by the utility function “the result is an even number.” Let B be the event “the
result is less than or equal to 3.” Find the conditional
U(x) = log(10 + x), probability P(A|B).
where x is the financial gain (in dollars) and U(x) 55. Two fair 6-sided dice are rolled: one red and
is the resulting utility. Suppose that this individual one white. You are told that each die shows a number
is offered a choice between a 60 percent chance of greater than or equal to 4. Given this information, what
gaining $1,000 and a for-certain payment of $100. is the probability that the numbers showing add up to
Which will they choose? Which would you choose in 10?
the same circumstance?
56. Three fair coins are flipped. Find the probability

51. Imagine that your community is considering the that all three come down heads, conditional on knowing
dam proposal described in Problem 9 on page 310.
that at least one of them comes down heads.
Because of your risk aversion, you decide to reject the
original proposal.
(a) You learn that because of climate change, the 57. A fair coin is flipped three times. Find the proba-
estimated probabilities of low. medium, high, and ex- bility of all three flips coming down heads, conditional
treme precipitation have been revised to 9 percent, on knowing that the first flip comes down heads.
22 percent, 67 percent, and 2 percent respectively.
58. Remember Marcus from Exercise 31 on result of the program?” Express this absolute
page 339? He’s the student who has an 8 a.m. class. risk reduction as a percentage probability.
There is a 70 percent chance that Marcus will remem- (d) Express the answer to (c) as a percentage of the
ber to set an alarm before he goes to bed the previous answer to (a). This is the relative risk reduction
evening. If he does not set an alarm, there is a 15 resulting from the program: it answers the ques-
percent chance that he will wake up and go to class tion, “By what percentage does the screening
anyway. If he does set an alarm, Marcus wakes when program reduce a participant’s chance of death
the alarm goes off and tosses two fair coins. If they in the next ten years?”
both land heads, he goes back to bed; otherwise, he (e) Imagine you are designing a website to encour-
goes to class. Suppose that on Monday morning, we age women to get screened for breast cancer.
observe Marcus heading to his 8 a.m. class. What is the Would you report the benefits of screening in
probability that his alarm went off this morning? terms of absolute risk reduction (c) or relative
risk reduction (d)?
59. Revisit Table 12 on page 324 in the solution Many health organizations speak in terms of relative
to Problem 3 on page 323, which shows outcomes risk reduction because, as you can see in the examples
related to breast cancer screening in 10,000 women. above, it yields much more dramatic numbers. But
Suppose that someone from this sample undergoes studies show that few people understand relative risk
mammography screening and tests negative. What is figures correctly. Many confuse relative and absolute
the probability that despite the negative mammogram, risk reductions, believing (for instance) that a relative
she does in fact have cancer? risk reduction of 50% means that 50 lives will be saved
for every 100 people screened. If a disease is rare,
60. Look again at Table 12 on page 324 of breast a large relative improvement in risk can make little
cancer screening outcomes. Cancer screening does not, absolute difference. See [128] for more about this.
by itself, treat disease: it allows for the possibility of
early treatment by detecting disease that would not
61. A rapid HIV test has a sensitivity of 97.7%,
otherwise have been recognized.
Make the following assumptions: a patient who has meaning that if a person has HIV, it will return a
cancer that is not detected by screening has a 50% positive result 97.7% of the time. It has a specificity
chance of dying of cancer in the next ten years; a of 90.7%, meaning that if a person does not have
patient who has cancer that is detected by screening HIV, it will return a negative result 90.7% of the time.
(and who follows up with treatment) has a 10% chance Researchers would like to use this test in a region in
of dying of cancer during the same period; and a “false sub-Saharan Africa where 4% of the population has
positive” patient (no cancer, but positive mammogram) HIV. If someone tests positive for HIV, what is the
has a 0.3% chance of dying during that period from side chance that they actually have the disease?
effects of unnecessary treatment. Answer the following
questions.
62. BSE (bovine spongiform encephalopathy) is a
(a) Out of 10,000 who are not screened, how many
prion disease of cattle commonly known as “mad cow
are expected to die of (breast) cancer in the
disease.” It is thought to be spread through infected
next 10 years? What is the probability (as a
percentage) that an unscreened patient will die meat products and to be potentially transmissible to
of breast cancer? humans. Assume that a test for BSE has a sensitivity
of 98.5% (that is, an infected cow has a 98.5% chance
(b) Out of 10,000 who are screened, how many are
of testing positive) and a specificity of 99.97% (an un-
expected to die of (breast) cancer or treatment
effects in the next 10 years? What is the prob- infected cow has a 99.97% chance of testing negative).
ability (as a percentage) that a screened patient Imagine a population in which 1 in every 120,000 cows
will die for these reasons? is infected with BSE. If a cow randomly selected from
this population tests positive, what is the probability
(c) Find the difference between the answers to (a)
that it actually has BSE?
and (b). This is the absolute risk reduction
resulting from the screening program: it answers
the question, “Out of 10,000 participants in a 63. Trash pickup at my house is normally on Mon-
screening program, how many are expected to days. When I wake up on a random Monday morning,
be saved from death in the next ten years as a there is a 85% chance that today is pickup day, and a
15% chance that the schedule has changed this week

for some reason that I have forgotten. 67. Suppose there is a test for a rare viral infection
If today is trash pickup day, there is a 95% chance
that is currently present in approximately one out of
that my neighbor will remember and put her trash out;
every 50,000 people in a certain population. Assume
if today is not pickup day, there is a 15% chance that
my neighbor will forget and put her trash out anyhow. the test for the virus has a sensitivity of 99.5% (an
One Monday morning I look out of the window and infected person has a 99.5% chance of testing positive)
see that my neighbor has left for work without putting and a specificity of 98.1% (an uninfected person has
her trash out. What are the chances that today is, in fact, a 98.1% chance of testing negative). If a randomly
trash pickup day? chosen person tests positive, what is the probability that
they have the infection?
64. For a certain manufacturer of laptop computers,

12% of the laptops it produces are not up to standard 68. Consider the following two-way frequency table
(that is to say 12% of the laptops produced are faulty in describing outcomes of antibiotics when compared
some way—whether or not the manufacturer can detect with a placebo to treat acute bronchitis.
a problem). N = 3496 Antibiotic Placebo
As part of the manufacturing process, the firm Improved (1 week) 1259 1154
performs a diagnostic test for quality assurance. A Did not improve (1 week) 514 569
laptop that fails the test is assumed to be faulty and is
scrapped. A laptop that passes the test is packaged for (a) What is the probability that a patient who im-
sale. proved after 1 week had taken an antibiotic (in
The diagnostic test has a sensitivity of 85% (85% of symbols, P(Took antibiotic|Improved))?
faulty laptops are detected by the diagnostic test) and a (b) Consider the “risk” that a patient will not im-
specificity of 95% (if a laptop is up to standard, there is prove after one week. What is the absolute
a 95% chance it will pass the diagnostic test). risk reduction arising from antibiotic treatment?
What is the probability that a laptop that passes the (See Exercise 60 for the terminology here.)
diagnostic test is not up to standard? (c) What is the relative risk reduction from antibi-
otic treatment?
65. A team of agro-ecologists are studying the in-
fluence of different local microclimates on the pro-
ductivity of winter wheat. They develop a computer 69. Ivynose College is a highly selective private
model that predicts the yield (in tons per acre) of winter school. It admits students if they possess one or both of
wheat in a given location. The model is applied in the following qualifications: (A) outstanding musical
112 different locations, and in each one the ecologists or artistic ability; (B) 99th percentile score on the
record (A) whether the model predicts a yield greater SAT. Assume that these qualifications are independent
than 4 tons per acre and (B) whether the actual harvest among the high school population from which Ivynose
is greater than 4 tons per acre. The results are shown in selects its students. Despite this independence, it is
the table below:
found that A and B are negatively correlated among the
Not A A Ivynose student body: that is, among Ivynose students,
Not B 18 24 P(A|Bc ) and P(B|Ac ) are both close to 1. Explain why
B 30 40 this is so.
(so that, for example, in 24 of the 112 cases the model
predicts a yield above 4 tons per acre, but the actual 70. According to the (very oversimplified) textbook
yield was below this amount). Based on this informa- model of eye-color and inheritance, everybody has one
gene pair for eye color, containing two alleles each of
tion, comment on the effectiveness of the ecologists’
which encodes either “blue” or “brown.” A blue-eyed
model.
person has two blue alleles; everyone else is brown-
eyed. A child inherits one allele randomly from each
66. A math course has two midterm exams; 80% parent.
of students pass the first midterm, 70% of students Assuming this model, which is more likely: that a
pass the second midterm, and 55% pass both. Find the mother has blue eyes given that her daughter has blue
probability that a student who fails the first midterm eyes, or that a daughter has blue eyes given that her
will pass the second one. mother has blue eyes?
reservoir (p) based on nine measurements. The result

71. Doctors studying the effect of hormone re- was a posterior distribution for p (part (a) of Figure 27
on page 334) whose mean, it was stated in the text,
placement therapy (HRT) offered in the 1980s and
is approximately 5.0. In Example 7 on page 335 we
1990s found evidence that those women who chose
HRT seemed to have a reduced risk of coronary heart imagined carrying out nine more measurements, lead-
disease (CHD). Express this statement in the language ing to a new posterior distribution (part (b) of the same
figure) whose “spread” is narrower and whose mean,
of conditional probability.
according to the text, is now approximately 5.1.
They hypothesized therefore that HRT might have a
Compare these numbers with (a) the mean of the first
protective effect against heart disease. A randomized nine measurements (in Example 6 on page 333) and
controlled trial (the scientific “gold standard”) was (b) the mean of all eighteen measurements (from both
carried out to test this hypothesis, and it found the examples together). What do you notice? What other
opposite effect, that HRT slightly increased the risk information does the posterior distribution convey, be-
of CHD. Can you explain this seeming contradiction? sides the mean value?
See [189]. This is a classic example of “correlation (The “coincidence” you notice here is actually a
does not imply causation.” feature of this kind of model. It can be shown math-
ematically, by the same kind of method as we used to
72. Consider Example 6 on page 333, where we “see” the peak flow rate rule (Rule 2 on page 213) using
inferred information about arsenic concentration in a algebra.)
C HAPTER 6
Deciding
You’ve arrived at the final chapter of this book.1 Congratulations! It’s time to take stock of what we
have learned on our journey through the mathematics of sustainability and to look forward quickly at
the issues that we will grapple with as we wrap up.
In the introduction we defined “a sustainable society” (Definition 1 on page vii):
A sustainable society is one that meets the needs of the present generation without
compromising the ability of future generations to meet their own needs.
Does this definition describe our society? In other words,

are we passing on to future generations an Earth system that
is as strong, vibrant, and full of life and resources as the
one that we inherited from our predecessors? So far in this
book we have learned about mathematical tools that help us
give specific, measurable answers to this question—and to the
similar, smaller-scale questions that are embedded in this Big
Question, questions not so much about the sustainability of
“the whole planetary system” as about “our town” or “my
industry” or “this campus” or “my home.” We’ve studied flows
and networks, and we’ve seen how the natural web of life
on Earth appears as a vast web of cyclic flows (powered by
the practically endless energy of the Sun), while many of
the processes on which we have founded our society appear
as one-way flows (from resource to waste). We’ve looked at
the ways that flows and networks can produce change, both
gradual and sudden; and we’ve taken a tour of the way risk Figure 1: The “blue marble.”
can be mathematically assessed and how those assessments are
updated when new evidence arrives. So we have the tools we
need. We have worked to learn how to apply them to specific
problems. Okay, then, how should we act?
That is a slippery question, and the slipperiest thing about it is the shortest word—the word “we.”
Who are “we”? Think about who “we” refers to in the following examples, and what it would involve
for “we” to make a decision and then to act on it.
1 At least of Part I. Remember, the book also contains many case studies in Part II, which give examples of how these ideas
can be applied in real-life situations.

https://doi.org/10.1007/978-3-319-76660-7_6
346 CHAPTER 6. DECIDING
(a) Bachelor in studio apartment, talking to his cat: “What shall we cook for dinner tonight,
Shadow?”
(b) One spouse to another: “What shall we cook for dinner tonight, love of my life?”
(c) Professor in “Math for Sustainability” class, to 30 students: “We need to work together on a
class final project. What shall we do?”
(d) Chair of local planning commission, to open meeting (commission members are present and
can vote, members of the local community can comment and observe but not vote): “Shall we
allow an easement so that Humungous Housing, Inc, can build on this tract of land that was
previously protected from development? We have to decide this tonight for the good of our
community.”
(e) Bystander witnessing an emergency (Section 3.2.2), to herself: “We really should do something
to help here.”
(f) President Obama speaking about climate change at Georgetown University in 2013 [243]:
“We’ve got to look after our children; we have to look after our future; and we have to grow the
economy and create jobs. We can do all of this. . . ”
(g) Pope Francis [123] speaking about global inequality: “We have to realize that a true ecological
approach always becomes a social approach: it must integrate questions of justice in debates on
the environment, so as to hear both the cry of the earth and the cry of the poor.”
(h) Worried oldster, thinking about sustainability: “Will we make it? I mean, will there be human
beings around five hundred years from now? What can we do?”
These examples become more complex as we move from (a) to (h)—more complex both in the size
of the community covered by “we” and in the difficulty of arriving at a common decision. Size and
difficulty move somewhat in tandem, but not all the time. For instance, the communities in (c) and
(d) are moderately large, but they also have fairly clear-cut methods for deciding things (after some
discussion, the professor will make the call on the class project, or the planning commission will take
a vote). By contrast, the community in (e) might be much smaller, but (as we saw in Section 3.2.2)
it can be paralyzed by each member waiting for another to make the first move, even if they all (in
theory) agree on what needs to be done.
There’s another feature of the slippery “we” that you might notice in examples (d) and (f) especially.
If you are a nonvoting community member in (d) you might feel excluded, even repelled, by the
commission chair’s “we” language in (d). Similarly, as Marshall [207, Chapter 6] explains with regard
to Obama’s speech (f):
The slippery “we” can be deeply alienating for people who do not consider themselves to
be included within it. . . No doubt if you are a supporter of Obama, it [his “we” language]
sounds wonderfully inclusive. [But if you oppose him], his avowal of common purpose
sounds deeply exclusive and you hear him saying me and my fellow global warming
zealot cronies are going to force you to do this.
Finally, in examples (g) and (h) the scope of “we” has been expanded further—to the whole world
community spread out in space and, in (h), in time also—and melded with concerns for justice and
fairness. However right this is from an ethical point of view, it makes it more and more difficult to
imagine a decision-making process that could bring this diverse “we” to an agreement about any
action that needs to be taken now. If there was one single decision-maker, one World Ruler, we could
perhaps imagine that person carrying out, on a gigantic scale, the calculations needed to maximize
everyone’s expected utility (Section 5.3.3) for the entire trajectory of world history. But there is no
CHAPTER 6. DECIDING 347
such person. And, what’s more, the history of centrally planned economies such as those of the former
Soviet Union cautions against believing that any actual World Ruler (as opposed to some theoretical
idealization) could in fact achieve the “greatest good for the greatest number,” even approximately.
No, what we have is a whole network (Definition 1 on page 129) of decision-makers, each of
whom has their own priorities and their own horizons of concern, trying to figure out decisions about
sustainability questions, some of which are on a scale, both in space and time, that the human race
has almost never faced before. When we talk of climate change, for example, we are talking about
decisions taken now that may affect the entire planet and every living thing on it for many thousands
of years—longer than the span of recorded human history. The only decision whose effects could
have had comparable scope was that faced by the United States and the Soviet Union in the period
between about 1960 and 1990: whether to engage in global thermonuclear war. It may be encouraging
to remember that the decision-makers faced by that question seem to have taken the right path. If they
had not, we would not be here today discussing sustainability.
So this final chapter is about how decisions are made when there are multiple decision-makers
connected in a network. We’ve already had a foretaste of some of these ideas in Section 3.2.2, and
the ideas of expected utility and risk aversion from Section 5.3.3 will help us clarify and extend them.
One issue that we have not addressed in detail is how to incorporate the values and interests of those
who do not presently have a voice into the decision-making process. Many, perhaps most, readers of
this book won’t have any grandchildren yet: does that mean that the interests of your (currently non-
existent) grandchildren should carry no weight in your sustainability decisions? It’s hard to answer
“no weight” to this question, especially in the light of Definition 1 on page vii; but if you give a
different answer, you are faced with the question of weighing the future against the present, which
does not have an obvious solution (take a look at the “think about it” quotation from Julian Simon on
page 209).
That’s where we’ll begin this chapter, in fact. The most prominent mechanism in our society for
“integrating” the decisions of individual decision-makers is the free market. In Section 6.1 we will
take a look at the market paradigm, at the way it handles “externalities” such as pollution, and
the way it can compare future values with present ones. Then in Section 6.2 we’ll look at choices
when there are two (or more) decision-makers whose value systems do not match up. In Section 6.3
we’ll apply these ideas to a classic decision problem related to sustainability: the “tragedy of the
commons” [153], where two or more decision-makers share a resource and each of them can gain an
individual advantage by over-exploiting it. Finally, in Section 6.4 we’ll remember that mathematics
can only take us so far. Mathematics cannot tell us what we ought to do: an ought conclusion requires
some ought inputs, coming from personal or shared ethical values, which may perhaps be codified in
the legal rules or tacit conventions of society. This section will ask you to think about what your own
core values are, how they might differ from those of others, and how people holding different values
might nevertheless work together to advance sustainability goals. We (and to avoid slipperiness, “we”
here means “the authors of this book”) believe that such self-understanding, which seems far from the
mathematics where our book began, is as important as mathematical understanding in building hope
for a sustainable future.
6.1 Market Perspectives and Large-Scale Change
Objectives
I understand the idea of a free market in goods or services.

I can define Pareto optimality and understand the arguments suggesting that a free
market should produce a Pareto optimal outcome.
I can define the terms market failure and externality.
I can carry out discounting of future costs and benefits.
I understand how a discount rate can be generated on the basis of revealed
preferences.
I can describe the objectives and philosophy of the Stern Review on climate change.
I know what is meant by intergenerational ethics.
I can describe the controversy around the Stern Review’s choice of discount rate.
6.1.1 The Market Paradigm

In the introduction to this chapter, we referred to one of the greatest challenges to sustainability
decision-making: to integrate the needs, decisions, preferences, and values of many different
individuals, spread over space and time, in a way that yields an outcome that is in some sense “the best
achievable for everyone.” One of the most powerful ideas that has shaped our modern world is about
precisely this question. It was given its classic form in 1776, with the publication of Adam Smith’s
Inquiry into the Nature and Causes of the Wealth of Nations [300]. Human societies, he stressed,
involve complex networks of support for every individual:
In civilized society he stands at all times in need of the co-operation and assistance of
great multitudes, while his whole life is scarce sufficient to gain the friendship of a few
persons. . . . man has almost constant occasion for the help of his brethren, and it is in vain
for him to expect it from their benevolence only. He will be more likely to prevail if he
can interest their self-love in his favour, and shew them that it is for their own advantage
to do for him what he requires of them.
In modern jargon, we might make Adam Smith’s point this way: you’ll get what you want (a “win”
for you) if you can show somebody else that providing it is also a “win” for them—if you follow a
“win-win strategy.” Some may believe that every “win” for me is a “lose” for you, that society is a zero
sum proposition, but Adam Smith was not among them: he saw abundant possibilities for “win-win”
transactions as lying at the heart of society. He then went further, however, in words that have become
famous:
[A certain trader] generally, indeed, neither intends to promote the public interest, nor
knows how much he is promoting it. . . . he intends only his own gain; and he is in this, as
in many other cases, led by an invisible hand to promote an end that was no part of his
intention. . . . By pursuing his own interest, he frequently promotes that of society more
effectually than when he really intends to promote it.
6.1. MARKET PERSPECTIVES AND LARGE-SCALE CHANGE 349
Here Smith argues, in effect, that if each market participant seeks only a “win” for themselves
individually, the effect of all these competing pressures will be to guide society as a whole toward a
good outcome (“the public interest”). The famous “invisible hand” is a metaphor for what we may
call the market paradigm.
Definition 1
The market paradigm is the idea that if

(a) goods and services can be traded freely and fairly among the participants in a society,
and
(b) those participants have regard only to their own perceived self-interest, then
the resultant exchanges will lead to an outcome that is, in some sense, best for all.
The market paradigm does not assume or require that everything be tradable in the market. For
example, the words “and fairly” in our definition have usually been held to include the existence
of a legal and judicial system that allows contracts to be entered into and enforced. If the judicial
system itself became part of the market—if verdicts in court could be sold to the highest bidder—
then the market paradigm would have been fatally undermined, not have notched up another success.
Nonetheless, the market paradigm is extremely seductive. Its language has colonized many important
areas of public discussion [284] that might previously have operated under other rules. In this section
we’ll take a look specifically at the application of market ideas to questions of sustainability.
First of all, though, let’s establish some of the language of the market paradigm. The things traded
in a market will generically be called goods, and the market participants will be called agents. (In
particular, money is just another good, although it has various special properties.) An allocation of
goods in a market is simply a list telling us how much of each good each agent has. If two agents
trade, that means that they exchange a certain quantity of various goods between them; e.g., “A gives
B a pig; B gives A three chickens, a bottle of champagne, and a hammer.” A trade changes the overall
allocation of goods in the market.
Definition 2
An allocation of goods in a market is Pareto optimal if there is no possible reallocation

that both parties will agree is a “win-win.”
In other words, a Pareto optimal allocation is one in which no single agent can be made better off
(in their own estimation) without someone else being made worse off (in their own estimation). In this
rather limited sense, a Pareto optimal allocation is “best for everyone.”
Think about it. . .
It is important to understand how weak the notion of Pareto optimality is. Suppose, for an
extreme example, that there is only one good in the market—money—and that I have all of
it (a trillion dollars) and no-one else has any. Then this is a Pareto optimal allocation; any
redistribution would make someone worse off (me). But do you imagine this state of affairs
should count as “best for everyone”? What factors have we not taken into consideration?
How might they be incorporated into decision-making?
Remark 1: Every economics textbook will include an argument to the effect that a “properly
functioning” free market should yield a Pareto optimal allocation (this is their version of Adam
Smith’s invisible hand). The argument goes like this, in outline form: Suppose the market settles at an
allocation that is not Pareto optimal. Then there is, somewhere, a possible win-win trade between two
parties (by definition of Pareto optimality). Since the market is properly functioning (whatever that
means exactly), these two parties will find and take the opportunity for this trade. But that means the
market had not settled at the previous allocation after all, QED.
Naturally, it takes a lot of formal mathematical work to make this argument into a real proof [92].
This work is also not always relevant to the real-world impact of the market paradigm, which is
more of an ideological one: to someone convinced by this paradigm, any instance of an actual market
seeming to lead to a nonoptimal outcome must be the result of the market’s not being “properly
functioning.” In the language of the market paradigm, this situation is referred to as a market failure.
Definition 3
A market failure occurs when market transactions (buying and selling) yield an allocation
of goods and services that is not optimal, that is, when things could be made better for
everyone, but it is not in anyone’s individual interest to make them better.
Example 1: Uncontrolled pollution (let us say, pollution of a watercourse by factory effluent) is

often cited as an example of a market failure. If a steel mill, located at a certain point on a river,
deposits polluted effluent into the river, everyone downstream from the mill bears a cost: their drinking
water may need additional treatment, crops may accumulate dangerous levels of pollutants and not
be salable, recreational opportunities may be curtailed, and so on. The benefits of generating this
pollution accrue to the mill’s owners: they are able to make their product more simply or more cheaply
than if they had had to control the pollution levels that they generated. But the costs of this pollution
are borne by everyone downstream, whether or not they have consented to it. This type of market
failure is called an externality.
Definition 4
An externality is a cost (or maybe a benefit) that affects someone who did not choose it and
is not compensated for it.
For another example, the traffic congestion created by your (and other people’s) cars is an
externality to me: I did not choose to travel on clogged freeways, and (under the conventions of
our society) nobody owes me any compensation for generating this congestion.
Remark 2: It is often argued that externalities like those described above reflect not so much a
“failure” of the market as a situation in which the power of the market does not extend widely enough.
For instance, the key point in the steel mill example is that the downstream residents can have costs
imposed on them without their consent. Suppose, though, that each resident’s “right” to clean water
was made a tradable thing, something that itself could be bought and sold in the market and enforced
by the legal system. Then, it is argued, the market would once again be “properly functioning” and
lead to an optimal conclusion: because the factory would need to buy those rights in order to operate,
its owners would have an incentive to operate in a way that required them to buy as few as possible—in
other words, to clean up their act. We will look in more detail at externalities of this kind, and at some
proposed market remedies, in Section 6.3. We will also see that market-based analyses of situations
like this are not the only possible ones, and may in fact erode the deeper values on which our hope to
maintain a sustainable world ultimately depends.
Early on we discussed ecosystem services as a way to conceptualize the interdependence between
human activity and the ecosystems on which this activity depends (see Definition 3 on page ix).
The ecosystem services concept has come under criticism for promoting a market-based vision of
conservation and a human-centric view of ecosystems. In a commentary that appeared in the journal
Nature, Douglas McCauley writes [211]:
Probably the most important trend in conservation science at the moment is ‘ecosystem
services,’ typically seen as economic benefits provided by natural ecosystems. . . .
[M]arket-based mechanisms for conservation are not a panacea for our current conser-
vation ills. If we mean to make significant and long-lasting gains in conservation, we
must strongly assert the primacy of ethics and aesthetics in conservation. We must act
quickly to redirect much of the effort now being devoted to the commodification of nature
back toward instilling a love for nature in more people.
In a recent review article, “Twenty years of ecosystem services,” Robert Costanza and co-authors have
responded to the criticism directed against the “ecosystem services” concept [82]:
Unless we recognize our interdependence with the rest of nature we are putting our
species’ well-being at risk, and at the same time we blindly endanger global ecosystems.
So, rather than implying that humans are the only thing that matters, the concept of
ecosystem services makes it clear that the whole system matters, both to humans and to
the other species we are interdependent with. If anything, the ecosystem services concept
is a ‘whole system aware’ view of humans embedded in society and embedded in the rest
of nature. ‘Centric’ with any prefix doesn’t really describe this complex interdependence.
Think about it. . .
Revisit our description of ecosystem services in Definition 3 on page ix and reflect on

the differing viewpoints of Costanza and McCauley. Do you think the ecosystem services
concept evokes a sense that human society is embedded in nature? Does it tend to promote
a human-centric view?
6.1.2 The Value of a Statistical Life

We have been talking quite abstractly so far about “costs and benefits.” However, in the public policy
sphere, costs and benefits frequently involve changes in risk to human life. How do we put a value
on benefits that include lives saved? Here we develop this idea through a series of questions. These
questions are ideas for you to reflect on: they do not have “right” and “wrong” answers.
Let’s start out with a concrete one. Suppose that you live in a town that has a population of 100,000
people. In your town there is a particular traffic intersection that is notoriously dangerous. On average,
accidents at this intersection are responsible for 1 fatality every year. No one is able to know when
such a fatality might occur or who will be the next victim, but if nothing is done to redesign the
intersection, everyone is confident that fatalities will continue at the same rate.
Question 1: Suppose that traffic engineers are confident that a complete reconstruction of the
intersection, using a new design, would eliminate the current hazard. The work will be paid for by
a new tax on each member of the community. How much additional tax are you willing to pay to
redesign the intersection, knowing that as a member of the community, your share of the total cost is
1
100,000 ? (For the purposes of this section, ignore anything you happen to know about the real costs
of building highways. You might suppose there is some special reason this particular intersection
might be very expensive to rebuild—our purpose here is to try to understand the interplay of costs and
benefits when lives are at stake.)
Think about your answer for a moment, and then read the following definition.
Definition 5
The value of a statistical life is the aggregate amount that a group of people would be
willing to pay for 1 less expected death among this group each year.
Example 2: Suppose that your answer to Question 1 is $100, and that you are representative of your
community. Then the value of a statistical life, as assessed by your community, is
$100
1 × 100, 000
person

persons
= $10 million.
Remark 3: The Environmental Protection Agency uses the concept of the value of a statistical life in
cost-benefit analyses of environmental policies. As of this writing, the EPA uses a value of about $9
million for the value of a statistical life.
Remark 4: We are intentionally framing the value of a statistical life in terms of reduced risk of
death among a group of people, but the previous definition can also be framed in terms of each
person’s willingness to pay for a reduction in their own personal risk (without regard to other people’s
risk) [18].
Question 2: Consider the benefits of reduced air pollution through the Clean Air Act. By the year
2020, the 1990 amendments to the Clean Air Act are expected to reduce early deaths attributable to air
pollution by 230,000 cases each year throughout the United States. Does your answer to Question 1
change if the hazard is somewhat more abstract, as in the case of the risk of early death from air
pollution? How does it change? Why does it change?
Question 3: Based on our definition, the value of a statistical life might be computed to be lower in
poorer regions of the world than in the United States, simply because such communities have fewer
resources to pay for safety improvements. (For example, one study [127] computed the value of a
statistical life in parts of Thailand as $0.25 million. Sadly, the “safety improvements” investigated in
this case were landmine clearances.) Does your answer to Question 1 change if the hazard is faced
not by you and other people in your community but by people who are in another part of the world?
Should it change?
Question 4: Finally, what if the hazard will instead be faced not by people living today, but by people
yet to be born? Again, does your answer to Question 1 change? Should it change?
Think about it. . .
In this brief section, we have confronted some deep questions: How much are we willing to
pay to save an individual human life? Why does the “price” vary so much between different
situations? Does justice demand equality in these “values” that we set? These questions
weigh heavily on sustainability, but they are highly significant in other contexts too. Can
you suggest some other situations in which they are important?
6.1.3 Discounting and the Time Value of Money
Question 5: Which would you prefer: a hundred dollars today, or a guarantee of a hundred and twenty
dollars one year from now?
Answer: There’s no “right” answer to this question, of course: “$100 now” and “$120 in a year” are
two different goods in the market, and whether you are willing to trade one for the other is governed
by your own individual preference. Nevertheless, if you hesitate in answering this question, you are
illustrating the principle of time discounting. A benefit that we receive today can seem more valuable
than the same benefit—or even than a greater benefit—that is guaranteed at some point in the future.
Partly this reflects our innate desire for immediacy (“I want it now”); partly it reflects our desire for
certainty (“How good is your guarantee anyhow?”); partly it reflects the fact that we could put the
benefit to use in the time period between now and next year, and so derive extra gain from it. (For
example, we could deposit the hundred dollars in an interest-bearing account.)
We introduce this concept now because it plays a central part in almost all market-based studies
of sustainability. Addressing sustainability questions requires that we compare costs and benefits to
people many generations in the future with costs and benefits to people living right now. As we’ll
illustrate in this chapter, the results of many sophisticated studies of the economics of climate change,
or of other large-scale sustainability issues, depend fundamentally on the assumptions that those
studies make about time discounting: assumptions that are not only mathematical, but that also reflect
ethical principles.
Standard economic theory uses an exponential decay model (Section 3.4) for discounting.
Definition 6
The (annual) discount rate d is the rate constant for the exponential discount model (with
time step 1 year). In other words, the present value of U units of utility, n years from now,
is
U(1 − d)n
units of present utility.
Before getting to sustainability applications, let’s work a few more standard financial examples.
Problem 1: Suppose that the discount rate is 3 percent. What is the present value of a guaranteed
payment of $120 five years from now?
Solution: Using the present value formula from Definition 6, the answer is
$120 × (1 − 0.03)5 ≈ $103
present dollars.
Problem 2: Suppose that the discount rate is 5 percent. I am considering whether to invest money in
a rental property that will produce an income of $10,000 at the end of each year for the next 25 years,
and after that period may be considered valueless. What is the greatest price that I should pay for the
property?
Solution: The income stream produced by the property at the end of year 1 has a present value of
$9,500 (that is, $10, 000 × (1 − 0.05)). The income stream produced at the end of year 2 has a present
value of $9,025 (that is, $10, 000 × (1 − 0.05)2 ). In general, the income stream at the end of year n has
a present value of $10, 000 × 0.95n . Thus the present value of all the future payments is
$10, 000 × (0.95)1 + (0.95)2 + · · · + (0.95)25 .

We can use the geometric progression formula, Rule 5 on page 198, to write this as
1 − (0.95)25

$10, 000 × 0.95 × ≈ $137, 000.
1 − (0.95)
Notice that this is quite a bit less than the $250,000 we would get by multiplying $10,000 per year by
25 years, without taking the discount rate into account.
Now let’s consider a question that is closer in structure to many sustainability problems.
Problem 3: Scientists at Starfleet University discover a new source of energy, dilithium crystals.
Professor Montgomery Scott reviews the prospects for establishing a dilithium power plant at its main
campus at University Park. He concludes that the power plant will generate a million dollars’ worth
of green energy every year for a hundred years; however, there is a 1.5 percent chance that at the end
of that hundred years it will explode, doing 20 billion dollars in damage. Should Scotty recommend
establishing the dilithium plant?
The details of this example are obviously fictional, but the basic structure—trading off assured
short-term gains against uncertain long-term risks—recurs in many environmental and sustainability
decisions, as we shall see. The key thing to notice is that the answer to the problem is strongly
dependent on our assumptions about the discount rate. To illustrate this, we first calculate a general
formula for the expected gains and losses, and then plug in two different values for d.
Solution: The gains from the plant are the present value of an income stream of one million dollars
a year for a hundred years. As in the previous example, this can be calculated using the geometric
progression formula
106 (1 − (1 − d)100 )
Gain = 106 × (1 + (1 − d) + · · · + (1 − d)99 ) = dollars.
d
The present value of the loss if the plant explodes is
2 × 1010 × (1 − d)100 dollars.
Since there is a 0.015 (1.5 percent) chance of this event, the expected present loss is
Expected loss = 0.015 × 2 × 1010 × (1 − d)100 = 3 × 108 × (1 − d)100 dollars.
Let’s calculate for two different values of d (Table 1). Using a 1 percent discount rate we see that
Table 1: Dilithium Power Plant Analysis: Present Values of Gain and Loss
Discount Rate Gain Loss

1% $6.3 × 107 $1.1 × 108
4% $2.5 × 107 $5.1 × 106
the present value of the expected net loss (from the small risk of an explosion) greatly outweighs the
present value of the energy stream. But using the 4 percent discount rate, the opposite applies. The
assumption of a 4 percent discount rate means that a risk 100 years in the future is weighted only
1.5 percent as seriously as an immediate risk, and that the benefits of the power plant now seem to
outweigh the costs.
Think about it. . .
How should we choose what discount rate to apply? This is a difficult question. Part of
the discount rate relates, as we have seen, to our sense that wealth is a growing thing (the
feedback loop of capital accumulation; see Figure 5 on page 195). Part, though, is what is
called a pure time preference: we value the interests of our generation over those of future
generations because they are close to us in time, just as we may value the interests of people
who are “like us” in other ways more highly than those of others. But is this right? That
is a question of fundamental moral values—“intergenerational justice” is the term that is
sometimes used. What do you think?
6.1.4 The Stern Review

One of the most ambitious attempts to apply the market paradigm to a huge sustainability question
was the so-called Stern Review, which was published in Britain in 2006. (There have been other similar
reviews, before and since, but this is one of the most widely known.) Nicholas Stern (born 1946) is
a British economist who became Chief Economist of the World Bank from 2000 to 2003. In 2005,
the British “Treasury Secretary,”2 Gordon Brown, commissioned Stern to carry out a review of the
2 Brits call this position Chancellor of the Exchequer.
economics of anthropogenic climate change, to help Britain and the world understand the challenges
ahead and the most effective way to respond to them. Stern’s report, which was 700 pages long, was
released in the fall of 2006. Its central conclusion was that strong, early, coordinated action on climate
change was justified on economic grounds: failure to act might lead to some short-term gains, but
they were far outweighed by the long-term costs. (This is like the situation we observed in Example 3
above, when we used the 1% discount rate.)
The analysis used by the Stern Review is referred to as integrated assessment
modeling, because it combines modeling techniques from several different
disciplines, notably climate science and economics. It tries to understand what
impact changes in climate may have on the human economy (for example, impacts
on food production, mass migration, or social instability) and also what impact
changes in the economy may have on climate (reduced fossil fuel consumption
owing to a recession, for example, may lessen future climate change). Here we
have another feedback loop. Stern treated climate change as an externality giving
rise to a market failure—in fact, to quote the Review, “the greatest and most wide-
ranging market failure ever seen.”
How does this externality arise? Recall the externality created by a polluting
Figure 2: Lord Nicholas Stern. factory on a river (Example 1 on page 350): the factory owners capture benefits
for themselves, while imposing costs on everyone downstream whether or not they
have consented to accept them. It’s the fact that people downstream have no say in
the matter—the phrase “whether or not they have consented”—that makes this an example of a market
failure. Stern argues that climate change imposes similar involuntary costs on those “downstream”
from the present generation in terms of time: our children and grandchildren and their descendants for
many, many years to come. They have not consented to bear the costs that result from our capturing
the benefits of fossil fuel use. A correct economic analysis should take these costs into account. The
Stern Review used a discounting approach, the kind we have looked at in Section 6.1.3, to estimate
the present value of these future costs.
Before we get into the details of Stern’s discounting approach, let’s take a look at the Review’s
conclusions. These are presented in the Executive Summary [309], and here are some of the key
points. As you study these, keep in mind that they were written in 2006.
• The benefits of strong and early action to mitigate climate change (most importantly, reducing
greenhouse gas emissions) far outweigh the economic costs of not acting.
• Climate change will affect the basic elements of life for people around the world—access to
water, food production, health, and the environment. Hundreds of millions of people could
suffer hunger, water shortages, and coastal flooding as the world warms.
• Using the results from formal economic models, the Review estimates that if we don’t act to
mitigate climate change, the overall costs and risks of climate change will be equivalent to
losing at least 5% of global GDP3 each year, now and forever. If a wider range of risks and
impacts is taken into account, the estimates of damage could rise to 20% of GDP or more.
• In contrast, the costs of action reducing greenhouse gas emissions to mitigate the worst impacts
of climate change can be limited to around 1% of global GDP each year.
• The investment that takes place in the next 10–20 years4 will have a profound effect on the
climate in the second half of this century and in the next. Our actions now and over the coming
decades could create risks of major disruption to economic and social activity, on a scale similar
3 GDP stands for gross domestic product, the value of the overall “output” of the economy.
4 That is, the period 2006–2026. We are in the middle of that period now.
to those associated with the great wars and the economic depression of the first half of the
twentieth century.
The discounting methodology used in the Stern Review combined a term related to future economic
growth (which itself was a variable in the model) with a pure time preference term. Stern argued on
ethical grounds that when studying a long-term social issue such as climate change, the pure time
preference rate should be set very low: we have no ethical grounds for valuing our own interests more
highly than those of future generations. Indeed, Stern took this to the point of saying that if we knew
for sure that the human race would survive indefinitely, then the pure time preference rate should be
zero: a pure time preference rate greater than zero is justified only, he claimed, by the remote but not
impossible chance that there won’t be any human beings at all in the future for us to worry about.
This decision led Stern to use an overall low discount rate, which in turn meant that many of the
“present values” of costs and benefits, carefully analyzed in the Review, arose from effects that were
anticipated in the long-term future: more than a hundred years from now.
Critical Thinking
Think about questions (a) and (b) below.

(a) Which would you prefer: 50 dollars now, or 100 dollars in six months?
(b) Which would you prefer: 50 dollars in three months, or 100 dollars in nine months?
The two questions involve the same sums of money and the same delay: in fact, question
(a) just is question (b) from the perspective of your three-months-future self. It’s natural to
argue, then, that a rational person ought to give the same answer to both questions, and the
exponential model of discounting is the only one that achieves that.
But many real people don’t give the same answer to both: they take the money now in (a),
and accept delay to get more money in (b). This is called hyperbolic discounting. If this
kind of preference is taken seriously, how would it affect the results of the Stern Report and
similar calculations? Ought it to be taken seriously?
While several leading economists supported Stern’s choices of discount rate, others disagreed,
sometimes sharply. Their argument was this: however ethicists might think that we should value the
lives of future generations, we can see how people actually do value the future relative to the present
by analyzing market transactions such as purchases and sales of investment assets. The results of
such an analysis are called revealed preferences, and they have been shown to be more consistent
with an approximately 3% pure time preference than with the very low value used by Stern. William
Nordhaus, a leading Yale economist who himself has written extensively on the economics of climate
change, commented [239]:
The Review’s unambiguous conclusions about the need for extreme immediate action
will not survive the substitution of assumptions that are more consistent with today’s
marketplace real interest rates and savings rates. Hence, the central questions about
global-warming policy—how much, how fast, and how costly—remain open. The Review
informs but does not answer these fundamental questions.
It is important to note, however, that Nordhaus is not an advocate of inaction. He also wants to
establish economic mechanisms that pressure us to take into account the future costs of the climate
change that we are generating; he simply thinks that the best amount of “pressure” is less than Stern
recommends. In the next couple of sections we will take a look at some of the proposals that have
been devised for such economic mechanisms and how they might work in a climate change context.
6.1.5 Is the Market Paradigm Appropriate?

Nordhaus and similar critics of the Stern Review agree with its use of market (that is, economic)
analysis, but disagree over some of the details of this analysis—details which, as we have seen, may
actually be freighted with ethical meaning. Some other criticisms of the Review take aim at more
fundamental issues, arguing that using the market paradigm as a basis for the review at all (whatever
adjustments are made to correct apparent market failures) already prejudges the situation and narrows
the range of acceptable solutions. Here are brief summaries of a few criticisms from this direction.
• Justice perspective. As we have seen above, technical assumptions about discount rates serve,
in a way, as a mechanism for including questions of intergenerational justice in the Review. But
there are other kinds of injustice also, and here the Review seems willing to accept the market’s
revealed preference that, for instance, the value of a statistical life is different in Thailand from
what it is in the U.S. (Question 3 on page 353). This feels inconsistent.
• Risk aversion issues. The Review’s economic analysis attempts to maximize an indicator
of human well-being—the present value of global expected utility. That word “expected”
(Definition 2 on page 303) is shorthand for an appeal to the law of large numbers: it refers
to the average over many trials. We don’t, however, get multiple trials of human history; we
get only one trial, because we have only one Earth. (This is perhaps analogous to the situation
of a gambler who is deciding whether to go “all in”: there is no “long-term average” that is
relevant; everything hinges on a single throw.) When we consider decision-making in such a
context, human risk aversion will come into play, as described in Section 5.3.4, and that may
lead people to prefer different decisions from those that apparently maximize expected utility.
• Valuation of nonhuman resources. Climate change is likely to have vast impacts on ecosystems
around the world, including those in which there is little direct human involvement (coral reef
bleaching caused by ocean acidification [25] is an example). It is difficult to assign a specific
monetary value to the “services” these ecosystems provide, and critics say this leads modeling
techniques such as those used by Stern to underestimate their importance and significance.
• Growth issues. The market paradigm doesn’t logically require unending economic growth; a
steady state market economy is perfectly conceivable [169]. Charts like those in Figure 1 on
page vi suggest that growth is bringing about what Speth [305] calls the “Great Collision”
between the human economy and the planet. Yet Stern is at pains to argue that his proposals
to stabilize climate change are entirely consistent with continued economic growth. Is this
realistic?
We should say a little more about the last of these critiques. Remember Figure 4 on page ix from
the introduction? This picture captures the intuition underlying arguments such as those of The Limits
to Growth [215], which we reviewed in Section 4.2. Former World Bank senior economist Herman
Daly, author of Beyond Growth [86], For the Common Good [88], and many other books, has spent
years developing an alternative paradigm for economics that does not depend on the idea of endless
growth. In fact, the idea of “growth forever” is a comparatively recent one. A reviewer of Daly’s work
wrote, “There is no more basic ethical question than the one Herman Daly is asking.” Nevertheless,
the Stern Review, which purports to take the longest view possible, seems determined not to ask it.
Clearly the work of the Stern Review is open to criticism from several directions. Nonetheless, it
attracted well-deserved praise because of its detailed attempt to answer the fundamental questions
posed by Gordon Brown: “what will be the economic, social, and environmental consequences of
climate change in both developed and developing countries?” and “what will be the costs and benefits
of actions to reduce the net global balance of greenhouse gas emissions from energy use and other
sources?”
Think about it. . .
Although his work is no longer part of the official machinery of government, Nicholas Stern
has continued to advocate for urgent and radical action to tackle climate change. His most
recent book is titled simply Why Are We Waiting? [310]. In his conclusion he states, “The
case for urgent and radical action is extremely strong, and the tools to make it happen are
firmly within our grasp” and “Today’s young people can and should hold their parents’
generation to account for their present actions.” Do you agree? How might such “holding to
account” look for you?
Summary of Ideas: Market Perspectives and Large-Scale Change
• A free market economic system is one in which prices for goods and services are
determined by voluntary trades between buyers and sellers.
• A state of the economy is Pareto efficient if there is no way to make any one person
better off without making someone else worse off.
• Classical economic theory argues that a free market should lead to a Pareto efficient
outcome. A market failure is said to occur when a market outcome is not Pareto
efficient.
• An externality is a cost that one person in a market can impose on another
involuntarily (that is, without compensation). Externalities are among the possible
causes of market failure.
• The discount rate tells us the present value of a future cost or benefit. A discount rate
inferred from people’s market behavior is called a revealed preference.
• The Stern Review was an integrated assessment of climate change mitigation costs
and benefits. It argued for strong, early intervention to mitigate climate change.
• The Stern Review’s conclusions were partly dependent on its choice of a low long-
term discount rate. This was justified on the grounds of intergenerational ethics,
rather than by revealed preference.
6.2 The Strange Behavior of Rational People
Objectives
I can describe how an economic interaction between two agents can be modeled in
terms of a game.
I understand the notions of strategy and payoff matrix for a game.
I can recognize when a player has a dominant strategy or a strictly dominant
strategy.
I understand the prisoners’ dilemma model, whereby each player pursues a domi-
nant strategy and the resulting outcome is worse for everyone.
I can use the idea of best response to understand what will happen in a game in which
only one player has a dominant strategy.
I know what is meant by a Nash equilibrium, and why such an equilibrium is likely
to be the result in a game with no dominant strategy.
I can define a mixed strategy, and know that Nash equilibria can always be found
when such strategies are allowed.
6.2.1 Game Theory
In the previous section, we discussed the concept of the market—a concept that seems to govern an
ever-increasing proportion of our lives. Viewed quite abstractly, the market is a device for integrating
the decisions and preferences of many different agents, with many different desires and priorities, in
order to arrive at an outcome that is “best for everyone” (at least in the weak sense of Pareto optimality,
Definition 2 on page 349). But will decision-makers always automatically arrive at such an “optimal”
outcome? The mathematical subject of game theory answers this question. Game theory studies the
behavior of decision makers who are linked by a network.
Example 1: In Section 3.2.2 on page 142 we looked at the bystander effect, which we modeled as an
information cascade. Each of the bystanders took “informational cues” from the responses of all the
other bystanders whom they could see. As a result, none of them were strongly motivated to “move
first” and intervene. We saw that the network structure of the group of bystanders had a powerful
influence on the outcome. In this section, by contrast, we will restrict ourselves to the simplest possible
kind of network, with two decision-makers (network nodes) who interact through a single link. The
advantage of this simplification is that we can then consider more complex and realistic models of
decision-making behavior than the simple “threshold” model we used in thinking about the bystander
effect.
6.2. THE STRANGE BEHAVIOR OF RATIONAL PEOPLE 361
Definition 1
The mathematical study of interconnected decision-making is called game theory. Game

theory applies whenever
(a) There are at least two decision-makers (called agents) who interact with one another.
(b) The agents must choose between alternative courses of action.
(c) Each agent’s satisfaction with the outcome of the game depends on the choices of all
the other agents.
(d) There is perfect information: each agent knows every agent’s level of satisfaction
with every possible outcome (in language that we will introduce below, each agent
knows the entire payoff matrix).
As mentioned above, we will restrict our attention to games involving only two agents.
Remark 1: The word “game” does not refer just to sports or board games (although some of these can
be analyzed quite precisely using game-theoretic ideas). Game theory can be applied to any decision-
making situation that has the characteristics noted in Definition 1. The crucial part of the definition is
(c): my satisfaction with the outcome of the game depends on your choices as well as on mine.
Example 2: A game that many of us learn to play in childhood is “Rock, Paper, Scissors.” The two
players, A and B, must each decide on one of three strategies: play Rock, play Paper, play Scissors.
Then both simultaneously reveal their chosen strategy. If both players have chosen the same strategy,
the game is a draw. Otherwise, the winner is determined according to the slogan
Scissors cut Paper; Rock blunts Scissors; Paper wraps Rock.
Mathematically this structure can be represented by a payoff matrix. This is a table that shows the
“reward,” or payoff, that each player will gain from each combination of strategies from both players.
If we assume that a win in “Rock, Paper, Scissors” brings a reward of +1, a draw brings a reward of 0,
and a loss brings a reward of −1, then the payoff matrix for “Rock, Paper, Scissors” will be as shown
in Table 2:
Table 2: Payoff Matrix for “Rock, Paper, Scissors”
B plays “Rock” B plays “Paper” B plays “Scissors”

A plays “Rock” 0, 0 −1, +1 +1, −1
A plays “Paper” +1, −1 0, 0 −1, +1
A plays “Scissors” −1, +1 +1, −1 0, 0
Notice that player A’s choices correspond to the rows of the matrix, player B’s to the columns.
In each cell of the matrix are two numbers, the first giving player A’s payoff, and the second giving
player B’s. For example, in the top right cell, corresponding to “A plays Rock, B plays Scissors,” we
find the entry +1, −1. That tells us that with these two choices of strategy, A’s payoff will be +1; B’s
payoff will be −1. In other words, A wins, B loses: “Rock blunts Scissors.”
Definition 2
The payoff matrix in a two-person game is a table showing the payoffs to each of the
players for each of the different strategies that may be adopted by each of them.
Remark 2: The payoff matrix for “Rock, Paper, Scissors” has a special feature that is not required
for all games: in “Rock, Paper, Scissors,” whatever payoff A gains, B loses (and vice versa). To
put it another way, the total payoff is zero, whatever strategies the players adopt. Games that have
this feature are called zero sum games. Not every game is zero-sum, however: many of the most
interesting examples have strategies that lead to gains for both players, and one of the fundamental
questions of game theory is whether strategic choices that are “rational” for each player individually
will lead them to behave together in a way that maximizes their joint gains.
Problem 1: You (A) and your partner (B) are both software engineers. Your company is working on
three projects code-named Particle, Neobase, and Kazam, all of which must be shipped in a month.
You (A) are assigned to work on Particle, and your partner (B) is assigned to work on Kazam. Because
Neobase is a more complex product, it requires you both to work on it to achieve its best possible state.
We assume that each of you must independently decide whether to allocate your effort in the coming
month to your own personal project (Particle or Kazam) or to the joint project (Neobase).
The company will pay a bonus to the developers of each piece of software, based on its quality
when shipped (which will depend on how much work you have put into it over the coming month).
These bonuses are expected to be as follows:
• For your personal project (Particle or Kazam), the bonus will be $150, 000 if you work on it
over the coming month, and $50, 000 if you do not.
• For the joint project (Neobase), the bonus (split equally between the two of you) will be
$300, 000 if you both work on it over the coming month, $150, 000 if only one of you works on
it, and $50, 000 if neither of you does.
Use the ideas of game theory to investigate what will happen.
Solution: The first step is to construct the payoff matrix. In this case, the “payoff” for each engineer
is the bonus they receive.
• If both engineers work on the joint project (Neobase), they each receive a payoff of 12 ×
$300, 000 = $150, 000 for their work on Neobase and $50, 000 for their work on their individual
projects: a total of $200, 000.
• If only one engineer, say A, works on Neobase and the other, B, works on their individual
project (Kazam), then they each receive a payoff of 12 × $150, 000 = $75, 000 for their work on
Neobase. For the personal projects, A receives $50, 000 and B receives $150, 000. Thus their
total payoffs are $125, 000 for A and $225, 000 for B.
• Finally, if both engineers work on their personal projects, they each receive 12 × $50, 000 =
$25, 000 for their work on Neobase and $150, 000 for their personal projects: total, $175, 000.
Thus we obtain the payoff matrix shown in Table 3 on the opposite page (payoffs are expressed in
thousands of dollars). Just as a reminder, the first figure in each cell represents the payoff to A in that
situation; the second figure represents the payoff to B. As far as A’s decision-making is concerned,
B’s payoffs are not directly relevant: A wants to achieve the highest payoff to A. Similarly, as far as
Table 3: Payoffs for Software Engineers Problem
B: joint project B: personal project

A: joint project 200, 200 125, 225
A: personal project 225, 125 175, 175
B’s decision-making is concerned, A’s payoffs are not directly relevant. (It’s important to stress this
point because people who are used only to zero sum situations sometimes find it confusing.)
What will happen in our example? If you (A) knew for sure what your partner (B) was going to
do, it would be easy for you to decide what you should do: you would just look down the column of
the table corresponding to your partner’s decision, and select the strategy that yielded the maximum
payoff for you. For example:
• If you knew that your partner was going to work on their personal project, you would look down
the right-hand column and compare your payoff for working on the joint project (125) with your
payoff for working on your personal project (175). Since 175 > 125, you would decide to work
on your personal project.
• If you knew that your partner was going to work on the joint project, then you would look down
the left-hand column and compare your payoff for working on the joint project (200) with your
payoff for working on your personal project (225). Since 225 > 200, you would decide to work
on your personal project in this situation also.
Thus, whatever your partner does, your best choice is to work on

your personal project. One says that working on the personal project
is a strictly dominant strategy. Your strictly dominant strategy is
represented by the bottom row of the payoff matrix, because for each
of the columns (that is, for each play by B) your payoffs in the bottom
row are greater than your payoffs in the same column and the top row.
Now, if you have a strictly dominant strategy, it is rational for you to
use it no matter what. Reasoning in this way, you will decide to work on
your personal project (Particle) rather than the joint project (Neobase).
Of course, your partner will also reason in the same way: the situation is
completely symmetric, and you both have the same information. Your
partner’s strictly dominant strategy is represented by the right-hand
column of the matrix, because the payoffs to your partner in this column
are greater than the corresponding payoffs in the same row and the other
column. Thus your partner will also decide to work on their personal
project (Kazam), and you will both end up in the bottom right corner,
for a bonus of $175, 000 each.
Rational though this all is, there is something strange about the Figure 3: The prisoners’ dilemma [344].
outcome. If you had both worked on Neobase, you would both have
done better ($200, 000 rather than $175, 000). But within the constraints we have set up, there is no
rational way for you to reach this improved outcome. Even if you made an unbreakable commitment
(an oath, or whatever is the equivalent in your personal ethical system) to work on Neobase, and your
partner knew that you could be trusted to carry out this commitment, he would still have an incentive
to work on Kazam instead. By doing so he would achieve a higher total bonus for himself. Even when
you are committed to cooperate, your partner still has an incentive to defect. This situation is called a
prisoners’ dilemma (see Figure 3).
In a game with a prisoners’ dilemma structure, “rational” play on both sides leads to an outcome
that is not the best possible for either. The “rational” software engineers, with complete freedom to
choose, inevitably arrive at an outcome that is not Pareto optimal (Definition 2 on page 349). This
contradicts the market paradigm (Definition 1 on page 349), which asserts that individuals following
their own self-interest will be “guided, as by an invisible hand” to an outcome that is in some sense
best for society. It is important to understand this example because it sets the basic structure for many
sustainability dilemmas, as we will see in Section 6.3 on page 373.
Critical Thinking
Some would say that the failure of the “market” to deliver the optimum solution in
this example comes about because there is no mechanism for making binding contracts.
(Suppose you had the opportunity to make a legal contract between you and your partner
that you would both work on Neobase. We’ll explore this idea, again using the methods of
game theory, in Exercises 15 and 16 at the end of this chapter.) What do you think of this
explanation? Notice what it admits: that the successful functioning of the market depends
on the existence of mutually agreed upon non-market mechanisms (something like our legal
system). This is a substantial step away from the “pure” market paradigm.
6.2.2 Best Responses and Dominant Strategies

We introduced several new ideas in our discussion above. Let’s make some definitions to formalize
them. We’ll think of a two-player game with players A and B; and we’ll use lowercase letters (like a,
a0 or b, b0 and so on) to denote the various strategies that are available to A and B. So in the example
above, one strategy available to A is “Work on Neobase”: we might call this strategy a. Another
strategy is “Work on Particle”: we might call this strategy a0 . A pair of strategies (a, b), one for A and
one for B, determines a “play” of the game and therefore a cell in the payoff matrix, yielding specific
payoffs for A and for B.
Definition 3
In a two-player game, a strictly best response by one participant (say A) to a given strategy
b of the other participant (B) is a strategy a such that A’s payoff for (a, b) is greater than A’s
payoff for any other choice (a0 , b) (that is, for any other choice of strategy by A, assuming
that B’s strategy remains the same).
In terms of the payoff matrix, we start by fixing a column—that is, a strategy b by B. A strictly best
response by A is a row a for which A’s payoff, in the column b, is greater than in any other row (and
the same column b).
Example 3: Consider the payoff matrix shown below, for a game in which player A has two strategies
(a1 and a2 ) and player B has three strategies (b1 , b2 , and b3 ). Suppose B plays b2 . Then A’s payoffs
are 3 for playing a1 and 0 for playing a2 . It follows that a1 is the strictly best response to b2 .
Suppose A plays a1 . Then B’s payoffs are 6 for b1 , 11 for b2 , and 9 for b3 . It follows that b2 is the
strictly best response to a1 .
Suppose A plays a2 . Then B’s payoffs are 4 for b1 , 8 for b2 , and 10 for b3 . It follows that b3 is the
strictly best response to a2 .
Table 4: Example Payoff Matrix
B plays b1 B plays b2 B plays b3

A plays a1 7, 6 3, 11 1, 9
A plays a2 5, 4 0, 8 0, 10
Definition 4
In a two-player game, a strictly dominant strategy for one participant (say A) is a strategy
that is a strictly best response to every strategy that the opponent (B) may play.
Example 4: In the game described by the payoff matrix of Example 3, you can easily check that a1
is the strictly best response to b1 , b2 , and b3 (we already verified this for b2 in the previous example).
Thus a1 is a strictly dominant strategy for A.
We already verified that B’s strictly best responses to a1 and a2 are different. Thus B has no strictly
dominant strategy in this game.
Remark 3: In a payoff matrix, a strictly dominant strategy for the first player (you) can be recognized
by the fact that the payoffs to you in the row corresponding to that strategy are greater than the
corresponding payoffs to you in all other rows. Similarly, a strictly dominant strategy for the second
player (your opponent) can be recognized by the fact that the payoffs to your opponent in the columns
corresponding to that strategy are greater than the corresponding payoffs in all other columns. Notice
that only the payoffs to you count in deciding what may or may not be a strictly dominant strategy for
you. Similarly, only the payoffs to your opponent count in deciding what may or may not be a strictly
dominant strategy for your opponent.
Strictly dominant strategies are important because, as we observed in our solution to Problem 1 on
page 362, a “rational” player who has a strictly dominant strategy will be sure to make use of it: it will
maximize their payoff whatever their opponent plays. We can formalize this as a rule.
Rule 1: Strictly Dominant Strategy Rule
A rational player who has a strictly dominant strategy will be sure to use it.
If both players have a strictly dominant strategy, we can predict the outcome of the game: they
will both use it. That is how we analyze Problem 1 and similar games with a prisoners’ dilemma
structure. But even if only one player has a strictly dominant strategy, we can usually predict the
outcome of the game. Consider, for instance, the game of Example 3. We have seen that A has a
strictly dominant strategy (namely a1 ), although B does not. It is part of the game-theory paradigm,
though, that games are played with perfect information—each player knows all the entries in the
payoff matrix. In particular, B can reason just as we did, that A’s strictly dominant strategy is a1 and
therefore A will play a1 . Knowing this, what should B do? A rational B should play the strategy that
is a strictly best response to a1 , that is, b2 . Thus we can predict the complete outcome of the game:
the strategy pair (a1 , b2 ).
Rule 2: Strictly Dominant Strategy Outcomes
• In a two-player game, if both players have strictly dominant strategies, they will be
sure to use them.
• If one player has a strictly dominant strategy and the other player has a strictly best
response to that strategy, those two strategies are sure to be the ones that are played.
There cannot be more than one strictly best response to a given play (because only one payoff can
be greater than all the others); therefore, there cannot be more than one strictly dominant strategy
(though, as we saw in Example 4 on the previous page, one or both players may not have a strictly
dominant strategy at all). This line of thinking suggests that we should give consideration to what
happens when two or more strategies are “tied” for best response. That leads us to the following
slightly weaker ideas.
Definition 5
In a two-player game, a best response by one participant (say A) to a given strategy b of the
other participant (B) is a strategy a such that A’s payoff for (a, b) is greater than or equal to
A’s payoff for any other choice (a0 , b).
It is the “or equal to” that makes the difference between a best response and a strictly best response.
Definition 6
In a two-player game, a dominant strategy for one participant (say A) is a strategy that is
a best response to every strategy that the opponent (B) may play.
There will always be a best response by A to a given strategy by B (possibly more than one if
there is a tie). There need not be any dominant strategies for a given player, though, as Example 4
on the previous page already shows. A dominant strategy that is not strictly dominant can be “tied”
as best option when considered against some opposing strategies. In such a situation, there may not
be a unique rational best choice of strategy; but playing a dominant strategy is never irrational. The
“dominant strategy” versions of our earlier rules are given in the following rules.
Rule 3: Dominant Strategy Rule
• A rational player who has just one dominant strategy will be sure to use it (whether
or not it is strictly dominant).
• A player who has a choice of dominant strategies may rationally play any one of
them.
Rule 4: Dominant Strategy Outcomes
• In a two-player game, if both players have just one dominant strategy, they will be
sure to use them.
• If one player has just one dominant strategy, they will play it, and the other player
will play a best response to that strategy.
Remark 4: In a payoff matrix, a dominant strategy for the first player (you) can be recognized by the
fact that the payoffs to you in the row corresponding to that strategy are greater than or equal to the
corresponding payoffs to you in all other rows. Similarly, a dominant strategy for the second player
(your opponent) can be recognized by the fact that the payoffs to your opponent in the columns
corresponding to that strategy are greater than or equal to the corresponding payoffs in all other
columns.
6.2.3 Nash Equilibrium

The notions of dominant strategy and best response allow us to predict the outcomes in many games,
but not in all. It is often the case that neither player has a dominant strategy. Consider the following
example:
Example 5: A system of lakes and rivers spans the national border between two countries, A and B.
Three lakes on the border, Lake Shadewell, Lake Darkwater, and Lake Moorspring, are polluted to
varying extents. Each country has the option to devote pollution-control efforts to one (and one only)
of the three lakes, and they will derive benefits from cleanup in the lake they focus on according to
the following rules:
• Lake Darkwater is the largest and most polluted of the the three. If both countries focus their
pollution-control efforts on Lake Darkwater they will each derive 70 units of benefit from the
resulting cleanup. But the pollution is so extensive that if only one country focuses on Lake
Darkwater, that country will get just 10 units of benefit.
• The other two lakes (Shadewell and Moorspring) are of similar size and are linked by a system
of rivers whose overall flow is from country A to country B. If one country focuses on cleaning
up (say) Shadewell, it will get 30 units of benefit if the other country is focusing on Darkwater,
and 35 units of benefit if the other country is focusing on Moorspring.5
• If both countries focus on the same one of the smaller lakes (Shadewell or Moorspring), the
total benefit from the cleanup of that lake is 60 units. But because of the direction of river flow,
this benefit is unevenly distributed: Country A will get 15 units of benefit and Country B will
get 45 units.
These rules give us the payoff matrix shown in Table 5 on the next page.
Neither country has a dominant strategy. We can see this by tabulating each country’s (strict) best
responses to the other one’s possible plays, as shown in Table 6 on the next page. Nonetheless, it
seems likely on reflection that both countries (having perfect information) will choose to clean up
Lake Darkwater. This is not only because this strategy offers the potential greatest benefit; as we’ve
seen in the prisoners’ dilemma example, “rational” play can steer players away from the apparently
optimal strategy. Rather it is because each play of Darkwater by one country is the best response to a
5 We imagine that there is an additional synergy coming from the interaction between the two lakes in this case.
Table 5: Payoff Matrix for Polluted Lakes Example
B: Darkwater B: Shadewell B: Moorspring

A: Darkwater 70, 70 10, 30 10, 30
A: Shadewell 30, 10 15, 45 35, 35
A: Moorspring 30, 10 35, 35 15, 45
Table 6: Table of Strictly Best Responses
If A plays B’s best response is If B plays A’s best response is

Darkwater Darkwater Darkwater Darkwater
Shadewell Shadewell Shadewell Moorspring
Moorspring Moorspring Moorspring Shadewell
play of Darkwater by the other country. You can easily check from the table above that (Darkwater,
Darkwater) is the only pair of strategies that has this property, that each one is the best response to the
other one. Thus, there is a kind of “equilibrium” or balance in this strategy pair: neither player has an
incentive to deviate from it. This kind of balance is called Nash equilibrium.
Definition 7
A Nash equilibrium in a two-player game is a pair of strategies (one for each player), each
of which is a best response to the other one.
The notion is named after the American mathematician John Forbes Nash Jr. (1928–2015), who
made fundamental contributions to pure mathematics as well as to game theory and economics. In
1994 he shared the Nobel Prize in Economics for his work on game theory, including the concept of
Nash equilibrium. Nash’s life, including his struggles with paranoid schizophrenia, was the subject of
the 2001 movie A Beautiful Mind, starring Russell Crowe as Nash.
Why should we expect players to arrive at a Nash equilibrium (if one exists)? The answer is a
bit more subtle than our previous discussion of dominant strategies: it depends on thinking about
what each player believes the other player will do. (Remember, they both have perfect information.)
Suppose that (a, b) is a strategy pair that is not a Nash equilibrium. Then one of the players (say A)
knows that their member of the pair (a) is not a best response to B’s play of b; therefore, it is not
rational for A both to play a and to believe that B will play b. In other words, a strategy pair that is
not a Nash equilibrium cannot rationally be believed by both players. The only strategy pairs that both
players can rationally believe are Nash equilibria.
Rule 5: Nash Equilibrium Rule
• When a game has a unique Nash equilibrium, we expect that both players will play
the strategies corresponding to that equilibrium.
• When a game has several Nash equilibria, we expect (but with less confidence) that
both players will play the strategies corresponding to one of those equilibria.
The reason for the “less confidence” is easy to imagine: even after each player has analyzed the
game and found the Nash equilibria, there may be no reason within the structure of the game to prefer
one over the other. For example, consider a modification of the pollution example:
Problem 2: Two countries, A and B, share a lake along their common border. They each have the
option to either let their industries pollute the lake, or not. If neither country pollutes, they will gain
benefits from having a clean lake; if just one country pollutes, it will gain economically while the
other country will lose some quality of life; but if both pollute, the lake will pass through a tipping
point into a eutrophic state (see Example 1 on page 247), and both countries will lose severely. The
payoff matrix is shown in Table 7. Identify the Nash equilibria in this problem. What do you think
Table 7: Payoff Matrix for Shared Pollution Example
B: don’t pollute B: pollute

A: don’t pollute 50, 50 40, 70
A: pollute 70, 40 −80, −80
will actually happen?
Solution: The payoff matrix shows that if B chooses “Pollute,” A’s best response is “Don’t Pollute”
and vice versa. There are no dominant strategies for either player, but both (Pollute, Don’t Pollute)
and (Don’t Pollute, Pollute) are Nash equilibria. Whether the countries will actually arrive at one or
other of these equilibria depends on the standards and ethics of their governments. Suppose country
B’s government is considering the “Don’t Pollute” option. They know that A will “rationally” respond
with “Pollute” and that B will therefore lose some quality of life (50 − 40 = 10 units of loss in our
example above). For some governments, accepting this loss (which would be entirely A’s fault) would
feel shameful and might be politically difficult. This would tend to push B toward the “Pollute” option.
But of course similar reasoning applies to A’s government. Without very careful negotiations (which
would perhaps explore other policy options than the simple yes/no to pollution that we have modeled)
there is a serious danger that both countries will end up in the “worst of all possible worlds,” the
(Pollute, Pollute) state.
A two-player game with this structure is sometimes called a chicken game, after the famous
“Chickie Run” scene in the 1956 James Dean movie Rebel Without a Cause.
Mixed Strategies
Let’s return to the example we used to start this whole section: the “Rock, Paper, Scissors” game
(Example 2 on page 361). Every child who has played the game knows that Rock is the strictly best
response to Scissors; Scissors is the strict best response to Paper; and Paper is the strictly best response
to Rock. There are no dominant strategies in this game and there are no Nash equilibria either, at least
in the way we have defined the concept so far (Definition 7).
That’s not surprising when we think about the underlying idea that Nash equilibrium represents
some kind of equilibrium of beliefs. After all, if your opponent in “Rock, Paper, Scissors” believes
they know what you will do, then they know how to find a strategy to beat you; but they must also
believe that you would know this and that you would therefore change the strategy they originally
believed you would use. No wonder a Nash equilibrium can’t be found! Children soon learn that the
best thing to do in “Rock, Paper, Scissors” is to play unpredictably. This introduces the idea of a
mixed strategy.
Definition 8
A mixed strategy, used by a player in a game, is a random process (Definition 1 on

page 278) whose sample space (Definition 2 on page 280) is the set of “ordinary” strategies
available to that player.
In other words, a mixed strategy is a way of choosing at random among the different ways of playing
the game. (In this context, the “ordinary” strategies or ways of playing the game are often called pure
strategies to distinguish them from the more complex notion of Definition 8.) An example of a mixed
strategy for player A in “Rock, Paper, Scissors” might be something like
I (Player A) will roll a die. If it comes up 1, 2, or 3, I will play Rock; if it comes up 4
or 5, I will play Paper; if it comes up 6, I will play Scissors. (We might call this “Mixed
Strategy α.”)
Because a mixed strategy is a random strategy, the payoff when one or more players are using mixed
strategies is a random variable. The notions of best response, dominant strategy, and so on can still be
defined though: the idea is to consider the expected payoff , that is, the expected value (Definition 2 on
page 303) of the payoff, instead of the payoff itself. For example, if Player A chooses Mixed Strategy
α above, and Player B responds with Paper, then Player B’s payoff will be +1 if A plays Rock (3 of 6
equally probable outcomes), 0 if A plays Paper (2 of 6 equally probable outcomes), and −1 if A plays
Scissors (1 of 6 equally probable outcomes). The expected payoff to B is therefore
3 2 1 2
× (+1) + × (0) + × (−1) = ≈ 0.33,
6 6 6 6
and A’s expected payoff is −0.33. What’s more, you can check that this is the best payoff to B from
any response (pure or mixed) to A’s use of Strategy α. In other words, B’s best response to α is Paper.
But A’s best response to Paper is not Strategy α, but Scissors, so by the same reasoning as before,
Strategy α cannot be part of any Nash equilibrium.
Nevertheless, there is a Nash equilibrium among mixed strategies. Consider the following variant
strategy by A:
I (Player A) will roll a die. If it comes up 1 or 2, I will play Rock; if it comes up 3 or 4,
I will play Paper; if it comes up 5 or 6, I will play Scissors. (We will call this “Mixed
Strategy α 0 .”)
Strategy α 0 (used by Player A) has the surprising property that whatever strategy, pure or mixed,
Player B chooses in response, B’s expected payoff is zero! Thus, by definition, any strategy whatever
by B is a best response to A’s choice of α 0 ; in plain language, nothing that B does can make any
difference. In particular, we could consider the exact same mixed strategy used by B:
I (Player B) will roll a die. If it comes up 1 or 2, I will play Rock; if it comes up 3 or 4,
I will play Paper; if it comes up 5 or 6, I will play Scissors. (We will call this “Mixed
Strategy β 0 .”)
Now β 0 is a best response to α 0 , because anything is a best response to α 0 . For the same reasons, α 0
is a best response to β 0 . Thus (α 0 , β 0 ) is a mixed strategy Nash equilibrium (in fact it is the only one).
Part of the work for which John Nash won the Nobel Prize was a mathematical proof of the fact—
which we have already illustrated in the “Rock, Paper, Scissors” example—that if mixed strategies
are allowed, there is always at least one Nash equilibrium. Indeed, kids who play “Rock, Paper,
Scissors” usually find this equilibrium quite soon for themselves, without the need for any elaborate
mathematical analysis!
Remark 5: We won’t discuss mixed strategies further in this book, but they are important in various
kinds of environmental and ecological modeling. One reason for this is the evolutionary interpretation
of the idea of a mixed strategy: instead of imagining a single player repeatedly randomizing over
the available pure strategies, we imagine a population of apparently identical organisms, each of
which is “programmed” to play one of the pure strategies, and whose members meet and interact at
random. Effectively, each encounter is then a mixed-strategy play, because it involves (on both sides)
a random choice of individuals programmed with one of the available pure strategies. We could also
imagine, if we like, that the payoffs in the game have some effect on the evolution of the population:
maybe a higher payoff is associated with leaving more offspring (greater “reproductive success,” as
the biologists like to put it). How, then, will the population change over time? Will it converge toward
proportions that represent a Nash equilibrium? These are fascinating questions, which we cannot
explore further here.
Think about it. . .
President Richard Nixon claimed to espouse the “madman theory” in foreign policy: that
is, he wanted to convince the leaders of hostile nations that his behavior was unpredictable
and turbulent. According to this theory, those leaders would then do all they could to avoid
provoking him, out of fear of a disproportionate response. In what ways is this “madman
strategy” similar to the game-theoretic concept of a mixed strategy, and in what ways is it
different?
Summary of Ideas: Game Theory
• A game models a decision-making situation involving two (or more) agents or

players whose choices of strategy influence one another’s outcomes.
• The payoff matrix for a game describes the outcomes for each of the agents involved
for each possible choice of strategy by each of them.
• In a two-person game, a strategy for Player A is strictly dominant if it leads to a
better outcome for A than any other, whatever strategy B may choose. It is dominant
if it leads to an outcome that is at least as good as any other, whatever strategy B may
choose.
• In the prisoners’ dilemma, each player pursues a strictly dominant strategy, but the
result is less than the best for each of them.
• A game in which only one player (say A) has a strictly dominant strategy can be
analyzed by assuming that B will choose the best response to a strictly dominant
play by A.
• A Nash equilibrium is a pair of strategies (one for A, one for B) each of which is a
best response to the other.
• A mixed strategy is one that involves random choices by the player.
• If mixed strategies are permitted, there is always at least one Nash equilibrium.
6.3. THE TRAGEDY OF THE COMMONS 373
6.3 The Tragedy of the Commons
Objectives
I know what is meant by a commons and can give examples.

I understand how the structure of a commons can drive individual users to over-
exploit it, making everyone worse off.
I can model this process (the tragedy of the commons) in terms of repeated two-
player games.
I understand some of the free-market and regulatory measures that have been
suggested to mitigate the tragedy of the commons. Specifically, I can explain the
strategies of privatization, regulation, cap and trade, and revenue-neutral use
taxes, and some of their advantages and disadvantages.
I understand the term regulatory capture and can describe the danger that it
represents.
I understand the additional difficulties involved in managing a commons when there
is no single body that has authority over it.
6.3.1 The Idea of a Commons

In this section we are going to look at a classic example of decision-making in a sustainability context:
the management of a commons.
Definition 1
A commons is a resource that can be freely used by each of the members of some group.
Each member of the group can use as much of the resource as they want, but once some of
the resource is used by one member, that part is no longer available for use by any other
member.
Example 1: . The name “commons” comes from the way grazing rights were allocated in medieval
societies in Europe. Grazing land might be held “in common” between a number of farmers in a
village. Each farmer had the right to graze their animals (however many) on the common land.
Example 2: A public highway is a “commons” in the sense of Definition 1. As a member of the

community, you have the right to drive your vehicle on the public highway for your own benefit, as
much as you want. This commons is paid for by taxation shared among community members: some
of this taxation may be related to how much you use the highway (e.g. gasoline taxes) but the rest of
it is not (income and property taxes, for example).
Example 3: Subsurface water (water in an aquifer, such as the Ogallala aquifer that underlies much
of the farmland of the Midwest) is a commons, where many landowners have the right to withdraw
water from a common aquifer that runs beneath each individual landowner’s property.
Example 4: International law treats ocean fisheries (outside of national waters) as a commons: anyone
suitably equipped can take as much fish as they like. National fisheries (in the waters of a particular
nation) have in the past been treated as a commons for fishing vessels that fly that country’s flag,
though many now are regulated to a greater or lesser extent.
Example 5: The atmosphere is a striking example of a worldwide commons. We all depend on the
atmosphere—with our every breath—as a source of the oxygen that our bodies need to function. We
also treat the atmosphere as a common sink: vehicles, homes, power plants, and factories emit various
kinds of pollution into the atmosphere, which is dispersed around the world. (In this case, the “benefit”
that members derive from the commons is the “ability to pollute.”) This function of the atmosphere is
particularly in view when we think about human-caused climate change: my ability to get around in
my CO2 -emitting car is a great benefit to me, whose costs (in terms of changing the world’s climate)
pass into the atmosphere and are spread over the whole world.
Problem 1: Making reasonable assumptions, compare the weight of CO2 waste produced by a car
carrying one passenger one mile with the weight of actual poop produced by a horse carrying one
rider the same mile.
Solution: Let’s start with the car, because it is more familiar (to most of us, at least), and let’s assume
that it gets about 25 miles to the gallon.6 Burning one gallon of gasoline releases about 20 pounds of
CO2 (see Section 8.2, “Carbon intensity of different fuels”) so traveling 1 mile in this car produces
× 1gal
20 lb
1
mi × 1 gal ≈ 0.8 lb
25mi
of CO2 .
Now the horse. According to [348], a 1000-pound horse will produce about 50 pounds of solid and
liquid waste per 24-hour day (see Example 5.5 in [345] for a way of estimating this without asking a
horse specialist!). If we imagine that the horse is walking with our passenger at 6 miles per hour, it
will take ten minutes (one-sixth of an hour) to travel 1 mile, and the share of its 50 daily pounds of
waste generated in those 10 minutes is
10min
1
hr
50 lb × ≈ 0.35 lb.

×
24
hr
60
min
Thus the car produces about twice as much carbon dioxide “waste” as the horse produces visible
“waste” to carry the same passenger the same distance.
The point of this calculation is not that we should all go back to traveling by horse! It does, however,
help us see very clearly the extent to which we make use of the “commons” of the atmosphere, every
day, to waft away wastes that would represent a disturbing accumulation if they happened to be visible
solids rather than invisible gases.
The term “commons” was introduced into discussions of sustainability and ecology by Garrett
Hardin [153]. Writing in 1968, Hardin pointed out that when each of the members of a group derives
some benefit from a commons, the interest each member has as an individual in maximizing their
benefit from the commons is different from the interest the group as a whole has in maintaining
the quality of the commons. This difference is liable to drive the group toward behavior that makes
everyone worse off—it does not even meet the low standard of Pareto efficiency introduced in
6 In 2017 the average fuel economy for new passenger cars was about 30 miles per gallon (mpg); the average fuel economy
for all passenger vehicles and light trucks on the road (including SUVs) was 25 mpg. The fuel economy of individual cars is
not regulated; manufacturers instead have to meet an average fuel economy (the CAFE standard) for the entire fleet of vehicles
produced in a given year [74].
50
Milk Production (gallons/day)

40
30
20
Total
Per Cow
10
0
0 5 10 15 20
Number of Cows
Figure 4: Productivity of the commons as a function of the number of cows grazing.
Definition 2 on page 349. In other words, Adam Smith’s “invisible hand” does not operate as it should
to achieve prosperity for everyone; the market paradigm (Definition 1 on page 349) fails.
The situation is similar to the prisoners’ dilemma that we studied in the previous section. Hardin’s
central interest was in population control,7 but he illustrated his idea with a very simple example, one
that we can study in detail using the game-theoretic ideas of the previous section.
The Example of Two Farmers

Now we’re going to apply these ideas in a sustainability context.
We imagine that there are just two farmers, A and B, who share some common grazing land. Both
A and B can put a certain number of dairy cows on the land; but as the total number of cows on the
land increases, the productivity of each individual cow becomes less. To be specific, let us assume that
when there are n cows on the land, the productivity of each one is 10 − n/2 gallons of milk per day.
Problem 2: What number of cows on the land will yield the greatest total volume of milk produced?
Solution: To answer this, we write the total amount of milk produced as (number of cows) times
(gallons of milk per cow). That is
n × (10 − n/2) gallons per day
if the number of cows is n.

The graph of this function is shown in Figure 4. It is a parabola that passes through the horizontal
axis at n = 0 and again at n = 20. The “nose” of the parabola is halfway between the two points at
which it crosses the axis, that is, at n = 10, where the total production is 50 gallons per day.
(You can see this by algebra also, if you like. Write
n(10 − n/2) = 10n − 12 n2 = 50 − 12 (n − 10)2 ,
using the process of “completing the square.” The squared term on the right of the equation is always
positive, so that n(10 − n/2) is equal to 50 minus a positive number, so its greatest possible value is
50, which occurs when the “positive number” is zero, that is, when n = 10. )
7 We will not follow him in this direction. See Case Study 7.4 for a more modern perspective on population and
sustainability.
Definition 2
The level of use of the commons that maximizes overall production (n = 10 in the example)
is called the social optimum.
If the land was all owned by a single farmer who started with nothing and whose objective was
to produce as much milk as possible, she would keep adding cows to her herd, one by one, until
production no longer increased—that is, at n = 10. In other words, a single farmer owning all the land
would automatically build up the herd exactly enough to reach the social optimum—there would be
no distinction between social goals and private goals in this case. But what happens when there are
two farmers? We can model their options, and their expected behavior, by means of game theory.
Problem 3: Suppose that two farmers, Farmer A and Farmer B, both have rights over the commons.
Each farmer, initially, has no cows. Each of them faces a choice of strategies: either to add a cow to
their herd, or to stand pat (with no cows). Their payoffs from these actions are in terms of gallons
of milk produced per day. Find the payoff matrix for this situation. What behavior does game theory
predict?
Solution: We represent the situation by a game in which the players are Farmers A and B and the
strategies are “Add a cow” and “Don’t add a cow.” If one farmer adds a cow and the other one does
not, the farmer who adds the cow will benefit by 9.5 gallons of milk daily (because with n = 1, the
productivity of each cow is 10 − n/2 = 9.5 gallons daily). If both farmers add a cow, they will each
benefit by a slightly smaller amount, 9 gallons of milk daily (because with n = 2 the productivity
per cow is 10 − n/2 = 9). The payoff matrix is shown in Table 8. Clearly, “add a cow” is a strictly
Table 8: Should I Buy a Cow?
B does not add a cow B adds a cow

A does not add a cow 0,0 0, 9.5
A adds a cow 9.5,0 9,9
dominant strategy for both of the farmers. So we expect8 that both of them will do that, and we will
reach the state in which each farmer has one cow on the land.
Now, here is the key idea. For each possible situation in which Farmer A has a cows and Farmer
B has b cows, we can model their choices by a game with similar strategies (“Add a cow to your
herd,” “Stand pat,” and also, possibly, “Remove a cow from your herd”) but with different payoffs.
The payoffs will be different because the productivity of each cow is governed by the total number
n = a + b of cows via the formula 10 − n/2. If the farmers “play the game” many times, we might
expect that they will end up in an “equilibrium” state in which game theory advises both of them to
stand pat. Where will this equilibrium be? At the social optimum? Or somewhere else?
We’ll make some simplifications in analyzing this question, so as to keep the key idea in focus while
making the algebra more manageable. First, we’ll neglect the “Remove a cow” strategy completely.
Second, we’ll look only at those cases in which each farmer has the same number c of cows (so
a = b = c and n = 2c). Before the game is played, then, each farmer has c cows; the productivity per
cow is 10 − n/2 = 10 − c gallons per day; so each farmer’s production before the game is played is
8 Of course, in the real world, cows do not come for free: adding a cow will be a significant capital expenditure for a farmer,
and the cost of that expenditure (suitably discounted over time) would have to be incorporated in the payoff matrix. Trying to
account for this would significantly complicate our analysis without changing its general structure. So we will just assume that
cows are available to our farmers by some mechanism that we don’t try to model.
c(10 − c) gallons per day. The payoffs for the farmers are their total milk production figures after they
play their chosen strategies:
• If neither farmer chooses to add a cow, nothing changes: both payoffs are c(10 − c) = 10c − c2 .
• If both farmers choose to add a cow, then the productivity per cow changes to 10 − (c + 1) =
9 − c. Each farmer now has c + 1 cows producing at this rate, so each farmer’s total production
(payoff) is (c + 1)(9 − c) = 9 + 8c − c2 .
• If one farmer (say A) adds a cow and the other does not, the productivity per cow changes to
10 − 12 (2c + 1) = 9.5 − c. This changes the total production of both farmers: A’s payoff becomes
(c + 1)(9.5 − c) = 9.5 + 8.5c − c2 (because A now has c + 1 cows) and B’s total production
becomes c(9.5 − c) = 9.5c − c2 . The same calculation can be applied if B adds a cow and A
does not.
We therefore obtain the general form of the payoff matrix shown in Table 9. To understand how
Table 9: Should I Buy a Cow? (We Both Have c Cows!)

A does not add a cow 10c − c2 , 10c − c2 9.5c − c2 , 9.5 + 8.5c − c2
A adds a cow 9.5 + 8.5c − c2 , 9.5c − c2 9 + 8c − c2 , 9 + 8c − c2
this works out for the farmers, we have to put in some specific values of c. When c = 0 we are back
in the setup of Problem 3: neither farmer has any animals. In this case, Table 9 reduces to Table 8
and as we saw, the strictly dominant strategy for both A and B is to add a cow. This brings us to
the situation in which both have one cow in their herd (c = 1). The payoff matrix is a little different
now. Computing the formulas from Table 9, we get the payoff matrix for c = 1 shown in Table 10.
Although the payoffs are slightly different, “Add a cow” is still a strictly dominant strategy for both
Table 10: Should I Buy a Cow? (We Both Have 1 Cow)

A does not add a cow 9, 9 8.5, 17
A adds a cow 17, 8.5 16, 16
farmers and is therefore a Nash equilibrium, and so it is the behavior we expect to see. What’s more,
the pair (“add a cow,” “add a cow”) is socially optimal (It maximizes total production) and is therefore
also Pareto optimal. Both farmers will add another cow, and indeed they will continue doing so until
we reach the social optimum at c = 5. At this point, things start getting interesting. The payoff matrix
is shown in Table 11. “Add a cow” remains a strictly dominant strategy for B, whose payoffs (25, 27)
Table 11: Should I buy a cow? (We both have 5 cows)

A adds a cow 27, 22.5 24, 24
in row 1 dominate the corresponding payoffs (22.5, 24) in row 2. Similarly, “add a cow” is a strictly
dominant strategy for A. Thus we expect that they will both play this strategy, increasing their herds
to 6 cows each (c = 6). This will happen even though this is not the socially optimal outcome: total
production will fall from 50 to 48 gallons per day as a result of this action.
Do you recognize the pattern here? It is the same pattern that we observed in our discussion of the
software engineers’ problem (Problem 1 on page 362), namely the prisoners’ dilemma. If both farmers
could stop adding animals at this point, each of them would be better off. But it is not “rational” for
them to cooperate in this way. Instead, the strictly dominant strategy for each of them is to add another
cow, because that is the best response to either choice by the other farmer (2 > 0 if the other doesn’t
add a cow, and −1 > −2.5 if they do). So “rational” behavior will push both farmers to add another
animal. We are now at c = 6. There’s a slight change here because “add a cow” is no longer a strictly
Table 12: Should I Buy a Cow? (We Both Have 6 Cows!)

A does not add a cow 24, 24 21, 24.5
A adds a cow 24.5, 21 21, 21
dominant strategy (see Table 12): if Farmer B chooses to add a cow, then Farmer A’s payoffs for
“add a cow” and “don’t add a cow” are equal (21 in each case). But “add a cow” is still a dominant
strategy for each farmer, in fact the only dominant strategy, so we can be confident that they will both
use it. This gets us to c = 7. Only at this point does it finally become “rational” for each farmer to
Table 13: Should I Buy a Cow? (We Both Have 7 Cows!)

A adds a cow 20, 17.5 16, 16
stop building up their herd: “do not add a cow” is the strictly dominant strategy for each of them (see
Table 13). They have reached a total stock of n = 14 cows and a total production of 42 gallons per day,
significantly below the 50 per day they would have achieved had they managed to stop at the social
optimum. (If there had been three or more farmers rather than two, the effect would have been more
extreme: their collective action would have driven the use of the resource still higher.) Hardin called
this behavior the tragedy of the commons.
Definition 3
The tragedy of the commons is the fact (contrary to what the market paradigm would
suggest) that “rational” exploitation of an unregulated commons often leads to overuse of
the resource and to results that make everyone worse off.
Can the “tragedy” of the commons be avoided? In the following section we’ll look at some
proposals that have been made to escape this trap.
6.3.2 Managing the Commons

As we have seen, when multiple agents have access to a resource that is an unregulated commons,
then limiting usage to the social optimum (best for everyone) requires agreement among the agents to
cooperate. Since they don’t have a rational reason to do this, they are likely to push past the optimum—
a course of action that harms everyone, including themselves. We can give several historical examples
of this, as well as potential future ones.
• The near-collapse of the Atlantic cod fishery (see Remark 1 on

page 151 and the sources referenced in the footnote there): “in
the 17th century, cod were said to be so abundant that you could
almost walk across the ocean on their backs.” For a similar
example on land, consider the hunting to near-extinction of the
American bison in the nineteenth century (Figure 5): from a
pre-1800 population of at least 60 million, the number of bison
dropped to less than a thousand before recovering after the end
of large-scale hunting and protection in National Parks. The
current population is around half a million.
Figure 5: Pile of bison skulls waiting to be
ground into fertilizer, ca. 1892 [216].
• Traffic on a highway—anyone has the right to drive on the
highway, but if too many do so, congestion makes the journey
worse for everyone. If everyone who was able to do so would
carpool to work (say), the resulting savings in congestion would more than offset the small time
inconvenience of carpooling; but there is no motivation for any single “rational” individual to
carpool, because their action will make so little difference to the congestion problem.
• The management of shared aquifers such as the Ogallala is an example of an unregulated

commons that poses a problem, which (for now) largely lies in the future (the present solution
being simply “drill deeper,” which is not sustainable, since the aquifer is not infinitely deep!).
• Climate change, caused by emissions from fossil fuel burning, is another example of a problem
related to the use of an unregulated commons, in this case the CO2 -absorbing capacity of
the atmosphere. Several of the solutions discussed below have been considered as policies to
address the issue of climate change.
Human beings have come up with several different solutions to the problem of an unregulated
commons. Some solutions focus more on the word “unregulated” and try to address the problem
through regulations or controls; others focus more on the word “commons” and try to address
the problem through creating clearly defined property rights. Broadly speaking, these two kinds of
approaches are associated respectively with the “left” and the “right” of the political spectrum. But
all of them aim to align the individual interests of the stakeholders with the interest of society as a
whole, so that the “tragic” effects of the prisoners’ dilemma will not occur. We will examine four such
solutions: privatization, use limits, cap and trade, and revenue-neutral use taxes.
Privatization
One way to resolve the misalignment between individual and collective interests that lies at the root
of the tragedy of the commons would be to remove the element of collective ownership completely—
to privatize the commons. In our example of the two farmers, suppose that the area of the common
land is 14 acres in total. We could divide it in two, giving Farmer A exclusive property rights over
seven acres and Farmer B rights over the other seven acres. Each of the farmers would then reach the
social optimum of 5 cows on a seven-acre holding, because reaching the social optimum would also
maximize his or her individual production.
This is not simply a theoretical idea: in fact, a large part of the

common agricultural land of England and Wales (nearly 7 million
acres) was privatized or “inclosed” between the seventeenth and
early twentieth centuries. Special legal proceedings were required
for these “inclosures” (Figure 6) , which were carried out because
a single landowner had the motivation to make improvements to
the land, such as drainage or fertilization, which no “rational”
commoner would be motivated to make on their own.
Privatization can be an effective solution if the commons can
be divided up in such a way that exclusive property rights can
be established over separate portions of it: in the language of
economics, if it represents an excludable good. But not all com-
mons can be divided up in this way. For example, how would
you establish an exclusive property right over your “share” of an
aquifer that ran under your land—or, for that matter, over your
“share” of the air that you need to breathe?
Figure 6: Title page of an Inclosure Act [29].

Regulation
A different way to address the tragedy of the commons would be
to retain the shared ownership principle but to limit the number
of cows that each farmer is allowed to keep. This limit could be a
legal one, or in a small community like a village it could be enforced by shared social convention. We
call this approach to managing the commons regulation.
1 15
0.8
(grams per kwh)
(grams per kwh)

NOx emissions
PM emissions
10
0.6
0.4
5
0.2
0 0
1985 1990 1995 2000 2005 1985 1990 1995 2000 2005 2010
Vehicle Model Year Vehicle Model Year
(a) Particulate matter (PM). (b) Nitrogen oxides (NOx ).

Figure 7: Tightening of U.S. emissions standards for heavy-duty compression-ignition (diesel) engines. Data from [32].
Example 6: We saw in Section 6.1 that pollution is often an externality for the polluter: one person
(or organization) gains profit by polluting a common resource, whereas the costs of the pollution
are spread among all the resource’s users. In the language of this section, we recognize this as
another tragedy of the commons, where the polluter is “rationally” motivated to overuse the pollution-
absorbing capacity of the common resource. Regulation is one of the most common approaches to
limit this sort of pollution: a suite of regulations that limit the amount of a pollutant that any one user
can discharge into the commons. For example, Figure 7 shows the tightening emissions standards for
U.S. heavy truck engines over time (there are two graphs, one referring to the emission of “particulate
matter” (PM), otherwise known as soot, and one referring to oxides of nitrogen, which are primary
constituents of smog).
Of course, regulating the pollution emissions of each truck may not be enough if the number
of trucks increases significantly. Similarly, in our cows-and-farmers model, limiting the number of
cows per farmer to five is fine when there are just two farmers, but if two more farmers arrive in the
community, the stage will again be set for a tragedy of the commons. The issue is that setting a limit
on each individual participant does not accurately reflect the nature of the problem. What we need to
limit is the overall effect of cows (or trucks), not the enterprise of individuals. This leads to two, more
nuanced, approaches that target the overall outcome more directly.
Think about it. . .
In his 1968 paper, Hardin raised the question of management of a commons by social
pressure—what he called “conscience”—and asserted that it could never succeed in the
long run, since those who disregarded “conscience” would always out-compete those who
paid heed to it. What do you think of this argument?
Cap and Trade

The cap and trade process for managing a commons is a hybrid of the property-rights and regulation
approaches. As in the regulation approach, it begins with an upper limit. But now, the upper limit is
for the overall use of the commons—the total number of cows, the total amount of pollution emitted,
or whatever. The regulator issues permits for the use of the commons up to the predetermined limit.
For instance, each permit might give the right to graze one cow on the common land, or to emit some
predetermined amount of pollution. Then (and this is the key “market” step) those permits are made
tradable. Owners of permits can sell them to others at a mutually agreed price. So, instead of having
a market in subdivided parts of the commons itself (as in option 1, privatization, with its attendant
problems), there is a market in permits to use the commons. See Figure 8.
Allowance Unused Allowance Allowance

5 tons 5 tons
3 tons 7 tons
Money
Figure 8: The cap and trade system.
Example 7: Let’s take a look at how this might work using the farmers-and-cows example again.
Suppose that the regulator has issued 10 tradable cow permits, 5 each to Farmer A and Farmer B. At
some point, let’s imagine that A has a herd of 5 cows and B has 4. The production per cow is then
10 − 92 = 5.5 units (gallons per day), so that A’s production is 5 × 5.5 = 27.5 and B’s production is
4 × 5.5 = 22.
There are two ways to reach the social optimum of 10 cows on the land: either B adds an extra cow,
so that their herds become 5 each, or A purchases B’s permit and then adds a cow, so that their herds
become 6 and 4. Because of the permit system, A has to negotiate with B to purchase the additional
permit if A wants to add an extra cow. That is the way that the cap and trade system avoids the tragedy
of the commons.
Let’s work out what would be a fair price for A to pay B for the permit. We make a table (Table 14)
comparing the farmer’s net production in three possible scenarios: their present state (A has 5, B has
4); the option whereby B adds a cow (both have 5); and the option whereby A has a cow (A has 6, B
has 4).
Table 14: Milk Production Scenarios

Scenario A’s production B’s production
Base case (A:5, B:4) 27.5 22
B adds (A:5, B:5) 25 25
A adds (A:6, B:4) 30 20
In each of the two “added cow” scenarios, there are now 10 cows on the land, so the production per
cow is 10 − 10
2 = 5. If B adds a cow, B’s production gain is 3 and A’s loss is 2.5; if A adds a cow, A’s
production gain is 2.5 and B’s loss is 2. These become equivalent if A pays B 5 units for the permit.9
In summary, cap and trade combines two philosophies: a regulatory approach that determines how
many permits should be issued (cap), and a market approach that allows the buying and selling of those
permits (trade). The poster child for the success of this approach is the U.S. Acid Rain Reduction
Program. “Acid rain” is largely caused by emissions of sulfur oxides from coal-fired generating
stations, and in the 1980s, it was threatening to severely damage forests in the eastern U.S. Acid rain is
a classical “commons” problem, because emissions are mixed in the atmosphere, and emissions from
one power plant may end up harming vegetation over a wide area. Beginning in 1990, the U.S. EPA
(Environmental Protection Agency), acting under Title IV of the 1990 Clean Air Act Amendments,
instituted a cap and trade program for managing sulfur oxide emissions. The program significantly
exceeded its goals, achieving 50 percent reductions and more in emissions even as the economy
was growing. The European Union currently operates a CO2 emissions trading scheme modeled on
the U.S. acid rain program, which captures about half of Europe’s total emissions. A proposal for a
similar CO2 emissions trading scheme in the U.S. (the Waxman–Markley bill) was introduced at the
beginning of the first Obama presidency, but was defeated in the Senate.
Critical Thinking
It has been argued that by making the “right to pollute” a tradable thing, cap-and-
trade obscures the ethical dimensions of pollution. For example, one of the principles of
wilderness travel is “pack it in, pack it out”—any waste that you generate on your journey
must be brought back. Suppose a regulatory agency decided that a wilderness area could
tolerate a certain limited amount of waste, and issued tradable permits allowing you to
pollute the wilderness to that extent; for example, to throw one empty soda can into the
Grand Canyon. Would you buy (and use) such a permit? What effect would doing so have on
your experience and understanding of what “wilderness” is? Is this an appropriate analogy
for the tradable permits of a “cap and trade” policy? We’ll return to these questions in
Section 6.4.
9 Strictly speaking, our units are daily production values. For an outright purchase of the permit, economic theory would
suggest that A pay B an amount of money that would equal the present value, at the prevailing discount rate (Definition 6 on
page 354), of a flow of 5 gallons of milk daily.
Remark 1: Cap and trade, and similar schemes, assign great power to the regulatory agency (the EPA
in the “acid rain” example above). The agency assigns the number of permits, and it is supposed to
do so on the basis of a purely scientific judgment about where the “social optimum” (the long-term
best interests of society) lies. In practice, however, there is always a danger that a regulatory agency
will become corrupted (either by explicit bribery, or by conformity and “groupthink”) by the very
industries it is supposed to regulate. This is called regulatory capture. Bruce Schneier [288] writes:
There’s one competing interest that’s unique to enforcing institutions, and that’s the
interest of the group the institution is supposed to watch over. If a government agency
exists only because of the industry, then it is in its self-preservation interest to keep that
industry flourishing. And unless there’s some other career path, pretty much everyone
with the expertise necessary to become a regulator will be either a former or future
employee of the industry, with the obvious implicit and explicit conflicts. As a result,
there is a tendency for institutions delegated with regulating a particular industry to
start advocating the commercial and special interests of that industry. This is known as
regulatory capture, and there are many examples both in the U.S. and in other countries.
Let’s now consider the fourth possible solution to the tragedy of the commons.
Revenue-Neutral Use Tax

In this scenario a tax is instituted on the use of the commons. All the proceeds of the tax are then
shared among all stakeholders as a dividend. In our cows-and-farmers example, the stakeholders are
the farmers with rights to use the commons. They might pay a certain tax rate per cow that they
own, with the proceeds divided equally among all farmers who have a right to the commons. The tax
therefore redistributes money among the stakeholders, but has no effect on their overall wealth—this
is what is meant by “revenue neutral.”
This approach to reducing carbon emissions is advocated by the Climate Leadership Council [34],
a recently formed nongovernmental organization. Martin Feldstein and Gregory Mankiw (who
chaired the President’s Council of Economic Advisors under Ronald Reagan and George W. Bush,
respectively) laid out their argument in a 2017 New York Times op-ed [112] entitled A Conservative
Case for Climate Action. In brief, their plan would involve
• A federal tax on carbon dioxide emissions, starting perhaps at $40 per ton and increasing
steadily over time.
• A commitment to return the entire proceeds of the tax equally to the American people. They
estimate that a family of four would receive a dividend of about $2, 000 in the first year,
increasing as the tax increases.
• A “border adjustment” process so that U.S. companies exporting to countries without similar
carbon pricing would receive rebates on the carbon taxes they have paid, and conversely imports
from such countries would be charged fees on the embedded carbon in their products.
• A rollback of regulations “made unnecessary by the carbon tax.”
Example 8: The Carbon Tax Center (CTC) is an advocacy organization that argues for the
implementation of a revenue-neutral carbon tax (they call it “carbon fee and dividend”) in the
United States. Two-thirds of Americans polled by Stanford University in early 2015 supported the
implementation of such a system (Figure 9 on the next page). See [185], where CTC’s Charles
Komanoff writes
For more years than I care to count, the Carbon Tax Center beseeched pollsters to take
Americans’ temperature on revenue-neutral carbon taxes. Time and again we explained
Figure 9: Carbon Tax Center. Polling data from [185].
that polling about carbon taxes had to incorporate the option of returning revenues to
households—as most carbon tax bills would do. Otherwise, the tax came off as all stick
and no carrot, and about as appealing to most folks as a cold shower.
Finally, a Stanford University–Resources for the Future poll asked that question. The
results. . . show that two-thirds of Americans support making corporations pay a price for
carbon pollution, provided the revenues are redistributed, i.e., made revenue-neutral.
In Example 9 below, we’ll use our cow example once again to see how a revenue-neutral use tax
could operate in practice. Before we dive into the details of how a revenue-neutral tax might work in
this particular example, let’s take a moment to review our assumptions about the model and its overall
purpose.
• Our cow example is a “toy model”; we want enough detail in the model so that we gain insight
into how a revenue-neutral tax would work in practice (in something like the carbon fee and
dividend proposal, for example),10 but not so much detail that the analysis becomes intractable;
see the comment about constructing environmental models (page 74) and also Section 2.1.4 on
page 77.
• In light of the previous bullet point, we are assuming there are just two stakeholders: the two
farmers with shared access to the commons. In a more realistic setting everyone in the village
might be considered a stakeholder, just as everyone on the planet is a stakeholder in policy
decisions concerning greenhouse gas emissions.
• Finally, our goal is in understanding how a revenue-neutral tax can be used to align the
individual incentives of each stakeholder with a socially optimal outcome. It turns out that
limiting ourselves to two stakeholders is sufficient for this purpose, while greatly simplifying
the analysis.
Example 9: Suppose that the village imposes a revenue-neutral cow tax of 4 units of production per
cow on each of the farmers A and B, with all the proceeds then being returned to both farmers equally
(revenue neutrality). Let’s once again investigate what happens as the farmers try to decide whether
or not to increase their stock. The revenue-neutral tax will change the payoff table in an interesting
way. We reproduce the old payoff table (Table 9 on page 377), in which each farmer had c cows, here
as Table 15 on the opposite page.
10 In fact, we have already a straightforward solution for our cow example: privatization, as noted on page 379. By
implementing a cap and trade system or a revenue-neutral tax in this familiar setting, we gain insight into more relevant,
but also more abstract settings that involve public goods and many more stakeholders.
Table 15: Should I Buy a Cow? (We Both Have c Cows!)

A does not add a cow 10c − c2 , 10c − c2 9.5c − c2 , 9.5 + 8.5c − c2
A adds a cow 9.5 + 8.5c − c2 , 9.5c − c2 9 + 8c − c2 , 9 + 8c − c2
Adding the revenue-neutral tax has no effect on the top left and bottom right entries of the payoff
table. In these positions both farmers have the same number of cows, so they each pay the same amount
of cow tax, and then they each receive the same amount back again because of revenue neutrality. The
net effect of the tax is zero.
But in the bottom left corner, say, Farmer A has c + 1 cows and Farmer B only has c cows. Thus
Farmer A pays 4(c + 1) = 4c + 4 units of tax and Farmer B pays 4c units. The total tax take is therefore
8c + 4 units, which is redistributed as 4c + 2 units to each of the farmers. The net effect is therefore
that A loses 2 units of payoff (compared to the no-tax situation) and B gains 2 units. The same thing
happens in the top right corner (with A and B exchanged). The resulting payoff matrix is given in
Table 16
Table 16: Should I buy a cow? (We Both Have c Cows and Tax is Payable!)

A does not add a cow 10c − c2 , 10c − c2 2 + 9.5c − c2 , 7.5 + 8.5c − c2
A adds a cow 7.5 + 8.5c − c2 , 2 + 9.5c − c2 9 + 8c − c2 , 9 + 8c − c2
To figure out when it will be a strictly dominant strategy for A to add a cow, we must compare
A’s payoffs under the “add a cow” and “don’t add a cow” strategies, for each of the possible strategic
choices by B.
• If B chooses not to add a cow, then A’s payoffs are 10c − c2 for “don’t add a cow” and 7.5 +
8.5c − c2 for “add a cow.” “Add a cow” is strictly preferred if 10c − c2 < 7.5 + 8.5c − c2 . This
simplifies to
7.5
0 < 7.5 − 1.5c, that is, c < = 5;
1.5
that is, “add a cow” is strictly preferred if c = 0, 1, 2, 3, 4; but if c = 5 then A is indifferent
between “add a cow” and “don’t add a cow” (because 7.5 − 1.5c = 0).
• If B chooses to add a cow, then A’s payoffs are 2+9.5c−c2 for “don’t add a cow” and 9+8c−c2
for “add a cow.” “Add a cow” is strictly preferred if 2 + 9.5c − c2 < 9 + 8c − c2 . This simplifies
to
7
1.5c < 7, that is, c < ≈ 4.67;
1.5
again, “add a cow” is strictly preferred if c = 0, 1, 2, 3, 4; but if c = 5, then “don’t add a cow” is
strictly preferred (because 1.5c > 7).
To summarize, “add a cow” is a strictly dominant strategy for A (and, equally, for B) if c ≤ 4; but
for c = 5, “don’t add a cow” is the unique dominant strategy. Thus, the (carefully chosen) tax regime
will motivate both farmers to stop the process of building up their stock exactly at the social optimum.
Remark 2: In the example above, we know the social optimum and have deliberately chosen a tax
rate that will bring it about. One of the advantages that a “revenue-neutral use tax” has over cap-
and-trade, however, is that it does not require some “expert” panel to determine the social optimum
at the beginning of the process. Instead, it could be phased in at a low level (as in the Feldstein–
Mankiw proposal above) and then gradually increased as necessary. Even a low level of revenue-
neutral taxation exerts market pressure toward solutions that don’t overuse the commons so much
(clean energy solutions, in the climate change example) and therefore does some good. With the costs
of many clean energy sources already near competitive with fossil fuels, many believe that even a
small amount of pressure from a revenue-neutral carbon tax could bring about a surprisingly rapid
transformation of the U.S. energy sector.
Remark 3: The revenue-neutral tax strategy works most simply when there is a government that
has full sovereignty over the commons. (Similarly, the cap-and-trade strategy assumes that there is a
regulatory body with full authority to manage the commons.) This worked for acid rain pollution—
the pollution traveled many hundreds of miles, but not many thousands, so it largely remained within
U.S. territory.11 But for carbon dioxide pollution and consequent climate change, the commons is the
atmosphere of the whole globe. There is no world government or supranational body with the authority
to manage the world’s atmosphere (and if you are waiting for one to be established, you might have to
wait a long time). Faced with this issue of border leakage, advocates for refundable carbon taxation
suggest a system of border tariffs: a nation operating a carbon tax would impose import tariffs on
goods coming from countries that did not have such a system, the tariff being equal to the estimated
carbon tax that would have been payable on the energy used to manufacture those goods. (This is the
third of the Feldstein–Mankiw bullet points, above.)
Think about it. . .
What do you think might be the effects of a “carbon border tariff” system as sketched in
Remark 3? What unintended consequences do you think might arise?
11 There was significant leakage across the border to Canada, though.
Summary of Ideas: The Tragedy of the Commons
• A commons is a limited resource over which several agents have potentially

unlimited usage rights.
• The phrase tragedy of the commons describes the result when agents make use of
a commons so that benefits accrue to each of them individually but costs are shared.
Each agent then has an incentive to overuse the commons, resulting in its degradation
for all.
• CO2 discharges into the atmosphere provide a worldwide example of the tragedy of
the commons.
• The tragedy of the commons can be modeled by repeated prisoners’ dilemma-type
games.
• Strategies that have been suggested to mitigate the tragedy of the commons include
privatization, regulation, cap and trade, and revenue-neutral use taxes.
• Regulatory capture describes a situation in which a governmental agency, intended
to regulate a certain interest group, instead becomes “captured” by those same
interests.
6.4 After Math: Decision-Making and Ethics
Objectives
I know my own core values and am able to relate my sustainability decisions to them.
I understand the importance of framing in the context of sustainability decisions.
I can describe a variety of different frames for large-scale decisions such as those
related to climate change.
I understand how math can sometimes illuminate and sometimes conceal the key
ethical issues involved in a sustainability question.
I can define the precautionary principle and its application in a sustainability
context. I am also aware of objections to the principle as it is commonly formulated.
I understand that there will be no “magic bullet” that answers the sustainability
question. A portfolio of responses, on different scales and by different agents, will
continue to be required.
I understand that such a portfolio is unlikely to deliver the best possible response in
some absolute sense, but that we may hope that it is good enough.
I understand the term the Anthropocene epoch and its ethical implications.
I will consider what virtues are appropriate to the Anthropocene, and I will seek to
cultivate them.
The aim of this final section is for us to ask ourselves, What happens after math? We may have
correctly carried out all the calculations related to some sustainability question, but what is the right
policy to pursue? Questions of right and wrong cannot be adjudicated by mathematics: they are the
province of ethics, that department of thought that deals with values, and which asks what makes
different actions right or wrong. In this section we’ll think about various ethical questions related to
sustainability that arise out of our mathematical analysis. In fact, we’re going to get more personal than
that and ask you to reflect specifically on your own values, your own sense of right and wrong, in the
context of such questions. Counselors sometimes refer to such reflection as a process of discernment,
and for that reason you’ll see a new kind of box that appears only in this section, a “discernment” box,
like this:
Discernment. . .
How do you react to the idea of bringing your own sense of right and wrong to bear in a
math class? Excited, anxious, angry, cheated, hopeful? Try to find one word (not necessarily
from the short list just given) that describes your deepest response.
Discernment is not something that happens in a hurry. So, when this chapter confronts you with a
discernment challenge, take time to answer sincerely. Do what it takes to get in touch with your true
self. If you have the opportunity and if this is helpful for you, take some time to be outside in a more
6.4. AFTER MATH: DECISION-MAKING AND ETHICS 389
natural environment—among the cycles of nature that we have described several times in this book
(especially in Chapter 2). Math is essential in identifying and analyzing the sustainability challenges
that your generation must face. But to answer them. . . There can be no answer, we believe, without
touching the deep places of the heart. In this section we will try to do that.
6.4.1 What Do You Value?

How do you—you, personally—go about deciding what is right and wrong?
That sounds like an unusual question for a math textbook, and perhaps it is. But let’s go back right
to the beginning. In the introduction, which is a letter addressed to you as “Dear Student,” we wrote
As citizens in a democracy, you will ultimately be the ones whose decisions will guide
your world toward a sustainable future.
The kind of mathematics we’ve learned in this book will help you understand the background and
context of those decisions. But it will not, by itself, tell you what to do. To figure that out—either on
the personal, individual level, or on the level of whatever group that you belong to that has decision-
making power—you need input from another level: your own sense of right and wrong, what some
people would call your core values, or your understanding of the purpose of human life. John Dewey
wrote in “On Democracy” [95]:
[D]emocracy is much broader than a special political form, a method of conducting

government, of making laws and carrying on governmental administration by means of
popular suffrage and elected officers. It is that, of course. But it is something broader and
deeper than that.
The political and governmental phase of democracy is a means, the best means so far
found, for realizing ends that lie in the wide domain of human relationships and the
development of human personality. . . . The key-note of democracy as a way of life may be
expressed, it seems to me, as the necessity for the participation of every mature human
being in formation of the values that regulate the living of [people] together:—which is
necessary from the standpoint of both the general social welfare and the full development
of human beings as individuals.
Example 1: I am the operating manager of a chemical plant, and I strive to produce goods that people
need and depend on (say, laundry detergents) in a responsible way. I discover that I can produce the
goods more cheaply, improving the plant’s profitability and incidentally earning myself a nice bonus,
if I change the production process in a way that will discharge a toxic compound into the local river.
This discharge will probably kill five people each year, though it is unlikely to be traced back to me.
What do I do?
Most people would not deliberately change a manufacturing process in a way that they know
would kill people. The mathematical information about the likely effect of the change interacts with
a fundamental ethical value that they hold: we should not deliberately harm others.
The text in blue above is an example of what we’re calling a core value, one of the most widely
held in our society. It is not, however, the only one.
Example 2: You are a manager for the National Park Service (NPS) at Yosemite National Park in
California. Despite its millions of visitors, the park is strapped for cash and could carry out many
more vital conservation projects if only it could fund them. An advertising company approaches you
with a proposal to support the NPS with huge payments in return for the right to project advertising
images, throughout the evening hours, onto the north face of Yosemite’s iconic Half Dome (Figure 10
on the next page). How do you answer?12
12 Set aside, for the moment, that the proposal might well be illegal under the Wilderness Act or other legislation.
Figure 10: North Face of Half Dome. Karl Bralich photo.
Many people would respond with an instinctual “no” to this proposal. But nobody will be (directly)
harmed by the suggested advertising: some other value is at work here. If you are minded to answer
“no,” and especially if you are a lover of Yosemite, you may well feel that what’s proposed would be a
violation of something that is very special, that ought not to be contaminated. The core value at work
is something like this: Sacred things should be kept pure.
One more example.
Example 3: For those desiring to immigrate to the United States, the goal is lawful permanent resident
status, colloquially known as the “green card.” (Green card holders can become U.S. citizens after a
period of time, usually five years.) Most routes to the green card involve extensive bureaucracy and
many waits and delays. But in 1990, Congress created the EB–5 immigration category. In return for
investing at least $500, 000 in the United States, a foreigner, together with his/her immediate family,
could receive a conditional green card (allowing them to immigrate) on a fast track. After two years,
the conditional status would be made permanent if they could demonstrate that their investment had
created sufficiently many jobs. What do you feel about this?
Many people are moved to anger by the EB–5 program. It feels wrong that rich people should be
able to “cut in line” for immigration and, effectively, for U.S. citizenship. At stake for those who feel
this way is another core value: fairness, or to put it another way, people should be treated equally. On
the other hand, some people don’t see anything wrong here at all. The market is doing its job (see
Section 6.1.1), they would say: allocating a good (the right to immigrate to the U.S.) to those who
most benefit from it (as measured by their ability to pay). Indeed, rather than setting a fixed price of
$500, 000, perhaps the government should auction off the visas, thus maximizing the value derived
from the sale of this “good”!
In the examples above we have drawn attention to three “core values” or “moral foundations” (the
text highlighted in blue). How many such values are there overall, and how do they interact? In a
famous paper [135] with a wide group of collaborators, social psychologist Jonathan Haidt identified
five core values that he proposed as fundamental. The resulting scheme, called moral foundations
theory and developed at greater length in Haidt’s popular book The Righteous Mind [147], suggested
that all people are sensitive to five different “moral scales”: Care versus Harm, Fairness versus
Cheating, Loyalty versus Betrayal, Authority versus Subversion, and Purity versus Degradation.
Haidt likened these to the different taste receptors in the mouth that respond to food flavors (sweet,
sour, salt, bitter, and savory). He then made the key point that just as people vary in the strength of
their response to food flavors—a tiny hint of sweetness is enough for one person, while another can
hardly ever get too much—so also people vary in the strength of their response to these “moral tastes.”
According to Haidt, differences in our levels of response to the moral foundations explain many of
the divisions in our society—not only why we have different ideas from those of others about morally
freighted issues, but also why we can’t understand how others can have different ideas from ours
about these same issues. At the YourMorals website [97], Haidt’s coworkers offer a simple online
questionnaire to help measure your level of responsiveness to each of these moral foundations. As
part of the process of discernment that we are calling for in this final section, we encourage you to
take the “Moral Foundations Questionnaire”) now.13
Discernment. . .
Take the Moral Foundations Questionnaire [97] and take some time to reflect on your results
on each of the scales presented there. How do they correspond to your own self-perception?
Are there other “moral foundations” (ways of evaluating) that you perceive as important to
your life but that are not represented in the questionnaire? If so, try to name them.
One possible comment about Haidt’s moral founda-

tions is that they are individualistic: they work best when Care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Harm
we have to evaluate a single act, by a single person, with Fairness . . . . . . . . . . . . . . . . . . . . . . . . Cheating
clear consequences. This is not necessarily an objection Loyalty . . . . . . . . . . . . . . . . . . . . . . . . . Betrayal
Authority . . . . . . . . . . . . . . . . . . . . . Subversion
from Haidt’s point of view: he believes that morality is
Purity . . . . . . . . . . . . . . . . . . . . . . . Degradation
primarily about evaluating actions (just as in the tasting
analogy we quoted above), and an “action” classically Figure 11: Haidt’s five moral foundations or scales (a sixth was
means something with a single actor, a single deed, and added later).
a single consequence (or a small set of consequences).
That is not simply an issue with Haidt’s particular framework but with many types of moral reasoning:
they tend to get stymied by large-scale issues involving many actors spread out over space and time.
Take climate change, once again, or any similar act of large-scale pollution that is a byproduct of
affluent lifestyles. One way of abstracting the moral content of such an act might be this: “Acting
somewhat independently, a large number of unacquainted people, mostly rich, set in motion a chain
of events that causes a large number of future people, mostly in another part of the world, to be
deprived of resources they might otherwise have—perhaps even resources that are vital for life.”
This description is adapted from [171, Example 6, Section 5.2]; the author, Dale Jamieson, goes on
(emphasis added):
[W]hat some people take to be the core of a moral problem persists [in this case]:
Some people have acted in a way that harms other people. However, most of what
typically accompanies this core has disappeared, and this is why others do not see this
case as presenting a moral problem. . . . [For them, it does not] appear that anyone has
intentionally deprived future people who will live in another part of the world. . . . [their
deprivation] is just a consequence of [other people] getting on with their lives. In these
circumstances it is difficult for the network of moral concepts that involve responsibility
and harm to gain traction.
13 As of this writing, you can find this questionnaire at http://www.yourmorals.org/. To take it you will need to create
an account (or login via Facebook). You can then select the “Moral Foundations Questionnaire”. Updated links are maintained
at this book’s website, http://math-for-sustainability.com/moral-foundations.
Discernment. . .
How do you feel that Haidt’s five moral foundations might be applicable to sustainability
problems that are spread out in time and space in the way described above? Pay particular
attention to those foundations that scored most highly for you in the questionnaire,
together with any other foundation that you perceived as significant to you in the previous
discernment exercise.
Remark 1: One way to address the concern that we’ve raised above—that “moral foundations”
understood as ways of evaluating acts tend to get bogged down when large-scale issues are addressed,
including many sustainability questions—is to say that ethics should focus less on how to evaluate
acts and more on how to build virtues. This is affirmed by the ancient tradition14 of virtue ethics,
which redirects the moral focus from “what should I do?” to “what kind of person should I become?”
From this point of view, sustainability questions push us to ask “What virtues do we, as individuals
and as societies, need to cultivate so that we will respond rightly to sustainability questions?” We’ll
return to this idea in the final section of this chapter (Section 6.4.3).
Framing the Question

Along with understanding our own core values, we need also to understand that a single sustainability
question can be presented—or “framed”—in different ways: ways that will activate different core
values and which will therefore resonate with the moral intuitions of different groups of people.
Sometimes, this observation is presented in a way that sounds simplistic or even cynical (as in, “if
you want people from party X to listen to your arguments on issue Y, make sure you use words A, B,
and C, not words K, L, and M”). It does not have to be so, though. Surely, it reflects both the complex
nature of sustainability problems and the still more complex variety of actual human beings to note
that there are different frames through which a given problem can be viewed. Let us give an example,
still thinking of anthropogenic climate change as the sustainability issue in question. What are some
of the ways this problem can be framed?
Mike Hulme’s book Why We Disagree About Climate Change [163] is an important resource for
this discussion. Among the potential framings he lists for climate change are these:15
• Climate change as a battleground between different ways of knowing. The idea of a

“controversy” between noble, truth-seeking scientists and acquisitive giant corporations (or,
if you prefer, between pseudoscientific prima donnas and wealth creators at the foundation of
society) is an attractive one to the media. Usually, questions about science lie on the surface of
much deeper currents—what should a future society look like, how should we use our wealth,
and who has the right to choose?
• Climate change as a justification for the commodification of the atmosphere (by means of
emissions-trading programs). In this framing, climate change justifies policies such as “cap
and trade” (Example 7 on page 381), which expand the rule of the market over what was
previously a global commons (the atmosphere, which we all share). We’ll say more about this
below (Example 4 on the opposite page).
• Climate change as a threat, even as “the sum of all fears.” Sometimes this framing compares
the threat of climate change unfavorably to the threat posed by international terrorism or by
14 In the Western tradition going back at least to Aristotle’s Nichomachean Ethics [27]. Interestingly, a section on virtue
ethics was cut from Haidt’s book during editing: it can now be found at the book’s website (see [147]).
15 See his preface, page xxvii, for the first four of these.
pariah states (“if scientists had discovered that North Korea was pumping greenhouse gases
into the atmosphere with the intent to destabilize the world’s climate, there would be immediate
political consensus to take action regardless of cost” [207]), and it leads to an understanding that
this threat should be mitigated by diplomats negotiating at the highest levels of government.
• Climate change as an inspiration for the emergence of a global network of new or re-
invigorated social movements. This framing, perhaps epitomized by Naomi Klein’s This
Changes Everything [182], sees climate change as generating new forms of activism that
challenge the global technocratic model and catalyze change in social, political, and economic
behavior.
• Climate change as a call to honor sacred values [163, Chapter 5], Many faith traditions, as
well as people who would not affiliate with any organized religion, see the natural world as
something that has a sacred value in and of itself. These traditions often read human relations to
the natural world in terms of the language of stewardship: a steward is responsible for managing
property on behalf of someone else and keeping it in good order, because they will have to
render an account. In this framing, fundamentally changing the climate is an abdication of
stewardship responsibility, a contamination of the sacredness of the natural realm.16
• Climate change as an issue of international development [163, Chapter 8] This framing

sees climate change as an additional layer of complication to the already difficult problem of
international development17 . This was the focus of the Brundtland Report [255]. Right after its
famous definition of sustainability (Definition 1 on page vii) that we cited in the introduction,
the report continues: “. . . the problems of poverty and underdevelopment cannot be solved until
we have a new era of growth in which the countries of the global south play a large role and
reap large benefits.” Climate change complicates this quest for justice because the rich Northern
nations have already appropriated and used a large proportion of a common resource—the
atmosphere’s CO2 -absorbing capacity—which is seen as vital for development. Contraction
and convergence [75] is one mechanism that has been proposed to achieve just sharing of this
resource going forward.
Example 4: The Case of Emissions Trading: Let us consider the cap and trade mechanism
(page 381) as a way of seeing how these different framings can lead us to evaluate policies in very
different ways. Recall the basic idea of cap and trade. An international scientific body decides how
much carbon dioxide (or other gases) can safely be emitted in a given year, and then issues “emissions
permits” up to that total. These emissions permits can be freely bought and sold, but every company
that wants to emit carbon dioxide in its manufacturing process (or in some versions of the idea,
to sell a product that will emit carbon dioxide when used, such as gasoline) must “pay” for this
with permits up to the total amount of carbon dioxide emitted. From the market perspective, this
is an excellent idea: thoughtful, well-qualified scientists, above the fray of individual industries and
associated policy-making, set a worldwide limit (appropriate to a worldwide problem). Following that,
the market mechanism allocates these emissions in an efficient manner. Because of market pressure,
those industries that can achieve their goals with lower emissions are motivated to do so; those that
cannot operate without high emissions have to raise their prices, and may go out of business.
Things look rather different through some of the other framings, though. If you are disposed to
see the science of climate change as a battleground, you might well be skeptical of the idea of a
scientific group being “above the fray” (as we said above) and able to set safe emissions levels in a
16 The perspective of a famous and controversial article by White [350] is related. White argues that Western Protestantism
“desacralized” the natural realm and thus paved the way for ecologically destructive exploitation.
17 Summary: Many citizens of the world’s rich nations lead lives whose luxury would have been unimaginable to previous
generations, whereas the world’s poorest 10 percent or so have an annual income below $500 per year. Nonetheless, global
prosperity—including the prosperity of the world’s poorest people—has substantially increased in the last decade [276].
“neutral” manner. Even if you are confident that the consensus of climate scientists is correct, imagine
the ethical pressure on this small scientific group, who are essentially being asked to take moral
responsibility for the future of the world while the rest of us go our ordinary, market-driven way. In
1942, during the Manhattan Project, Edward Teller asked whether there was a chance that the first test
atomic explosion would cause runaway fusion in the atmosphere—turning the Earth itself into a giant
hydrogen bomb and incinerating everything on it. Hans Bethe, together with Teller, soon calculated
that the chance of this process happening was vanishingly low. Imagine the moral weight of actually
performing such a calculation. Something like that is what the “Emissions Board” is being asked to
take on.
Or again, consider the “sacred value” framing. From this perspective, buying and selling permits
to pollute the atmosphere might be seen as analogous to buying and selling permits to toss one’s
trash into the Grand Canyon on a hike to save the trouble of carrying it out again. Michael Sandel is
professor of government at Harvard University, and his lectures on justice have been viewed, online
and on television, throughout the world. In his book What Money Can’t Buy: The Moral Limits of
Markets [284] he writes
From the standpoint of the heavens, it doesn’t matter which places on the planet send less
carbon to the sky.
But it does matter morally and politically. Letting rich countries buy their way out of
meaningful changes in their own wasteful habits reinforces a bad attitude—that nature
is a dumping ground for those who can afford it. Economists often assume that solving
global warming is simply a matter of designing the right incentive structure and getting
countries to sign on. But this misses a crucial point: norms matter. Global action on
climate change may require that we find our way to a new environmental ethic, a new set
of attitudes toward the natural world we share. Whatever its efficiency, a global market
in the right to pollute may make it harder to cultivate the habits of restraint and shared
sacrifice that a responsible environmental ethic requires.
Sandel expressed these ideas in a New York Times op-ed at the time of the Kyoto conference on
global warming. He reports, “The Times was flooded with scathing letters, mostly from economists.”
Here is a clear example of how the same issue looks through different frames: is emissions trading a
simple, efficient way to allocate costs and benefits, or an evasion of moral responsibility?
Discernment. . .
Remind yourself again about your core values or moral foundations that you explored in
the previous discernment exercises. How do they lead you to frame the idea of emissions
trading, and how do you evaluate the ethics of such a practice?
The Anthropocene Epoch

Geologists have divided the history of our planet into various time periods, based on stratigraphy
(the evidence obtained from the deposition of rock in layers, and the ability to correlate those layers
between different sites). The most recent of these acknowledged time divisions, or epochs, is the
Holocene epoch, which begins about 11,700 years ago with the end of the last glacial period (or
“Ice Age”). As the glaciers retreated, a relatively warm, stable period of climate began. All of written
human history, almost all its agriculture, the rise and fall of its great civilizations, and the recent rapid
expansion of humanity both in numbers and power, have taken place during the Holocene.
In fact, this expansion has been so great that there are

many parts of the Earth’s ecosystem on which human
beings and their activities are the single most significant
impact. A striking example is shown in Figure 13 on
the next page, which represents all the Earth’s land
mammals by weight (with each square representing a
million tons). As can be seen very clearly in the figure,
we humans, together with our food animals18 and our
pets, constitute the vast majority of mammalian life. The
great beasts that fill our stories and our imaginations—
the lions, tigers, bears, and elephants—form a minuscule
fraction of the mammalian universe we have brought
about. (To be fair, though, we should note that insects Figure 12: Skipping stones.
outweigh us by orders of magnitude—see Example 8 on
page 39.) Because of this expansion of human impact,
a number of scientists have suggested that we are living through the beginning of a new geological
epoch, the Anthropocene, or “epoch of human influence.” Future geologists would be able to detect
the signature of this influence in the fossils of our domesticated animals, the disappearance (extinction)
of numerous species, the reorganization of water flow patterns by human construction of roads
and dams, and the pulse of exotic radionuclides deposited from the H-bomb tests of the fifties and
sixties. At the time of writing, the International Union of Geological Sciences is considering a formal
recommendation from its Anthropocene Working Group that indeed, a new geological epoch should
be declared to have begun. The proposal is a controversial one that may not be accepted, but the fact
that it has been raised at all shows the breadth and impact of human-caused change to our planet.
You might recall that we opened this book by quoting Gus Speth [305], dean of environmental
studies at Yale, about the road that humanity is now traveling between the “old world” and the “new
world” (page vii). In the “old world,” nature is a huge, abundant, teeming, and dangerous background.
Human life skips on its surface, having no more lasting impact on it than the ripples left by a stone
that a child “skips” over the surface of a pond (Figure 12). In this “old world,” our greatest literature
was written, our religions and moral intuitions were developed. The scientists who are calling for the
designation of an Anthropocene epoch are saying emphatically that for good or ill, that “old world”
is past. The Earth systems that we have relied on as a background for human life and culture, and to
provide for our needs for food, shelter, and communication, are now themselves affected by human
beings more than by any other single cause.
People respond to this thought in different ways, of course. Diane Ackerman [3] is optimistic
about the prospects of the new “world shaped by us.” Stuart Chapin and Erica Fernandez [64] see
the designation of the Anthropocene epoch as an opportunity for their fellow ecologists to develop
new paradigms for science and to take a more proactive role: “defining problems and possibilities
in a rapidly changing world” and committing to “a new ethic of environmental citizenship.” Less
sanguine responses tend to emphasize that there is great difference between being able to influence
something and being able to manage that thing for a good outcome. (When one of the authors of
this book was a toddler, they let off the parking brake on their father’s car at the top of a steep
driveway. They had considerable influence on the result, but exactly zero ability to successfully
manage it!) Roy Scranton [292] draws on his experiences in the Iraq War to frame the challenge
of the Anthropocene as “learning to die.” Clearly there is a huge gap between these perceptions of
the challenges or opportunities of the Anthropocene epoch. These and other writers have in common,
however, the thought that the unprecedented power of humanity symbolized in the “Anthropocene”
language brings with it an unprecedented ethical challenge: the challenge to recommit to principles
18 Now largely kept in factory farms or “confined animal feeding operations,” CAFOs.
Figure 13: Weight of mammals on earth (from xkcd.com).
to which we perhaps have paid only lip service, to emphasize values that were not previously central
to our ethical reflection, or even to understand and exhibit completely new values that match this new
context.
Discernment. . .
Reflect on what the beginning of the Anthropocene means to you. Are your core values
strong enough to face the challenges that it will present? Are there other values you will
need to develop? (You may wish to review some of the ideas presented in [334].)
6.4.2 The Role of Mathematics

What part does mathematical reasoning, which we have emphasized so much in this book, play in the
process of ethical reflection and discernment that we are advocating in this final section? There are
many possibilities. Here are a few—you may be able to think of some more yourself.
• Mathematics helps us understand questions by drawing attention to their fundamental structure.

One good example here is the distinction between one-way and cyclical flows that we drew on
page 80. Having understood that structure in a mathematical, technical way, we can begin to
reflect on it ethically: what does it mean that our society has so valorized “consumption”19
that it has redefined human beings as “consumers,” as facilitators of the one-way flow from
19 In the nineteenth century, “consumption” referred to tuberculosis: a wasting and ultimately fatal disease.
resource to waste (see [223])? Another example is the mathematics of climate change itself (see
Section 2.4). It is true that the models scientists use to make detailed predictions are hugely
complex: mathematics helps ethical reflection by reminding us that the basic structure of the
issue (the Sun keeps the Earth warm; carbon dioxide acts as a “blanket”; we are making that
blanket thicker) is not complex and indeed has been known for well over a century.
• Mathematics pushes us to quantify and compare effects and risks. In doing so it directs our
attention to the often unwelcome fact that real choices will always involve trade-offs of one
kind or another, like the trade-off between false positive and false negative errors in medical
testing (Section 5.4.2). You can see this process at work in our discussion of the Stern Review
(Section 6.1.4). The Review arrives at a definitive conclusion: but by allowing us to look “under
the hood,” as it were, mathematics lets us see that the process by which this conclusion is
reached contains inbuilt ethical principles about how we “trade off” present goods against
future ones. Marshall [207] suggests that one way to overcome the “stultifying indifference
to future loss” represented by our consumption choices is to make this language of trade-off
more explicit: “talk less about the costs of avoiding climate change and more about the lousy
deal we are getting in return for a marginally higher standard of living.”
• In the previous point we emphasized how understanding—even at a simple level—some of the

mathematics behind the Stern Review helped us to see that certain ethical choices went into
its modeling. To his credit, Stern offered a justification of these choices in ethical terms, terms
that anyone can engage with. But the negative aspect of this same point is that sophisticated
modeling can hide ethical assumptions that if brought out into the open, might be controversial
or hard to justify. In [256], Cathy O’Neil cites the example of recidivism risk models, which are
used in several American states to estimate the risk that a convicted prisoner will re-offend—
estimates that then may be used in the sentencing process. As she points out, if a judge baldly
said, “You’re a rich kid from a good neighborhood, so you’re less likely to offend again; I’ll give
you a lighter sentence,” most people’s sense of justice would be outraged. But when the same
assumptions are buried inside a mathematical model—especially one that its manufacturers
claim is intellectual property, not open to public inspection20 —then they are able to work
unnoticed. Oddly enough, that is another reason for a book like this: not only to teach you to use
mathematics where that is appropriate, but also to equip you with a certain skill in sniffing out
inappropriate uses of mathematics, where math is simply being used to generate results from
assumptions already predetermined for ideological reasons. In sustainability work, the so-called
“IPAT equation” and its variants (page 441) are a notorious example of this.
• Finally, mathematics itself could not be done without its own internal ethical principles. One
of the most basic of these, one that we perhaps learn in grade school, is that when we do
mathematics we respect the truth: 2 + 3 = 5, and no matter how much we would like the answer
to be 6, mathematics will not bend to fit our desires. Sometimes this principle gets mistranslated
into “everything in mathematics is 100% certain and there is no room for error or imprecision.”
We hope that after completing a book that includes extensive sections on measuring imprecision
(1.3.1) and uncertainty (5.2), you will not make that mistake. But the basic commitment to truth
is a value that it would be good for us to internalize in sustainability work. In particular, the
unpalatable truth is that we are all caught up in huge patterns of production and consumption
that depend deeply on those unsustainable one-way flows mentioned in an earlier bullet point.
We can, and should, push against them; but no reader (or author!) of this book stands outside
them.
20 This is not true of the models used in scientific research. Openness is part of the essence of the scientific process.
Discernment. . .
Reflect on the above discussion in the light of your own experience of reading and using this
book. In what ways has it equipped you for ethical engagement with sustainability issues?
Example 5: We referred above to the way that mathematics, by pushing us to quantify and compare
things, encourages thinking about trade-offs. Some people’s understanding of ethics, however, is that
it is a realm that does not allow for trade-offs: there are laws that are hard boundaries (Thou shalt not
do X) and “trading off” between them takes us outside the province of ethics altogether. An example
of an attempt to set such a hard boundary in the sustainability context is the so-called precautionary
principle.
Definition 1
The precautionary principle suggests that if an action or policy has a possible risk
to society or the environment, especially a risk of irreversible damage, then it is up
to those proposing the policy or action to prove that it is not harmful. Principle 15
of the Declaration of the Rio Earth Summit (1992) stated: “In order to protect the
environment, the precautionary approach shall be widely applied by States according to
their capabilities. Where there are threats of serious or irreversible damage, lack of full
scientific certainty shall not be used as a reason for postponing cost-effective measures to
prevent environmental degradation.”
It’s certainly hard to quarrel with the precautionary principle, especially in the light of the
evidence [257] that some businesses (such as the tobacco industry) have deliberately worked to sow
public uncertainty (see Section 5.2.5) about the risks of their products in order to continue producing
profitable but harmful substances (such as cigarettes). Nevertheless, the precautionary principle does
include a strong element of status quo bias: if in doubt, don’t do it. For those, like many Westerners,
who benefit from the status quo and are not threatened by it, at least in the short term, this may look like
a good principle. But what about the members of the Alliance of Small Island States (AOSIS [333]),
for whom the “status quo” will mean that many of their countries disappear under rising seas? Or the
3 billion people who rely on open fires for cooking (which substantially elevates their risk of death
from air pollution, see Example 2 on page 416) and whose lives would be greatly improved by access
to modern energy sources, fossil fuel powered or not? Helping these people may be ethically right
even though it may require risky action! It seems difficult to defend the “hard boundary” set by the
precautionary principle in cases like this.
Discernment. . .
How do you feel about the precautionary principle and the objection to it articulated above?
Can you discern a way of formulating the principle that might be less open to this kind of
objection?
Doing Well Enough

One idea that is strongly embedded in much mathematical analysis is the desire to optimize—to find
the best possible answer to a question. Think of the “economic person” we discussed in Section 6.1,
who makes every decision on the basis of maximizing time-discounted expected utility. Popular advice
books and websites have the same message: to decide between various options, write down a list of
reasons pro and con, “add them up” somehow, and choose the one for which the “pro”s most heavily
outweigh the “con”s. In thinking about sustainability, this pushes us toward the kind of top-down
solutions suggested by the Stern Review: every factor is taken into account (at least in theory!) and
the result gives the optimum trajectory for the entire world.
We’ve mentioned one clear difficulty with this kind of plan: there is no single body that has the
authority to implement it. But here’s another: many of us are not, in fact, the “optimizers” that
economics assumes us to be [290]. Imagine yourself in the toilet paper aisle at your local supermarket.
Are you really going to review all twenty brands on offer, comparing them for price, softness,
absorption, and whatever other qualities are important to you, and then making of these factors some
global “utility” index that you then maximize? Highly unlikely. What most people will do is have
some idea of what is “good enough” and then load up their carts with the first thing they see that is
“good enough.” If past experience helps them to judge (“We used this brand last time and it was fine”),
so much the better.
The kind of behavior described here has a technical name: satisficing.
Definition 2
Satisficing describes a decision-making strategy that accepts the first good enough option.
It is contrasted with the optimizing (also called maximizing) strategy that seeks the best
possible outcome to each decision.
Research suggests that satisficers are, on the whole, happier than optimiz-
ers. One possible reason for this (which is highly relevant to many questions
of sustainability ethics) is that satisficing is simply easier to implement than
optimizing, especially in a context in which we have a whole portfolio of
choices to make. To explain this term, think about the various relationships
you are in and communities that you belong to from the point of view of
sustainability decision-making. Some decisions you can take by yourself
(you switch to using an LED bulb in your desk lamp). Some involve a small
community (you and your roommates decide to recycle mixed office paper;
you, a professor, tell your colleagues you won’t be flying to that conference Figure 14: The TP aisle.
because you want to reduce your carbon footprint) or a larger one (you run
for election to the local Board of Supervisors on a platform that includes
strengthening protections for water quality). Some relationships, like those involving voting, may be
national or even wider in scope.
Definition 3
We refer to this whole complex of sustainability choices that you have to make, together
with the relationships within which they are embedded, as your portfolio of sustainability
choices.
Once we appreciate that there is no one “sustainability manager” for the whole world, or even
for our country or our town, we come to see that we all face a portfolio of sustainability choices
every day. Whether our planet heads down a sustainable track will be the result of whether people
at organizations at different structural levels choose decisions from their portfolio that head in a
sustainable direction. There is not going to be a master plan, and some of the subsidiary choices we
have to make may compete or even conflict with each other. So be it. As long as we are satisficing—
trying to be “sustainable enough”—across our whole portfolio of sustainability choices, the world is
going to move in a more sustainable direction. “Optimizers” may worry that sustainability problems
are so hard that only the best possible solution will do, but such a worry may lead to “paralysis by
analysis,” where nothing gets done at all because we cannot determine the “best.” “Satisficers” must
be on their guard against settling too early, but they will at least accomplish something, and that
accomplishment will bring personal joy: joy in doing right.
Discernment. . .
Would you see yourself more as an optimizer or a satisficer? Whichever you are, try to
discern both the benefits that your stance can bring to sustainability decision-making and
the characteristic temptations to which it is exposed.
6.4.3 Virtues for an Anthropocene Future

We have seen that we are all going to face a portfolio of sustainability choices: choices and possibilities
that will reflect not only who we are, but also the various relationships and structures of power in which
we are embedded. Our portfolio of possible choices will change as we move through life. Mathematics
will provide vital information to help us with our choices, but the choices themselves derive from our
own inner character or virtues. Virtue ethics (Remark 1 on page 392) remind us that what we do
matters not only because of its effect on the world, but also because of its effect on ourselves: we
ask not only, What kind of world do we want to build? but also, What kind of people do we want to
become?
In [305], Gus Speth quotes from the “Great Transition Initiative,” written by Paul Raskin. This
optimistic outlook on sustainability is written from the perspective of a historian looking back from
the late twenty-first century on the changes humanity went through to achieve a sustainable world. It
is quite explicit that the foundation of all this “history” was a fundamental shift in virtues (or “values”
as he puts it). Speth writes:
Raskin’s device is to write from the vantage point of someone in the second half of this
century looking back on the dominant value shifts that occurred earlier. His is a history
of the future. Here is what he sees: “The emergence of a new suite of values is the
foundation of the entire edifice of our planetary society. Consumerism, individualism,
and domination of nature—the dominant values of yesteryear—have given way to a new
triad: quality of life, human solidarity, and ecological sensibility.”
Discernment. . .
How do you react to Raskin’s idea that a transformation of our virtues is needed before a
sustainable world can emerge? What would you list as the virtues most necessary for an
Anthropocene future?
The Anthropocene age will indeed call all of humanity to a new understanding of virtue. Some
of these “Anthropocene virtues” might be ones we already know well, reconfigured for a new age:
for example, “human solidarity” (one of Raskin’s triad) surely incorporates many elements of the
old-fashioned virtue of justice. Others may be less familiar: when Raskin opposes “quality of life”
to “consumerism,” there may be some connection to temperance (not taking more than you need—a
very old-fashioned virtue indeed) but also to something that, at least to many Westerners, is quite new:
the concept of mindfulness, of acting with full awareness in each thing you do. Jamieson [171] writes:
Much of our environmentally destructive behavior is unthinking, even mechanical. In

order to improve our behavior we need to appreciate the consequences of our actions
that are remote in time and space. A [mindful person] would see herself as taking on the
moral weight of production and disposal when she purchases an article of clothing (for
example). She makes herself responsible for the cultivation of the cotton, the impacts of
the dyeing process, the energy costs of the transport and so on.
Finally, “ecological sensibility” circles back to one of humanity’s oldest virtues, humility in the face
of nature, the “dust” that gave us birth and to which we shall return. Others might construe Raskin’s
list in different ways, or wish to change it or add to it. That’s okay. As we close this chapter and this
book, we want to share a gift and offer a challenge. The gift is the suite of tools offered in this book—a
gift that we hope will help you to gain insight into many sustainability questions. The challenge is to
become a person who meets these questions not just with appropriate math but also with appropriate
virtue. The future of our beautiful world may depend on it. Good luck!
7. Redo Exercise 6, where now in addition you expect

6.1. MARKET PERSPECTIVES AND electricity prices to remain constant for 4 years and
LARGE-SCALE CHANGE thereafter rise each year at a rate of 5% per year.
8. Suppose a city is considering making an invest-

1. Suppose that each participant in a market measures ment in a flood control project. Most of the costs would
the value of their allocation of goods by attaching some be incurred up front in the project construction, while
numerical “utility” to it (compare Section 5.3.3), and most of the benefits are gained over time, as potential
that they consider a trade a “win” if and only if it floods are avoided. A cost-benefit analysis over a 50
increases their utility. We could define the total utility year period using a discount rate of 5% shows that the
of an allocation of goods to be the sum of the individual costs slightly exceed the benefits. How would (a) an
utilities of all the participants, and say that an allocation increase or (b) a decrease in the discount rate be likely
is socially optimal if it maximizes the total utility. to affect the conclusion of the cost-benefit analysis?
Show that a socially optimal allocation must be Pareto
optimal. (The reverse is not the case.) 9. In Exercise 48 on page 340 from the previous
chapter, we estimated expected annual damages to the
2. Question 1 on page 352 involved personal risk to U.S. economy attributable to climate change at the end
of the twenty-first century. Call this figure L.
yourself (a 1/100,000 probability of death each year)
as well as risk to your larger community. In considering (a) First suppose that we are concerned with the
this question, did you think of this in terms of your own damages in a particular year 80 years in the fu-
ture. Estimate the present value of the expected
personal risk, or in the overall risk to your community?
damages L, assuming a discount rate of 3%.
Discuss this difference with other members of your
(b) Now (using the same discount rate) estimate the
class.
present value of damages L per year beginning
80 years in the future and continuing every year
3. Calculate the present value of $10 million, 100 after that forever.
years from now, assuming a discount rate of (a) 1%,
(b)3%, (c) 8%, and (d) 10%. 10. One model that has been proposed for hyperbolic
discounting (“Critical Thinking” on page 357) is this:
the present value of 1 dollar (or other unit of utility), n
4. Relate your results from the previous problem
years in the future, is
and the material on the value of a statistical life in
Section 6.1.2 to the present value of a statistical life 1
$ ,
100 years in the future. How much do you personally 1 + rn
believe the present generation should be willing to where r is a constant (analogous to the discount rate).
spend for a reduction in mortality risk equivalent to one Draw a table comparing hyperbolic discounting ac-
statistical life 100 years in the future? cording to this formula, with r = 0.05 = 5% per year,
with exponential discounting at various rates (1%, 2%,
3%, 4%, 5% per year), for various numbers of years n
5. Calculate the present value of a guaranteed pay- in the future. What do you notice?
ment of $25,000 per year for the next 10 years if the
discount rate is 4%. 11. Consider a person who discounts hyperbolically
according to the formula of the previous exercise, with
6. Consider Problem 8 on page 308 from the pre- r = 5% per month. On January 1, 2018, this person is
vious chapter, about the economics of investing in offered a choice between two options:
solar panels. Suppose that future costs and benefits are (a) Receive $100 on January 1, 2019.
discounted with a discount rate of 2%. How does that (b) Receive $150 on January 1, 2020.
affect the result of the problem? Explain why they will choose (b).
12. Again consider the person described in Exer- relationship if he doesn’t work on your shared project,
cise 11. If both options remain open, this person will that could change his behavior. But this is simply to
change their mind at some point during the year 2018 say that your overall “payoff” from this interaction
and decide they prefer option (a) instead. When will is not just the bonus you receive—it involves other
this change happen? factors as well. Imagine that your partner perceives
that the damage to your relationship if you work on
13. Someone who changes their mind in the way de- Neobase and he works on his personal project, Kazam,
scribed in the previous exercise may be said to exhibit is equivalent in value to a $20, 000 loss to him (and
time-inconsistent discounting. The only discounting that you have a symmetric perception of the damage
methodology that is never time-inconsistent like this that you will do if you “defect” on him). Make a new
is the standard exponential one of Definition 6 on payoff matrix incorporating these assumptions. Does
page 354. Can you explain why? the outcome change? If not, what value would you
each need to set on your relationship in order that the
outcome would change?
6.2. THE STRANGE BEHAVIOR OF RA-
TIONAL PEOPLE 18. (Adapted from [100].) A penalty kick in pro-
fessional soccer can be (approximately) analyzed as
a two-player game. The designated kicker confronts
14. Can a player have more than one dominant
a goal, which is defended by the opposing team’s
strategy? If so, give an example; if not, say why not.
goalkeeper (“goalie”); no other players are involved.
The kicker’s best option is to aim for one corner (left
15. Revisit the software engineers’ example (Prob- or right) of the net; the goalie’s options are to dive left
lem 6.2.1 on page 362). Suppose that the two engineers or right. These decisions are made at almost exactly the
A and B enter into a legal “contract to cooperate,” same time—the goalie does not have time to wait to see
agreeing that if one of them works on their own which way the ball is headed before diving. In [260],
personal project and the other works on the joint based on analysis of over a thousand penalty kicks in
professional soccer, the author derives the following
project (Neobase), the person who broke the contract
payoff matrix:
by working on their personal project will pay the
other $60, 000. Legal fees for preparing this contract Goalie
are $5, 000 split equally between the two engineers. Left Right
Draw up the new payoff table. What are the dominant
strategies now? What will be the outcome? Left 0.58,−0.58 0.95,−0.95
Kicker
16. Imagine now that the two software engineers Right 0.93,−0.93 0.70, −0.70
from Exercise 15) are trying to decide whether they
want to enter into the legal contract suggested there. Penalty Kick Game
The legal agreement will come into force only if they
(a) Suppose you are the kicker (A) and want to
both agree to it, in which case they will be playing the
choose a mixed strategy α p , of the form “With prob-
game described in Problem 15; if one or both refuse,
ability p I will kick to the left, and with probability
they will be playing the game described in 6.2.1 on 1 − p I will kick to the right.” The object of your choice
page 362. Both engineers know what the outcomes of is to make the expected payoffs to the goalie (B) equal
both those games will be. Draw up a payoff matrix for whether they dive to the left or the right. Explain why
the meta-game in which each engineer has two possible this requirement leads to the equation
strategies: “Agree to the contract” and “Do not agree to
−0.58p − 0.93(1 − p) = −0.95p − 0.70(1 − p).
the contract.” What will be the result?
Solve this equation for p.
(b) Carry out a similar calculation from the goalie’s
17. Another “solution” to the problem raised by the point of view, finding the value of q such that the mixed
software engineers’ example is to point out that in the strategy βq , “With probability q I will dive to the left,
real world, other motivations are in play besides the and with probability 1 − q I will dive to the right,”
amount of your bonus. If (for example) your partner is makes the kicker’s expected payoffs equal whether they
concerned about the damage that he might do to your kick to the left or the right.
(c) Explain why (α p , βq ) (with the values of p and q Player 2

you calculated) is a mixed strategy Nash equilibrium.
(d) In the real-world data analyzed in [260], kickers L R
kick to the left 40% of the time and goalies dive to the
U 10,10 12,4
Player 1
left 42% of the time. Compare this real-world data with
your calculated Nash equilibrium.
D 10,25 12,0
19. The tables below give the payoff matrices Game 4

for various two-player games. Player 1’s payoffs are
always shown first. In each case, answer the following Player 2
questions:
“War” “Peace”
(a) Identify dominant or strictly dominant strategies
for each player (if they exist). “War” 6,0
Player 1
−10,−10
(b) Identify best responses to any dominant or

“Peace” 0,6 4,0
strictly dominant strategies.
Game 5
(c) Identify any Nash equlibria and say what you
can about the expected outcome of the game.
Player 2
Here are the games: “War” “Peace”
“War” 6,0
Player 1
−10,−10
Player 2
L R “Peace” 0,6 6,6
Game 6
U 10,10 15,5
Player 1
D 5,15 6,6 Player 2

L R
Game 1
U 10,5 5,4
Player 1
Player 2 M 11,3 1,2

L R
D 8,4 4,3
U 10,10 5,15
Player 1
Game 7
D 15,5 6,6
Game 2 6.3. THE TRAGEDY OF THE COMMONS
20. Refer back to the discussion on privatization in

Player 2 Section 6.3.2 and give additional examples of non-
L R excludable goods.
U 50,20 70,30 21. Consider Example 9 on page 384 again, but

Player 1
suppose now that the villagers decide to impose a

D 30,70 10,20 revenue-neutral tax of 2 units per cow, rather than 4.
What will be the final outcome in this case?
Game 3
22. Can you think of some examples (large or tax revenue is then divided equally between Alice and
small) of regulatory capture as Schneier describes it in Bob.
Remark 1 on page 383? What do you think can be done Consider the case in which Alice and Bob have each
to avoid this problem? fished for 6 hours (so they are at the social optimum).
If the tax rate is T = 2, is “Stop Fishing” a dominant
23. Consider the “two farmers” model for the tragedy strategy for each player? What is the dominant strategy
of the commons modified in the following way: with if instead T = 8?
n cows on the land, the amount of milk per cow is 10
gallons if n ≤ 10 , and is 20 − n gallons if n > 10. What 28. Suppose that the number of automobiles in the
number n of cows now gives the social optimum? world doubles because of greater demand, whereas
the average fuel economy of the world’s vehicle fleet
24. Consider our basic “two farmers” model for the improves by 40% because of better regulation. Will
tragedy of the commons, but suppose this time that the total amount of fuel consumed worldwide increase,
Farmer A starts with 2 cows and Farmer B with none. decrease, or stay the same?
How do the various stages of the game play out, and
what is the final outcome? Repeat your analysis for 29. According to a complete life-cycle analysis, a
several other cases, e.g., Farmer A starts with 4 cows wind turbine is responsible for about 12 grams of
and B with none, Farmer A starts with 6 cows and B CO2 emissions per kilowatt-hour of electrical energy
with 3, and so on. supplied, while a coal fired power plant is responsible
for 900 grams of CO2 emissions per kilowatt-hour of
electrical energy supplied.
25. Again consider the “two farmers” model, but this According to estimates from the U.S. Energy Infor-
time suppose that we start in an overgrazed state: for mation Administration, the cost of electricity from coal
example, Farmer A has 10 cows and Farmer B has 8. is forecast to be $0.10 per kilowatt-hour in 2018, while
We allow a third kind of move in the game—either the cost of wind is forecast to be $0.086 per kilowatt-
farmer can “retire” a cow as well as adding one. Use hour.
the same kind of analysis to investigate what will be Suppose that the U.S. instituted a carbon fee of $60
the ultimate outcome in this case. per ton of CO2 emitted for all power sources.
Estimate the costs per kilowatt-hour for both coal
26. Two people, Alice and Bob, fish from the same and wind with the addition of this fee. (Note that
lake. The number of fish a person catches per hour under the Carbon Fee and Dividend program, these fees
depends on the total number of hours of fishing effort. would be refunded with an equal share to all citizens.)
If H is the total number of fishing hours, then the
number of fish caught per hour is 24 − H. For example, 30. CO2 emissions in the United States during 2013
if each person fishes for one hour, then H = 2 and there are estimated to have been 16.5 metric tons per capita.
are 22 fish caught per hour. Suppose that under a carbon fee and dividend pro-
The total fish harvest is maximized when H = 12.
gram, the U.S. imposes a fee of $80 per metric ton of
Local authorities implement a tradable permit sys-
CO2 . Estimate your annual cost due to this fee if your
tem, whereby Alice and Bob are each issued 6 permits.
carbon footprint is 40% below average (i.e., 40% below
One permit gives the holder the right to fish for one
the per capita amount). Estimate your annual dividend
hour. If Alice and Bob each use their permits to fish,
(ignoring any costs of administering the program).
they each fish for 6 hours and each harvests 72 fish.
Suppose instead that Alice wishes to fish for 7 hours
and Bob trades one of his permits to Alice and fishes 31. Burning 1 gallon of gasoline emits about 20
only 5 hours. How many fish is the permit worth in this pounds of CO2 . If a carbon fee were instituted of $60
trade (if we ignore any costs of fishing)? per ton of CO2 emitted, how much would this raise the
cost of gasoline at the pump? (We are ignoring the fact
that there will later be a refund.)
27. Refer to the previous problem. This time local
authorities implement a revenue-neutral tax. Alice and
Bob are each required to pay a tax T for each hour 32. In the U.S., a vehicle is driven 12,000 miles per
that they fish. They pay the tax in fish, so if T = 1, year (on average) and average fuel economy is about
Alice would pay 6 fish if she fishes for 6 hours. The 22 miles per gallon. A Toyota Prius gets about 55 miles
per gallon and a Hummer H3 gets about 14 miles per 37. The European Union set up an emissions trading
gallon. market in 2005 (which continues to operate). Because
Suppose that a carbon fee and dividend program of a quirk in the way payments were assessed, from
is implemented at the level of $60 per ton of CO2 . approximately 2010 to 2013, several companies in
Compute the annual cost of the carbon fee for (a) a China were deliberately creating large quantities of
vehicle getting average gas mileage, (b) a Toyota Prius, HFC–23, a greenhouse gas thousands of times more
and (c) a Hummer H3. Assume that each vehicle is powerful than carbon dioxide, in order to gain credits
driven the average mileage. (which could then be resold in the EU market) for
destroying it again. Does this information affect your
33. Referring to the previous exercise, assume that ethical evaluation of emissions trading in any way?
under a carbon fee and dividend program, each vehicle Why or why not?
owner receives a dividend equivalent to the fee charged
on a vehicle that gets average gas mileage. (This is
an oversimplification, since an actual program would 38. Imagine that you are responsible for managing the
include carbon emissions from other sources besides Earth’s carbon budget under some “cap and trade” type
vehicle transport.) Determine the net loss or gain from proposal. Scientists tell you that your budget is 5×1011
the carbon fee and dividend program for each of the 3 tons of carbon dioxide—Earth’s humans cannot afford
vehicles in the previous problem. to emit more than that total from fossil fuels, ever,
in order to have a reasonable chance of maintaining
a stable climate. Now some countries have already
6.4. AFTER MATH: DECISION-MAKING emitted substantial amounts of carbon dioxide in the
AND ETHICS past (e.g., Britain, 5.5 × 1010 tons over the twentieth
century, or the United States, 3.1 × 1011 tons), whereas
34. Think about Example 1 on page 389 again, but others have emitted essentially no carbon dioxide up
with a slight change. Instead of thinking about a future till now because the process of industrial development
possibility, imagine this time that you, as plant man- has scarcely begun (for example, Bangladesh, 5 × 108
ager, somehow learn that the process you are already tons, will suffer enormously from climate change).
operating produces a toxic discharge that kills five What do you think is the ethical way to take account
people per year. Is there a morally relevant difference of past emissions? Should they be “grandfathered in”
between this situation and the one in Example 1 on somehow, or should those countries that have enjoyed
page 389 when you were considering introducing a the benefits of high emissions in the past be required to
harmful process at some future date? If so, what is it? reduce emissions more rapidly in the future?
35. Regarding climate change, some people have

39. In 2001 the United States joined with 90 other
suggested that we are morally responsible only for
countries in agreeing to the Stockholm Convention,
those of our greenhouse gas emissions that are blame-
a United Nations treaty addressing persistent organic
worthy in some way. Do you agree with the idea that
pollutants (POPs). What are POPs and why are they
some emissions are blameworthy and some not? If
considered a “global threat?” Who benefits from their
you do, how do you reconcile it with the fact that all
use? Who is harmed? See [13].
emissions have the same physics: a “virtuous” carbon
dioxide molecule warms the Earth just as much as a
“blameworthy” one? 40. Impact from climate change can often feel far off
for but there are vulnerable communities around the
36. In the context of the previous question, there are globe and in the United States that are vulnerable to
different possible ways of understanding “blamewor- climate change impacts today. Read about the resettling
thy”: for example, knowing versus unknowing emis- of the people from Isle de Jean Charles, Louisiana [91]
sions, or deliberate versus accidental, or luxurious and the people of Lake Poopó in Bolivia [62]. What
versus necessary. Do you find any of these distinctions is the role of climate change in each case? Are there
convincing? If not, can you think of a way of assigning other contributing factors? What other communities are
“blame” that is convincing (to you)? vulnerable today?
41. Can you account for the apparent fact that some fix” for climate change, also deny that climate change
of the very same people who advocate geoengineering is actually occurring?
(Exercise 23 on page 254 of Chapter 4), a “technical
Part II
Case Studies
C HAPTER 7
Case Studies
In Part I we developed mathematical tools for investigating sustainability. The many short examples
and exercises provided throughout Part I are there to help build familiarity with these methods. Now
it’s time to put them to use within the context of persuasive writing. Our aim is for you to use
quantitative information, appropriately and effectively, in your own writing, and also to develop skills
in assessing quantitative reasoning when it is used by others.
Here in Part II we provide a series of case studies that demonstrate how to use quantitative
reasoning—the mathematical methods of Part I—as part of an extended discussion, something that
might be a newspaper article or a blog post or simply your part of a conversation with your roommate.
Each case study is about a specific, practical question that is also sustainability-related. Some of
them are purely analytical pieces (discussing the question “impersonally,” without recommending a
particular answer), but others are examples of advocacy: writing about a public policy question that is
intended to persuade, motivate, or influence people toward a specific solution. These range in scope
from the local (recycling in your dorm on at your college campus) to the global (worldwide policy
designed to address climate change).
Quantitative argument can be a powerful tool in advocacy. Yet at the beginning of this chapter
we want to make clear that as mathematically literate people, your first loyalty should be to the
truth. This book will have failed in its intention if all you take from it is some impressive-sounding
technical language that you can deploy to argue for conclusions you’ve already reached by other
means—whether those conclusions are “deep green” or the exact opposite. In fact, we should expect
quantitative (mathematical) reasoning to sometimes challenge our preconceptions, whatever they are,
and it’s exactly for that reason that this skill is so important. One reason we’ve provided the case
studies in this chapter is to help you see how it is possible for the rigorous use of quantitative reasoning
to coexist with passionate advocacy for a particular cause. For this to work, the advocacy has to emerge
from the reasoning, not the other way around.
Several of the case studies also provide supporting exercises. These give you a chance to develop the
argument further but also to criticize the case studies themselves: Could the arguments be improved?
Could the presentation be made more effective? Are there logical gaps and fallacies? The ability to
criticize an argument—which means first to understand it in the most favorable way you can, and then
to point out any problems that still remain—is extremely important, for other people’s arguments but
especially for your own. Being able to improve through self-criticism is one of the most valuable life
skills any person can possess.
The methods and ideas of Part I can be applied to many different situations: these case studies
represent only a small sample. As time permits, we will include additional case studies on the
resource site for this book, http://math-for-sustainability.com. If you have a suggestion
for a case study (especially one that you have worked on some yourself), feel free to send it to us at
comments@math-for-sustainability.com.

https://doi.org/10.1007/978-3-319-76660-7_7
412 CHAPTER 7. CASE STUDIES
The first section (Section 7.1) of this chapter is a little bit different from the others. Rather than being
a case study of its own, this section gives some brief advice about writing effectively using quantitative
reasoning. Of course, a college writing class will teach you much more about writing in general (and
we encourage you to take one), but here we address a few specifics about incorporating mathematical
material in your written work. At the end of Section 7.1 we list some additional resources that you
could consult about this kind of writing, or about writing in general.
7.1. MATHEMATICS AND PERSUASIVE WRITING 413
7.1 Mathematics and Persuasive Writing

Remember Section 1.4? There we talked about various strategies for presenting a single set of
quantitative information. We discussed the advantages and disadvantages of verbal, tabular, and
graphical presentation, as well as some of the ways that some styles of presentation can be misleading.
Having completed Section 1.4, you’ll have a good idea about the most appropriate and effective way
to communicate such a single “data set” to your audience.
Here’s the thing, though: a single set of data does not make an argument by itself. In this section we
want to take a step back and see how quantitative reasoning—including the presentation of quantitative
data—can fit into the larger structure of an effective and persuasive argument. In the courses from
which this book was developed, we have required students to write extended essays or blog posts1
that use quantitative reasoning to make a case for or against some course of action in the sustainability
arena. What can be done about the number of disposable water bottles wasted on our campus? How far
on average has my food traveled before it reaches my plate, and what will be the consequences if I try
to eat more locally? Should our nation make a major investment in new nuclear power plants? These
are examples of the kinds of questions that our students have addressed. Addressing them involves the
ideas of measurement, networking, flow, change, and risk that we have learned about in the first part
of this book, but a persuasive answer requires an extended piece of writing, not just a few numbers
and calculations. In this section, we’ll briefly review some strategies for producing a written piece of
this sort: a persuasive essay that incorporates quantitative reasoning.
7.1.1 Principles of Persuasive Writing

First, let’s think about general principles for crafting effective writing, whether or not it incorporates
mathematical ideas. This is the subject matter of rhetoric, which originates with the ancient Greeks,
who believed that the skills of persuasion were fundamental to democratic institutions.
Definition 1
Rhetoric is the art of effective or persuasive speaking or writing [270].
Here are some important skills.
Gathering ideas: Imagine that you have chosen to write about the issue of disposable water bottles
on campus, one of the examples we mentioned above, It goes without saying that you can’t write
if you have no ideas to share. So your first step must be to gather a good selection of ideas around
the issue. These will include your own initial thoughts on the topic (“I really hate wasting resources
unnecessarily,” or “It’s so convenient to be able to buy a bottle of water whenever I need one”); your
questions about those thoughts (“I wonder how much waste is really involved?” or “What alternatives
might be equally convenient?”); calculations or research that you might do to generate relevant
quantitative information (“How could I estimate how many water bottles are thrown away each year?”
or “How can I find out what happens to a bottle when it’s discarded on our campus?”); summaries
of sources (articles, books, online materials) that you’ve reviewed, including your own assessment
of their likely relevancy and trustworthiness (“This information looks helpful, but it comes from a
company that makes its profits from selling bottled water. How far can I trust it?”); and comparables
(“Here’s some information about what this other campus did about this issue”). At the end of this
process you’ll have a whole heap of different ideas. That, however, does not make an argument, any
1 Details about the specific assessment process that we have used are presented on the book’s website, http://
math-for-sustainability.com.
more than a heap of bricks make a house. Your job now is to build the “house.” If some of the “bricks”
don’t get used in the end, that doesn’t matter. In fact, good writers consider many more ideas than
they finally use.
Thesis and Framing: Once you have reviewed your

ideas, you should have an idea of the main point of the
argument that you want to make (“I’m going to argue
that we should phase out the sale of disposable water
bottles, and at the same time install a network of refilling
stations for reusable bottles. Oh, and every freshman
should get a reusable bottle with the university logo
when they arrive.”) That is what some people call the
Figure 1: Bottled water for sale on a store shelf. thesis of your argument. You’ll also want to decide the
context or framing that you’re going to use to present
your thesis. Continuing with the water-bottle example,
you could choose a financial framing (“I’m going to show how this will save everyone money in the
long run”). Or an environmental framing (“Look at the reduction in campus energy use if everyone
follows this policy”). Or a more personal framing (“Here are the changes that I’ve made to lower the
amount of waste I generate, and here’s why you should consider making those changes too”). The way
you frame a topic will affect the overall impact of your argument. In general, a framing that connects
with the reader’s experience (compare Section 1.1.5) and is active (calling for a specific response from
the reader) is likely to be more effective. If you choose, say, the personal framing, you are deciding
the overall direction of your written piece: you probably will still use financial and environmental
calculations, but you’ll use them in support of the framing you have chosen.
Outlining: Now sketch out the overall structure of your piece. No need for super-elaborate outlining
schemes here: list the main idea for each paragraph, and, if you need them, a few bullet points
underneath that main idea for other points that you want to bring out in the paragraph. Check that
the outline flows logically: each paragraph rests on the previous ones, building up your argument
step by step. Again, the picture of building a house out of bricks is helpful; you can’t put bricks in
for the second floor until the first-floor bricks are in place underneath. It’s quite usual to realize at
this point that there’s some piece of information or idea that you are missing. No problem: go back
to the “gathering ideas” step for a while and add the new materials that you need to your “ideas”
folder. Some people will tell you that after adding new ideas you should restart your outline from the
beginning. That may be a piece of advice that is more often given than followed; but at least, having
added new ideas, you should check that your outline up to this point is still consistent with them.
Understanding and Using Rhetorical Appeals: Remember, rhetoric is the art of persuasion. The
ancient Greeks developed a list of different types of persuasive skills, called rhetorical appeals (see
Table 1 on the opposite page). Our emphasis on mathematical and quantitative reasoning falls squarely
into the class of the logos appeal—the appeal to reason. But mathematical reasoning is only a part
of your argument. A strong argument will probably make use of many or even all of the appeals: it
should be based on sound logic (logos), but imagine the strength of the impact if you also establish
yourself as a credible source (ethos), make an emotional connection with the reader (pathos), and
explain why now is the time to act (kairos). Now you’re persuasive!
Writing and Revision: Now it’s time to start writing. Following your outline, pour it out on the
page; don’t stop to think too hard (the things where you have to think hard, like mathematical
calculations, you’ve done already in the “gathering ideas” step, right?). Great! Now you have a first
Table 1: Rhetorical Appeals
Name Meaning Example

Logos Appeal to reason “We’ve done the calculations based on the best available
data. The results are completely clear.”
Pathos Appeal to emotion “I love to hike at night under the stars, and I want my children
to be able to enjoy hiking too. Make no mistake: unless we
control carbon pollution, our children’s future is threatened.”
Ethos Appeal to credibility “Dr. X has spent her entire professional career in climate
science, writing three books and over 200 peer-reviewed
journal articles. You can trust what she has to say.”
Kairos Appeal to timeliness “The decisions we make today will determine the future for
generations to come. Let’s be sure we get them right.”
draft. But no piece of writing is perfect the first time—or even the second or third. Skilled writers like
George Saunders [286] and Susan Sontag [304] know that excellent work is the product of a continual
process of revision and improvement, of asking oneself again and again, “Does this phrase or sentence
or paragraph do the best job that it possibly can in sharing what I want to say with the audience I want
to reach?” Keep revising until you can answer “yes” to this question, on every level from the phrase
to the paragraph to the structure of the entire piece.2
Analysis and Advocacy: In the “thesis and framing” step above, we said that the best framing
often includes an active component—something for the reader to do. As you revise and review your
piece, don’t let your readers walk away wondering what actions you wanted them to take. Your final
paragraph is the place to make this clear. If your writing is what we call an analysis piece, one that
is crafted to help your reader understand a specific topic in greater depth, your conclusion might be
something like, “Thank you for following along with me through our discussion of [whatever the issue
is]. The key points were [such-and-such], and I hope that next time you have to take a decision about
[some relevant matter], they’ll help guide you.” Or your writing may be an advocacy piece, one that
argues that the reader needs to take some specific action right now. A great way to close out such
an appeal is to include a call to action in the final paragraph, something like, “Here are the changes
I’ve made [name them]. But I’m only one person. Won’t you join me?” Or on a broader level, “We
need our government to take [such-and-such action]. Does that mean there’s nothing we can do as
individuals? No! We live in a democracy. I’ve written to my congressional representative to share the
arguments I’ve made. If you’re convinced, won’t you do so too? Here’s how you can. . . ”3
7.1.2 Incorporating Mathematics

In this section we’ll talk about some of the specific issues that come up when you are writing the
quantitative part of your piece:
Use human terms: We’ve talked about this in Sections 1.1.5 and 1.4.1, as well as in other places.
Our minds receive quantitative information more effectively if it is related to units that are on a human
or imaginable scale, using numbers that are not too large or small (Rule 1 on page 17). Let’s give one
more example to review the point.
2 It can be helpful to have someone else’s opinion during this revision process—in our classes, a “double blind peer review”
process is implemented through software: neither reviewer nor reviewee know the other’s identity. If you’re in the position of
reviewer, remember that your goal is to help the writer make their work more effective.
3 Contacting U.S. Representatives and Senators is made much easier by apps like Countable [84].
Example 1: The loss of glacial ice in Alaska, owing to warming temperatures, has resulted in an
increased frequency of large landslide events. One such landslide in June 2016 was reported on in the
Alaska Dispatch News. Here’s an excerpt from that article [150]:
According to their preliminary analysis of the seismograms and available imagery, the
landslide started at 8:21 a.m. Tuesday when the rock face collapsed on a high, steep
slope. For nearly a minute the debris accelerated down the mountain, hitting the ice on
Lamplugh Glacier and pushing up snow and ice as it continued across the glacier, [Dr.
Colin Stark] said.
Stark said rough estimates put the size of the slide at about 130 million tons, comparable
to roughly 60 million medium-size SUVs tumbling down the mountainside.
“That’s a lot of SUVs,” Stark said. “It’s huge.”
Dr. Stark is surely doing his best to express things in human terms, but he doesn’t altogether
succeed. (To be fair, he probably didn’t have much time to come up with a comparison!) His example
replaces one difficult to comprehend quantity, 130 million tons of rock, with another that is equally
abstruse. If one cannot picture 130 million tons of rock debris, it’s difficult to imagine that the image
of 60 million SUVs tumbling down a mountainside is much more helpful. Here’s an alternative
comparison:
In the landslide described above, 130 million tons of debris fell in less than one minute.
The flow rate of water over Niagara Falls averages 3000 tons per second. At that rate, it
takes about 12 hours for a comparable weight of water to go over Niagara Falls.
Because your reader has almost certainly seen pictures of Niagara Falls (Figure 2), and may even
have visited there, this “human terms” expression is likely to be more effective than the image of 60
million SUVs. We have effectively used rescaling, as in Example 10 on page 20, to compare the flow
rate of the Alaskan landslide with the already-massive flow rate of Niagara Falls.
Use graphics and tables: Though we are talking about a “written

piece,” that doesn’t mean that you can present quantitative information
only in words, sentences, and paragraphs. Often, graphics or tables are
more effective ways to present the information and don’t require the
reader to work so hard to understand the point. In Section 1.4.3 we
reviewed how to use graphical techniques such as bar charts, pie charts,
scatter plots, and line graphs to vividly convey a summary of complex
quantitative data. When it’s really important that your readers have the
Figure 2: Niagara Falls. actual numbers, tabular presentation, discussed in Section 1.4.2, may
be the best way to go. You can even think about including an equation
if you’re writing for the right audience: for those who have the relevant
mathematical background, an equation can be a shorter and clearer way of explaining the relationship
between two quantities than a lot of complicated writing. (But don’t do this unless you’re sure it will
really help your readers to understand your argument.)
Let’s review one more example in which an appropriate use of graphics will help make a point more
clearly than anything else can.
Example 2: Worldwide, about 3 billion people rely on open fires or simple stoves burning biomass
(wood, straw, or dung) to cook their food or heat their homes. Burning biomass produces smoke, which
accumulates in homes or cooking spaces and gets into the lungs—especially the lungs of those who
are traditionally responsible for food preparation, usually women and children. Fine smoke particles
remain in the lungs and cause a variety of diseases including pneumonia, stroke, chronic obstructive
pulmonary disease (COPD), and lung cancer. The World Health Organization estimates that over
4 million premature deaths a year result from indoor air pollution from biomass burning. These
lives would be saved by access to heating technologies taken for granted in the developed world—
cookstoves using electricity or natural gas, for example. Western environmentalists who worry about
the pollution that our sources of energy produce (compare Section 2.2.2) ought also to remember this
and other ways in which access to these sources of energy can be a literally life-and-death matter for
half the world’s population.
Western Pacific 1,620

Southeast Asia 1,692
Europe 117
Eastern Mediterranean 201
Americas 81
Africa 581
0 200 400 600 800 1,000 1,200 1,400 1,600 1,800

Annual deaths (thousands)
Figure 3: Annual premature deaths attributable to household air pollution (from [314]).
Suppose you want to write about this. It’s a deeply important and seemingly intractable problem,
combining social, environmental, and justice issues. You can find much useful data from the World
Health Organization and other groups working in the field. But how to present the basic numbers to
your reader in an attention-grabbing way? One effective answer, if you choose to frame your article in
terms primarily of justice, might be the chart of Figure 3 (adapted from [314]), which shows how may
deaths from indoor air pollution arise in different regions of the world (not all of the world is covered
by the data, but most is). The bar chart makes it immediately clear how stark the divide is between rich
and poor areas of the world. Once you have conveyed this, you have got your justice-minded reader’s
attention, and you can go on to a more detailed discussion.
Relevance and inference: What quantitative evidence is needed for your argument? Be sure that the
information that you provide is relevant to the argument you want to make; quantitative information
that is not directly relevant will distract from your purpose. Bear in mind, however, that evidence of
any kind that runs counter to your argument is highly relevant. If counter-evidence of this kind exists,
your initial idea-gathering process should have found it. The best response may well be to confront
the counterargument head-on (see “Addressing counterarguments” below).
Inference is the art of correctly drawing conclusions from evidence. Once you have assembled
and discussed your relevant quantitative evidence, you want to make sure that your argument uses
it in a correct way. There may be attractive or popular conclusions that the evidence does not
directly support; avoid drawing them. For example, a scatter plot using data from many years will
show that there is a correlation between atmospheric carbon dioxide levels and mean global surface
temperatures. But this does not suffice to prove that increases in atmospheric CO2 cause increased
Earth surface temperatures (see page 323): someone could argue that the causation is the other way
round (hotter Earth causes more CO2 to be released), or that both CO2 levels and temperatures are
caused by something else, or that these two factors are part of a complex system in which they both
influence each other (which is close to the truth as understood by modern climate science), or even
that the apparent correlation is a coincidence. To present your argument honestly, you need to be able
to explain not just the observed correlation but also the mechanism whereby CO2 levels can affect the
temperature (see Section 2.4).
Addressing counterarguments: It is always a good idea to think about possible counterarguments

to your thesis, as part of the revision process: this will make your own argument stronger. Sometimes,
though, you will want to do more than this: you will want to “counter-argue” as part of your actual
writing. Counter-arguing involves two moves, sometimes described as the “turn away” and the “turn
back.” In the turn away, you imagine an unconvinced reader, or maybe cite an actual written source,
who makes an objection to your line of reasoning. This might be a piece of evidence that points in
a different direction, or another explanation of the facts that seems more convincing, or a practical
difficulty with the action you propose. Whatever it is, you address it head-on: “Some people feel that
a carbon fee and dividend proposal is bound to make transportation unaffordable, especially for those
who are already living in relative poverty.” (And go on to develop this objection for several sentences.)
Then comes the “turn back”: in this example, something like “Remember, though, that the proposal
is revenue neutral: all the carbon fees are returned to the people as dividends. Studies show that these
dividends will make the biggest difference to the lives of poorer people, and most of the time will
more than offset their increased travel costs. . . .” Anticipating and disarming objections like this helps
convince the reader, not only by the power of argument itself (logos) but also by showing that you
are someone who can think clearly about both sides of an issue, and therefore someone whose word
carries weight (ethos).
Citing sources: In an academic paper, you will need to “cite” or refer to all the sources that you used
in preparing your work (there are many conventions about how this should be done; for an example,
look at the bibliography at the end of this book). Even when you are not writing for an academic
audience, though, it pays to give details about where you got your quantitative information—and
these references had better be authoritative and credible. You could write, “United States emissions
of greenhouse gases have been declining since 2007.” But how is the reader to trust this unattributed
statement? Much better to write, “The U.S. Environmental Protection Agency maintains an annual
inventory of greenhouse gas emissions. This shows that U.S. greenhouse gas emissions peaked in
2007 and have been declining (though not steadily) since then.” And if your writing will be available
online, for instance as a blog post, it is hardly any extra effort to add a hyperlink to the text “annual
inventory of greenhouse gas emissions,” which will send your readers to the EPA report itself. Doing
this is an important way in which you establish a relationship of trust with your readers; please don’t
neglect it!
Further reading: In this brief section we have only had time to hit a few highlights about writing
about sustainability from a mathematical perspective. The remaining case studies give examples of
writing, both analysis and advocacy (see page 415), that uses the mathematics of Part I of this book to
look at specific sustainability questions. If you want more general ideas about writing skills, though,
we’ll have to point you to the other excellent resources that are available. Here are a few:
• General guides about college-level writing: [161], [158].
• Specifically about writing in the quantitative disciplines: [38], [218].
• About quantitative writing and democracy: [307].
• Famous writers talk about writing: [181], [286], [304].
Happy writing!
7.2. THE CHANGING RISKS OF WILDFIRES 419
7.2 The Changing Risks of Wildfires
About this case study
• This case study is an advocacy piece. It argues that communities need to update their
assessment of risk from natural disasters and prepare for previously unanticipated
risks due to climate change.
• It uses ideas from Sections 5.1 (data and distributions), 1.4 (verbal communication,
graphical presentation, and tables), 4.4 (tipping points in the global climate sys-
tem), 5.3 (risk aversion), and 6.4 (the precautionary principle).
On November 23, 2016, two teenage hikers started a small fire along the Chimney Tops trail in the
Great Smoky Mountains National Park. The fire slowly grew over the next several days, eventually
covering 2–3 acres. It was monitored by park officials and thought to be contained within a limited
area inside the park.
On Sunday, November 27, weather conditions became both drier and windier; the fire grew to 35
acres, but remained within the containment area. By Monday morning, however, embers carried on
the wind were igniting new fires up to a mile away. During the afternoon the winds intensified and
changed direction, driving the fire toward the nearby town of Gatlinburg, Tennessee. The situation
soon turned catastrophic and deadly, with residents and vacationers driving down forested mountain
roads, with wildfires and homes burning on both sides of the road. The fire would eventually cover
17,000 acres, taking 14 lives and consuming over 2400 homes and commercial structures.
In this study we’ll examine the connection between climate change and the changing risks of
wildfires. We’ll make the case that communities must assess how climate change affects their
vulnerability to natural disasters and that steps should be taken to prepare for potential impacts and
for adapting to these changes.
Changing Conditions
We think of climate change as a gradual process; increases in global average temperature might be
in the range of 1–2 ◦ C over a century; sea level rise is measured in millimeters per year. These
changes do not sound dramatic. However, as we saw in Section 4.4, slowly changing parameters can
sometimes precipitate abrupt changes. Wildfires are not unknown in the southeastern United States,
but the severely dry conditions that allowed a 3-acre contained fire to quickly transform into a deadly
conflagration may very well be. What we are interested in here are changes in risk and our ability to
update our assessment of those risks.
Consider the following quotation from a summary report of the Chimney Tops fire by the National
Park Service [22]:
Spot fires 1/2 to 1 mile away from the fire front are extremely rare in this part of the country
and in these fuel types.
More recently, a member of a national panel reviewing the National Park Service’s handling of the
fire, after interviewing officials involved in the fire was quoted as saying:
people with a lot of experience ... have never seen anything like this before in Tennessee.
While the fire was closely monitored and thought to be contained, the prevailing conditions were
outside of the experience of the region. This likely led officials to underestimate the risks to the
community outside the park boundary.
Changing Patterns of Wildfire in the United States

Let’s step back and take a look at wildfires throughout the United States. Table 2 shows data from
the National Interagency Fire Center on the number of wildfires and millions of acres burned between
2005 and 2016 [36]. In 2015, over 10 million acres were burned in wildfires, an area about 1/3 the
size of Pennsylvania. In Table 3 we present the same data, sorted by the acres burned variable. Using
the techniques presented in Section 5.1, we can compute the median, upper and lower quartiles, and
the interquartile range. The results of these calculations are given in Table 4 on the opposite page
Table 2: U.S. Wildfire Data, 2005–2016
year no. of fires acres burned (millions)

2016 67,700 5.5
2015 68,200 10.1
2014 63,300 3.6
2013 47,600 4.3
2012 67,800 9.3
2011 74,100 8.7
2010 72,000 3.4
2009 78,800 5.9
2008 79,000 5.3
2007 85,700 9.3
2006 96,400 9.9
2005 66,800 8.7
In Table 3, we’ve sorted the data by the acres burned variable.4
Table 3: U.S. Wildfire Data 2005–2016, Sorted by Acres Burned
year no. of fires acres burned (millions)

2015 68,200 10.1
2006 96,400 9.9
2012 67,800 9.3
2007 85,700 9.3
2011 74,100 8.7
2005 66,800 8.7
2009 78,800 5.9
2016 67,700 5.5
2008 79,000 5.3
2013 47,600 4.3
2014 63,300 3.6
2010 72,000 3.4
A box plot (Figure 4) provides a visual representation of the same statistics.
4 6 8 10
Figure 4: Box plot for the data from Table 2 on millions of acres burned by wildfires in recent years (2005–2016).
4 For large data sets it’s helpful to use a spreadsheet like Microsoft’s Excel or Google Sheets.
Table 4: Summary Statistics, 2005–2016
Statistic Value (millions of acres)

Median 7.3
Lower Quartile 4.3
Upper Quartile 9.3
IQR 6.6
Although more acres were burned during the 2015 wildfire season than in any other recent year, it
doesn’t particularly stand out in this presentation. In 6 of the 12 years in our data set, more than 8.5
million acres were burned, and the median value is 7.3 million acres. There are no outliers in this data
set (recall that an outlier is a value that is significantly far outside the usual range, formally expressed
in Definition 7 on page 266).
To put this recent period of wildfires into context, though, we might want to look further back in
the record. Table 5 gives the summary statistics for millions of acres burned going back to 1960 (this
larger data set is from the same source [36]).
Table 5: Summary Statistics, 1960–2016
Statistic Value (millions of acres)

Median 4.1
Lower Quartile 2.9
Upper Quartile 5.5
IQR 2.6
Look at the box plot for our larger data set (Figure 5). You will notice that now there are outliers!
These correspond to the 9.9 million acres burned in 2006 and the 10.1 million acres burned in 2015.
What’s more, the value at the end of the “right whisker,” 9.3, corresponds to the years 2012 and
2007. The interquartile range is much narrower, with the values centered on the median of 4.1 million
acres, and the upper quartile is 5.5 million acres. In fact, almost every value above the upper quartile
represents a year after 2000, and in 10 of the 12 most recent years, the acres burned fall above the
median. The year 2015 is not remarkable when compared to other recent years, but the recent period
itself stands out markedly from the past.
2 4 6 8 10
Figure 5: Box plot for summary statistics from Table 5 on millions of acres burned by wildfires (1960–2015).
The Role of Climate Change

What is the role of climate change in the changing patterns of wildfires? Warmer and drier conditions
make forest fires more likely to spread once ignited, and climate change has already been implicated
as one factor in the recent increase in the number and extent of wildfires [2]. Climate forecasts predict
substantial increases in wildfires in the U.S. (and elsewhere) in coming decades.
Perhaps paradoxically, some areas of the eastern United States could see both an increase in rain
and an increase in the risk of drought [21]. How is this possible? Warming temperatures allow the
atmosphere to hold more water vapor, and this, along with changing weather patterns, may lead to
higher-intensity rain events. Warmer temperatures also increase the rate of evaporation, moving water
from soils to the atmosphere more rapidly. In a nutshell, the eastern U.S. may see fewer but more
intense storms, interspersed with longer dry spells; a higher risk of both flooding and drought!
There are other natural variations in climate that also play a role in wildfire risk. El Niño events
are associated with warmer and drier autumn weather in the southern U.S., increasing fire risk in
those years. But climate change also has a role here, affecting the frequency and severity of El Niño
events [262].
The Keetch–Byram Drought Index

Let’s return our focus to the southeastern
United States and the Chimney Tops fire.
We noted above that the dry conditions that
allowed the fire to spread may have been
unprecedented for this region of the coun-
try. The Great Smoky Mountains do not owe
their name to a propensity for wildfires, but
rather to the blue mists and fog that are often
present. The region receives 55–85 inches of
rain per year on average, with the higher
rainfall amounts at higher elevations [295]
(one of the authors once became somewhat
lost after missing a trail marker while hiking
in thick fog here). Drought-like conditions are
not typical.
We can put a number on these dry con-
Figure 6: View from Mt. LeConte in early April, Great Smoky Mountains
ditions. The Keetch–Byram Drought Index
National Park.
(KBDI) serves as a proxy for the moisture
content of forest soils and is used for assessing wildfire risk. The accumulated debris of leaf litter
and other dead plant matter, called the duff layer of forest soils, is a potential fuel source for forest
fires; the KBDI gives an indication of how likely the duff layer is to burn in a fire event. For a given
day and location, the scale depends on the maximum temperature, the amount of precipitation, the
mean annual rainfall, and the previous day’s index. It ranges from 0 to 800, divided into four intervals
indicating low, moderate, high, and extreme fire risk, with 0 indicating saturated soils [317]. See
Table 6. The U.S. Forest Service’s Wildland Fire Assessment System publishes a daily map of the
Table 6: Keetch–Byram Drought Index
Range Fire Risk Interpretation

0–200 Low High moisture in duff layer, unlikely to actively burn
201–400 Moderate Drying duff layer beginning to contribute to fire intensity
401–600 High Duff layer will actively burn
601–800 Extreme Intense, deep burning fires, downwind spot fires likely
KDBI for the entire country. A graphical summary of the KBDI ratings for Nov 26, 2016, is shown
in Figure 7 on the opposite page. The Chimney Tops fire was active (less than 3 acres in size) on this
day, in the Great Smoky Mountains National Park, along the North Carolina-Tennessee border. In this
region, the index ranged from just over 600 at the Gatlinburg weather station to over 750 at nearby
stations. Throughout much of the southeastern U.S., the KBDI was over 700, indicating severely dry
conditions.
Figure 7: KBDI for the continental U.S. on Nov 26, 2016. Throughout much of the Southeast, the KBDI was above 700. In the
Great Smoky Mountains National Park along the border of TN and NC, the KBDI ranged from 500-750 [317].
How do these conditions compare with past Novembers? Typical values for the KBDI in this region
during the fall range from 200 to 400 (low to moderate risk) [197]. The year 2016 is the only year
going back to 1960 in which the KBDI exceeded 600 during the month of November. In 2007, the
index came close, reaching 598 in early November (Nov 5, 2007). The KBDI map for this date is
shown in Figure 8. Prior to the fall of 2016, this represented some of the driest conditions experienced
in decades during November. In this map, areas where the KBDI exceeds 600 are more limited in
extent (parts of Georgia and South Carolina), with no regions in Tennessee rising to this level and no
regions throughout the South where the index exceeded 700.
Figure 8: KBDI for the continental U.S. on Nov 5, 2007. The KBDI for the region around Great Smoky Mountains National
Park is above 500; there are no extensive regions in the Southeast where the index exceeds 700 [317].
In contrast, the KBDI in November 2016 exceeded 600 throughout most of Tennessee and exceeded
700 throughout much of the South. What will the future bring? Climate research suggests that these
conditions are likely to become more common in the coming decades. Figure 9 shows a projection of
expected changes in the KBDI for the years 2041–2070, compared with the period 1971–2000; the
red areas indicate the largest increases (a change of +200 in the index). Particularly in the fall season,
the largest increases are concentrated in western mountain states and the Appalachian Mountains of
the eastern U.S. Much of the region in the southeastern United States, where fire risk currently ranges
from low to moderate during the autumn are expected to see a shift to moderate to high fire danger—as
typical conditions. The extreme conditions seen in 2016 will become only more likely.
Figure 9: Predicted changes in the Keetch–Byram Drought Index, 2041–2070 (change relative to the years 1971–2000).
Reprinted from [198].
Summary
Almost a year after the Chimney Tops fire, the U.S. Department of the Interior’s Division of Fire and
Aviation released their review of the fire and the National Park Service’s response to it. The review
panel concluded “that the firefighting decisions made by the personnel involved were commensurate
within their knowledge and experience fighting wildland fires in the region,” and that a lack of
preparedness in handling the prevailing drought conditions “overwhelmed [the] National Park Service
response” [251].5 The Chimney Tops fire provides a warning that in changing conditions, past
experience may not serve as a reliable guide. Communities and states must take an active role not
just in evaluating current natural disaster risks, but in assessing how those risks will be affected by
changes to the climate; this is an essential aspect of building and maintaining community resilience
and in adapting to the impacts of climate change.
5 While the fire review acknowledges that conditions were outside of the historical experience of the region and calls for
building the capacity of federal agencies to respond to “unprecedented” events, it does not directly indicate the role climate
change may play.
While we cannot hope to make long-range predictions about individual events such as the Chimney
Tops fire, we do have the tools to understand how climate change is affecting the likelihood of these
events. Using this understanding we can take steps to mitigate the increasing risks to human life and
property from a range of threats that may be exacerbated by climate change, wildfires included. We
need to recognize this and act on it.
7.3 Is Recycling Worth It, Really?
• This case study is an analysis piece. It examines the value of current recycling
programs in the United States and the potential benefits of increasing the recycling
rate.
• It uses ideas from Sections 1.1 (unit-factor method), and 1.4 (graphical presentation
and using tables),
Very few objects of modern consumption were designed with recycling in mind. If the
process is truly to save money and materials, products must be designed from the very
beginning to be recycled. . . [54].
When we put an item in the trash, we typically say we are “throwing it away.” For most of the trash
generated in the United States, “away” is a potentially distant landfill.6 A little over one-third of our
trash is recycled or composted; a smaller fraction is sent to waste-to-energy incinerators. See Figure
10(a). In this case study we’ll take a closer look at trash and recycling in the United States.
Combustion and 2.2% Food
Energy Recovery 2.9% Wood
12.8% 6% 3.5% Plastics
3.3% Glass
Metals
8.8%
52.6% 34.6% 49.7%
Recycling and Paper &
Landfilled Paperboard
Composting 23.6%
Yard
Waste
Management of municipal Recycling and

solid waste in the U.S. (2014). composting, by material (2014).
Figure 10: Management of municipal solid waste (MSW) in the U.S., 2014. Adapted from [15].
The Value of Trash

At its heart, recycling is about capturing the value in discarded materials for use as feedstocks in
the manufacturing process. Ideally, recycled materials would displace raw materials that must be
extracted or harvested from the environment, and serve the same role again and again. This is the
case with aluminum, which can be repeatedly recycled with no loss in quality. In practice, many
materials collected for recycling are actually downcycled to lower quality products. Plastic bottles
and food containers, for example, are often used as feedstocks in manufacturing products that are not
required to meet food safety standards, including carpeting, fabrics, and plastic composite lumber.
6 Trash from New York City is sent to landfills as far away as South Carolina (about 750 miles).
7.3. IS RECYCLING WORTH IT, REALLY? 427
For some materials, the salvage of scrap has long been a part of the industrial process.7 Recycling
and reuse are also linked to times of shortages and to regions where poverty rates are high—situations
in which the relative value of discarded materials will tend to be higher. There are an estimated
300,000 rag-pickers in Delhi and Mumbai, India (some as young as 5 years old), who try to scrape
together a living by searching through waste, collecting materials that they can sell to scrap dealers.
Figure 10(b) on 426 shows a breakdown of what is currently recycled or composted in the U.S. Most
of what we currently recover can be categorized as materials that are either high in value (metals), easy
to recycle or compost (cardboard and yard waste), or for which there exist well-developed markets
for recycled scrap (metals, paper, cardboard, and some plastics). What of the remaining materials
that head to landfills? Can they be economically recycled or composted? What are the benefits to the
environment for doing so? In the next section we look to electronic waste as an example of high-value
materials that are difficult to recover.
The Gold in Your Smartphone

Let’s consider a material that you probably aren’t used to thinking of as trash:
gold. There is gold in your smartphone and in most other electronic devices. In
addition to gold, your smartphone contains many valuable platinum group metals
and rare-earth elements. Unlike the aluminum in a soda can, however, these high-
valued materials are not so easy to recover. According to the EPA, in 2010 about
11% of discarded mobile devices in the U.S. were recycled. The rest went to
landfills. How much gold are we talking about here?
A typical smartphone contains about 0.001 troy ounces of gold; a troy ounce is
approximately 31 grams. If we can recover all the gold from each one, then the Figure 11: A one ounce gold
bar, the amount of gold in
gold from 32 smartphones adds up to about 1 gram of gold: about 28 tonnes of low-grade
ore from an open pit gold mine,
1 smartphone 1 troy
oz

or in about 900 smart phones.
× ≈ 32 smartphones/ g.
0.001troy
oz
31 g
To get a better sense of the value of the gold in smartphones, it’s helpful to look at what’s involved
in mining one gram of gold.
Modern gold mines are strip mining operations that often rely
on a chemical process called heap leaching to recover gold from
low-grade ore. The low-grade ore from an open pit mine contains
about 1 part per million of gold by weight. Ore from the Sunrise
Dam open pit mine in Australia (pictured in Figure 12) is about 2
parts per million gold by weight [71]. Since one tonne is equivalent
to one million grams, each tonne of low-grade ore mined yields
about one gram of gold. The ore is crushed and placed in a heap;
a cyanide leach solution is used to leach gold and other minerals
from the heap. In the final stage, gold is recovered from the leach Figure 12: Sunrise Dam open pit gold mine, Aus-
tralia.
solution.
In 2010, Americans sent 135 million mobile devices to landfills. Using the estimates on gold content
discussed above, we can estimate that this amounted to about 4200 kilograms of gold that might have
been recovered:
1
g 1 kg
135 × 106 phones
×
× 1, 000 g ≈ 4200 kg.
32
phones

7 According to a discussion that one of the authors had with a colleague in the competitive metal fastener industry,
recovering and recycling of scrap metal produced during the manufacturing process is an important source of revenue for
firms in this industry
In May 2017, 4200 kg of gold had a market value of over $165 million, and would represent the
mining of over 4 million tonnes of ore.
Closing the Loop
Waste
Extraction Manufacture Distribution Use Disposal
Energy
Figure 13: Cradle-to-grave life cycle of a plastic water bottle.
We can imagine a smartphone along its entire life cycle. At one end, raw materials are extracted and
processed into feedstocks for manufacturing; at the other end the phone is discarded into the trash,
into the environment, or recycled. We refer to the linear life cycle of products that are landfilled after
their useful life is through as the cradle-to-grave life cycle. See Figure 13.
Each stage in the life cycle requires inputs of energy and additional materials and also generates
waste. If the phone ends up in the landfill, we lose the value of any materials that could be recovered.
Furthermore, when we recover materials for reuse in manufacturing new phones, we avoid the costs
and impacts of extracting and processing new raw materials.
When we compare the amount of gold in smartphones with the gold content of ore in mining
operations (see the exercises), it’s not surprising to learn that recovering gold and other materials
from electronic waste is economically viable. In 2015, Apple recovered 61 million pounds of material
from recycled Apple products, including $40 million worth of gold. What may be surprising is that
most of those recovered materials are not used to manufacture new Apple products. Most recycling is
a long way from a true cradle-to-cradle process, or what is called closed-loop manufacturing.
In closed-loop manufacturing, materials from recycled products are reused as inputs in manufactur-
ing the same products. Ideally, we could envision the economy consisting of two interlinked cycles:
a technical cycle of materials that can be continually reused and remain isolated from ecosystems,
and a biological cycle of organic materials that can safely interact with, and will breakdown in, the
environment; see [73].
There are examples of manufacturing processes that approach the closed-loop ideal: we’ve already
mentioned the recovery of scrap aluminum from cans and other products; corrugated cardboard
provides another example.8 You’ll notice that these products are not too hard to separate from other
materials in the waste stream. In contrast, separating out the various materials in an automobile or a
computer for reuse is more challenging.
Sorting materials isn’t the only challenge. A person who spends hundreds of dollars on the newest
smartphone model is not likely to be very concerned about the value of the small amount of gold
in their discarded phone. It is only in the aggregate that discarded materials have enough value to
8 About 70% of recycled corrugated cardboard is manufactured into new corrugated cardboard. The remaining 30% is
downcycled to other paper products.
make recovering those materials worthwhile; recycling as it is currently practiced thus depends on
encouraging the voluntary participation of a high proportion of citizens. Usually this involves an
appeal to the environmental benefits of recycling and making participation in recycling programs as
easy and convenient as possible.
Extended Producer Responsibility

One way to improve participation is by moving the burden of participation from the buyer to the
product manufacturer. This idea goes by the name of extended producer responsibility (EPR). Thirty-
two U.S. states currently have EPR laws concerning one or more products. These are primarily for
products containing hazardous wastes such as lead-acid batteries, fluorescent lighting, and mercury-
containing thermostats [165]. These laws make producers responsible for the recycling or safe disposal
of wastes at the end of the product’s life, and usually involve take-back programs.
In the European Union, the framework for EPR is more extensive, with most countries having EPR
laws that extend to packaging. Under these laws, manufacturers are legally responsible for meeting
national recycling targets. In practice, recycling is funded by fees that manufacturers pay based on the
amount of packaging they produce [118].
EPR laws not only increase recycling participation, they can create incentives for manufacturers
to consider recycling and product end-of-life as part of the design process. Some manufacturers who
use voluntary take-back programs have been able to greatly increase recycling and make a profit on
waste reduction. In the 1990s, Xerox became a pioneer in toner recycling and in take-back programs
for copiers and printers. Toner and ink-jet cartridge recycling has since become a successful example
of closed-loop manufacturing [98].
Creating Markets for Recycled Materials

Recycling programs cannot be successful without markets for scrap material. Closed-loop manufac-
turing depends on manufacturers choosing to use recycled waste materials as feedstocks. Paper and
cardboard make up half the material recovered from the municipal solid waste stream. The markets
for scrap paper and cardboard are well established, and high-quality paper products made with post-
consumer waste are readily available. How were the markets for waste paper developed?
In 1993, the U.S. federal government established minimum standards for post-consumer recycled
content in paper purchased by federal agencies. This mandate established a consistent demand for
post-consumer waste paper. The paper industry responded by improving technologies for using waste
paper as a feedstock, including deinking and removing contaminants [125].
For most products, we should expect closed-loop manufacturing to involve technical hurdles in
reusing waste materials. Some combination of consumer demand for recycled materials, extended
producer responsibility laws, and mandates for minimum standards for recycled content may be
necessary to give manufacturers the incentives to address these challenges.
The Environmental Benefits of Recycling

So far we’ve focused on the economics of recycling, and what it takes to make recycling programs
work. What are the environmental benefits of recycling? Is it beneficial to use recycled materials, or
does it actually waste energy to collect and sort discarded materials?
Recycling has a number of critics. In 1996, John Tierney famously questioned the benefits of
recycling in an opinion piece for the New York Times entitled Recycling Is Garbage [321]. A recycled
version of his argument was published more recently in the Times as The Reign of Recycling [322].
Daniel Benjamin, a senior fellow with the Property and Environment Research Center (PERC) has
also challenged the environmental benefits of recycling [47].
The main claim of Tierney, Benjamin, and others is that collecting and using recycled materials
is more wasteful than sending waste to the landfill. In terms of upfront time and energy use this is
certainly true. In curbside collection of recyclables, there is some duplication of effort; each house is
visited by both a garbage truck and a recycling truck. Additional energy and effort must be spent in
sorting and preparing materials for reuse. As a result, for most materials, collecting and sorting trash
to prepare it for recycling requires more energy and effort than the energy required to send it to the
landfill.9 The amount of extra energy involved will vary by material and depends on many factors,
including the distance that materials are shipped to be processed or landfilled.
But this upfront cost should not be the only measure of the quality of a program! According to the
EPA and other assessments that consider the entire life cycle of materials, the main environmental
benefits of recycling stem from offsetting the use of virgin materials. Since these benefits outweigh
the costs associated with collecting and transporting recycled material, recycling is a net benefit for all
materials for which an adequate market for recycled scrap currently exists. In other words, our current
recycling efforts are not counterproductive in terms of their environmental impacts.
Figure 14: Comparative energy usage for recycled vs. virgin content products. Reprinted with permission from [228].
The overall benefit varies by material. Making aluminum from scrap cans uses less than 5% of the
energy required to make aluminum from bauxite ore (in other words, it represents a 95% saving). At
the other end of the spectrum, making glass bottles from scrap results in an energy saving of only 20%
over using virgin materials. Even for glass, however, the energy costs of collecting and transporting
recycled glass are outweighed by this benefit. Table 7 from the EPA [15] shows these benefits in terms
of greenhouse gas emissions, measured in millions of tons of CO2 equivalent10 .
Plastics
Plastics have become ubiquitous in the products we buy and use. They are useful, versatile, and
inexpensive to manufacture [24]. Unfortunately, over 10 million tonnes of plastic trash enters the
world’s oceans each year. Once in the ocean environment, this plastic slowly breaks down into tiny
fragments that can be ingested by small marine invertebrates, causing physical or chemical harm along
the entire food chain of ocean ecosystems. Physical entrapment is another significant source of harm
to ocean dwellers.
As we noted in a “think piece” in the first chapter (page 17), the U.S. accounts for a very small
proportion of the plastic entering the oceans, largely because of an energetic and effective waste
management culture. In many developing countries, such a culture is lacking, and plastic waste readily
reaches the ocean. Without improvements in waste management around the world, the amount of
plastic entering the ocean each year is expected to rise as the use of plastics continues increasing,
possibly resulting in as much as a 10-fold increase by the year 2025 [170].
9 A possible exception is yard waste, which does not require sorting, and may require no more energy than regular trash
pickup. Since most municipalities need to pay private firms to accept waste at the landfill, it is usually more cost effective for
municipalities to collect yard waste separately for composting.
10 “Equivalent” figures of this sort are used because methane and other greenhouse gases are emitted that have more powerful
heat-trapping effects than CO2 .
Table 7: Greenhouse Gas Benefits Associated with Recovery of Materials (2014)
Material Weight Recovered GHG Benefits (millions

(millions of tons) of tons of CO2 equiva-
lent)
Paper and paperboard 43 149
Glass 3.2 1
Steel 5.8 9.5
Aluminum 0.7 6.4
Other nonferrous metals 1.37 5.9
Plastics 3 3.6
Rubber and leather 1.24 0.6
Textiles 2.3 5.8
Wood 2.47 3.8
Food 1.84 1.7
Yard trimmings 20.6 1.04
One of the largest uses of plastic is in packaging; there may be opportunities to lessen the
environmental impact of plastic packaging by switching to plastics that can safely break down in the
environment. An innovative edible six-pack ring has been developed by Saltwater Brewery in Delray,
Florida. The six-pack rings are manufactured out of barley waste from the brewing process [281, 45].
The firm Ecovative Design manufactures biomaterial substitutes for protective plastic packaging and
thermal insulation that are grown from mycelium (mushrooms) [14]. There now exist bioplastics made
from plant-based materials that are biodegradable, although it’s not clear that these plastics will safely
and quickly degrade in marine environments. There are also no current standards for biodegradable
plastics—it is an area where there may be substantial potential, but it should also be viewed cautiously
as “still under development” [303].
Composting
What ends up in U.S. landfills? Figure 15 shows the makeup of
what is discarded after accounting for recycling and composting. 8.1%
At 21%, food waste is the largest share of the waste stream. Less Wood
21.6%
than 2% of the recyclable materials recovered from municipal Food
waste are food waste suitable for composting. Composting food 18.5%
waste therefore presents a significant opportunity to move toward Plastics
a more “closed loop” society. 10.8%
Composting food waste reduces greenhouse gas effects by Textiles 5.2% Glass
avoiding methane and nitrous oxide production in the anaero- 4.2%
bic (that is, low-oxygen) conditions of the landfill. These are Other 9.4%
14.3% Metals
much more powerful greenhouse gases than carbon dioxide (Sec- Paper & 7.9%
Paperboard Yard
tion 4.1.4), and therefore avoiding them is a significant benefit. Waste
Composting does produce carbon dioxide emissions, but overall
this is a carbon neutral process: carbon dioxide is absorbed from
Figure 15: Materials discarded in municipal solid
the atmosphere when food crops are grown, and then re-released waste (percentage by weight), 2014. Adapted
when food and agricultural waste decays. from [15].
In terms of CO2 equivalent, each ton of food composted has a
benefit of about 1 ton of avoided CO2 emissions (see Table 7).
Composting also returns otherwise wasted food resources to the
agricultural cycle, reducing other inputs [56]. Municipalities, as well as other large organizations like
colleges, can gain significant environmental benefits from a composting program.
Summary
Over the past 40 years, the percentage of waste that is either recycled or composted has steadily
increased. This has involved changes in citizen behavior, municipalities and waste management
companies rethinking and redesigning waste collection, manufacturers beginning to consider product
end-of-life in the design process, the development of new technologies to efficiently sort and
process recycled waste, and the development of markets for recycled scrap. Some materials have
well-developed markets with high rates of participation and are true examples of closed-loop
manufacturing, but in most cases, there are more improvements to be made. Moving toward a zero-
waste or circular economy saves valuable resources and is beneficial for the environment, but that will
require continued innovation in product design and in handling and processing materials at product
end-of-life.
7.4. WORLD POPULATION GROWTH AND THE DEMOGRAPHIC TRANSITION 433
7.4 World Population Growth and the Demographic Transition
• This case study is an advocacy piece (see page 415). It argues that consumption
patterns in the developed nations are a greater threat to climate stability than
overpopulation driven by population growth in developing nations.
• It uses ideas from Sections 1.1 (percentages), 1.4 (graphical presentation), 2.1 (stock-
flow models and dynamic equilibrium), and 3.4 (exponential growth and the doubling
time approximation).
Overpopulation is often among the first concerns brought up in discussions about sustainability.
At the time of this writing, there are some 7.5 billion people living on Earth and the population is
expanding at a rate of one billion people every 12–13 years. However, this rate is not fixed; it has been
declining over time as populations throughout the world have followed a predictable pattern known
as the demographic transition. In this case study we’ll take a closer look at this pattern and what
it means for future population growth. We’ll argue that those of us living in the United States and
other developed nations have reason to be concerned by our own disproportionate impact on climate
stability and the environment. We can and should focus our efforts on bringing these impacts more
in line with our share of the global population while continuing to support efforts that have increased
the well-being and security of those living in developing nations—efforts that appear to be leading
to smaller family sizes by choice, lowering the risks of overpopulation, and marking the start of a
transition toward a global population in dynamic equilibrium.
Will There Be Enough Food?

Modern concern about overpopulation dates back to Thomas Malthus’s 1798 treatise, An Essay on
the Principle of Population. Malthus theorized that population growth would inevitably outpace any
growth in food production, consigning the bulk of humanity to a perpetual state of privation [206]:
...when subsistence is scarce in proportion to the number of people, it is of little
consequence whether the lowest members of the society possess eighteen pence or five
shillings. They must at all events be reduced to live upon the hardest fare and in the
smallest quantity.
In Malthus’s view, this state of affairs served as the only reliable means of curbing population growth,
and he feared that attempts to alleviate the plight of the poor would only temporarily remove this
check on population, worsening the problem in the long run:
To remedy the frequent distresses of the common people, the poor laws of England have
been instituted; but it is to be feared, that though they may have alleviated a little the
intensity of individual misfortune, they have spread the general evil over a much larger
surface.
What Malthus observed in the England of the 1790s is what we now refer to as stage one of the
demographic transition. It’s a world that few of us in the developed nations would recognize today.
Population growth was minimal, with high birth rates matched by high rates of death; food shortages
were a reality for many. A typical mother might have given birth to 5–6 children throughout her life,
with one or more of those children dying before reaching age five.
In the early nineteenth century, childhood mortality rates ranged from 20% to 35% throughout
Europe; globally the childhood mortality rate is estimated to have been higher than 40% [277]. That
is to say, out of every 1000 children born, more than 400 did not live beyond age 5. Malthus’s
perspective on the relationship between food and population was accurate for eighteenth century
England and indeed for most of human history [63]. But Malthus did not fully grasp the changes
that were unfolding; with the start of the industrial revolution, stage two of the demographic transition
was just getting under way.
40
(per 1,000 people per year)
Birth Rate
Birth and death rates
30 Natural Increase
20 Death Rate
10
0
Stage 1 Stage 2 Stage 3 Stage 4
Total Population
Stage 1 Stage 2 Stage 3 Stage 4

Figure 16: Four stages of the demographic transition. Adapted from [279].
The demographic transition refers to changes in population growth that accompany the shift from
a preindustrial to an industrial society. See Figure 16. Prior to the transition, the population growth rate
is low because birth and death rates are both very high. In the post-transition, the population growth
rate is low because birth and death rates are both low. As the transition progresses, mortality rates
fall first, leading to a surge in population growth. Birth rates eventually follow until the population
levels off or possibly even begins a decline. As development has proceeded, populations throughout
the world have followed this pattern with remarkable consistency [76].
The Population Bomb

By 1970, most of the developed nations had passed through the demographic transition and were
experiencing low birth and death rates and slowing rates of population growth. But concern was
growing about the potential threat of overpopulation in developing nations. The biologist Paul Ehrlich
drew popular attention to the issue with the publication of The Population Bomb in 1968, pressing the
need for population control [101]:
. . . many lives could be saved through dramatic programs to “stretch” the carrying
capacity of the earth by increasing food production. But these programs will only provide
a stay of execution unless they are accompanied by determined and successful efforts at
population control.
In an article published in Science the same year, Garrett Hardin cast the problem of overpopulation as
a tragedy of the commons [153]:
there is no prosperous population in the world today that has, and has had for some time,
a growth rate of zero.
Ehrlich foretold of the possibility of mass starvation. Was this concern about overpopulation
overblown? Let’s take a closer look at what the world population looked like at the time; doing so
puts this concern fully in perspective. In the early 1960s the global population growth rate was 2.1%
per year. That may not seem excessive, but it was the highest rate of population growth the world had
ever known. Recall the doubling time approximation rule for an exponentially growing stock from
Section 3.4.4:
70%
≈ 33 years.
2.1% yr−1
The global population in 1970 was 3.6 billion people. With a growth rate of 2.1%, the population
would double in 33 years, reaching 7.2 billion by 2003; had that rate continued, the world population
would stand at over 9.5 billion today and would reach almost 30 billion people by the year 2070:
3.6 billion people × (1.021)100 ≈ 29 billion people.
Furthermore, the global population growth rate itself had been steadily rising [279]. If this growth
rate had continued its upward trend, these estimates would be exceedingly low. Even if we could
expect developing nations to pass through a demographic transition, this process had taken more
than 150 years to occur in the developed countries. The Kissinger Report, a classified report by the
U.S. National Security Council provided to President Ford in 1974 [83] (declassified in 1989), serves
as an indication of how pressing the problem of overpopulation was considered to be. The report
analyzed the implications of world population growth in terms of the risk of malnutrition and famine
in poorer regions of the world and its bearing on U.S. national security interests, strongly advocating
for programs that would lead to lower fertility rates in lesser developed countries (through voluntary
means). Global population projections out to the year 2075 in the report range from a low of 8.5
billion to a high of 103 billion people, reflecting a high level of uncertainty about the world’s future
population.
20
Child Deaths per year (millions)
15
10
0
70
80
90
60
65
00
75
10
85
95
05
15
19
19
19
19
19
20
19
20
19
19
20
20
year
Figure 17: Number of deaths per year of children under age five, 1960–2015. Data from [277].
Fortunately, just as this concern about overpopulation was mounting, many developing nations were
entering stage three of the demographic transition. The population growth rate has steadily fallen from
its peak of 2.1% seen in 1961. As childhood mortality rates have dropped, birth rates have tended to
follow. We’ve also seen most developing nations pass through a demographic transition at a greatly
accelerated pace relative to Europe and North America. All of this should not be taken to mean that
concern about overpopulation in the 1960s and 1970s was necessarily overblown; that concern led
to policies and programs that have played a role in facilitating demographic transitions in regions
throughout the world..
The Fertility Rate
So far we’ve been expressing population growth as a percentage increase per year. There’s another
way of viewing population that is directly related to family size: the (total) fertility rate. This is the
number of children born per woman (on average) throughout a woman’s life (or more specifically
through a woman’s childbearing years). The replacement rate is the fertility rate that would exactly
replace each generation. If the population were in a dynamic equilibrium, the replacement rate is the
fertility rate that would keep it there. This replacement rate is estimated to be 2.1 children per woman.
We discuss the global fertility rate below; here we want to highlight the striking changes in fertility
rates by nation. See Figures 18 and 19 on the opposite page. In 1960, the fertility rate was above 3
children per woman almost everywhere outside of Japan, Europe, and the former Soviet Union and
above 6 children per woman in many countries. In contrast, as of 2015 the fertility rate has dropped
below 3 in most regions and below replacement level in most developed countries. Niger stands out
with a fertility rate of 7.3 children per woman, the highest in the world. Niger also highlights the
importance and effectiveness of family planning programs in driving the transition to lower fertility
rates. In many African nations, modern contraceptive methods are used by more than 50% of women
of childbearing age. In Niger, this figure is less than 15% [1].
Fertility Rate
0–1
1–2
2–3
3–4
4–5
5–6
6–7
7–8
8+
Figure 18: Fertility rates by nation, 1960. Data from [278].
While there are many factors that have contributed to lowering fertility rates around the world,
research points to women’s access to education as perhaps the most important. With a higher level
of education comes an increase in opportunities; women with more education may be more likely
to choose to start families at a later age or choose to raise fewer children in order to pursue these
opportunities. Education also reinforces other factors that lead to lower fertility rates including better
outcomes in child health and more knowledge of contraception and family planning [278].
Fertility Rate
0–1
1–2
2–3
3–4
4–5
5–6
6–7
7–8
8+
Figure 19: Fertility rates by nation, 2015. Data from [278].
The Great Famine and China’s One-Child Policy
At the start of 2016, China’s one-child policy ended (replaced by a two-child policy). Events in China
during the latter half of the twentieth century may be taken as the embodiment of both the worry
about what overpopulation might mean for humanity and concern about repressive population control
measures.
Between 1958 and 1961, China experienced a terrible famine; estimates of the death toll range from
20 to 40 million people. While drought and adverse weather played a role, the famine is regarded as a
consequence of disastrous economic and governmental policy implemented as part of the Great Leap
Forward under Mao Zedong’s government [355]. In the years preceding the great famine, the Chinese
government viewed a larger population as an economic resource. Birth control measures were banned
and population growth was actively promoted [316].
In the wake of the great famine this perspective on population shifted at a time when overpopulation
was a rising concern around the world. The fertility rate in China was above 6 children per woman in
the years following the famine, but it began a rapid decline in 1965, so rapid that it is unclear exactly
how important China’s one-child policy has been in lowering China’s population growth rate. See
Figure 20 on the next page. China’s fertility rate has been below replacement since the early 1990s,
but it had already dropped to about 2.7 children per woman by 1979, when the one-child policy
went into effect, and followed a trajectory similar to that of Thailand, whose national population
policy, adopted in 1971, is based on education and the entirely voluntary use of family planning and
contraception [174].
Population Momentum
Globally, the fertility rate is currently 2.5 children per woman, and the United Nations Population
Division projects the fertility rate to reach the replacement rate by 2070, although there is considerable
uncertainty about this [329]. Nonetheless, the global population will continue to rise for many decades
after the replacement fertility rate is reached. Why is this? The reason is due to what is called
population momentum. We’ve now looked at population change as a percentage of the total population
and from the viewpoint of children per woman. The next level of detail to consider is the age structure
of the population. Figure 22 on page 439 gives a picture of the global population by age and gender.
You can see from this figure that the largest share of the population is under 30 years of age.
7 China
Children per Woman (total fertility rate)
Thailand adopts national
population policy Thailand
6
2
China’s one-child
1 policy begins
0
0
0
5
197
198
199
200
201
195
195
196
196
197
198
199
200
201
year
Figure 20: Fertility rates in China, which instituted a one-child policy in 1979, and in Thailand, which adopted a national
population policy in 1971 based on education and voluntary family planning, 1950–2015. Data from [278].
7 Least Developed
Children per Woman (total fertility rate)
Less Developed
6 More Developed
World
5
1
70
80
90
00
50
10
55
60
65
75
85
95
05
15
19
19
19
20
19
20
19
19
19
19
19
19
20
20
year
Figure 21: Fertility rates by level of development, 1950–2015. Data from [278].
Let’s look at a highly simplified example to see how this will affect population growth in the future.
Let’s suppose that a population is divided into just four age groups, as shown in Figure 23 on the
opposite page. We make a few (somewhat unreasonable) simplifying assumptions:
1. Only women in the age group 26–50 bear children, and (after the initial distribution) they do so
with a fertility rate of 2 children per woman.
2. Every person in the first three age classes survives to the next age class, and no one in the final
age class survives beyond age 100.
Notice that in Figure 22 on the opposite page, the youngest age class is also the largest (the same is
true of the age structure of the actual global population). There are half as many people in the next
age class, meaning that the fertility rate for this “starting generation” was 4 children per women.
Global Population by Age and Gender (in millions) - 2015
100+ years 0.09 0.35

95 - 99 years ♂ 0.8 2.2 ♀
90 - 94 years 4 9
85 - 89 years 14 23
80 - 84 years 30 42
75 - 79 years 51 64
70 - 74 years 72 82
65 - 69 years 103 113
60 - 64 years 144 151
55 - 59 years 168 172
50 - 54 years 202 202
45 - 49 years 229 227
40 - 44 years 246 240
35 - 39 years 254 247
30 - 34 years 281 272
25 - 29 years 313 300
20 - 24 years 309 292
15 - 19 years 306 287
10 - 14 years 317 296
5 - 9 years 333 311
0 - 4 years 348 325
Males Females
Figure 22: Population structure of the the world population (2016). The total population in 2016 was estimated at 7.4 billion
people.
After one generation (25 years), the people in each age class have shifted to the next age class, and
the previous 25 and under population class has reproduced itself. The fertility rate is at the replacement
level, but the total population has grown from 75 million to 110 million people. This growth will
continue until the population reaches a dynamic equilibrium.
76 - 100 years 2.5 2.5

51 - 75 years ♂ 5 5 ♀
26 - 50 years 10 10
0 - 25 years 20 20
Males Females
Figure 23: Population structure in first generation. The total population is 75 million people.
76 - 100 years 5 5
51 - 75 years ♂ 10 10 ♀
26 - 50 years 20 20
0 - 25 years 20 20
Males Females
Figure 24: Population structure in the second generation. The people in the age 0–25 class in the first generation have moved
to the second age class (and reproduced themselves). The total population is 110 million people.
Increasing Agricultural Production

Modernization of agriculture has allowed increases in food production to outpace the rise in
population, but it isn’t obvious that this trend can continue. The best agricultural lands are already
in use; at some point our productivity must yield to the law of diminishing returns. Nevertheless, there
remains a considerable gap in agricultural yields between industrialized countries and developing
nations; narrowing this gap would allow considerable gains in food production, sufficient perhaps
to meet the demands of a rising population, for a time. On the other hand, a growing population
necessarily narrows the gap between what we can produce and what we must produce. The narrower
this gap, the less resilient human populations are to crop failures and natural disasters.
Concern over Declining Populations

While low fertility rates should be taken as good news for anyone concerned about overpopulation,
in many developed nations concern is cutting in the other direction. The below replacement fertility
rate in Europe has been described as a perfect demographic storm and is seen as a crisis that threatens
future prospects for economic growth [178].
Japan, with a fertility rate of 1.4 children per woman, has recently set a target of achieving a fertility
rate of 1.8 and is attempting aggressive measures to achieve it [306]. Falling fertility rates do present
social and economic challenges; when the average age of the population increases, this can strain
resources available for taking care of the elderly, and population growth is a driver of economic
growth. But instead of looking to turn around fertility in an attempt to avoid these challenges, we
will need to face them head on. There is little doubt that as fertility rates drop below replacement in
more countries, more nations will institute “family friendly” policies in an effort to return to an era of
higher birth rates. But such policies do not seem to have had much effect on fertility where they’ve
been attempted thus far. Perhaps a greater cause for concern are the rising tensions in the United States
and Europe over immigration. With falling birth rates, we can anticipate that future population growth
in developed countries will be driven primarily by immigration. Unfortunately, this is an unwelcome
prospect for many, and we will almost certainly face a rough time ahead unless people in developed
regions can adjust to accommodating a greater influx of immigrants.
Quantifying the Impact of Earth’s Human Population

We’ve demonstrated that overpopulation does not present the same concern that it did a generation
ago; we must now come back around to an idea that was stressed in the introduction: the
disproportionate impact of those living in the developed nations—where fertility rates are already
low—is a primary concern. Our aim here is to demonstrate just how disproportionate this impact
really is. But how exactly do we go about quantifying human impact on the environment and how do
we compare the impact of someone living in the U.S. with someone living in a less developed region?
One approach we could take is using what is called the ecological footprint (see Definition 3 on
page 209). This is an estimate of how much productive land is dedicated to meeting the needs of
a typical person or region. Another tack we could take is to estimate the per capita greenhouse gas
emissions for different regions. Instead, we will rely on electrical energy as a proxy for impact; the
data is reliable and easily accessible, and as we’ll see, the differences between the developed and
developing world are stark enough to make the difference in resource use clear.
According to estimates, in 2014, the United States produced 3,900 terawatt-hours of electricity,
amounting to 12,000 kilowatt-hours per person. Let’s compare this with Niger; as we’ve already
mentioned, this is one of the only places on Earth where the fertility rate remains above 7 children
per woman. In Niger, electrical energy consumption for 2014 is estimated to have been around 65
kilowatt-hours per person. This means that the lifestyle of a typical U.S. citizen demands almost 200
times the energy as that of the typical citizen of Niger (at least as measured by electrical energy—not
the only form, of course). Niger’s population, which is under 19 million people, uses about 3,000
times less electricity each year than the United States [77].
Even in the unlikely event that Niger maintained its currently high population growth of 4% per
year, it would take 75 years for Niger’s population to reach the current population of the United
States. To reach the current level of electrical energy use as the United States, Niger would need
to double its population 4 times over (a 16-fold increase) and simultaneously increase its per capita
electrical energy use by a factor of 200. One might object to this analysis on the grounds that Niger is
a smaller country (by area) than the United States. If we instead use the geographic sizes of Niger and
the United States as the basis for comparison, Niger’s energy use would still need to rise by a factor
of 400 to match the U.S.
Undoubtedly people in developing nations desire the same advantages those of us in developed
nations already enjoy. This is precisely why we must assume a leadership role in lowering our
dependence on fossil energy and in reducing our harmful impacts while maintaining a high quality of
life. We must do this if we are to have any hope of extending a higher standard of living to the rest
of the world while drawing down greenhouse gas emissions to safe levels. While we’ve emphasized
electricity, energy use in Niger is dominated by biomass burning (78% of energy use) [124]. Over 90%
of households burn wood for cooking, and electricity accounts for less than 3% of total energy used
(in the U.S. electricity accounts for about 40% of the total energy used; see Figure 18 on page 93).
The IPAT Equation

Another method for trying to quantify human impact—one that has received considerable attention—
is called the IPAT equation, after the variables involved [338]. This idea for quantifying human
impact arose out of debates in the late 1960s and early 1970s over the relative importance of
population growth on environmental impact [102, 69]. We’ve already seen that because of historically
high population growth rates during this time, overpopulation had become a pressing concern. This
equation sounds ready-made for just the sort of analysis we are interested in here. Why, then, are we
not using it?
The IPAT equation purports to quantify the environmental impact of human activity:
Environmental Impact = Population × Affuence × Technology,
or
I = P × A × T,
where affluence, A, is measured by GDP per capita, and technology, T, is measured by environmental
impact per unit of GDP. The proponents of the equation do not always specify very precisely in what
units “environmental impact” is to be measured, but so long as the same units (whatever they are)
are used on the left and the right, this does not matter. The equation is what mathematicians call a
tautology: it is true simply because of the properties of the symbols involved, like the associative law
of multiplication,
(a × b) × c = a × (b × c),
which is true whatever numbers are represented by the symbols a, b, and c.
Why is this a concern? Because a tautology—something that is “true by definition”—can have no
scientific content: a scientific statement must be open to the possibility of being wrong (Section 2.1.4).
The IPAT equation turns the idea of using mathematics to understand our environmental impact
completely on its head. The desired result (rising populations are a concern) is assumed up front
and then couched in mathematical terms that attempt to make it self-evident. This, we would argue,
is a misuse of mathematics. If we are concerned about the environmental impacts that result from our
activities, then we really need to understand these impacts. Such understanding cannot come from a
simple tautology. It requires mathematical and scientific models that are testable (and a tautology is
never testable, because it can never be wrong).
Take climate change as an example. If we are concerned about carbon emissions, then we should
understand the carbon intensity of the various fuels we use (see Table 5 on page 488 for fossil fuels).
We can use mathematics to understand the consequences of burning coal versus burning natural gas
versus using wind and solar to power our homes. An attentive reader might notice that while we
indeed compared the per capita energy use of the United States with Niger in the analysis above,
we never assigned a single number to “impact.” We know that in the United States, the majority of
our electricity is provided by burning fossil fuels (primarily coal and natural gas; see Figure 18 on
page 93). Even if the carbon intensity of electricity production is higher in Niger, we can still see from
the analysis above that the United States is responsible for far greater carbon emissions, both in total
and on a per capita basis (burning biomass is treated as carbon neutral). But the quantity of carbon
emissions marks just one of many dimensions along which we might quantify impact. Some industrial
pollutants have a global reach and can have a negative impact on far more than the people who receive
the benefits from the activity that produced the pollution. Other pollution issues relevant to Niger (for
instance, degraded indoor air quality arising from cooking over open fires) [258] are primarily local.
Addressing the sustainability question requires us to tackle many complex problems with many
different solutions that will necessarily involve efforts to develop a careful understanding on a case
by case basis. The IPAT equation, in its attempt to summarize “the big picture,” instead oversimplifies
this process while offering no real clarity or understanding on effective approaches to sustainability
questions.
Summary
In summary, the growing global population is one component of the sustainability question—a
question with many different linkages: climate change, resource scarcity, energy production, and
water resources, among others. If we are able to meet the technical challenges of shifting to low
carbon energy sources, and if we can extend modern agricultural achievements to developing nations,
then overpopulation itself will prove to be a lesser concern. We must continue to make progress on
improving access to those resources that have been demonstrated to have an effect on fertility rates—
namely women’s access to education, adequate health care, and family planning resources, but we can
be thankful that history has demonstrated that we can expect the global population to peak without
recourse to coercive population control measures. As families feel more security over the well-being of
their children and are given opportunities beyond subsistence-level agriculture, many can be counted
on to make a choice for smaller family sizes.
7.5. GENETIC ENGINEERING AND THE FUTURE OF FOOD 443
7.5 Genetic Engineering and the Future of Food
• This case study is an advocacy piece (see page 415). It argues that evidence favors
the use of genetically modified (GM) crops in food production, as long as a robust
and transparent regulatory framework is in place.
• It uses ideas from Sections 1.1 (percentages), 1.4 (graphical presentation and how it
can mislead), 4.1 (absolute versus relative change), 5.2 (random mutations) and 6.3
(regulatory capture).
Over the past two decades, farmers in the United States and Canada growing corn, soy, rapeseed
(canola), and cotton have increasingly turned toward the use of genetically modified crops (GM crops),
while in Europe farmers are restricted from growing such crops. See Figure 25 on the next page.
Meanwhile, public debate over the safety and efficacy of GM crops seems mired in misrepresentation
and misperception of the risks and benefits. In October 2016, the New York Times published an article
that cast doubt on the value of GM crops [148].
[A]n extensive examination by The New York Times indicates ... genetic modification in
the United States and Canada has not accelerated increases in crop yields or led to an
overall reduction in the use of chemical pesticides.
In this case study we’ll take a critical look at this piece from the New York Times and examine the use
of GM crops in agriculture. We expect genetically modified foods to play an increasing role in food
production, and we’ll argue the case that research and development in genetic engineering should
certainly continue, accompanied by a robust and transparent regulatory process for approving new
genetically engineered crop varieties.
What Is a Genetically Modified Crop?
In traditional crop development, breeders cross-breed existing crop varieties and then select from
among the resulting offspring for desirable traits. This process, called selective breeding, can continue
for many generations; in fact, we’ve been systematically altering food crops in this way for 11,000
years. Sometimes this process yields offspring that have undesirable traits alongside, or instead of,
the desirable ones; such offspring are then not selected for production. An interesting example is the
Lenape potato developed at Penn State in the late 1960s. This potato was selected for use in potato
chips and indeed was planted commercially before it was discovered that it contained unsafe levels
of naturally occurring toxins (glycoalkaloids). These toxins also exist in the potatoes we enjoy today,
but at low enough levels that they are not a concern for human health [70]. (In fact, many potato
chips today are made from descendants of the Lenape variety: further selective breeding was able to
preserve the positive qualities of this variety while reducing the glycoalkaloid content to safe levels.)
This selection process can be greatly accelerated through the modern technique of mutation
breeding, whereby random mutations are introduced to crop varieties by exposure to DNA-altering
chemicals or radiation. This provides the selective breeder with a much wider range of “offspring”
to work with, and much more quickly, than traditional cross-breeding. However, it is similar to the
traditional process in that there is no control in advance of the range of “offspring” obtained. The
breeder simply has to work with what chance gives them.
All of these breeding processes produce

new varieties with different genomes from
those of their parents (the genome is the
“program” written in DNA at the heart of
every living cell, which tells it how to act,
reproduce, and respond to its environment). In
fact, the genome “is” the crop variety: “differ-
ent variety” (on the macro-scale) is the same
is “different genome” (on the micro-scale).
The big difference with genetic modification
(also known as genetic engineering) is that it
relies on our newly acquired understanding
of the genetic code to allow for intentional
editing of the genome of existing crops to
produce new varieties. For example, a genetic-
modification technologist can deliberately add
some desirable property (say, resistance to
some kind of pest) to an existing crop vari-
Figure 25: Percentage of planted acres in the U.S. for GE crops of different ety by “splicing” new genes into its genome.
types. Bt crops contain an insecticide. HT crops are herbicide resistant. Though the method is completely different
These traits can be “stacked” in the same plant [248].
and much more technically complex, the re-
sult is the same as the result of conventional breeding: a new genome, a new variety.
It’s important to emphasize that there is no evidence showing that GM crops pose a general risk
to human health. GM crops that are currently commercially available meet the same safety standards
we apply to conventional foods; new varieties should continue to be held to the same standard. As
happens in conventional plant breeding, experimentation may sometimes produce varieties that pose
unintended risks (such as the Lenape potato, mentioned above), but the purpose of a well-developed
regulatory process is to guard against these risks and ensure that dangerous food products do not make
it to market. A detailed review of the human health implications of GM foods is given in the National
Academy of Science’s recent report, Genetically Engineered Crops: Experiences and Prospects [233,
Ch. 5]. We’ll refer to this report again in sections below. The important point here is that genetic
engineering does not pose risks we don’t already encounter from conventional foods. In fact, we cite
below research that suggests that genetic engineering in some cases may present an opportunity to
lower these risks.
Western Europe
Canada
3,000
Yield (Pounds per acre)
2,000
1,000
0
84
88
92
94
98
00
04
86
08
10
12
14
96
06
9
0
19
19
19
19
19
19
20
20
20
19
20
20
20
20
19
20
year
Figure 26: Comparison of rapeseed (canola) yields in Canada and Western Europe. Data from FAO [109].
A Misleading Comparison
Let’s now take a look at the New York Times article mentioned in the introduction. Some parts of
its analysis involve misleading comparisons. We’ll take a closer look at just one of these. Look at
Figure 26, which shows the trend in yields for rapeseed production11 in both Western Europe, where
GMOs are banned, and Canada, where GMOs now comprise over 95% of what is grown. We’ve
plotted annual data from the Food and Agriculture Organization of the UN [109] and added trend lines
(the black dashed lines in the figure); a similar figure appears in the New York Times [148]. The trends
appear to be similar, but the rapeseed yields in Western Europe are much higher than those in Canada.
The reason for this difference is that Western Europe has a milder climate and grows a winter variety
of rapeseed with a long growing season. The variety grown in Canada (canola) is a summer variety,
suitable to Canada’s shorter growing season. Thus it is inappropriate to say without qualification
that “Western Europe achieves higher per-hectare yields without GMO crops than Canada does with
them”; that is not a comparison of like with like. If we are willing to look a little more closely into
the data, we’ll see something else interesting. The trend for Western Europe shows an increase from
about 2,500 to about 3,300 pounds per acre over the time frame shown, representing an increase of
about 30%. The data for Canada, on the other hand, show an increase from roughly 900 to 1,800
pounds per acre. This is a 100% increase in yield! The almost-parallel trend lines conceal the fact
that in percentage terms, the Canadian yield increase is far more impressive. (See Example 2 on
page 186 and the surrounding discussion for why we focus on relative—that is, percentage—rather
than absolute comparisons here.)
An additional factor that it’s important to consider is the total land area dedicated to rapeseed
production in Europe and in Canada. Looking first at Europe, one finds that the land area dedicated to
rapeseed has varied little over the period shown in Figure 26, and since 2014 it has actually declined,
giving way to increased soy production.12 In contrast, the total land area in canola production in
Canada has grown significantly: from about 12 million acres in the year 2000 to over 20 million acres
in 2015.
Why might this detail matter? Because when a farm (or a nation) is expanding the acreage devoted
to a particular crop, it will always harvest the most productive land first: newly added land will tend to
be less productive (poorer soil, worse drainage, and so on) than that which was already under crops.
It’s helpful to consider a hypothetical example. Imagine two farms that each have 1000 acres that
could be in production.
On Farm A, 500 acres produces 600 tons of canola each year, while the remaining 500 acres
produces only 400 tons each year. Averaged over the whole farm, the yield is 1 ton per acre. Farm A
makes a change, moving the less productive half from canola into soy. Even with no change in the
productivity of the land planted in canola, the yield per acre for canola rises from 1 ton to 1.2 tons per
acre.
Farm B is similar to Farm A, but on farm B only the most productive 500 acres are in production,
producing 600 tons each year. The average yield is 1.2 tons/acre. With existing practices, the output
of the remaining 500 acres would be an additional 400 tons per year, just as on Farm A. See Figure 27
on the next page.
Now suppose, though, that Farm B decides to put a new, higher-yielding variety of canola into
production, and expand to use 1000 acres. The yield on both halves of the farm rises. The yield on
the more productive part of the farm rises to 700 tons per year, while the remaining half of the farm is
yielding 500 tons per year (up from the unused potential for 400 tons prior to the change). No one can
doubt that the new variety is making a big difference. However, the overall yield per acre remains the
same as before, 1.2 tons per acre; total production has exactly doubled to 1200 tons, even though the
yield on both halves of the farm has risen.
11 The FAO reports data on yields in hectograms per hectare. A hectogram is one-tenth of a kilogram; a hectare is 10,000
square meters. We’ve converted this to the conventional units used in the U.S.: pounds per acre.
12 In recent years, Europe has started importing rapeseed products from Canada [280].
500 acres 500 acres 500 acres 500 acres

600 tons/yr 600 tons/yr 600 tons/yr 700 tons/yr
500 acres 500 acres 500 acres 500 acres

400 tons/yr no prod. no prod. 500 tons/yr
1 ton/acre 1.2 ton/acre 1.2 ton/acre 1.2 ton/acre

(a) Farm A. (b) Farm B.
Figure 27: Hypothetical changes in yields based on changes in the amount and quality of land in production.
In short, if the least productive land goes out of production and no other changes are made, then the
average yield will automatically go up. Conversely, if less productive land is brought into production
and average yields do not go down, that already shows that the quality of the varieties planted must
be improving. This hypothetical example doesn’t prove that GM rapeseed has been a boon, but it
demonstrates another reason to treat the claims made about data in Figure 26 on page 444 with
skepticism. In fact, additional research into the use of GM canola in Canada has shown that it has
improved economic outcomes (hence its increase in production) and also led to a reduction in the use
of herbicides [301].
Genetically Modified Crops in Use Today

Consistent with the experience of Canadian growers using GM rapeseed, the National Academy of
Sciences Committee on Genetically Engineered Crops recently concluded that the use of genetically
engineered crops has been beneficial for growers [233]. As of 2015, the engineered traits in
commercially available seeds are primarily plants that produce a bacterial insecticide (Bt) and plants
that are resistant to herbicides, especially glyphosate (commercially known as Roundup). It should be
noted that Bt is an insecticide approved for use in organic agriculture and is not harmful to humans.
Its use as a GM trait in crops has generally lowered losses to insect pests and has also coincided with
a significant decrease in the use of insecticides. Interestingly, in some cases these Bt crops have had
spillover effects, benefiting conventional crops by lowering insect pest pressure in neighboring areas.
The reduced use of sprayed insecticides has also resulted in increased insect biodiversity on farms
using Bt crops.
Although the yield increases attributable to GM crops have in most cases been modest, we should
keep in mind that continuous improvements in agriculture prior to the advent of genetic engineering
had already produced dramatic gains in agricultural yields. See Figure 28 on the opposite page. The
yield per acre cannot be expected to increase indefinitely. What’s more, increased yield represents just
one dimension along which to measure the impact of genetic modification. We’ve already mentioned
another: reduced use of insecticides. Several more benefits are listed in the section “Future Prospects”
below. Many of these are already in development, while others are more long-term bets.
“Big Ag”
Much of the debate on GM crops has centered on the role of large agribusinesses, Monsanto in
particular. The experiences of past battles over environmental and public health crises have given
the public ample reason to be skeptical of corporate motives. Business interests have cast doubt on the
Figure 28: Corn yield in bushels per acre [236]. Reprinted with permission.
science on the effects of lead exposure from leaded gasoline and the impacts of tobacco, and delayed
important health regulations in these areas by decades [50, 180]. More recently, we’ve learned that
the same has been true for oil companies and the science of climate change [99]. So what about the
management of genetically modified crops? In the U.S., agencies tasked with regulating agriculture—
the USDA, FDA, and EPA—need to establish and maintain trust in their independence through open
and transparent regulatory processes at a time when public faith in government has been in short
supply. The risk of “regulatory capture” (that is, the subversion of regulatory agencies by the very
interests they are supposed to be regulating, see Remark 1 on page 383) is real, and only openness
to public scrutiny can guard against it. For their part, agribusinesses need to anticipate this public
skepticism and actively support these open regulatory processes, which are in their long-term interest.
Superweeds
A well-founded concern relates to crops engineered to be resistant to glyphosate (“Roundup ready”
crops). The point of this engineering, of course, is to allow the use of Roundup on growing crops
at higher concentrations, which will then more effectively kill undesired plant species in the field.
Over time, however, the weeds themselves may become resistant to the herbicide through selective
pressure; any weeds that manage to survive an herbicide application confer that genetic advantage to
their offspring. This mechanism is not unique to GM crops; it can also lead to herbicide resistance in
conventional crop systems. See Figure 29 on the next page. Monsanto here has some culpability in
the rise of glyphosate-resistant crops, having relied on flawed methodology studying these risks and
claiming that glyphosate resistance was unlikely to become a problem [129].
It is also possible for genes to be transferred from one plant species to another through cross-
pollination; genetically engineered genes can pass in this way from GM crops to their wild relatives.
The general thinking is that the risk of GM traits persisting in the wild is low, but in at least one
case, researchers demonstrated that a “Roundup Ready” gene in rice could be passed to a wild and
weedy cousin, leading directly to the development of an herbicide-resistant “superweed” [268]. Both
Figure 29: Herbicide-resistant weeds [129]. Reprinted with permission.
herbicide-resistant weeds and gene transfer are thoroughly discussed in the National Academy of
Sciences review [233, Ch. 4]. Addressing these problems will ultimately require the use of integrated
pest management strategies. This may be an area in which tools that are typically used in organic
agriculture, including crop rotations and multicropping systems, may serve conventional agriculture
as well [266].
Future Prospects
To date, GMOs are dominant in only a few commercially available crops, and the main applications
have been insect resistance (Bt toxin made by the plant) and herbicide resistance. Another early
example is resistance to the crop-damaging ringspot virus in papaya. But the techniques used in
genetic engineering are undergoing rapid change; this is still very much an emerging technology.
A position paper from the Task Force on Global Food Security of the International Society for Plant
Pathology gives an overview of additional crop varieties that may soon be available with similar
insecticide- and herbicide-resistant traits. They also outline a number of interesting applications in
development including drought and disease resistance, salt tolerance, and nitrogen efficiency, among
others [291]. Other applications that have been proposed or are in development include enhancing
adaptation to increased levels of carbon dioxide in the atmosphere [186], gene regulation of fungi
that cause crop losses through fungal infections [320], varieties of rice that can survive prolonged
flooding events [274], enhanced production of pharmaceuticals [297], and crops that address vitamin
and nutritional deficiencies [66].
Summary
In the coming years we are very likely to see the increased use of genetically modified crops. Like
conventional crop breeding, this process will probably both enhance agricultural productivity and
expand the variety of foods that are available to us. We should support the research that will improve
our understanding and use of genetic engineering and advocate for a robust and transparent regulatory
process that the public can have confidence in. Let us do more than just hope that this process hastens
a transition to a sustainable agriculture—let’s actively advocate for it and help it achieve that goal!
Resources and Further Reading

There are additional issues related to genetic engineering and genetically modified organisms that we
have not addressed in this case study. We provide here additional resources for exploring this topic.
• Doubts About the Promised Bounty of Genetically Modified Crops, New York Times [148].
• Genetically Modified Crops: Experiences and Prospects, National Academy of Sciences [233].
• Signal to Noise Special Edition: GMOs and Our Food, Science in the News, Harvard Medical
School [31].
• Genome Editing, An Ethical Review, Nuffield Council on Bioethics [33].
• EU Policy for sustainable use of pesticides [67].
• Intellectual Property and Genomics, National Institutes of Health [250].
• FAOSTAT, Data on Pesticide Use, Food and Agricultural Organization of the United Na-
tions [109].
7.6 Nuclear Power Is a Commitment to the Future
• This case study is an advocacy piece. It examines the prospects and risks of nuclear
power as a source of reliable energy that does not contribute to global warming.
It concludes that nuclear power makes sense—but only under certain conditions,
conditions that test our faith in humanity’s future.
• It uses ideas from Section 2.2 (Energy Stocks and Flows), Section 4.2 (Limits to
Growth), Section 5.3 (Expectations and Payoffs), Section 5.4 (Assimilating New
Information), and Section 6.4 (Decision-Making and Ethics).
Too Cheap to Meter?

The nuclear age began in war. Thousands of scientists had worked for years on secret bases at Los
Alamos, New Mexico, and elsewhere. On July 16, 1945, they detonated the Earth’s first nuclear
explosion in the New Mexico desert. This was followed by nuclear attacks on the Japanese cities
of Hiroshima (August 6) and Nagasaki (August 9), bringing the Second World War to an end at a cost
of 130,000–200,000 deaths, mostly of civilians. As we have seen, the geological signature of these
events and subsequent nuclear tests—an unmistakable blend of artificial isotopes—has been taken to
mark the boundary of the new epoch of the Anthropocene (see Section 6.4.1).
However much they might have accepted its necessity, this outcome was not what many nuclear
scientists had hoped for. Could not this hugely powerful energy source be developed for peaceful
rather than warlike purposes? That was the theme of President Eisenhower’s “Atoms for Peace” speech
(December 8, 1953) [105] and of subsequent remarks by Lewis Strauss, chairman of the Atomic
Energy Commission, to the National Association of Science Writers on September 16, 1954 [312].
Toward the end of his speech Strauss summed up his hopes:
Transmutation of the elements, unlimited power, ability to investigate the working of

living cells by tracer atoms, the secret of photosynthesis about to be uncovered, these
and a host of other results all in 15 short years. It is not too much to expect that our
children will enjoy in their homes electrical energy too cheap to meter, will know of great
periodic regional famines in the world only as matters of history, will travel effortlessly
over the seas and under them and through the air with a minimum of danger and at great
speeds, and will experience a lifespan far longer than ours, as disease yields and man
comes to understand what causes him to age. This is the forecast for an age of peace.
The context makes it clear that Strauss’s words “too cheap to meter” were part of a long-term vision,
not a prediction for the immediate future. Nevertheless, the hope expressed by Strauss remains active
today. Can nuclear power provide sustainable, long-term, inexpensive energy—energy that, by the
way, produces no CO2 emissions and therefore does not contribute to global warming? In this case
study we’ll examine this question in the context of global energy needs in the second decade of the
twenty-first century.
Nuclear Energy
As explained in Section 2.2, nuclear energy is energy that is stored in the bonds between the particles
of the atomic nucleus. These bonds are enormously stronger than the chemical bonds that unite
7.6. NUCLEAR POWER IS A COMMITMENT TO THE FUTURE 451
Xe–142
U–235
Sr–90
Figure 30: Schematic diagram of the fission process. At left, a neutron (blue) strikes a fissile nucleus (magenta). The results
of fission are shown at the right: two fission product nuclei of slightly different sizes (orange and yellow) and several more
neutrons (blue).
atoms into chemical compounds, and therefore the energies associated with manipulating them are
enormously greater.
In order to release the energy in nuclear bonds, though, one must somehow carry out nuclear
reactions (that is, reactions that transform one kind of nucleus into another), in the same way that
in order to release chemical energy one must carry out chemical reactions (like burning natural gas
or another fossil fuel). Nuclear reactions can release energy because the energy per nuclear particle13
incorporated in the bonds that stick them together is least for elements with medium-sized nuclei
(round about 56 nucleons), such as iron and nickel. For an element with a much bigger nucleus than
this, energy could be released by breaking the nucleus apart into two pieces of roughly equal size: this
is nuclear fission, the process that powered the original atomic bomb and that also powers modern
nuclear power reactors. Energy can also be released (in fact, a great deal more energy) by combining
two nuclei of size much less than 56 nucleons to form a single, larger nucleus: this is nuclear fusion,
the process that has powered the Sun for billions of years and that has given us the enormously more
destructive H-bomb but has not, as yet, despite years of research, provided a controllable way of
generating power on Earth. Until the very end of this case study, we will talk about fission energy
only.
The key to nuclear power (or to a bomb) is to find nuclei that can participate in the so-called
fission chain reaction, called fissile nuclei for short. When a fissile nucleus is struck by a single
neutron (moving at a suitable speed) it will break apart (releasing energy) into two roughly equal-
sized pieces (fission products) together with several more neutrons. (See Figure 30.) In a mass of the
fissile material, these neutrons can then strike other nuclei, causing them to break apart and release a
second generation of neutrons, which will then strike and break apart yet more nuclei, and so on in a
process of exponential growth (Section 3.4). Do this in an uncontrolled fashion, and the whole process
is over, with a massive release of energy, in a fraction of a second: you have an atomic bomb. But the
process can also be carried out in a controlled way, interspersing small masses of fissile material with
other devices that manage and regulate the flow of neutrons. When this is done, you have a nuclear
reactor: a chain reaction that can be kept “simmering” for a long time, releasing energy at a steady
rate, until eventually the depletion of the original fissile material and the accumulation of fission
products forces a shutdown for refueling.
There is just one fissile material that occurs naturally, uranium-235 (U-235). (A uranium nucleus
has 92 protons; the number 235 refers to the total number of nucleons, that is protons and neutrons
13 Protons and neutrons, collectively known as nucleons.
together, in a single nucleus.) Two other fissile materials, plutonium-239 and uranium-233, can be
produced by artificial means. Naturally occurring uranium contains less than 1 percent of the fissile
U-235, with nearly all the remainder being the nonfissile U-238. To build a nuclear reactor, the
concentration of U-235 must be increased, or enriched, from the natural level to about 3–5%. The
resulting material is suitable for nuclear fuel but is of no use for building a bomb, which requires a
much higher level of enrichment. (This is the theoretical foundation for the hope that nuclear power
technology could be widely disseminated while still keeping a tight control of nuclear proliferation,
that is, the spread of nuclear weapons.) In commercial U.S. reactors, the enriched uranium fuel is
contained in ceramic pellets about the size of a thumbnail. Each pellet can release the same amount
of heat energy as a ton of coal. The pellets are then stacked in special alloy tubes (fuel rods) about 4
meters long. The core of an operational power reactor contains tens of thousands of these fuel rods.
What is the payoff for nuclear power? Using standard transportation methods, the fuel rod
assemblies needed to power a nuclear plant with 1 gigawatt (109 W) electrical output for a year and
a half could be delivered on seven large trucks. To generate the same amount of electricity from coal
would require a trainload of coal (90 train cars, 100 tons of coal per car) every day of operation! The
nuclear plant would release no pollution during its operation: the equivalent coal plant would release
over 4 million tons of CO2 and other pollutants and generate over 350,000 tons of ash. According
to the Environmental Protection Agency, heavy metal emissions (especially mercury) from coal-fired
power plants account for thousands of premature deaths each year in the United States—to say nothing
of the climate-changing effects of the CO2 produced by such plants.
Despite the obvious advantages of nuclear power, many people have reasonable questions about it.
Here are some of the most common.
• Is nuclear power sustainable?
• Can we manage radioactive waste safely?
• What about the danger of nuclear accidents, like the one at Fukushima in 2011?
• Will greater use of nuclear power inevitably lead to the proliferation of nuclear weapons?
• What about fusion?
In the following subsections we’ll try to answer some of these questions.
Sustainability
The straightforward answer to the question “is nuclear power sustainable?” is, of course, “no.” The
technology is much more exotic, but in the end, the uranium fuel for nuclear power plants has to
be mined just as coal is mined, and nuclear energy generation, just like coal burning, takes the fuel
on a one-way journey from resource to waste (see page 80). The question is, how much uranium is
available and for how long will it power a nuclear program?
In a lengthy report [241], two international agencies evaluated this question and concluded that
present-day known reserves are sufficient for 150–200 years at present rates of consumption. This is
a static reserve index in the language of Definition 4 on page 200. But this is a situation in which
(even without global growth assumptions) a static reserve index may well be highly inappropriate.
One of the key reasons to invest in nuclear energy is that it could provide a stable source of power
that does not involve burning fossil fuels and therefore generating greenhouse gas emissions. How
much investment would it take to replace fossil fuels by nuclear power? Take a look at Figure 18 on
page 93, which shows energy flows in the United States. You will see that nuclear power provides
less that one-fourth of U.S. electricity generation, with fossil fuels supplying the great majority of the
remaining three-fourths (not to mention nearly all the fuel in the transportation sector, which rivals
electricity generation in size). To significantly reduce the use of fossil fuels we would not be looking
at “present rates of consumption,” but at increasing the overall use of nuclear energy by a factor of
four or more. If this should happen worldwide, those 150–200 years of reserves would start to look
like less than 50 years. Very definitely not long enough to be called “sustainable.”
Nuclear technologists can offer a potential solution to this problem in the form of the breeder
reactor. As mentioned above, the great majority of naturally occurring uranium is not fissile U-
235 but nonfissile U-238. When a U-238 nucleus is struck by a neutron it does not split apart, but
something interesting does happen nevertheless. The U-238 absorbs the neutron, becoming U-239;
and U-239 naturally undergoes radioactive transmutations that end up with the fissile nucleus Pu-239
(plutonium). Thus a reactor can “breed” fissile plutonium out of the originally nonfissile U-238. Using
this process systematically could, in effect, multiply our usable uranium reserves by a factor of 100 or
more, pushing the possibility of exhaustion far into the future.14
In a sense, this idea is already used today. The process we described is the way plutonium for
nuclear weapons is made. Since the 1990s, both the United States and Russia have decommissioned
many nuclear weapons. Some of the plutonium from these weapons has been used to replace U-235 in
nuclear reactor fuel (the resultant “blend” is called mixed oxide fuel, or MOX). This “unintentional”
recycling of plutonium is very different, though, from demonstrating a breeder reactor that will
function effectively and safely as part of a civilian power program. Many countries, including the
United States, that had at one time planned to build breeder reactors have now shuttered their breeder
programs as safety and proliferation concerns have multiplied and overall support for nuclear power
has declined. At the time of writing the only operational commercial-scale breeder reactor in the world
may be the BN-800 reactor at Zarechny, Sverdlovsk Oblast, in Russia.
Why have so many breeder programs failed? The reasons are partly economic: demand for nuclear
power did not grow as expected, and therefore the need for a new and more sustainable source
of nuclear fuel was not so urgent. But concerns about nuclear proliferation have also been very
important. Breeder reactors deliberately produce large quantities of plutonium-239, which is the basic
material for building nuclear weapons. A “breeder reactor economy” is one in which large quantities
of weapons-grade material are distributed across many different facilities. Do we want to do this? It
seems fair to say that a decision to do so would have to be based on a high level of confidence in
global institutions that can secure nuclear materials and protect them from rogue actors (terrorists,
hostile governments, or criminal gangs). Does such a level of confidence exist at the moment?
Nuclear Waste
After a fuel rod has been in a nuclear reactor for five years or so, much of its fissionable material has
been burned up. What’s more, some of the fission products that it has generated are neutron absorbers,
which slow down and would eventually stop the chain reaction. The fuel rod is “spent” and needs to
be removed.
Before being used in a reactor, a fuel rod is radioactive, but only mildly so. You could safely pick
up fuel pellets with your bare hands. After use, though, it is a different story. A spent fuel rod is
loaded with highly radioactive fission products and other dangerous materials that have been created
during its years in the reactor core. Arthur Motta, professor of nuclear engineering at Penn State, told
a class, “If a spent fuel rod were somehow magically to appear on the desk in front of me, we would
all instantly receive a lethal dose of radiation.”
In fact, nuclear waste is one of the most toxic substances known. Its dangers do become less over
time: strontium 90, a typical fission product, has a half-life (Definition 4 on page 174) of about 29
years, so that its radioactivity will be reduced by a factor of about 10 over a century. Nevertheless,
it is estimated to take 100,000 years or so before the radioactivity of spent fuel (from the current
generation of reactors) will decline to roughly the same level as that of unmined uranium.
14 Advocates claim that power from breeder reactors would allow for the effective extraction of the very low concentrations
of uranium in seawater. In principle this would extend the reserve lifetime to millions of years.
The radioactivity of the spent fuel rods produces

heat—a lot of heat, at first. Spent fuel is usually stored
under water (to keep it cool) at the site where it was
produced, at least until its high initial radioactivity (pro-
duced by elements with short half-lives) has subsided a
good deal. Then it is sealed in special casks, which by
now are lined up in the back lots of most U.S. nuclear
power stations. This is called interim storage. And then
what? Once the waste has “cooled off” sufficiently it is
possible, in principle, to manipulate it chemically, ex-
tracting uranium and plutonium for reuse (for example)
and separating out the less dangerous parts from the
more dangerous.15 The final destination of the waste
should be isolation and containment in a permanent
underground disposal facility.
You’ll notice we said “should be.” The amount of ma-
Figure 31: Schematic of Yucca Mountain site. 1. Canisters of
waste, sealed in special casks, are shipped to the site by truck or terial involved is quite small (all the many tons of high-
train. 2. Shipping casks are removed, and the inner tubes with level waste produced in the U.S. over the last forty years
the waste are placed in steel multilayered storage containers. 3. would cover a football field to a depth of less than 10
An automated system sends storage containers underground to feet). The Nuclear Waste Policy Act of 1982 mandated
the tunnels. 4. Containers are stored along the tunnels, on their
sides. (Nuclear Regulatory Commission.) that the federal government construct a permanent, deep
underground storage facility for the high-level waste that
had accumulated at various temporary storage facilities
during the first forty years of the nuclear era. Twenty
years later (!), the U.S. Congress approved funding to construct such a site at Yucca Mountain, Nevada,
on federal land adjacent to a former nuclear weapons testing site. Construction proceeded for nearly a
decade with an expenditure of more than $12 billion, but the project was abandoned in 2011 under the
Obama administration16 (for reasons widely reported to be more political than technical [240]). This
failure leaves the U.S. once again without any long-term plan for the safe disposal of nuclear waste.
This is deeply worrying. Acceptance of nuclear power, we would argue, carries with it an ethical
obligation to dispose of its waste products safely—and that includes building a repository that is, as
far as possible, “future proof.” It is hardly surprising that no one wants nuclear waste permanently
stored in a deep hole close to their backyard. But the very purpose of federal government is to take
the difficult decisions that have to be made nonetheless. If it cannot do this, perhaps because it is too
indebted to special interests of one kind or another, it is failing to exercise the virtues needed in the
Anthropocene epoch (Section 6.4.3).
Nuclear Accidents
No nuclear scenario evokes more fear than a reactor meltdown. Suddenly, whether through operator
error (Three Mile Island, Chernobyl) or natural disaster (Fukushima), the core of a power reactor
begins to overheat. Before operators can regain the upper hand, the core becomes hot enough to
deform slightly, so that neutron-absorbing “control rods” can no longer be inserted to stop the reaction.
As temperatures increase, the reactor fuel deforms more and finally melts, continuing to fission and
becoming a white-hot, highly radioactive pool at the bottom of the containment vessel, or even burning
15 This reprocessing is a key component in plans for breeder reactors. It can also make the resulting waste safer, since the
most long-lived components of reactor waste are the transuranic elements, which can serve as fuel for breeder reactors. What
is left decays much more quickly, perhaps reaching background levels in 1,000 years rather than 100,000.
16 A bill to reopen the repository (HR3053) was advanced from committee in the House of Representatives at the time of
writing, June 28, 2017. It is too early to say what will become of it.
through it. Side reactions release hydrogen, a gas that can produce dramatic explosions when it mixes
with the outside air.17 Meanwhile, through containment breaches or other leaks, a plume of radioactive
elements—including iodine, which accumulates in the thyroid, and strontium, which accumulates in
the bones—spreads, silently and invisibly, to contaminate and poison farms, schools, and communities
for miles downwind. Life will never be the same again.
This story has all the psychological features needed to activate our risk-aversion mechanisms
(Section 5.3.4): a beginning in a dramatic and spectacular disaster, endless media coverage, and
an uncanny and invisible threat to ourselves and our families. Without minimizing this sort of
response, let’s try to carry out a “cooler,” more disengaged analysis of radiation risks. This begins
with acknowledging that we are all exposed to radiation every day of the year.
Table 8: Natural Radiation Exposures
Source Radiation (millisieverts)

Background radiation (U.S. average) 6 per year
Background radiation (Denver, CO) 12 per year
Eating 1 banana 10−4
Eating 4 ounces of Brazil nuts 0.01
Chest X-ray 0.02
Single flight from New York to 0.08
London
Average additional exposure for 0.18 per year
nuclear power plant worker
CT scan of head 1.4
Average additional exposure for 2 per year
airline flight crew
Whole body CT scan 10–20
ICRP action level for radon 10 per year
ICRP on-the-job exposure limit 50 per year, and 100 in any 5-year period
Dose that would kill within a month about 5,000
There are a number of different units for measuring the “quantity of radiation” to which a human
being may be exposed. Some of them are defined in a purely physical way, others in a way that takes
into account the fact that some kinds of radiation exposure are more dangerous to humans than others.
The standard unit that takes human susceptibility to radiation into account is the sievert (Sv).18 A
variety of causes expose us to radiation all the time: radon in our homes (Problem 3 on page 74),
natural radioactive elements in rocks and in our food, and radiation from outer space (“cosmic rays”),
which is greater at higher altitudes because there is less thickness of atmosphere to provide a shield.
These causes add up to expose the average American to about 6 × 10−3 Sv (6 millisieverts) per year,
although there is wide regional variation. Table 8 lists the additional radiation doses to which we are
exposed by some ordinary activities. (In this table, “ICRP” refers to the International Commission
on Radiological Protection, an independent, international organization whose mission is to “prevent
cancer and other diseases and effects associated with exposure to ionizing radiation, and to protect the
environment.”)
You’ll notice that we are prepared to accept the risks of quite significant amounts of additional
radiation, especially for medical purposes in the form of CT scans. Having established this scale, we
can take a look at the risks from a nuclear accident. The Three Mile Island (TMI), Chernobyl, and
17 The explosions seen worldwide on videos from Fukushima were of this sort. They were not nuclear explosions.
18 You may also encounter the rem, equal to 10−2 Sv, in older writing. Other units such as “becquerels” and “grays” are
physical measures of radiation level that do not take into account varying human sensitivities.
Fukushima incidents were carefully studied, and information about their radiation effects is readily
available.
The first thing to say is that no one died from acute radiation effects either at Three Mile Island19
or at Fukushima. Fukushima is a particularly telling example of the fearmongering bias of the news
media. The Tokohu earthquake and tsunami of March 11, 2011, led to the Fukushima meltdown; but it
also left nearly 20,000 people dead or missing, damaged nearly a million buildings, displaced about a
quarter of a million people, and amounted to the costliest natural disaster in history. Yet somehow all
this loss and human tragedy was obscured by the drama associated with the word “nuclear,” a drama
that directly killed nobody and that current models estimate may cause at most a few hundred excess
cancers. As we saw in Exercise 36 of Chapter 1, the radiation level in the open at one of the most
highly contaminated locations after the accident was 30 microsieverts per hour. Supposing that this
level did not decrease over time (which of course it did), one would have had to stand in the open 24
hours per day for two weeks to accumulate the amount of radiation involved in a single CT scan. There
are many lessons to be learned from Fukushima, especially related to the siting of the backup diesel
generators whose inundation was the immediate cause of the disaster. The most basic lesson, though,
is that the containment structure of a 1960s-vintage nuclear plant survived an enormous earthquake
and that the subsequent meltdown, the worst-case scenario for a plant of that design, killed no one.
Things are different when we consider the Chernobyl accident (April 26, 1986). A bungled safety
test on Reactor Number 4 at Chernobyl, in what was then the Soviet Union, led to the worst-ever
civilian nuclear accident. The RBMK reactor design used at Chernobyl did not have a containment
structure in the same sense as Western-designed reactors; the explosion and subsequent fire ejected
plumes of radioactive material into the atmosphere that traveled with the wind over the western USSR
and much of Europe. Heroic efforts by almost half a million special workers known as “liquidators”
succeeded in bringing the fire under control and entombing what was left of the reactor, as well as
the radioactive materials ejected during the fire and explosions, in a hastily built “sarcophagus.” Some
liquidators were allowed to work only 40 seconds at a time on the most dangerous tasks. About
50 liquidators later died from acute radiation poisoning. The World Health Organization estimates
approximately 4,000 excess cancer deaths worldwide from the aftereffects of Chernobyl. An area
30 kilometers in radius around the plant was permanently evacuated, and it now reportedly has a
background radiation level of about 10 millisieverts per year (though highly variable and much greater
in some “hot spots”). In the absence of a human population, it has become a huge wildlife reserve.
What do we learn from studying these nuclear accidents? Each was a hugely expensive disaster,
both in financial and in human terms; unless the risk of such accidents can be reduced, nuclear power
may simply not be financially viable. Reactor designs need to be improved to make them intrinsically
safe, so that they cannot melt down even if all power for cooling systems is lost. Such designs do exist,
and some (like the Integral Fast Reactor) have even been tested under conditions very similar to those
that caused the Fukushima meltdown—the IFR shut itself down without any human intervention.
We also learn, though, that in the case of radiation risks, our natural fear-management instincts may
have misled us. The radiation levels to which almost all members of the public were exposed by
these terrible disasters were no greater than those we happily accept as part of routine medical testing.
Rather than focusing on the silent and invisible risks of radiation, perhaps we should turn our attention
to the silent and invisible risks of climate change: another danger that slowly and yet inexorably is
altering the world in which we live.
Conclusion
Our world needs abundant energy supplies which are steady and reliable and that do not involve
emissions of climate-changing CO2 . Where are they to be found? Traditional energy sources (fossil
fuels) involve CO2 emissions that we must curtail in order to pass on a habitable planet to our
19 In fact, no one has ever died in the U.S. from acute radiation poisoning owing to an accident at a civilian nuclear plant.
successors. Natural gas is not a long-term answer here: it may halve CO2 emissions compared to
coal (see Table 5 on page 488), but in the long term, halving emissions is not good enough. Solar and
wind power are growing rapidly, and coming down in cost too, but the problem of intermittency (the
sun does not always shine and the wind does not always blow) has yet to be solved. Nuclear power
answers both needs: it is both clean and steady.
However, if nuclear power is to supply a substantial portion of our energy needs for long into the
future, we have to move on from present-day power reactor technology. Such technology uses only
a tiny fraction of the energy that is available in uranium, and so if we scale it up to meet our needs,
presently known uranium supplies will soon be gone. The breeder reactor answers this need, extending
the reach of our uranium supplies by 100 times or more. At the same time it offers the possibility of
recycling nuclear waste, which right now is sitting in cask storage all over the United States, into
new fuel. If we make the environmentally sound choice to ramp up nuclear power, we need to invest
in breeder technology. (The fusion reactor is a longer-term prospect, but one that has not yet been
successfully demonstrated; breeder reactors have been experimentally proven.)
This is not a simple choice, however. To invest heavily in nuclear reactors, especially breeder
reactors, is to believe that we can find intrinsically safe designs; manage radioactive waste for
hundreds or thousands of years; and avoid the risks of proliferation even as we breed, for reactor fuel,
the plutonium that can also be used in atomic bombs. Can we trust ourselves (and our successors) to
make the right choices here? The fate of the Yucca Mountain project is not an encouraging precedent.
Alvin Weinberg, a famed nuclear physicist from the Manhattan Project era, wrote [343],
We nuclear people have made a Faustian bargain with society. On the one hand we offer—
in the catalytic nuclear burner [i.e., the breeder]—an inexhaustible source of energy.
Even in the short range, when we use ordinary reactors, we offer energy that is cheaper
than energy from fossil fuel. Moreover, this source of energy when properly handled is
almost nonpolluting.
....
But the price that we demand of society for this magical source is both a vigilance from
and longevity of our social institutions that we are quite unaccustomed to.
Weinberg’s point is that the ultimate “containment vessel” for nuclear power (if we decide to
pursue it) is not the steel and concrete of the reactor buildings and disposal sites, but the network
of institutions that must grow up around them and continue to function for many centuries. Are we
capable of building and maintaining such institutions? If so, nuclear power may be the answer to our
needs. If not, our investments in nuclear power may prove a dangerous legacy for future generations.
Think about it. . .
The 2013 documentary Pandora’s Promise argues that nuclear power, and a breeder
reactor program in particular, are vitally important to protect the environment, reduce CO2
emissions, and build a sustainable future. The controversial documentary has been described
by different people as “impactful and important,” as based on “half-truths,” and as a “big
lie.” After you have read this case study, watch the documentary and see what you think.
7.7 Using Electricity Efficiently at Home
• This case study is an analysis piece. It examines two specific ways in which we can
use electricity more efficiently in our homes and shows how to calculate the savings
that might result.
• It uses ideas from Section 2.2 (Energy Stocks and Flows).
Electric power has been supplied to U.S. homes for a little over a hundred years. Having energy
available instantly, on demand, wherever it is wanted has transformed every aspect of our lives. A
famous 1946 poster from the Electric Institute of Washington is titled “Leisure through Electricity.” It
shows an impeccably dressed “housewife” relaxing by reading a magazine while electric appliances
take care of all her domestic chores. That is not exactly how things have worked out—in accordance
with the Jevons paradox (page 94), the cheaper electricity has become, the more things we have
found to do with it, including extra work for ourselves. Now, however, we are more aware of the
environmental cost of that electricity. It may be “clean” at the point of use (our homes), but it has to
be generated somewhere. The generation process may involve various sorts of pollution. Moreover,
the further away the generator is from the point of use, the more energy is lost in transmission through
the electrical grid. Making sure we use electricity more efficiently at home, then, is good for everyone.
The consumer has more money to use for other purposes; the nation improves the security of its energy
supply; the climate change threat to future generations is reduced. In this section we’ll discuss how to
estimate the savings that can result from various energy efficiency measures.
Home Electricity Use

Let’s begin by asking what are the biggest users of electricity in an American home. The Energy
Information Administration [9] provides the data in Figure 32.
Water heating
Interior lighting
Refrigeration 9.1% 9.5%

Air conditioning
8.9%
17.5%
TV, computer and
entertainment
8.1%
6.8%
Space heating 32.9%
4.9%
2.3%
Clothes washing and drying
Cooking Other
Figure 32: Use of electricity in U.S. homes, from [9].
The “other” category is rather broad, including all appliances and small electrical devices not
specifically listed (dishwashers, furnace motors, portable heaters, outdoor lights. . . ). Still, it is helpful
to see the broad pattern of power usage and to notice, in particular, the fact that air-conditioning is
7.7. USING ELECTRICITY EFFICIENTLY AT HOME 459
by far the largest single identified use. (As the climate warms, the demand for air-conditioning is
only expected to grow.) In this case study we will look at two examples: lighting, where there is a
comparatively easy and cheap “technical fix,” and air-conditioning, where there isn’t.
Remark 1: There is a substantial difference between this pie chart and the one in Figure 22 on page 99,
which shows the breakdown of all energy use in U.S. homes. The reason for this difference is that the
earlier chart considers all energy sources, not just electricity. Other energy sources delivered to U.S.
homes, like natural gas, oil, and solid fuels, are used primarily for space heating and water heating.
That is why the pie segments for these items in Figure 22 on page 99 are so much larger than the
corresponding segments in the chart that shows electricity use only.
If we are going to estimate the savings from various efficiency measures, we need also to know
the cost of electricity. There are two ways of framing the cost: in terms of money, and in terms of
environmental impact. Both of these vary significantly depending on where you are located in the
country and how you choose to purchase your power. Electricity prices vary from a low of about 10
cents per kilowatt-hour (kWh) in Idaho (where most power comes from hydroelectric dams that were
completed and financed years ago) to a high of about 30 cents per kWh in Hawai’i (where most power
comes from oil, which must be delivered by ship). An average price is about 13 cents per kWh.
As a measure of environmental impact, we use the amount of CO2 emitted in generating a given
amount of electricity (called the carbon intensity of electric power). At present, the carbon intensity
of U.S. electricity overall is about 0.96 pounds (of CO2 ) per kWh. This represents a reduction of
almost 25% in the last decade, a reduction that has been brought about by a move away from coal and
toward cleaner energy sources such as natural gas (see Table 5 on page 488), wind, and solar power.
Lighting
We’ll start by thinking about lighting. Though not the biggest sector of the pie, lighting’s share of U.S.
energy consumption is substantial, and there are simple ways to save energy here.
Let’s recall the definition of efficiency (Definition 6 on page 90): the efficiency of an energy process
is the quotient
Useful power output
.
Power input
When we consider lighting, the efficiency of a bulb or other lighting device is therefore the fraction
of the electrical power input that is converted by the device into useful visible light.
Thomas Edison’s light bulb (1879) was the beginning of the electricity revolution. It was an
incandescent bulb, that is, it produced light by heating a carbon filament to a very high temperature. If
you have ever put your hand near an incandescent bulb, you will know that it is not very efficient—it
produces a great deal of heat relative to its light output!
Problem 1: According to the General Electric Company, a standard 60-watt incandescent bulb
produces 840 lumens (see below) of light output. Estimate its efficiency.
Solution: What is a lumen? It is a measurement of the amount of radiant energy produced, calibrated
according to the human eye’s sensitivity to light of different wavelengths; in fact, it is exactly a
measure of what we called “useful power output” in the definition of efficiency. Lumens are related
to watts by the so-called luminous coefficient, which we will take to be 683 lumens per watt. (The
exact coefficient depends on the color composition of the light we are talking about—how yellow or
blue it is. That is because the definition of a lumen takes into account the varying sensitivity of the
human eye to different colors. For simplicity, we will use the value of 683 lumens per watt in all our
calculations.)
Armed with the luminous coefficient, we can calculate the efficiency of our 60-watt incandescent
bulb. The useful power output is
1W
840
lumens
×
= 1.22 W.
683
lumens
By definition of efficiency, then,
Useful power output 1.22

W
Efficiency = ≈ 0.02 = 2%.

=
Power input 60W

So the efficiency of the incandescent bulb is 2%. To put it another way, 98% of the energy used by the
bulb is wasted.
Because the incandescent bulb is such an inefficient device, replacing incandescent bulbs by more
modern, efficient alternatives such as compact fluorescent (CFL) and light-emitting diode (LED)
bulbs is a quick and easy way to make substantial savings in energy use. This was the thinking
behind one part of the “Energy Independence and Security Act of 2007,” which (among other
things) mandated significant energy efficiency improvements for domestic lighting. On signing the
act, President G.W. Bush said [57],
Taken together, all these measures will help us improve our environment. It is estimated
that these initiatives could reduce projected CO2 emissions by billions of metric tons.
At the U.N. climate change meeting in Bali last week our nation promised to pursue
new, quantifiable actions to reduce carbon emissions. Today we’re doing just that. The
legislation I’m signing today will lead to some of the largest emission cuts in our nation’s
history.
General Electric’s modern replacement for the 60 W incandescent bulb is their 11-watt A19 LED
bulb.
Problem 2: The 11-watt A19 LED bulb has a light output of 800 lumens. Estimate its efficiency, using
the same methods as in the previous solution.
Solution: The useful power output is

1W
800
lumens
×
= 1.17 W.
683
lumens
By definition of efficiency, then,
Useful power output 1.17

W
Efficiency = ≈ 0.11 = 11%.

=
Power input 11W

Still not perfect, but a great improvement over 2%!
Problem 3: The average U.S. home contains 40 light bulbs. Suppose that each bulb is used for 3
hours per day. Calculate the amount of energy saved per year if the householder switches from 60-
watt incandescent bulbs to their LED equivalents. By how much will CO2 emissions be reduced by
this change?
Solution: The incandescent bulb uses 60 watts, and its LED equivalent uses 11 watts. The difference
is 60 − 11 = 49 watts for each bulb. Multiplying by 40 bulbs gives a power saving of 49 × 40 ≈
2, 000 W = 2 kW for all the bulbs taken together.
The bulbs are on for 3 hours per day. Thus the power saved by switching to modern bulbs is
2 kW × 3( hr/ day) = 6 kWh/ day.
Multiplying by 365 days in a year gives an annual saving of
6 kWh/
day
× 365
day
≈ 2, 200 kWh.
To find the reduction in emissions we us the carbon intensity of U.S. electricity, which, as we
mentioned in the previous subsection, is approximately 0.96 pounds of CO2 per kWh. Thus the annual
emissions savings are
2, 200 × 0.96 lb ≈ 2, 100 lb ≈ 1 t.
kWh

kWh

The worldwide average carbon dioxide footprint is about 4 tons per person, so that this saving is
equivalent to one-fourth of the average person’s emissions. It is a significant amount.
In addition to saving energy, LED bulbs last far longer than their incandescent counterparts. The
incandescent bulb costs about a dollar and should last about 1,200 hours. The LED bulb costs about
$8 but should last 25,000 hours.
Problem 4: Compare the total cost of lighting our 40-bulb house with incandescent and with LED
bulbs, over a fifteen-year period.
Solution: At 3 hours per day over a 15-year period, each fitting must run for 15 × 365 × 3 ≈ 16, 500
hours. That is well within the lifetime of a single LED bulb, but using incandescents, we will have to
replace the bulb 16, 500/1, 200 ≈ 14 times.
Comparing the bulb costs only, then, we have
Cost for LEDs = 40 × $8 = $320,

Cost for incandescents = 40 × 14 × $1 = $560,
giving a saving of $560 − $320 = $240 by using the LEDs. To this we must add the cost of the
electricity saved. We save 2, 200 kWh/ yr (see the previous solution), and U.S. average electricity
costs are $0.13 per kWh. Thus the electricity cost savings over a 15-year period are
kWh
$0.13
2, 200 × 15
yr × = $4, 290.

y r 1
kWh

The overall savings over the 15-year period are $4, 290 + $240 = $4, 510; a substantial sum.
Remark 2: This is not necessarily the end of the story for energy efficiency, or even for incandescent
bulbs. Contrary to much that was said at the time, the EISA did not “ban” incandescent bulbs, but
simply required that all domestic bulbs meet higher energy efficiency standards. Halogen bulbs (used
in some specialized applications) are examples of incandescent bulbs that are more efficient than the
traditional technology. In 2016, researchers at MIT demonstrated a prototype of an incandescent bulb
that uses “nano-mirrors” to recycle waste heat from the filament back into light. If this works at a
commercial scale, incandescent bulbs with LED-like efficiencies could yet make a comeback. This is
an example of how regulation can accelerate progress by providing incentives for the development of
new technologies.
Heat Pumps and Air Conditioning

As we saw in Figure 32 on page 458, the largest single consumer of home electricity is air-
conditioning. Before thinking about ways we might improve the efficiency of this process, let’s think
about how it works in terms of the laws of thermodynamics.
You may recall that the Carnot limit (Rule 4 on page 91) tells us the maximum possible efficiency
of a heat engine, in terms of the temperatures (in kelvins) of its “hot” and “cold” sides. In a heat
engine, the flow of heat from hot to cold gives rise to an output of usable energy in a different form,
such as mechanical or electrical. But there are also machines that work like heat engines in reverse: an
input of mechanical or electrical energy generates a flow of heat from cold to hot. The most familiar
example of such a machine is the domestic refrigerator, which takes in electrical energy and uses it to
pump heat out of the refrigerator cabinet to the surrounding air. Other machines of the same type are
the air-conditioner (cooling your home by pumping heat out to the atmosphere) and the heat pump
(warming your home by pumping in heat either from the atmosphere (“air source heat pump”) or from
the subsoil (“ground source heat pump”)).
Problem 5: U.S. government standards express the efficiency of an air-conditioning unit in terms of
the SEER (Seasonal Energy Efficiency Ratio), defined as the average number of British thermal units
of heat that the AC unit pumps out of the house for each watt-hour of electrical energy that it consumes
as input. Thus the SEER is measured in British thermal units per watt-hour. Both Btus and watt-hours
are measures of energy, so the SEER is a pure number (Definition 2 on page 15), as we would expect
for an efficiency measure. Find the coefficient by which we must multiply to convert the SEER to a
“pure number” rating of efficiency.
Solution: Using our conversion tables, we can write

1 Btu 1050 J
= ≈ 0.29.
1 Wh 1 J/
sec × 3600
sec

Thus a SEER of 15 Btu per watt-hour (on the low end of reasonable for a modern central AC)
corresponds to an efficiency of 15 × 0.29 ≈ 4.3; over 4 times as much heat is pumped out of the
house as energy is supplied to the AC unit.
You might think that the previous answer can’t possibly be right. Didn’t we explain at the start of
Section 2.2.4 that there is no free lunch in the energy sector? So how can an ordinary domestic air-
conditioner have an apparent “efficiency” greater than one? As with any comparison, it all depends
on what you are comparing! Converting the movement of heat into high-grade energy must have an
efficiency of less than one: that’s the second law of thermodynamics (Rule 3 on page 89). But using
high-grade energy to move heat is a different matter; especially if the temperature difference is low,20
you can move a lot of heat for a fairly small expenditure of high-grade energy.
Example 1: Let’s think about home heating for a moment. It is possible to heat a home by electricity,
and electricity can be converted to heat with 100 percent efficiency. But from a larger perspective, this
is not such an efficient idea! Recall that the electricity has to be generated somewhere, at a power plant
that (if it uses fossil fuels) turns heat into electricity with an efficiency of maybe 40% (Problem 2 on
page 91), and then transmitted to your house over a long distance, which also involves energy losses.
From an overall energy-use perspective, it is more efficient to burn the fossil fuels in your own home,
for instance in a natural gas furnace—avoiding the losses that come from turning fossil fuel energy
into electricity and then back again. As we saw in Remark 1 on page 459, the majority of American
homes use this approach.
The efficiency tradeoff changes, though, if we take advantage of the fact that we can use high-grade
energy to move heat with an apparent efficiency greater than 1. The heat pump uses an energy input
(electricity) to pump heat from an outside source, either the atmosphere or the ground, to warm your
house in winter. Just as with the air-conditioner example, the amount of heat energy pumped into your
house can be much greater than the amount of electrical energy that you use. This works particularly
well for ground-source heat pumps, since in winter the deep subsoil is usually much less cold than the
atmosphere, so that the heat for your house does not have to be pumped across such a wide temperature
range. Indeed, a well-designed ground-source system can work as an air-conditioner (pumping heat
into the ground) in the summer months, and can then be reversed to act as a heater (pumping that heat
back out of the ground again) in the winter.
20 The SEER is an average over a range of temperature differences intended to represent a typical “cooling season”; hence
the seasonal in “seasonal energy efficiency ratio.”
Just as there is a Carnot limit for the efficiency of a heat engine, so is there a Carnot limit for the
efficiency of a heat pump, and indeed they are reciprocals of each other:
Rule 1
• Consider a heat pump that has a “hot side” and a “cold side.” Let Th equal the
temperature of the hot side of the engine and Tc the temperature of the cold side,
both measured in kelvins (see Remark 4 on page 13).
• Then the efficiency with which the engine converts high-grade energy input into “hot
side” heat output can be no greater than the fraction
Th
,
Th − Tc
called the reverse Carnot limit for the given temperatures Th , Tc .
Example 2: For domestic heating you might have Th = 300 K and Tc = 270 K, so that the reverse
Carnot limit 300/(300 − 270) = 10 shows that theoretically, the heat pump can give you ten times
as much heat out as you put energy in. Heat pumps are expensive devices, and the electricity to run
them is not generated with perfect efficiency either, but this calculation shows their great potential for
saving energy.
The thermodynamics is interesting and important, but we have not yet begun to address the
following basic question: how can we reduce the amount of energy expended on air-conditioning our
homes? Unfortunately, the answers to this question are not as simple as changing a few lightbulbs.
It is possible to install high-SEER air-conditioning equipment: SEERs up to 25 are commercially
available, though the high-efficiency equipment tends to be much more expensive.21 However, this
is not really the right place to start. Instead of working to pump heat out of our homes with greater
efficiency, perhaps we should ask how so much heat got into them in the first place?
Before air-conditioning existed, home designers knew many ways to keep summer heat out of a
home: orienting the building correctly to the sun (so that few windows face east or west, which are
the biggest heat collectors in summer months); using awnings or external shutters on south-facing
windows; planting shade trees; cooking outside in hot weather. Many of these techniques have been
forgotten, or are not widely used, since the advent of central air. Instead, we can just do the lazy thing:
put down a big box of a house any which way, and then install enough AC to keep it cool. But quite
apart from the huge energy consumption this generates, do we really want to build houses that are
uninhabitable without air-conditioning?
In the modern era we may add more ideas to the traditional ones: high R-value insulation
(Definition 1 on page 100), especially in the attic, which will keep heat out as well as in; specialized
window materials that transmit less solar radiation; lighter-colored roofs that reflect more solar energy;
high-efficiency whole-house fans that are well insulated and effective at bringing cool nighttime air
into the house. What’s more, reducing the energy consumption of lightbulbs and other appliances
helps reduce AC loads. We previously calculated that using incandescent bulbs in an average house
“wastes” 6 kWh, or 20,000 Btus, of energy per day. That energy does not just disappear—it ends up
in the house as heat, which (in summer) adds to the load on the AC unit.
21 The reverse Carnot limit gives the maximum theoretically possible SEER as about 50–100.
Remark 3: Some of these ideas, like choosing the house’s orientation, have to be applied at the design
stage. Let’s emphasize a few that are not intrinsic to design, but are possible retrofits. It is usually
possible to improve attic insulation, which will save energy both in summer and winter. Awnings
or shades can be added to south-facing windows, and shade trees planted if the land is available.
Modern whole-house fans are quiet and efficient and can be equipped with remote control. And as
we mentioned, reducing energy consumption elsewhere in your home will also help keep it cool in
summer.
Problem 6: The attic of my house has a total area of 1350 square feet. On a hot summer day, the
attic temperature reaches 125 ◦ F for 11 hours (attic temperatures can get much higher than outside air
temperatures). Assume that a summer season is made up of 65 such days. My air-conditioner has a
SEER of 12 Btu per watt-hour, and I try to maintain a temperature of 78 ◦ F inside my house. How
much electricity for air-conditioning will I save (in kWh per summer season) if I upgrade my attic
insulation from its present R-value of 13 hr ◦ F ft2 / Btu to a new value of 60 hr ◦ F ft2 / Btu?
Solution: This complex problem is best handled in stages. First, let’s calculate
how many Btus of heat flow from the attic into the house every “hot” hour, both
with the old and with the upgraded R-values. The temperature difference across
the insulation, ∆T in the notation of Definition 1 on page 100, is 125 ◦ F − 78 ◦ F =
47 ◦ F. Thus, with the old insulation the heat flow rate is
2 1 Btu
47 ◦ F × 1350
ft × 2
≈ 4900 Btu/ hr.
13 hr ◦ F
ft
The corresponding flow rate for the new insulation is
2 1 Btu
47 ◦ F × 1350
ft × 2
≈ 1100 Btu/ hr.
Figure 33: Whole-house fan re- 60 hr ◦ F
ft
mote controller.
Thus installing the new insulation reduces the heat flow into my house by 4900 −
1100 = 3800 Btu/ hr for every “hot” hour. The terms of the problem specify that
there are 11 hot hours per day and a 65-day season,22 giving a total amount of heat saved (per season)
of
Btu hr ≈ 2.7 × 106 Btu.
3800 × 11 × 65day

hr day

To convert this to kilowatt-hours of cooling electricity we use the SEER:
2.7 × 106 × 1 Wh ≈ 2.3 × 105 W hr = 230 kWh.

Btu
12
Btu

Insulating the attic results in energy savings of 230 kilowatt-hours every year. Interestingly, changing
all the bulbs in the home to LEDs produces energy savings that are almost ten times larger (Problem 3
on page 460). Would you have guessed that?
Problem 7: Consider the same house as described in Problem 6. As an alternative to upgrading the
insulation, I consider installing a solar powered attic ventilator. This will draw exterior air through
the attic, reducing its “hot” temperature from 125 ◦ F to 105 ◦ F. The cost of the solar powered attic
ventilator is 70% of the cost of upgrading the insulation. Which offers better value?
22 Of course this is a simplifying assumption—the attic does not spend 11 hours hot and then suddenly switch to room
temperature—but it gives us an approximation to work with.
Solution: We can use some of the work we did in the previous solution. The solar ventilator reduces
∆T from 47 ◦ F to 105 − 78 = 27 ◦ F. Thus the heat flow rate with the ventilator (but still with the “old”
insulation) is
2 1 Btu
27 ◦ F × 1350ft × 2
≈ 2800 Btu/ hr.
13 hr ◦ F
ft
The heat savings from installing the ventilator are 4900 − 2800 = 2100 Btu/ hr for every “hot” hour.
Calculating as in the previous solution, we get a total amount of heat saved (per season) of
Btu hr ≈ 1.5 × 106 Btu,

2100 × 11 × 65
day

h

r day

and an amount of cooling electricity saved of
1.5 × 106 × 1 Wh ≈ 1.3 × 105 Wh = 130 kWh.

Btu
12
Btu

The ratio is
Electricity saved by solar ventilator 130
kWh

=
Electricity saved by upgraded insulation 230 ≈ 0.57 = 57%.
kWh

Since the solar ventilator costs 70% as much as the upgraded insulation, but saves only 57% as much
energy, we conclude that the insulation upgrade offers better value.
Conclusion
We have examined only two of the many uses of electricity in the home (and we have not gone as
deeply as we could have even into those). In one case (lighting) we have seen that modern research
has developed an effective technical fix: lighting energy consumption can be very greatly reduced by
using LED bulbs rather than traditional incandescents, and the change is simple to make and pays for
itself over a short period. In the other case (air-conditioning), improving efficiency of the AC itself
is possible, but difficult and expensive. In many situations, using a mixture of traditional and modern
ideas on “heat management” will save you more money and energy on air-conditioning than you
would save by making the AC itself more efficient. The emphasis is on making the home a place that
is adapted to its environment, rather than one that must continually use energy to push back against
its environment via air-conditioning.
7.8 Growth and Payback Time for Solar Energy
• This short case study is an advocacy piece. It examines the notion of payback time
for solar energy installations and how the rapid growth of the solar industry can
paradoxically lead to longer payback times for the industry as a whole than for an
individual solar installation. On this basis, it advocates that now, when nonrenewable
energy is relatively cheap, is the right time to invest in renewables.
• It uses ideas from Sections 2.2 (Energy Stocks and Flows), and 3.4 and 4.1
(exponential models).
Payback Times
Solar energy installations (in the U.S. and elsewhere) are growing at a rapid pace. The majority of
these projects are utility-scale (solar “power plants,” distributing energy over the electricity grid), but
residential solar photovoltaic installations (that is, solar panels on your roof or in your yard) are also
growing at a rapid clip, as the chart below illustrates.
2016
2014
Year
2012
2010
0 500 1,000 1,500 2,000 2,500
New capacity (megawatts) installed in year
Figure 34: U.S. residential installed solar PV capacity, from [293].
If you talk to a solar installer, one of the figures that they are sure to share with you is the expected
payback time of your installation: how long will it take before you recoup the cost of the installation
from the savings that you make by generating your own power for free? Payback times have come
down fast as the cost of solar cells has fallen: current payback times in the U.S. can be as low as 4
years (in Hawai’i, where there is plenty of sun and conventional electricity costs are high), though
8–10 years is more typical.
In this discussion we will focus on a different “payback time,” which from a sustainability point
of view may be even more significant. Solar panels of course produce energy, but the process of
manufacturing them consumes energy—quite a lot of energy, in fact. So from the point of view of a
global energy account, one has invested a lot of energy upfront in manufacturing the solar panel—in
the same way that the homeowner needs to invest money upfront in a solar installation. How long, we
can ask, will it take before the panel repays that “energy investment”?
7.8. GROWTH AND PAYBACK TIME FOR SOLAR ENERGY 467
Definition 1
The energy payback time for a solar installation is the amount of time that the installation
takes to generate the same amount of energy that it took to manufacture.
To manufacture a solar panel requires a certain amount of energy. Let’s call this the commissioning
energy E. Then the solar panel will produce energy at a certain rate, the power production rate P. At
the end of the panel’s lifetime there may be a decommissioning process that can also be quantified in
energy terms; but to keep things simple, we will neglect the relatively small amount of energy involved
here.
The energy payback time, then, is the length of time it takes the solar panel to generate as much
energy as was used in its production. That is,
E
Energy payback time = .
P
According to a fact sheet from the U.S. National Renewable Energy Laboratory [234], the energy
payback time for domestic solar photovoltaic technologies in 2004 was between 2 and 4 years (and
was anticipated to improve to 1–2 years with newer technologies). A more recent research article [133]
suggests that solar cells based on perovskite rather than silicon could shorten the energy payback time
to a few months. The rated lifetime of solar panels is usually 25 or 30 years, so over a panel’s life it
will generate many times the amount of energy that was used to manufacture it.
Growth and Industry-Wide Payback

But even if each individual panel “pays back” many times in this way, it is possible for the energy
balance of the whole solar industry to be negative! How can this be? Two facts combine to produce
this negative effect: the front-loading of costs (the commissioning energy E must be “paid” in full
before the panel produces any power at all) and the very rapid exponential growth of the solar industry
(illustrated in Figure 34).
Let’s give a simple example using made-up figures. Suppose that each panel takes 2 units of energy
to produce (E = 2 units) and generates 1 unit of energy each year (P = 1 unit per year). Then the
payback period for each installation is 2 years. But suppose also that starting from a baseline of 1
panel installed, the industry grows exponentially, doubling in size each year, so that 1 new panel is
installed in year 1, 2 panels in year 2, 4 panels in year 3, and so on.
Then in year 1, the one panel we started from generates 1 unit of energy, but it costs 2 units to
produce the 1 new panel we install: a net loss of 1 unit.
In year 2 we have two panels operational, generating 2 units, but it costs 4 units to produce the 2
new panels we install: a net loss of two units. In year 3 we have a total of four panels operational,
generating 4 units, but it costs us 8 units to install 4 new panels. And things only get worse: using the
geometric progression formula (Rule 5 on page 198) we see that in year n, the number of operational
panels is
2n−1 − 1
1 + 1 + · · · + 2n−2 = 1 + = 2n−1 .
2−1
Also in year n we install 2n−1 new panels at an energy cost of 2 × 2n−1 . So our net energy balance
(energy generated minus energy cost of new installations) in year n is 2n−1 − (2 × 2n−1 ) = −2n−1 ;
negative and getting worse.
This example shoes that if the industry is growing sufficiently fast (specifically, with doubling
time less than the payback period), then it can give rise to an overall energy drain even though each
individual panel has a positive energy balance over its lifetime. There are two ways out of this trap:
the growth rate can slow (increasing the doubling time) or the technology can improve (reducing the
payback time).
Of course, our calculations above were simply an example, to show how the combination of
exponential growth and front-loading of costs can yield a negative overall energy payback. But what
about the real world? In their paper [85], environmental scientists Michael Dale and Sally Benson
carry out exactly this kind of analysis. They analyze historical information about the installed base of
six different generations of solar panel technology (each of which has different commissioning costs
and power production rates). The aggregate result of their analysis is shown in Figure 35 below (taken
from their paper):
Figure 35: Energy balance of the global solar PV industry.
This shows, using the two key variables that we identified above (growth rate and payback time),
how the net energy balance of the world’s solar installations looks in different years. For example,
the point labeled “2000” tells you that in the year 2000, the average energy payback time of a solar
installation was about 5 years, and the annualized growth rate of solar power was a little over 20
percent. As you can see, although technology improved considerably during the 2000–2010 period
that the authors study (reducing the payback time from 7 years to less than 2), the accelerating rate of
growth means that in 2010 (the end of the analysis period), the net energy balance was still negative.
After ten years, the world’s solar installations as a whole had consumed more energy than they had
produced! (The paper, published in 2013, suggests 2015 as the most likely year in which the solar
industry will achieve global break-even.)
Conclusion
What does this mean in practice? Not that solar is worthless: there are many situations in life where
we need to make an up-front investment in order to obtain a long-term return. It does tell us something
about the timing of that investment though. Building a solar (or other renewable) energy base requires
significant investment of energy from other sources—perhaps, nonrenewable sources—to get started.
With rapid growth, it takes a significant amount of time to accumulate these initial investments to
make the overall program profitable in energy terms, much longer than might be expected given
the payback period for an individual solar panel. Therefore, the time to make those investments in
renewable energy is now, when there is still plenty of nonrenewable energy around. If we wait until
nonrenewable energy is scarce, we won’t have the surplus we need to invest in getting renewable
energy started at a significant scale. University of California physicist Tom Murphy, writing on his
blog “Do the Math,” calls this the energy trap [231].
At the time of writing, fossil fuel prices are close to historic lows. Many voices claim that we should
therefore ditch public support for “uncompetitive” renewable energy. The analysis above suggests that
this would be exactly the wrong response. Instead, we should take advantage of this huge opportunity
7.8. GROWTH AND PAYBACK TIME FOR SOLAR ENERGY 469
to invest energy in producing energy—to ramp up renewables to a scale that will allow them to
shoulder the burden of our energy needs when nonrenewable sources decline, as they surely will.
After all, that is what “nonrenewable” means.
7.9 Energy Return on Energy Invested
• This case study is an analysis piece. It examines the notion of energy return on energy
invested (EROEI) for different energy sources, explaining why this number is both
important and difficult to calculate.
• It uses ideas from Section 2.2 (Energy Stocks and Flows) and Section 4.2 (Models
and the Limits to Growth).
Energy Investments and Returns

In Section 2.2.2 we gave a series of reasons why questions about energy stocks and flows are central
to nearly every discussion of sustainability. In fact, energy is of such critical importance that it is often
natural to reverse the money analogy (Rule 1 on page 83) and think of energy as the “currency” in
which sustainability matters are transacted.
This naturally leads us to consider our various technologies for obtaining energy in the language
of energy investment. Think about a “wind farm” (an array of wind turbines), for example. To build
the wind farm, we have to fabricate steel into giant turbine blades, pour concrete, build gearboxes and
generators, lay down roads and transmission lines, and so on. All of these use energy. There will also
be additional energy-using processes to maintain the wind farm during its lifetime and to clear the
site once it has reached the end of its usefulness. Let’s call the total amount of energy expended in all
these processes our energy investment in the wind farm.
We chose a renewable-energy example above, but the same principle can be applied to fossil fuels
also. Drilling an oil well requires an energy investment—not so much if the well is on land, a great
deal more if the well is in deep water. So does mining coal or fracking for natural gas. Obtaining oil
from Alberta’s tar sands requires a huge energy investment (the “oil” is so thick and, well, tarry, that
it has to be heated by burning natural gas so that it becomes liquid enough to separate from the sands
in which it is found).
We make these investments in energy technologies because we want to get some energy back—
electricity from the wind farm, coal from the mine, oil from the well, whatever. And naturally we
want to get more energy back than we invest: an energy project that yielded 1 joule of useful energy
for every 2 joules that we put in would be worse than useless. In fact, like classical investors in
the financial markets, we’re interested in our return on investment. Except that now, this return on
investment is defined strictly in energy terms.
Definition 1
The energy returned on energy invested, or EROEI, of an energy supply process is the
pure number
Amount of energy delivered by a certain source
EROEI = .
Amount of energy invested to deliver that energy
A process with a high EROEI is the “low-hanging fruit” of energy supply: it can easily be “picked”
for little effort. In many areas of life, the low-hanging fruit will be picked first. In the world of energy
7.9. ENERGY RETURN ON ENERGY INVESTED 471
supply this gives an important principle: nonrenewable energy supplies will generally be exhausted in
order of decreasing EROEI. (We already used this idea in Example 2 on page 214.)
For example, on January 10, 1901, Captain A.F. Lucas drilled to an un-
precedented depth—almost 1200 feet—into Spindletop Hill, near Beaumont,
Texas. That certainly required some investment of energy (and money!). But
the resulting oil “gusher” produced 100,000 barrels of oil per day for nine
days straight. No pumping was required; the oil simply sprayed into the air
until the well was brought under control by an early blowout preventer (one
of Lucas’s own inventions). It is estimated that the EROEI of those early
Texas wells was far above 100. By 1970 (approximately the Hubbert peak
for U.S. oil production; see page 218), the average EROEI of U.S. oil is
estimated to have fallen to about 30. Current “tight” oil production in the
U.S. has an estimated EROEI between 10 and 20. There are no more gushers:
more and more energy has to be invested to force the reluctant oil out of the
ground, or to find it in remote and inhospitable environments such as deep
water.
This is certainly a cause for concern. If energy supplies were available only Figure 36: The Lucas gusher.
at an EROEI less than 1, then our society would quickly cease to function:
we could not obtain any energy without spending a greater amount of energy
to obtain it. But in practice, we need some “margin” for society to operate. The more complex a
society is, the more energy is wasted in the processes of transmission and distribution and the higher
the EROEI of its primary energy sources needs to be. It has been estimated [230, 164] that our society,
the most complex the world has ever seen, requires energy inputs with an EROEI of at least 3–5 to
function effectively.
Arguments About Definitions

Before we look at estimates of EROEI for specific fuels, though, we should acknowledge that
calculating EROEI is not only difficult but also controversial. Its apparently straightforward definition
has several aspects that need to be clarified. These ambiguities have led to some strongly worded
exchanges in the scientific literature. For example:
Serious methodological errors in a paper by Scientist X and others invalidate their

results. The origin of such errors is investigated and explained. Scientist X’s results are
rebutted.
And in reply:
Scientist Y’s claims are based on sophisms.23 . . . the comments of Scientist Y show
how deeply certain methodologies and reasonings are already entrenched in politically
motivated energy evaluations—They are, however, not scientifically justified.
What is at stake here? Let’s try to explain some of the key questions whose answers will affect the
results of any EROEI calculation.
(a) What form of energy? In Section 2.2 we saw that energy comes in high-grade forms (like
electricity) and low-grade forms (like heat). High-grade energy can easily be converted to heat,
but converting heat to high-grade energy can be done with only limited efficiency (Rule 4 on
page 91). Now, some energy investments (like solar panels) produce high-grade energy directly,
whereas others (like natural gas wells) produce fuel that can be converted to high-grade energy
only by first burning it (yielding heat) and then converting the heat to high-grade energy (with
23 Dictionary definition: “a fallacious argument, especially one used intentionally to deceive.”
limited efficiency). Should we measure “energy return” in terms of raw joules (treating all forms
of energy the same) or in terms of high-grade energy equivalent? Some researchers choose the
first alternative and some the second, and the choice significantly affects the calculated EROEI
of solar and wind projects in comparison to thermal energy sources (those that produce heat
first).
(b) Should energy from the project itself count as part of the investment? If a tar sands project
brings in natural gas or electricity from outside to provide the heat needed to separate the oil
from sand, that is clearly an investment. But suppose the project burns some of the sands,
underground, in order to liquefy the rest: should the energy released by this burning count as
part of the “energy investment” in the project? Again, some say yes and some say no, and for
certain projects, which way of measuring we choose can make a significant difference.
(c) Where do we draw the boundaries of the project when calculating investment? Imagine
constructing a utility-scale solar photovoltaic plant in the Nevada desert. What should count
as the “energy investment” for this project? Clearly the energy used in actually fabricating the
solar panels should count. What about the energy used in building the factories that make the
solar panels, which after all did not spring up from nowhere? (And if you agree that some
fraction of this energy should count, how do we estimate what fraction?) The energy used to
ship the panels to the site? Probably yes. The energy used in building transmission lines to
take the power from the site? Probably yes again—a power plant in the middle of a desert is of
little use by itself. What about the energy to feed and house and equip construction workers,
maybe in a temporary camp? What about the cost of energy storage facilities (giant batteries,
pumped storage, or whatever) that might need to be added to the electrical grid to cope with the
variability of solar energy? And so on. We learned in Section 2.1.1 that “whenever we try to
pick out anything by itself, we find it hitched to everything else in the Universe,” so the point
where we draw the boundary of our project must be a matter of definitional choice. Different
choices here will lead to different denominators in the EROEI calculation.
Remark 1: Beyond these points (and beyond the EROEI idea itself) lies the further question of
the time when the energy investments are made and the returns gained. In the financial context,
investments and returns at different times would be reduced to present value using the discount rate
(Definition 6 on page 354). Most researchers working on EROEI would reject the idea of using a
discount rate for energy. However, calculations like those in Case Study 7.8 make it clear that the
timing of investments and paybacks can be critically important. An appropriate and agreed-upon
methodology to take account of these timings remains to be developed.
Some EROEI Values

Despite the difficulties outlined in the previous section, several researchers have tried to compile
EROEI values for different energy sources available now. The tables below give results from two
summary papers, [230] and [164], and a few specific numbers from other literature. To help with
the “what form of energy” question ((a) above), there are two different tables: one for liquid fuels
(producing thermal energy) and one for electricity production (a form of high-grade energy). A
thermal energy source that has an EROEI of X to produce heat will have an EROEI of roughly X/3
to produce electricity, because power generation and transmission is, very roughly, 33% efficient in
converting thermal energy to electrical. To put that another way, an “apples to apples” comparison of
the two tables (the first of which gives the thermal EROEI values of liquid fuels and the second the
electrical EROEI values of electricity sources) would require the numbers in the second table to be
multiplied by 3.
Here are a few observations based on these numbers:
7.9. ENERGY RETURN ON ENERGY INVESTED 473
Table 9: Thermal EROEI for Liquid Fuels
Energy Source EROEI from [230] EROEI from [164] Other

Conventional oil 11–18 16
Tar sands oil 2–4 5 6 (from [53])
Ultra-deepwater oil 4–7 (from [221] )
Shale or heavy oil 5 4
Biodiesel 1.3 5.5
Corn-based ethanol 0.8–1.6 0.4
Sugarcane ethanol (from Brazil) 1–10 9
Table 10: EROEI for Electricity Sources
Energy Source EROEI from [230] EROEI from [164] Other

Coal 25 18
Hydroelectric > 100 > 40
Nuclear 5 2–5 75 (from [347])
Natural gas 10 7
Wind energy 18 20
Solar photovoltaic 7 6 9–34 (from [49])
• EROEI is a qualitative measure, but one should also take quantity into account. For instance,
hydropower is by far the highest-quality electricity source as measured by EROEI, but its
quantity is strictly limited. In the other direction, tar sands oils are of marginal EROEI, but
they are available in enormous amounts.
• In the EROEI realm, conventional oil still leads the pack among liquid fuels. Biodiesel (made
from soy) is the only “renewable” liquid fuel that is presently competitive on an EROEI basis.
• Renewable sources such as wind and solar are already competitive with natural gas for
electricity generation (on an EROEI basis) and may soon be competitive with coal.
• Estimated EROEI numbers for nuclear power are highly variable. These numbers are based
on current reactor technology, using enriched uranium, a significant component of the energy
investment is the fuel enrichment process used.24 Hypothetical new reactor technologies have
been estimated to have EROEIs in the thousands, but such technologies have not yet been
demonstrated.
• While EROEIs for fossil fuels such as oil has declined over time, in accordance with the “low-
hanging fruit” principle explained above, EROEIs for renewable technologies such as solar PV
have increased over time as the technologies have matured.
EROEI is a key metric for a sustainable society, as the papers cited above underline. For example,
from [230]:
We believe that comprehensive EROI analysis can and should serve as a critical vetting
platform on which different energy sources are compared.
24 Centrifugal enrichment, the current standard technology, uses far less energy than the gaseous diffusion process that was
the foundation of the Manhattan Project and was used until the 2000s. This may account for some of the variability of the
numbers for nuclear power.
But the idea is not well known, and there are few places where it is clearly explained for the ordinary
reader ([164] is an exception here). This can lead to expensive mistakes, such as the one the U.S. may
have made with corn ethanol.
The U.S. Ethanol Issue
15
The U.S. is the world’s largest producer of ethanol for
fuel (see Figure 37), almost all from corn. The Energy
Policy Act (2005) and the Energy Independence and
10 Security Act (2007) mandated that a certain amount of
Production
ethanol biofuel must be included in the U.S. gasoline

supply: the target was 7.5 billion gallons by 2012 and
increasing to 36 billion gallons by 2022. These mandates
5 were supported by an improbable coalition of environ-
mentalists who believed that they would foster renew-
able energy development and reduce carbon emissions
0
and conservatives who believed that they would reduce
2000 2002 2004 2006 2008 2010 2012 U.S. dependence on unreliable foreign oil suppliers.
Year From the beginning of the ethanol mandate, its critics
Figure 37: U.S. ethanol production, in billions of gallons. pointed to its low efficiency in terms of energy. But these
criticisms were often formulated in a less than effective
way—that corn ethanol “consumes more energy than it
produces,” that is, in our terms, that its EROEI is less than 1. As you can see from Table 9 on the
previous page, it is now widely agreed that in this form, the criticism of ethanol was wrong: the
EROEI is greater than 1, though not by much.
This is a good example of a criticism that backfires by being too extreme. It turned out that the
simple memorable slogan—corn ethanol consumes more energy than it produces—was wrong. But
even so, an EROEI of 1.3 is not at all impressive. It is the lowest by far of all the liquid fuels listed in
Table 9 on the previous page, and lower than any estimate of what’s needed for a fuel that can sustain
a complex civilization. Bio-ethanol at a much higher EROEI is available from Brazilian sugarcane
(sugarcane cannot be grown in most of the U.S.). And the diversion of something like 20 percent of
the U.S. corn crop into fuel production has had significant effects on world food prices, including in
the poorest countries [282]. The authors of this article write, “By putting pressure on global supplies
of edible crops, the surge in ethanol production will translate into higher prices for both processed and
staple foods around the world. . . with potentially devastating implications for global poverty and food
security.” That seems a high price to pay for the most inefficient liquid fuel around.
Think about it. . .
The ethanol mandate is a huge boost to U.S. corn farmers. Some argue that however hard it
is to justify on sustainability grounds, the mandate is politically invulnerable because Iowa,
a major corn-growing state, has a disproportionate weight in the process that the Democratic
and Republican parties use to select their presidential candidates. Do you agree?
(c) The total amount of energy that would be saved

7.1. MATHEMATICS AND PERSUASIVE over the typical life of an LED light by switch-
WRITING ing from the current lighting in your home to
LEDs.
1. Work through the “gathering ideas” step (see
Section 7.1.1) as if you were intending to write a piece
on one of the following topics. As part of this process,
try to formulate two specific mathematical questions 7.2. THE CHANGING RISKS OF WILD-
you could answer that would provide support for a FIRES
specific idea you would develop.
(a) “Disposable Water Bottles” (discussed in Sec- 4. Identify the use of any of the rhetorical appeals in
tion 7.1.1). this piece (See Table 1 on page 415). Explain.
(b) “Meatless Mondays: The impact of reducing
your consumption of meat.” 5. Identify examples in this piece of quantities put
(c) “Net metering laws (laws requiring utilities to into human terms. Can you think of other compar-
purchase surplus residential solar power) and isons?
their impact on the development of residential
solar.”
6. Describe the skewness of the distribution for the
“acres burned” variable in Figure 5 on page 421.
2. Develop an argument directly refuting or support-
ing one of following claims. Your argument should
7. An issue not discussed in this piece, but relevant
make use of quantitative evidence from a credible
to changing risks from wildfires and other natural
source.
disasters, is the changing pattern of land use. When
(a) “If everyone throughout the United States un- communities grow, extending their reach into lands
plugs their smartphone chargers when they are
where wildfires occur naturally, those communities
not in use, we could save 50 megawatts, enough
face increased risks from these events, even if there is
to power 50,000 homes. This is one of simplest
and most effective actions we can take to reduce no underlying change in the frequency of such events.
greenhouse gas emissions.” Research this issue. Are there examples of states or
local governments accounting for these types of risks
(b) “You’d have to recycle 40,000 water bottles
to offset the greenhouse gas emissions of one (wildfires, flooding, or other natural disasters) in plan-
round trip flight from New York to Paris. Recy- ning and zoning?
cling just isn’t worthwhile.”
(c) “As our use of fossil fuels has increased over the 8. Investigate what natural disasters are considered
past 50 years, the death rate from flood events potential risks in your community. What are the rec-
has fallen. We should be less concerned with the ommendations for disaster preparedness in each case?
threat of sea level rise due to CO2 emissions and The U.S. Department of Homeland Security provides
more concerned with continued development of information on disaster preparedness at the website
the resources that have directly led to preserving http://www.ready.gov/.
life.”
9. In an effort to deal with mounting debt in the U.S.

3. Work out an answer for each of the following National Flood Insurance Program, the U.S. Congress
quantities; then put the result in human terms. passed the Biggert-Waters Flood Insurance Reform
(a) The amount of precipitation that falls on your Act in 2012. The act was intended to remove federal
roof or yard in one year. subsidies for flood insurance and allow flood insurance
(b) The total CO2 emissions from driving a car with premiums to match market rates [78]. Research the
U.S. average fuel economy for one year (if you National Flood Insurance Program (NFIP) and the
own a car, use your own driving for one year). effects the Biggert-Waters Flood Insurance Reform Act
had on homeowners in flood prone areas (for informa- smartphones. Compare the amount of gold per tonne
tion on how changes to the flood insurance program of smartphones with the 1 gram of gold per tonne of
have affected Pennsylvania, see the Center for Rural low-grade ore typically seen in large-scale gold mine
Pennsylvania’s study led by Lara Fowler [122]). operations.
17. In 2010, U.S. gold production was about 230

7.3. IS RECYCLING WORTH IT, REALLY? tonnes. Compare the amount of gold that was discarded
in mobile devices in 2010 with the amount of gold
produced from mining in the same year (see page 427).
10. Identify examples in this piece of quantities put
into human terms. Can you think of other compar- 18. In 2014, 136 million tons of waste was sent
isons? to landfills in the U.S. Use Table 7 on page 431 and
Figure 15 on page 431 to
11. Solid waste management in the United States (a) estimate the amount of discarded food that was
involves much more than handling the stream of mu- sent to landfills in 2014.
nicipal solid waste (MSW) that we have focused on in (b) estimate the potential GHG emissions benefits
this case study. Investigate other sources of solid waste. that might have been realized if we instead
Under what policies are solid wastes in the United composted this food waste.
States regulated? How does the amount of municipal
solid waste generated each year compare with other
sources of solid waste? The Wikipedia page on “Solid 19. According to the EPA, avoiding 1 million tons
Waste Policy in the United States” is a good starting of CO2 is equivalent to taking about 190 thousand cars
point [81]. off the road for 1 year. Put the GHG benefits from the
previous exercise in terms of cars taken off the road.
12. In the U.S., some wastes generated by households
are classified as household hazardous waste (HHW). 20. Use Table 7 on page 431 and Figure 15 on
Find out what materials are included under this classi- page 431 to estimate the potential GHG emissions
fication. Regulations on the disposal of these materials benefits of recycling plastics that are currently sent
vary by state. What are the regulations for disposal of to the landfill. Compare this with your results from
HHW in your state? Exercise 18. If you were charged with setting policy
for reducing GHG emissions, would you choose to
prioritize composting or plastics recycling? What other
13. In 2014, 34.6% of municipal solid waste was
factors might need to be considered?
recovered and 3.5% of recovered materials were plastic
(by weight). Estimate the amount of plastic that was
recovered in 2014. 21. One way of reducing the amount of food waste
that is sent to landfills is to compost more of that that
14. Investigate packaging as a source of municipal waste. Another way may be to reduce the amount of
solid waste (in the U.S. and elsewhere). Is there evi- food that is wasted in the first place. Investigate food
dence that the rules in Europe on extended producer waste and efforts to reduce it.
responsibility (EPR) for packaging have been effective
in reducing waste? 22. Relate the notions of “cradle-to-cradle” and
“cradle-to-grave” manufacturing discussed in this
15. A smartphone contains roughly 0.001 troy ounces piece to the distinction between one-way and cyclic
of gold. Estimate the value of this gold at current systems introduced on page 80.
market prices, assuming it could all be recovered from
the phone.
7.4. WORLD POPULATION GROWTH
16. The average weight of a smartphone is about AND THE DEMOGRAPHIC TRANSITION
150 grams. Estimate the amount of gold per tonne of
23. Identify the use of any of the rhetorical appeals in environmental impact per lecture in mathematics. Just
this piece (See Table 1 on page 415). Explain. like the IPAT equation, this equation is a tautology
(true by definition). Do you think there is a significant
24. Identify examples in this piece of quantities put difference between the IPAT and ILUV equations? If
into human terms. Can you think of other compar- so, try to explain what it is.
isons?
25. Although we have argued that the threat of

overpopulation has diminished and conditions have 7.5. GENETIC ENGINEERING AND THE
improved for many, the toll of childhood mortality in FUTURE OF FOOD
the developing world remains significant and heart-
breaking.
31. Identify the use of any of the rhetorical appeals in
The rate of childhood mortality among children
this piece (See Table 1 on page 415). Explain.
under 5 years old in India has dropped from about 250
deaths out of every 1,000 children in 1960 to about 48
deaths out of every 1,000 children in 2015. There were 32. Identify examples in this piece of quantities put
25 million children born in India in 2016. Using the into human terms. Can you think of other compar-
child mortality rate for 2015, about how many of these isons?
25 million children will not make it to age 5?
33. Read the New York Times article Doubts About the
26. Under what conditions would you expect popula- Promised Bounty of Genetically Modified Crops cited
tion growth to follow an exponential model? in the case study. Develop three or more substantive
follow-up questions about the article and attempt to
27. Continue the analysis of the population momen- find answers for them. Answering each question should
tum example presented on page 438 and in Figures 23 require investigating sources beyond what is immedi-
and 24 on page 439). How long will it take for the ately available in the article. These can be sources cited
population to settle down into a dynamic equilibrium? in the article from the New York Times itself, sources
What will the total population be in this equilibrium? cited in the case study, or additional sources that you
Compare the equilibrium population with the initial find on your own.
population presented in Figure 23.
34. Review the graphical information provided in the
28. On page 440 we briefly discussed the yield New York Times article cited in the case study. Analyze
gap in agricultural production between industrialized the graphics comparing total pesticide use in France
and developing countries. Visit the World Bank’s open with total pesticide use in the United States. What
data website [40], do a search for “Cereal Yield”, and criticisms do you have concerning the presentation of
explore grain yields for various regions. How does the the data. Explain. Assuming you had access to any level
cereal yield in the United States compare with other of detail about pesticide use, can you think of other
regions? How has the global average changed over comparisons that might be more appropriate?
time?
35. According to the data shown in the article
29. Do some research on the expected impacts of from the New York Times, fungicide use in France is
climate change on agricultural yields. Discuss the role substantially higher than herbicide and insecticide use,
of climate change in helping or hindering the closing while herbicide use in the United States is substantially
of the yield gap from the the previous exercise. higher than insecticide and fungicide use combined.
What might this difference be attributed to? Is this
30. The “ILUV equation,” invented by one of the difference an important consideration in the analysis?
authors, says I = L × U × V , where I is Impact (as
in the IPAT equation), L is the number of Lectures in 36. Data on pesticide use by country is available
mathematics per year in the typical university, U is the from FAOSTAT, the food and agricultural data center
number of Universities, and V is the average global of the FAO (Food and Agriculture Organization of the
United Nations). Go to the website [110]. Choose sev- 42. In some areas of Canada, electricity comes
eral countries from the drop-down menu and compare mostly from hydroelectric power (a zero carbon energy
their pesticide use over time. How do other nations in source), whereas home heating is provided mostly by
Western Europe compare with the data presented about natural gas. The climate is cold, so that there are many
France? How might it have affected the comparison days when heating is required and few, if any, when air-
made in the article from The New York Times if the conditioning is required. In these circumstances it has
authors had reported on total pesticide use throughout been claimed that switching from incandescent bulbs
Western Europe instead of selectively focusing on to LEDs will increase overall emissions of CO2 , since
France alone? the zero-carbon “waste” heat from incandescent bulbs
must be replaced by furnace heat generated by burning
37. The European Union has a “sustainable use natural gas. Analyze this claim.
directive” concerning pesticides, which you can read
about in the document EU Policy for a sustainable 43. Calculate the reverse Carnot limit for the effi-
use of pesticides [67]. In accordance with this directive ciency of an air conditioner where the outside tem-
France has sought to dramatically cut its use of pesti- perature is 95 ◦ F and the inside temperature is 78 ◦ F.
cides under a program entitled “Ecophyto 2018.” Com- (Don’t forget to convert the Fahrenheit temperatures to
ment on the potential importance of this information kelvins before you do the calculation!) Express this as
in relation to the comparison of pesticide use between an energy efficiency ratio in Btu per watt-hour. How
France and the United States. does this compare with the typical SEER of 15 for a
real AC unit?
7.6. NUCLEAR POWER IS A COMMIT- 44. The Washington Post of May 31, 2016, re-
MENT TO THE FUTURE ports, The world is about to install 700 million air
conditioners. Here’s what that means for the climate,
available at [225]. Write a critical review of this article.
Pay particular attention to the meaning of the phrase
“emissions avoided.”
39. Identify examples in this piece of quantities put

45. David MacKay [202] estimated that if everyone
into human terms. Can you think of other compar-
in some British cities used ground-source heat pumps
isons?
to heat their homes, they would pump out so much heat
that they would cause a “permafrost” layer to develop
40. Research the thorium cycle, a “breeder” reactor such as now exists only in Arctic regions. Investigate
program relying on uranium-233 as the fissile mate- this claim and see whether it seems plausible.
rial. Uranium-233 can be “bred” from thorium-232,
a relatively abundant element, in the same way that
46. Why must the reverse Carnot limit (for a heat
plutonium-239 can be “bred” from uranium-238. What
pump) be no more than the reciprocal of the Carnot
are the potential advantages and disadvantages of this
limit (for a heat engine)? (Hint: Suppose not, and try
process when compared to current reactors?
to design a perpetual-motion machine.)
7.7. USING ELECTRICITY EFFICIENTLY 7.8. GROWTH AND PAYBACK TIME FOR
AT HOME SOLAR ENERGY

41. A new Kickstarter project, the NanoLight, claims
that it will revolutionize home lighting. The NanoLight
is said to produce a light output of 1100 lumens while
using “less than half a watt” of electricity. Would you 48. The analysis in Case Study 7.8 considers costs
fund this project? Explain your reasoning. and benefits spread out over time, but makes no men-
tion of discounting (Section 6.1). Do you think this
is correct? If you believe that discounting needs to be systems. The ratings are on the basis of electrical
included in the analysis, how would you do it? energy that can be produced from the stored energy
[41].
49. According to a life-cycle analysis of vehi- Storage ESOEI
cles [354], it takes about 120,000 megajoules (MJ) of Compressed air energy storage 240
energy to produce a typical American car. One gallon Pumped hydroelectric storage 210
of gasoline contains 120 MJ of energy. About how Pebble bed thermal storage 62
many miles does a car with a fuel economy of 25 mpg Solar salt thermal storage 47
need to drive before it has used more energy in its use Lithium-ion battery 10
than in its production? Sodium-sulfur battery 6
Lead-acid battery 2
50. Referring to the previous problem, suppose you (a) What are some reasons why compressed
currently drive a car with a fuel economy of 25 mpg. air storage and pumped hydroelectric storage
Calculate the energy payback time of purchasing a new would have much higher ESOI ratings than
fuel-efficient car that gets 45 mpg. That is, how many battery storage systems?
miles do you need to drive the new car before the (b) Discuss potential benefits and limitations of the
fuel savings offset the energy used to produce the new various storage systems.
car? How long will this take (if you maintain the same
driving habits)?
53. An energy storage alternative in development
(with no ESOI rating as yet) is rail storage: train-cars
7.9. ENERGY RETURN ON ENERGY full of rocks are moved uphill when there is a surplus
INVESTED of energy; this energy can be put back into the grid by
allowing the train to move downhill, generating power
when needed. The first such project, in Nevada, has a
51. U.S. ethanol production from corn was almost 15
storage capacity of 12.5 MWh [209]. This is a relatively
billion gallons in 2012. How does this compare with
small project that can help smooth transitions between
the amount of gasoline consumed in the U.S. in 2012? different fuels when renewable sources are intermittent.
Can you put this in human terms? The entire 12.5 MWh of energy can be delivered over
the course of 15 minutes as the train moves downhill.
52. Energy storage is an important component of any What is the power output of the train as it descends
renewable energy system. For energy storage systems (assume a constant power output over the 15 minutes)?
there is a measure analogous to EROEI that allows us
to compare different systems, ESOI: energy stored on
energy invested. This is a ratio of the total amount of 54. We could also define a measure like EROEI and
energy that can be stored over the life of the system ESOI for comparing investments in energy efficiency.
over the amount of energy required to build and main- Describe how this might work in the case of choosing
tain the system. The table below gives estimates of to invest in switching to LED lighting (see Section 7.7
the ESOI for batteries and several other energy storage on page 458).
Part III
Resources
C HAPTER 8
Resources for Student and Instructor
This part of the book brings together additional information for both student and instructor. We begin with a few
suggestions for further reading or different ways in which you could pursue some of the themes of this book
more deeply. Then a series of tables lists useful data about our Earth and about various environmental influences.
These are followed by a list of the book’s illustrations (and their sources); by the bibliography, which contains all
the references that are cited in the text; and finally by the index for the book.
8.1 Resources for Further Reading and Writing

This book is intended to be self-contained. But for the student (or the instructor for that matter) who wants to
delve deeper, here are some suggestions for additional resources. Many more resources are cited in the individual
chapters.
• On sustainability in general: The Bridge at the Edge of the World [305]; Earth: An Operator’s
Manual [23]; Environmental Issues: Looking Towards a Sustainable Future [212].
• Chapter 1: Guesstimation [345].
• Chapter 2: Thinking in Systems [214].
• Chapter 3: Networks, Crowds and Markets [100].
• Chapter 4: Consider a Spherical Cow [156].
• Chapter 5: Probability for the Enthusiastic Beginner [227].
• Chapter 6: Game Theory and the Humanities [52].
• On writing: Student Writing in the Quantitative Disciplines [38].
• More advanced texts: Ecological Economics [87]; Environmental Physics [51]; Mathematics and
Climate [177]; Principles of Planetary Climate [265].
8.2 Useful Numbers for Sustainability Calculations

The idea for this section comes from the appendix to John Harte’s book [156], titled simply “Useful Numbers.”
As Harte does, we remind the reader of the need to observe appropriate precision in working with these numbers.
The precision of the figures given here is no greater than that implied by the number of significant figures used.
(See Section 1.3.1 for general information about the idea of precision, and Section 1.3.2 for the specific idea of
significant figures.) It may be less, in which case we will try to say so. Sources for the information contained
herein are listed at the end.
8.2.1 Units and Conversions

This section repeats some information from the tables in Chapters 1 and 2, and gives some additional conversions
as well. In this section only, when we use an equality sign (=) it refers to exact equality. We will use an

https://doi.org/10.1007/978-3-319-76660-7_8
484 CHAPTER 8. RESOURCES FOR STUDENT AND INSTRUCTOR
approximation sign (≈) to indicate equality up to the number of significant figures shown. For reference purposes,
some of the approximations are given here to a higher degree of precision (more significant figures) than the data
in Chapters 1 and 2 above.
Time
The fundamental unit of time is the second (abbreviated s or sec). Other units are derived from the second as
follows:
1 minute (min) = 60 sec,

1 hour (hr) = 60 min = 3600 sec,
1 day = 24 hr = 8.64 × 104 sec,
1 year (yr) ≈ 365.24 day ≈ 3.1557 × 107 sec.
(Astronomically, one year is the length of time the Earth takes to complete one orbit around the Sun. Because
this is not a whole number of days, our standard calendar year is 365 days with leap years of 366 days, roughly
every four years, to make up the difference.)
Length
The fundamental unit of length is the meter (abbreviated m). Other metric units of length are obtained by applying
prefixes, like millimeter (mm, 10−3 m), centimeter (cm, 10−2 m), kilometer (km, 103 m) and so on.
U.S. units of length are
1 inch (in) = 0.0254 m,

1 foot (ft) = 12 in ≈ 0.305 m,
1 yard (yd) = 3 ft ≈ 0.914 m,
1 mile (mi) = 1, 760 yd = 5, 280 ft ≈ 1.61 km.
The conversion from inches to meters above is exact (it is now the definition of an inch). The other metric
conversions are approximations, valid to the number of significant figures shown.
Area
As explained in Section 1.1, area units are obtained by “squaring” corresponding length units (that is, raising
them to the power 2). Thus the fundamental unit of area in the metric system is the square meter, denoted by m2 ,
and in U.S. units we have the square foot ( ft2 ), and its counterparts for the other length units.
In addition:
1 acre = 4, 840 yd2 ≈ 4, 047 m2 ,

1 hectare (ha) = 104 m2 ≈ 2.47 acres.
Volume
As explained in Section 1.1, volume units are obtained by “cubing” corresponding length units (that is, raising
them to the power 3). Thus the fundamental unit of volume in the metric system is the cubic meter, denoted by
m3 , and in U.S. units we have the cubic foot ( ft3 ), and its counterparts for the other length units.
In addition:
1 U.S. gallon (gal) = 231 in3 ≈ 3.7854 × 10−3 m3 ,

1 U.S. pint = 0.125 gal ≈ 4.73 × 10−4 m3 or 473 cm3 ,
1 barrel of petroleum (bbl) = 42 gallons ≈ 0.159 m3 ,
1 liter (l) = 10−3 m3 ≈ 2.11 pints.
8.2. USEFUL NUMBERS FOR SUSTAINABILITY CALCULATIONS 485
Weight or Mass
The standard metric unit of mass is the kilogram (kg), equal to 1,000 grams (g). A metric ton or tonne is 1,000
kilograms.
U.S. units are:
1 ounce (oz) ≈ 28.35 g or 2.835 × 10−2 kg,

1 pound (lb) = 16 oz ≈ 0.4536 kg,
1 U.S. ton = 2, 000 lb ≈ 907 kg.
The U.S. ton is also sometimes called the short ton. A U.S. ton and a metric ton are not the same, but they are
within 10% of each other, which is often close enough for the kinds of calculations we are doing.
Energy and Power

The standard metric unit of energy is the joule (J). Power is the rate of flow of energy; the metric unit of power is
a watt (W), meaning an energy flow rate of one joule per second.
There are a variety of U.S. units both for energy and power. In the list below, all except the last two are energy
units; the horsepower and Btu/hr are units of power:
1 foot-pound ≈ 1.356 J,
1 British thermal unit (Btu) ≈ 1, 055 J,
1 (food) calorie (Cal) ≈ 4, 184 J,
1 kilowatt-hour (kWh) = 3.6 × 106 J ≈ 3412 Btu,
1 Quad = 1015 Btu ≈ 1.055 × 1018 J,
1 horsepower ≈ 746 W,
1 Btu/ hr ≈ 0.293 W.
Several of these conventional units (British thermal unit, calorie) are defined in terms of the properties of water.
See the section “Water” below for more details. Information about specific energy sources is in the “Energy”
section below.
Temperature
Temperature can be measured on the Celsius, Fahrenheit, or Kelvin scale. The formulas for conversion between
these scales are given in Table 1.
Table 1: Temperature Conversions

◦C ◦F K
◦C C =C C = 59 (F − 32) C = K − 273.15
◦F F = 32 + 95 C F =F F = 95 K − 459.67
K K = C + 273.15 K = 59 F + 255.37 K=K
Stefan’s law (see Rule 1 on page 109) gives the thermal power output, per unit area, of an idealized physical
object at a temperature of T kelvins. The formula is σ T 4 , where Stefan’s constant σ is approximately 5.67 ×
10−8 W/ m2 K4 .
Metric Prefixes
Listed in Table 2 are some of the more commonly used standard prefixes in the metric system and the powers of
10 that they represent. For example, 1 gigajoule = 109 joules and is abbreviated 1 GJ.
Table 2: Metric Prefixes

Prefix nano micro milli centi kilo mega giga tera peta
Power of 10 10−9 10−6 10−3 10−2 103 106 109 1012 1015
Abbreviation n µ m c k M G T P
8.2.2 About Our Planet

Source for physical data: NASA Earth Fact Sheet [232].
Radius of Earth ≈ 6, 380 km ≈ 3, 960 miles,

Mass of Earth ≈ 5.97 × 1024 kg ≈ 6.58 × 1021 U.S. tons,
Radius of Orbit ≈ 1.50 × 108 km ≈ 9.30 × 107 miles,
Surface Area ≈ 5.10 × 1014 m2 ≈ 2.0 × 108 square miles,
Land Surface Area ≈ 1.48 × 1014 m2 ≈ 5.7 × 107 square miles,
Ocean Surface Area ≈ 3.6 × 1014 m2 ≈ 1.4 × 108 square miles,
Albedo ≈ 0.306,
Solar constant ≈ 1361 W/ m2 ,
Mean surface temperature ≈ 288 K,
Mass of atmosphere ≈ 5.1 × 1018 kg,
Pressure of atmosphere, at surface ≈ 14.7 lb/ in2 ≈ 1 kg/ cm2 ,
Mass of all water ≈ 1.4 × 1021 kg,
Gravitational constant at surface ≈ 9.8 J/ kg m.
The gravitational constant is the amount of energy it takes to raise one kilogram through one meter (or the
amount of potential energy released when that one kilogram descends by one meter). Some of you will know that
this is also the acceleration due to gravity—a falling object increases its speed by 9.8 meters per second every
second—but the formulation in terms of potential energy is the one we use most often in the text.
8.2.3 Water
Physical Properties
Basic physical properties of water (H2 O). Density is weight for a given volume; heat capacity, the amount of
heat energy it takes to heat a given weight of water by a given amount; latent heats of melting or evaporation,
the amount of extra heat it takes to turn a given weight of ice (at freezing point) to liquid water, or turn a given
weight of water (at boiling point) to steam.
Density ≈ 103 kg/ m3 ≈ 8.3 lb/ gal ≈ 62 lb/ ft3 ,

Heat capacity = 1 Btu/ lb ◦ F ≈ 4, 200 J/ kg ◦ C,
Latent heat of evaporation ≈ 2.3 × 106 J/ kg ≈ 971 Btu/ lb,
Latent heat of melting ≈ 3.3 × 105 J/ kg ≈ 143 Btu/ lb.
Water on Earth
Data from the U.S. Geological Survey [264]. Antarctic data from [201].
Volume of all water ≈ 1.39 × 1018 m3 ,

Volume of all liquid fresh water, including groundwater ≈ 1.06 × 1016 m3 ,
Volume of surface fresh water (lakes, rivers) ≈ 9.3 × 1013 m3 ,
Volume of (fresh) water in all ice caps/sheets ≈ 2.4 × 1016 m3 ,
Volume of Greenland ice cap ≈ 2.8 × 1015 m3 ,
Volume of West Antarctic ice sheet ≈ 2.2 × 1015 m3 ,
Volume of all Antarctic ice sheets ≈ 2.6 × 1016 m3 ,
Volume of water in atmosphere (water vapor, clouds) ≈ 1.3 × 1013 m3 ,
Density of sea water ≈ 1.03 × 103 kg/ m3 ,
Annual precipitation ≈ 5 × 1014 m3 / yr.
8.2.4 Energy
Energy Flows
Table 8.2.4 data gives rates of energy flow for various natural and human-related processes, measured in terawatts
(1 terawatt (TW) = 1012 W). Sources: [103], [191], [145], [156], [352].
Table 3: Energy Flows
Process Energy flow (TW)

Power output of Sun 3.8 × 1014
Total solar radiation hitting the top of Earth’s atmo- 170,000
sphere
Total solar radiation hitting Earth’s surface (approxi- 80,000–120,000
mate)
Net primary energy productivity (solar energy taken up 75–125
by photosynthesizing organisms)
Human appropriation of primary energy productivity 12–20
Global energy consumption 18–20
U.S. energy consumption 3
Energy content of food consumed by Earth’s human 1
population
Nonrenewable Energy Resources

These estimates for the world’s nonrenewable energy resources are mostly obtained from the work of the World
Energy Council [37], an international organization founded in 1923. For reasons explained in the footnote
on page 253, the figures provided are, at best, approximations. This applies especially to the estimates for
“unconventional” oil (shale oil, tar sands, extra-heavy oils) and gas (shale gas, especially “fracked” gas). The
numbers given refer to the amount of heat energy available from these fuels: multiply by the relevant efficiency
(Definition 6), which is a number less than 1, to find the amount of energy available for any other purpose.
Available energy is given in quads; to convert to joules, multiply by 1018 .
A note on nuclear energy: The only nuclear-energy technology in commercial use today makes use of only
a fraction of the energy contained in natural uranium. Use of uranium (and possibly thorium also) in breeder
reactors could increase the amount of available energy by a factor of 100 or more, but the technology has not been
demonstrated on a commercial scale. Similarly, if nuclear fusion should ever become practicable, the lithium and
deuterium content of the world’s oceans could yield a trillion quads or more of energy. But no controlled fusion
process has yet been exhibited that generates more energy than it consumes.
For the general notion of energy return on energy invested (EROEI), see Case Study 7.9.
Table 4: Estimated Global Energy Reserves
Energy resource Global reserves (quads)

Coal 33,000
Oil 9,500
(Unconventional and shale oil, estimated) 30,000
Natural gas 6,800
(Shale and fracked gas, estimated) 14,000
Uranium (used in conventional reactors) 4,700
Facts About Fossil Fuels

Table 5 gives useful facts about various fossil fuels. The density of the fuel is the mass per unit volume; the energy
content is the amount of heat energy released by burning a fixed amount of the fuel. The carbon intensity of the
fuel tells us how much carbon dioxide is emitted by burning enough of the fuel to yield a fixed amount of heat
energy. It is measured in pounds of CO2 per million Btu (using U.S. units). Data from [156] and U.S. Energy
Information Administration [104].
Table 5: Fossil Fuel Facts

Fuel Density Energy content Carbon intensity,
in pounds CO2 per
million Btu
Coal 1200–1800 kg/ m3 or 3 × 107 J/ kg 205–230 lb/MBtu
about 100 lb/ ft3
Heating oil or 850 kg/ m3 or 7 4.3 × 107 J/ kg or 161 lb/MBtu
diesel oil lb/ gal 1.3 × 105 Btu/ gal
Gasoline 750 kg/ m3 or 6.2 4.3 × 107 J/ kg or 157 lb/MBtu
lb/ gal 1.2 × 105 Btu/ gal or
1.2 × 108 J/ gal
Propane 580 kg/ m3 (lique- 5 × 107 J/ kg or 2 × 139 lb/MBtu
fied) 105 Btu/ lb
Natural gas 450 kg/ m3 (lique- 5 × 107 J/ kg or 2 × 117 lb/MBtu
fied) 105 Btu/ lb
The carbon intensity figures refer only to the carbon dioxide emitted at the point of use (where the fuel is
burned). For example, burning hydrogen does not release any CO2 directly; but the production of hydrogen by the
most standard method (reforming natural gas) releases quite a bit. You can find estimates of the full cycle carbon
intensity of various fuels (including these “production costs”) at various places online. These are important data,
but we have not included them here because they are not purely physical facts like those in the table above.
Note that 27.3% of CO2 by weight consists of carbon.
8.2.5 Living Things

Human Beings
Population: Data on worldwide human population through time; earlier figures are estimates. From the
Population Reference Bureau, a Washington, DC-based nonprofit organization.
Date Human Population

2017 7.5 × 109
2012 7 × 109
2000 6 × 109
1960 3 × 109
1800 1 × 109
1500 5 × 108
0 3 × 108
8000 BCE 5 × 106
Table 6: Facts About “Standard” Human Adults

Characteristic Males Females
Weight 70 kg or 155 pounds 58 kg or 128 pounds
Weight (U.S. average) 88 kg or 195 pounds 75 kg or 165 pounds
Percentage water by mass 60% 50%
Food energy intake (U.S. adults) 2700 calories per day 2200 calories per day
Food energy intake (global estimate) 2300 calories per day 1900 calories per day
Carbon dioxide exhaled 1 kilogram per day or 2 pounds per day (approximate)
Sources: [156], [352], [121], [26]. One must be careful with data about “caloric intake” found online. Much
commonly cited data is based on the amount of food that passes through national accounts (sometimes called
“food disappearance measures”), and therefore does not take account of wastage, which is usually significant—
by some estimates up to 30% of food supplies are wasted. The U.S. data above comes from 24-hour recall
surveys [121] modified for underreporting [26]. The global figures are estimated from food disappearance
data [352] modified for wastage.
Other Living Things

The table below gives estimates for the total weight of various kinds of living things on Earth. The weight is
calculated “as is” (including water). Information from [140], [351].
Type of Organism Total Weight (in millions of tons)

Humans 400
Domesticated animals 700
Wild mammals 30
Termites 500
Fish 800–2,000
Cereal crops 2,000–3,000
All plants 400,000 or more
Bacteria May be of similar magnitude to plants
Note that many living organisms carry substantial populations of bacteria. For instance, the human colon is
estimated to contain 220 g of bacteria, representing over 3 × 1011 individual bacterial cells.
8.2.6 Risks
Chances of various “risky” events, U.S. average. Sources: [287], [252], [187], [272]:
Event Probability
Accidental death traveling by car 1,000 miles 7.3 × 10−6
Accidental death traveling by motorcycle 1,000 miles 2.12 × 10−4
Accidental death traveling by rail 1000 miles 1.5 × 10−7
Accidental death traveling by (commercial) air 1000 miles 7 × 10−8
Getting lung cancer over a 1-year period 7 × 10−4
Death in terrorist attack, over 1 year 7 × 10−8
Accidental death from falling, over 1 year 2 × 10−5
Major earthquake somewhere in San Francisco Bay Area 0.03 (approx)
over next year
Major hurricane (Category 3 or higher) strikes Miami, FL, 0.04 (approx)
over next year
Major solar storm, global impact, strikes during next year 0.01 (estimated)
U.S. stock prices rise by more than 50% in a calendar year 0.02–0.03
(empirical probability from past 200 years)
U.S. stock prices fall by more than 40% in a calendar year 0.01–0.02
(empirical probability from past 200 years)
Creative Commons and GNU Free Documentation Licensing: In the lists of figures and tables, the term “Creative
Commons License” or “GNU Free Documentation License” refers to one of several different licenses as indicated in the
list below. The details of any particular license are available at the url provided.
• Creative Commons License (CC BY 2.0): “Creative Commons Attribution 2.0 Generic,” http://creativecommons.
org/licenses/by/2.0/
• Creative Commons License (CC BY-ND 2.0): “Creative Commons Attribution-NoDerivs 2.0 Generic,” http://
creativecommons.org/licenses/by-nd/2.0/
• Creative Commons License (CC BY-SA 3.0): “Creative Commons Attribution-ShareAlike 3.0 Unported,” http://
creativecommons.org/licenses/by-sa/3.0/
• GNU Free Documentation License (GFDL v1.2): http://www.gnu.org/licenses/old-licenses/fdl-1.2.

en.html
List of Figures
Figures for the Introduction ix

1 Some measures of the “size” of humanity (from [308]). Source: From [308]. Reprinted with permission. . . . . . x
2 The “blue marble.” Source: NASA. Public domain. http://tinyurl.com/y6vmy47o . . . . . . . . . . . . . . xi
3 Marine rescue operation in Galveston, Texas, following Hurricane Harvey, August 31, 2017. Source: Sgt. Ian
Ferro. Public domain. Marine Corp Photo 170831-M-TI204-002.JPG. http://tinyurl.com/ycsxqhb9 . . . . xii
4 Sustainability and limits. Redrawn from [65]. Source: From [65]. Redrawn with permission. . . . . . . . . . . . xiii
5 Pope Francis has said: “These ancient stories...bear witness to a conviction which we today share: that everything
is interconnected, and that genuine care for our own lives and our relationships to nature is inseparable from
. . . justice and faithfulness to others” [123]. Source: a katz/Shutterstock. Standard license. Stock photo ID:
320559005. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi
Figures for Chapter 1 3

1 Measuring instruments. Source: pio3/Shutterstock. Standard license. Stock vector ID: 80904334. . . . . . . . . 3
2 Boeing 747 at takeoff. Source: Adrian Pingstone/Wikimedia Commons. Oct 21, 2007. Public domain. http:
//tinyurl.com/y9d3x75a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3 Illustrating the formation of units of area. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . 9
4 Illustrating the formation of units of volume. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . 11
5 Celsius and Fahrenheit comparison. Source: Gringer/Wikipedia. Public domain. http://tinyurl.com/gku4uqd 14
6 A grown elephant can weigh 4 tons. Source: Ikiwaner/Wikipedia. GNU Free Documentation License version 1.2.
http://tinyurl.com/y9xc3azr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7 All the water on Earth would fit into a “raindrop” 430 miles in radius. Source: Produced by author using TIKZ.
Photo of Earth from NASA. Public domain. http://earthobservatory.nasa.gov/resources/blogs/
earthday_day_lrg.jpg. Image of water texture. Public domain. . . . . . . . . . . . . . . . . . . . . . . . . . 20
8 Standing by the Sun on the Sagan Planet Walk. Source: Author photo. . . . . . . . . . . . . . . . . . . . . . . . 21
9 A smart phone calculator app. Turning the phone on its side changes the display and the set of buttons available.
Source: Screen captures by author. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
https://doi.org/10.1007/978-3-319-76660-7
492 List of Figures
10 A smart phone calculator app. To enter numbers in scientific notation, use the E button. Results can be displayed
in the usual way (“FIXED” mode, shown on the left) or using scientific notation by choosing the “FLOAT” mode
(center and right). Source: Screen captures by author. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
11 Dinosaur skeleton. Source: Davide Costanzo/Flickr. Nov 11, 2011. Creative Commons license (CC BY 2.0).
http://tinyurl.com/y97zsbqs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
12 Planes at San Francisco Airport. Source: Erik Pesik/Flickr. Jun 29, 2007. Creative Commons license (CC BY
2.0). http://tinyurl.com/y77e7vsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
13 Fireworks. Source: TMAB2003/Flickr. June 27, 2009. Creative Commons license (CC BY-ND 2.0). http://
tinyurl.com/ycy54gva . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
14 Ocean currents and garbage. Source: Greenpeace [167]. 2012. Reprinted with Permission. . . . . . . . . . . . . 46
15 Bar chart representation of the data in Table 11 on page 47 on U.S. peak generating capacity from various sources
(gigawatts). Source: Produced by author using TIKZ. Data Source: Table /vrefCD1 . . . . . . . . . . . . . . . . 49
16 Pie chart representation of the data in Table 11 on page 47 on U.S. peak generating capacity from various sources
(gigawatts). Source: Produced by author using TIKZ. Data Source: Table 11 on page 47 . . . . . . . . . . . . . . 49
17 Scatter plot of data from Table 12 on page 47 on U.S. natural gas production (billions of cubic feet per quarter).
Source: Produced by author using TIKZ. Data from Table 12 on page 47 . . . . . . . . . . . . . . . . . . . . . . 50
18 The Keeling curve, which shows atmospheric CO2 concentrations in parts per million by volume (ppmv). Source:
NOAA/Wikimedia Commons. May 12, 2013. Creative Commons license (CC BY-SA 3.0).http://tinyurl.
com/y9vsxaxy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
19 Scatter plot of the data from Table 12 on page 47 on U.S. natural gas production (billions of cubic feet per quarter).
Source: Produced by author using TIKZ. Data from Table 12 on page 47 . . . . . . . . . . . . . . . . . . . . . . 51
20 Obama administration (2015) graphic of high school graduation rates. Source: White House tweet [245]. . . . . . 52
21 Senate Budget Committee (2012) graphic of welfare rolls. Source: Senate budget committee graphic [294].
Accessed via [149]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
22 Body temperature plot. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
23 Body temperature plot starting at zero. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . 52
24 Prevalence of West Nile virus. Avoid the use of three-dimensional charts. Source: From [319]. Original source
unknown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
25 U.S. annual energy consumption by fuel source as a percentage of total (2016). Avoid using 3-dimensional charts.
Source: Produced by the author using TIKZ. Data source: [191] TIKZ . . . . . . . . . . . . . . . . . . . . . . . . 54
26 Average temperatures in PA plotted against time. Source: Produced by author using TIKZ . . . . . . . . . . . . . 54
27 Global monthly mean temp anomaly1997–2014, data from [119]. Source: Produced by author using TIKZ. Data
Source: [119] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
28 Sen. J. Inhofe carries a snowball into the Senate to “refute” global warming. Source: Wikimedia Commons.
Public Domain. Uploaded April 28, 2017. http://tinyurl.com/yan47y2w . . . . . . . . . . . . . . . . . . . 55
29 Global monthly mean temp anomaly 1880–2016, data from NASA [119]. Source: Produced by author using
TIKZ . Data Source:[119] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
30 Five-pronged industrial AC plug. Source: ER 09/Shutterstock. Standard License. Stock photo ID: 63751945. . . 60

1 Heraclitus, painting by Johannes Moreelse. Source: Wikipedia. Public domain. http://tinyurl.com/y8layo4b 67
2 A bathtub is a standard example of a stock-flow system. Source: Erica Nicol/Flickr. Sept 25, 2006. Creative
Commons license (CC-BY-SA 2.0). http://tinyurl.com/ybl7xvzn . . . . . . . . . . . . . . . . . . . . . . 70
3 System diagram for the bathtub. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . 70
4 Stock-flow system for a primary cell, such as an AA battery. Source: Produced by author using TIKZ . . . . . . . 71
5 The Earth’s carbon cycle from [271]. The white numbers in the figure refer to stock levels, measured in gigatons
(billions of tons) of carbon. The yellow numbers refer to natural flows, in billions of tons of carbon per year. The
red numbers refer to human-caused flows, or human-caused changes to the natural flows. Source: NASA Earth
Observatory [271]. Public Domain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6 Simple stock-flow model of radon concentration in a home. Source: Produced by author using TIKZ . . . . . . . 75
7 A student ready to graduate. “Residence time” in college is the average length of time from matriculation to
graduation. Source: Courtesy of R. Hoellman . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
8 A globe. Source: Wendy Cope/Flickr. Aug 15, 2015. Creative Commons license (CC BY 2.0). http://
tinyurl.com/nxjn2k8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
9 The Wason four-card task. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . 78
10 Rowing machine. Image courtesy of Concept 2. Source: Concept 2. Reprinted with permission. http://www.
concept2.com/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
11 Energy stock-flow diagram for athlete and rowing machine. Source: Produced by author using TIKZ . . . . . . . 82
List of Figures 493
12 Rappel device. Source: Roylindman/Wikipedia. May 17, 2011. Creative Commons License (CC BY-SA 3.0)
http://tinyurl.com/yayg6nqt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
13 The Hoover Dam (Ansel Adams photo). Source: [4] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
14 Perpetual Motion by Norman Rockwell. Source: Wikimedia Commons. Public domain. http://tinyurl.com/
y73c4g7u . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
15 Nuclear thermoelectric power system for Mars rover. Source: Idaho National Laboratory/Flickr. Feb. 7, 2009.
Creative Commons License (CC BY 2.0) http://tinyurl.com/yc8m4kxd . . . . . . . . . . . . . . . . . . . 89
16 Cooling towers. Source: zootalures/Wikimedia Commons. May 22, 2004. Creative Commons License (CC BY-
SA 3.0). http://tinyurl.com/y96aeykj . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
17 Incandescent bulb. Source: James Bowe/Flickr. Mar. 8, 2009. Creative Commons License (CC BY 2.0) http:
//tinyurl.com/ze4ypqn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
18 Energy flows in the U.S. in 2016. Source: Lawrence Livermore National Laboratory [191] . . . . . . . . . . . . 93
19 Stock-flow model of the bathtub example. The thick blue arrows denote flows. The red control arrow indicates a
relationship between the water level and the outflow through the drain. Source: Produced by author using TIKZ . 96
20 Equilibrium for the bathtub model. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . 96
21 Lakeside factory. Source: ttarasuik/Flickr. Nov. 20, 2009. Creative Commons License (CC BY 2.0). http://
tinyurl.com/ybhdotgh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
22 Energy consumption in U.S. homes by end use (2009). Source: Data from EIA Residential Energy Consumption
Survey [5] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
23 Thermographic image, showing heat loss through uninsulated walls. Source: Ivan Smuk/Shutterstock. Standard
License. Stock photo ID: 543798988 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
24 Stock-flow diagram for a model of home heating. Source: Produced by author using TIKZ . . . . . . . . . . . . . 100
25 Furnace rating plate. Source: Author photograph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
26 Stock-flow diagram with outside temperature as a parameter. Source: Produced by author using TIKZ . . . . . . . 103
27 Two-box model for hemispheric ethane flow, after Harte [156]. Source: Produced by author using TIKZ . . . . . 103
28 Stock-flow diagram for two-layer insulating sandwich. Source: Produced by author using TIKZ . . . . . . . . . . 104
29 The naked planet model. Notice that the wavelength of the incoming radiation is governed by the surface
temperature of the Sun (relatively high, hence short wavelength), whereas the wavelength of the outgoing radiation
is governed by the surface temperature of the Earth (relatively low, hence long wavelength). Source: Produced
by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
30 The naked planet model as a stock-flow system. Source: Produced by author using TIKZ . . . . . . . . . . . . . 112
31 Equilibrium in the naked planet model occurs at about 255 K. Source: Produced by author using TIKZ . . . . . . 113
32 A greenhouse. Source: Lilla Frerichs/PublicDomainPictures.com. Public Domain. http://tinyurl.com/
yczp2zbp, created by Lilla Frerichs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
33 Svante Arrhenius. Source: Wikimedia Commons. Public Domain. http://tinyurl.com/y8d2alqv . . . . . . 114
34 The “glass ball” greenhouse model as a stock-flow system. Source: Produced by author using TIKZ . . . . . . . . 115
35 Stock-flow climate model with variable greenhouse effect. Source: Produced by author using TIKZ . . . . . . . . 116
36 Figure for Exercise 47 on page 121. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . 122

1 A network. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
2 Two representations of the same network. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . 130
3 One author’s Facebook neighborhood. Source: Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4 Two paths from A to E in the network of Figure 1 on page 127. Source: Produced by author using TIKZ . . . . . 131
5 A network with several components, colored differently. Blue might be considered a “giant component.” Source:
Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
6 Indian vulture. Source: Goran Eckstrom, from [141]. Creative Commons License (CC BY 2.0) . . . . . . . . . . 133
7 Predation links in the Indian vulture crisis. The solid arrows show direct predation; the dashed arrow represents
the effect of dogs on humans via the transmission of rabies. Source: Produced by author using TIKZ . . . . . . . 133
8 Aftermath of Smokey Spur fire, Wisconsin. Source: Wisconsin Department of Natural Resources/Flickr. Aug. 4,
2012. Creative Commons License (CC BY-ND 2.0) http://tinyurl.com/yc39e87q . . . . . . . . . . . . . . 134
9 Figure 1 on page 127, redrawn. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . 135
10 Weak ties link different clusters in a network. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . 136
11 Explanation of the three components of the resilience measure β : from Gao et al. [126]. Source: [126]. Reprinted
with permission. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
12 Vaccination. Source: Daniel Paquet/Flickr. Oct. 22, 2010. Creative Commons License (CC BY 2.0). . . . . . . . 141
13 The bystander effect. Source: Wikimedia Commons. Creative Commons License (CC BY-SA 3.0) http://
tinyurl.com/ydflbmys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
14 The bystander effect. Source: Produced by author using TIKZ. . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
15 Restructuring the social network by appealing to a specific bystander. Source: Produced by author using TIKZ . . 145
494 List of Figures
16 Percentage agreeing with “The Earth is growing warmer, mostly because of human activity” by self-described
political affiliation [61]. Source: Produced by the author using TIKZ. Data Source: [61] . . . . . . . . . . . . . . 146
17 Harary’s triangles. Source: Produced by author using TIKZ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
18 Dynamic behavior of bathtub model. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . 152
19 Examples of oscillation and overshoot and collapse. Source: Produced by author using TIKZ. . . . . . . . . . . . 153
20 Stock-flow model of heat in home. There are two feedback loops. Source: Produced by author using TIKZ . . . . 154
21 Feedback loop for bacteria population. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . 154
22 Growth of bacteria population, starting with a single cell. Source: Produced by author using TIKZ . . . . . . . . 155
23 System diagram for heat in home, signs marked. We mark the red thermostat control arrow with a − because an
increase in temperature tends to decrease the inflow of heat (by shutting off the furnace). We mark the red control
arrow to “heat loss” with a + because an increase in temperature increases the rate of heat loss. Both feedback
loops are stabilizing. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
24 Stock-flow model for metabolism of caffeine in the body. Source: Produced by author using TIKZ . . . . . . . . 157
25 Stock-flow diagrams for the Carters’ electric blanket. Source: Produced by author using TIKZ. . . . . . . . . . . 158
26 Climate model with parameter g that governs the strength of the greenhouse effect (GHE). An increase in g
decreases the outflow of long-wave radiation from the planet. Source: Produced by author using TIKZ . . . . . . 158
27 Stock-flow model for phosphorus in a lake, with a pollution parameter. Source: Produced by author using TIKZ . 159
28 Basic demographic model. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . 160
29 A Canadian lynx. Source: Keith Williams/Wikimedia Commons. Creative Commons License (CC BY 2.0).
http://tinyurl.com/yab287ts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
30 Interaction between lynx and hare populations. Source: Produced by author using TIKZ . . . . . . . . . . . . . . 160
31 Simulation and historical data for hare and lynx populations. Source: Data from [94] based on [107] and [200] . . 161
32 Inventory system with delay. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . 162
33 St. Matthew Island. Source: Carport/Wikimedia. May 24, 2010. Public Domain. Screenshot from NASA World
Wind. http://tinyurl.com/y7tvhjau . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
34 Model for the reindeer of St. Matthew Island. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . 163
35 Growth of $100 at different annual interest rates over 20 years. Source: Produced by author using TIKZ . . . . . 169
36 Basic demographic (population) model. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . 170
37 Demographic model indicating the strength of each control. Source: Produced by author using TIKZ . . . . . . . 171
38 Find nodes that exhibit triadic closure and calculate clustering coefficients. Source: Produced by author using TIKZ 176
39 Find nodes that exhibit triadic closure and calculate clustering coefficients. Source: Produced by author using TIKZ 176
40 Find the clustering coefficients and identify bridges. Source: Produced by author using TIKZ . . . . . . . . . . . 177
41 Stock-flow model for Exercise 31 on page 179; the ice-albedo feedback. Source: Produced by author using TIKZ 180
42 Stock-flow model for Exercise 35 on page 179; the Arctic methane release feedback. Source: Produced by author
using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
43 Stock-flow model for Exercise 37 on page 179; the chemical rock-weathering and biosequestration feedback.
Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

1 A peach. Source: jrmclovin/Flickr. Aug. 2, 2008. Creative Commons License (CC BY-ND 2.0). http://
tinyurl.com/l9wr8b3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
2 John Napier, 1550–1617. Source: Wikimedia Commons. Public Domain. From engraving by Samuel Freeman
(1835) based on portrait (1616) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
3 Graph of y = 3x . Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
4 Plot of log of soybean production. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . 192
5 Simple model for capital formation. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . 195
6 Moore’s law: transistors per microprocessor have doubled about once every two years. Source: Data from [80] . . 196
7 William Stanley Jevons. Source: Wikipedia. Public Domain. http://tinyurl.com/ycae3c9j Public Domain 197
8 Your new residence awaits. Source: rSnapshotPhotos/Shutterstock. Standard License. Stock Photo ID: 238032067 199
9 Radiative forcing (relative to the year 1750) over time, according to the U.S. Environmental Protection Agency.
Source: [19] Data from [247] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
10 Likelihood of various estimated values for the climate sensitivity, after [273]. Source: From [273]. Reprinted with
permission. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
11 Simulation of the constrained growth of a population. Source: Produced by author using TIKZ . . . . . . . . . . 208
12 Donella Meadows, environmental scientist and lead author of The Limits to Growth. Source: The Donella
Meadows Project. Reprinted with permission. http://donellameadows.org/donella-meadows-legacy/
photos-dana/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
13 The logistic model is a feedback loop, but the control function is nonlinear. Source: Produced by author using TIKZ 212
14 Control function for the logistic model. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . 212
15 Stock (blue) and flow rate (red) in a logistic model. Source: Produced by author using TIKZ . . . . . . . . . . . . 213
List of Figures 495
16 Logistic model for nonrenewable resource extraction, e.g., mining. Source: Produced by author using TIKZ . . . 214
17 Graphical representation of growth rates in the logistic model. Source: Produced by author using TIKZ . . . . . . 215
18 Plots of stock against the growth rate for data from a logistic model. Source: Produced by author using TIKZ. . . 216
19 Fitting a logistic model to U.S. oil production, 1930–1999. Source: Produced by author using TIKZ . . . . . . . . 218
20 One example of a “peak oil” production forecast. Source: Gralo/Wikimedia Commons. Modified from [120].
Copyright holder allows unrestricted redistribution. http://tinyurl.com/y8ea8jxk . . . . . . . . . . . . . . 219
21 Bathtub example indicating the strength of the control arrow. Source: Produced by author using TIKZ . . . . . . 222
22 Stock-flow model for caffeine metabolism. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . 223
23 Population model for the U.S. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . 223
24 Stock-flow model for fish farm. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . 225
25 Graph of inflow and outflow of phosphorus when the pollution level is P = 15 tons per year. Source: Produced
26 Plots of the inflow and outflow of phosphorus, indicating stable and unstable equilibrium points when the pollution
level is P = 15 tons/yr. The gray arrows indicate the direction in which the phosphorus stock level will move.
Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
27 The strength of the control is the slope of the red line. Source: Produced by author using TIKZ . . . . . . . . . . 229
28 Stock-flow diagram for lake phosphorus model, showing control strengths at equilibrium A. Source: Produced by
author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
29 Kelp forest at Monterey Bay Aquarium. Source: Tom Bridge/Flickr. Aug. 24, 2010. Creative Commons License
(CC BY-ND 2.0). http://tinyurl.com/mn6u4qx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
30 Kelp forest stock-flow model. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . 231
31 Plots of the inflow and outflow of phosphorus when the pollution rate is P = 25 tons/yr. Source: Produced by
32 Stock-flow diagram for phosphorus at new equilibrium (P = 25 t/ yr). Source: Produced by author using TIKZ . . 234
33 Amplification factor. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
34 Cover of NAS report. Source: National Academies Press . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
35 The naked planet climate model, from Section 2.4. Source: Produced by author using TIKZ . . . . . . . . . . . . 240
36 Graphical representation of equilibrium in the “naked planet” model. Source: Produced by author using TIKZ . . 240
37 Graphical representation of equilibrium taking the greenhouse effect into account. Source: Produced by author
using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
38 Climate model with greenhouse parameter. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . 241
39 Equilibrium at various values of g. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . 242
40 Climate model with ice-albedo feedback. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . 243
41 The ice-albedo feedback. Source: Produced by author using TIKZ. . . . . . . . . . . . . . . . . . . . . . . . . . 244
42 Graphical representations of the climate model with ice-albedo feedback as it approaches a tipping point. Source:
Produced by author using TIKZ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
43 An ordinary light switch is an example of “tipping point” behavior. Source: Steve Cukrov/Shutterstock. Standard
License. Stock Photo ID: 95322187. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
44 Representations of climate model with Earth in a “Snowball Earth” state. Source: Produced by author using TIKZ. 246
45 Graphical representation of model for phosphorus pollution as it approaches a tipping point. Source: Produced
by author using TIKZ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
46 Stock-flow diagram for phosphorus in a lake showing the strength of each control when P = 27 t yr−1 . Source:
47 Critical slowing down in an ecosystem, from van Nes and Scheffer [332]. Source: [332]. Reprinted with permission.249
48 Risks of various climate tipping points at various levels of global warming, from Lenton et al. [192]. Source:
[192]. Reprinted with permission. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
49 Visual representation of Earth at a tipping point. Illustration by Cheng “Lily” Li. Used with permission. Source:
From [42]. Attrib to Cheng (Lily) Li . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
50 A phosphorus model with an unrealistic inflow, used in Exercise 51 on page 257. Source: Produced by author
using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257

1 Is this a risky activity? Source: szanyierika97/Pixabay. Public Domain. http://tinyurl.com/ycahh8ql . . . 259
2 Is this a risky activity? Source: U.S. Department of Transportation/Wikimedia Commons. Public Domain. http:
//tinyurl.com/yauv8wkc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
3 Box plot for heights in example data set. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . 265
4 Box plot of phosphorus data set, with outliers. Data from [130]. Source: Produced by author using TIKZ. Data
from [130] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
5 Histogram of class heights corresponding to frequency table given in Table 3 on page 268. Source: Produced by
496 List of Figures
6 Histogram of phosphorus concentrations corresponding to Table 4 on page 269 [130]. Source: Produced by author
using TIKZ. Data from [130]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
7 Box plot of employee salaries (in thousands of dollars). Note the diamond indicating the mean. Source: Produced
8 Box plot of student scores. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . 272
9 A distribution can be skewed to the left or the right. Source: Produced by author using TIKZ . . . . . . . . . . . 274
10 Illustrating skewness via box plots. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . 275
11 Body mass index dataset, from [263]. Source: [263] Reprinted with permission. . . . . . . . . . . . . . . . . . . 275
12 The MIT “Greenhouse Gamble” spinners. “No policy” at the left; “with policy” at the right. From [219]. Source:
[219]. Reprinted with permission. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
13 Venn diagram for Problem 11 on page 290. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . 290
14 Result of flipping ten virtual coins at random.org [144]. Source: Produced by author using TIKZ. Trial and
individual coin images from “Coin Flipper” at [144]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
15 Probability distributions for obtaining heads in (a) 20 and (b) 200 flips of a fair coin. Source: Produced by author
using TIKZ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
16 A lottery ticket (it lost). Source: Jeffrey Beall/Flickr. July 15, 2006. Creative Commons License (CC BY-ND
2.0). http://tinyurl.com/lenv2dn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
17 Illustration of the random variable W , the payoff for a certain bet on a coin flip. Source: Produced by author using
TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
18 A random variable X based on the roll of a die. Source: Produced by author using TIKZ . . . . . . . . . . . . . . 301
19 Illustrating the random variable that gives the total number of heads in two coin flips. Source: Produced by author
using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
20 Total loss. Source: Dedyukhin Dmitry/Shutterstock. Standard Licensed. Stock photo ID: 39504259 . . . . . . . . 305
21 A risk-averse utility function. Notice how the graph “bends over to the right.” As explained in the text, this
“bending” is a sign of risk aversion. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . 312
22 A utility function that might be used in prospect theory. Source: Produced by author using TIKZ . . . . . . . . . 313
23 Extreme rainfall in Beckley, West Virginia, from [194]. Source: [194] . . . . . . . . . . . . . . . . . . . . . . . 315
24 Venn diagram for problems 1 on page 318 and 2 on page 318 Source: Produced by author using TIKZ . . . . . . 318
25 Cover of [289], which describes the Sally Clark case. Reprinted with permission. Source: [289]. Reprinted with
permission from Hachette Book Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
26 The process of (Bayesian) statistical inference. Source: Produced by author using TIKZ . . . . . . . . . . . . . . 330
27 Posterior distributions for arsenic concentrations. Source: Produced by author using TIKZ . . . . . . . . . . . . . 334
28 Bar chart for Exercise 5 on page 337. U.S. rates of death by age group, 2004. Data obtained from [90]. Original
source: [117]. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338

1 The “blue marble.” Source: NASA. Public domain. http://tinyurl.com/y6vmy47o . . . . . . . . . . . . . . 345
2 Lord Nicholas Stern. Source: London School of Economics. Courtesy of Lord Stern. . . . . . . . . . . . . . . . 356
3 The prisoners’ dilemma [344]. Source: Saturday Morning Breakfast Cereal [344] . . . . . . . . . . . . . . . . . 363
4 Productivity of the commons as a function of the number of cows grazing. Source: Produced by author using TIKZ 375
5 Pile of bison skulls waiting to be ground into fertilizer, ca. 1892 [216]. Source: [216] Public Domain. . . . . . . 379
6 Title page of an Inclosure Act [29]. Source: Adapted from [29] (cropped image). Public Domain. . . . . . . . . 380
7 Tightening of U.S. emissions standards for heavy-duty compression-ignition (diesel) engines. Data from [32].
Source: Produced by author using TIKZ. Data from [32]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
8 The cap and trade system. Source: Produced by the author using TIKZ. Factory graphic from Studio G/Shutterstock.
Standard License. Image ID: 334409558 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
9 Carbon Tax Center. Polling data from [185]. Source: Carbon Tax Center. Reprinted with permission. Polling Data
from [185] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
10 North Face of Half Dome. Karl Bralich photo. Source: Karl Bralich . . . . . . . . . . . . . . . . . . . . . . . . 390
11 Haidt’s five moral foundations or scales (a sixth was added later). Source: See [147]. . . . . . . . . . . . . . . . 391
12 Skipping stones. Source: Cathleen A. Clapper/Shutterstock. Standard License. Stock Photo ID: 828125. . . . . . 395
13 Weight of mammals on earth (from xkcd.com). Source: Webcomic “xkcd.” http://xkcd.com/1338/ . . . . . 396
14 The TP aisle. Source: Krista/Flikr. Creative Commons License (CC BY 2.0) http://tinyurl.com/ybquvyaz 399

1 Bottled water for sale on a store shelf. Source: Yusuf C./Flickr. Mar. 3, 2010. Creative Commons License (CC
BY 2.0).http://tinyurl.com/hsyr7o7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
2 Niagara Falls. Source: jgorzynik/Shutterstock. Standard License. Stock photo ID: 112275293 . . . . . . . . . . . 416
List of Figures 497
3 Annual premature deaths attributable to household air pollution (from [314]). Source: Produced by author using
TIKZ . Data from [314] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
4 Box plot for the data from Table 2 on page 420 on millions of acres burned by wildfires in recent years (2005–
2016). Source: Produced by author using TIKZ. Data from [36] . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
5 Box plot for summary statistics from Table 5 on page 421 on millions of acres burned by wildfires (1960–2015).
Source: Produced by the author using TIKZ. Data from [36] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
6 View from Mt. LeConte in early April, Great Smoky Mountains National Park. Source: Aviator31/Wikipedia.
April 7, 2007. Public Domain. http://tinyurl.com/yd5xcsnm . . . . . . . . . . . . . . . . . . . . . . . . . 422
7 KBDI for the continental U.S. on Nov 26, 2016. Throughout much of the Southeast, the KBDI was above 700. In
the Great Smoky Mountains National Park along the border of TN and NC, the KBDI ranged from 500-750 [317].
Source: U.S. Forest Service Wildland Fire Assessment System [317] . . . . . . . . . . . . . . . . . . . . . . . . 423
8 KBDI for the continental U.S. on Nov 5, 2007. The KBDI for the region around Great Smoky Mountains National
Park is above 500; there are no extensive regions in the Southeast where the index exceeds 700 [317]. Source:
U.S. Forest Service Wildland Fire Assessment System [317] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
9 Predicted changes in the Keetch–Byram Drought Index, 2041–2070 (change relative to the years 1971–2000).
Reprinted from [198]. Source: Future U.S. wildfire potential trends. Reprinted with permission from [198] . . . . 424
10 Management of municipal solid waste (MSW) in the U.S., 2014. Adapted from [15]. Source: Produced by the
author using TIKZ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
11 A one ounce gold bar, the amount of gold in about 28 tonnes of low-grade ore from an open pit gold mine, or in
about 900 smart phones. Source: Stanslavs/Shutterstock. Standard License.Stock photo ID: 746851675 . . . . . 427
12 Sunrise Dam open pit gold mine, Australia. Source: Calistemon/Wikipedia. Oct 12, 2010. Creative Commons
License (CC BY-SA 3.0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
13 Cradle-to-grave life cycle of a plastic water bottle. Source: Produced by author using TIKZ. Images (left
to right): pan demin/Shutterstock. Stock photo ID: 412353814. 06photo/Shutterstock. Image ID: 133014653.
Yvvtt 1b/Flickr. July 13, 2012. Creative Commons (CC BY-SA 2.0). http://tinyurl.com/y9hupk3v Andrey
Arkusha/Shutterstock. Image ID: 484717678. mraoraor/Shutterstock. Stock photo ID: 428424577. . . . . . . . . 428
14 Comparative energy usage for recycled vs. virgin content products. Reprinted with permission from [228].
Source: Comparative LCAs for Curbside Recyling [228] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
15 Materials discarded in municipal solid waste (percentage by weight), 2014. Adapted from [15]. Source: EPA [15] 431
16 Four stages of the demographic transition. Adapted from [279]. Source: Produced by author using TIKZ. Adapted
from [279]. Creative Commons License (CC BY-SA 4.0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
17 Number of deaths per year of children under age five, 1960–2015. Data from [277]. Source: Produced by the
author using TIKZ. Date from [277] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
18 Fertility rates by nation, 1960. Data from [278]. Source: Produced by author using TIKZ. Data from [278] . . . . 436
19 Fertility rates by nation, 2015. Data from [278]. Source: Produced by author using TIKZ. Data from [278] . . . . 437
20 Fertility rates in China, which instituted a one-child policy in 1979, and in Thailand, which adopted a national
population policy in 1971 based on education and voluntary family planning, 1950–2015. Data from [278].
Source: Produced by author using TIKZ. Data from [278] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
21 Fertility rates by level of development, 1950–2015. Data from [278]. Source: Produced by author using TIKZ.
Data from [278]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
22 Population structure of the the world population (2016). The total population in 2016 was estimated at 7.4 billion
people. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
23 Population structure in first generation. The total population is 75 million people. Source: Produced by author
using TIKZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
24 Population structure in the second generation. The people in the age 0–25 class in the first generation have moved
to the second age class (and reproduced themselves). The total population is 110 million people. Source: Produced
25 Percentage of planted acres in the U.S. for GE crops of different types. Bt crops contain an insecticide. HT crops
are herbicide resistant. These traits can be “stacked” in the same plant [248]. Source: USDA Economic Research
Service [248]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
26 Comparison of rapeseed (canola) yields in Canada and Western Europe. Data from FAO [109]. Source: Produced
by author using TIKZ. Data from [109] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
27 Hypothetical changes in yields based on changes in the amount and quality of land in production. Source:
28 Corn yield in bushels per acre [236]. Reprinted with permission. Source: Historical Corn Grain Yields for Indiana
and the U.S. [236] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
29 Herbicide-resistant weeds [129]. Reprinted with permission. Source: A hard look at GM crops [129] . . . . . . . 448
30 Schematic diagram of the fission process. At left, a neutron (blue) strikes a fissile nucleus (magenta). The results
of fission are shown at the right: two fission product nuclei of slightly different sizes (orange and yellow) and
several more neutrons (blue). Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . . . . . . . . 451
31 Schematic of Yucca Mountain site. 1. Canisters of waste, sealed in special casks, are shipped to the site by
truck or train. 2. Shipping casks are removed, and the inner tubes with the waste are placed in steel multilayered
storage containers. 3. An automated system sends storage containers underground to the tunnels. 4. Containers
are stored along the tunnels, on their sides. (Nuclear Regulatory Commission.) Source: U.S. Nuclear Regulatory
Commission, http://www.nrc.gov/images/waste/hlw-disposal/yucca-drawing.jpg . . . . . . . . . 454
32 Use of electricity in U.S. homes, from [9]. Source: Produced by author using TIKZ . . . . . . . . . . . . . . . . 458
33 Whole-house fan remote controller. Source: Author picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
34 U.S. residential installed solar PV capacity, from [293]. Source: Produced by author using TIKZ. Data from [293] 466
35 Energy balance of the global solar PV industry. Source: [85] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
36 The Lucas gusher. Source: Public domain. Original photo by John Trost . . . . . . . . . . . . . . . . . . . . . . 471
37 U.S. ethanol production, in billions of gallons. Source: Produced by author using TIKZ . . . . . . . . . . . . . . 474
List of Tables
Tables for Chapter 1 3

1 Units of Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Units of Length (U.S.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3 Units of Length (Metric) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4 U.S. Units of Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5 Units of Volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
6 Units of Mass (U.S.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
7 Units of Mass (Metric) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
8 Unit Conversions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
9 Temperature Scale Conversions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
10 Volume of Water on Earth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
11 U.S. installed electric generating capacity by source in 2015 (gigawatts) [7]. Source: Produced by the author.
Data Source: [7] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
12 U.S. quarterly natural gas production (billions of cubic feet) [8]. . . . . . . . . . . . . . . . . . . . . . . . . . . 47

1 Amounts of Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2 Conventional Units of Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3 Efficiency of passenger cars by type. Data from the U.S. Department of Energy [331]. . . . . . . . . . . . . . . . 90
4 Notations for Simple Climate Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

1 Calculation of β in Figure 1 on page 127 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
2 Result of Bystander Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
3 Homophily on Global Warming Concern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
4 Time Steps and Rate Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5 The Law of Exponentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
https://doi.org/10.1007/978-3-319-76660-7
List of Tables 499

1 Forested areas in two countries (square miles). Source: UN FAO [109] . . . . . . . . . . . . . . . . . . . . . . . 185
2 Soybean production figures and logarithms. Source: Data from [109] . . . . . . . . . . . . . . . . . . . . . . . . 192
3 U.S. population data. Source: U.S. Census Bureau [160]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
4 U.S. population data, with growth rates. Source: 1900-2000 data from [160], 2010 data from [204] . . . . . . . . 207
5 ACME Corp sales figures. Source: Author example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208

1 Example student data. Source: Randomly generated by authors . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
2 Measured phosphorus concentrations and land-use data for lakes used to evaluate phosphorus loading by
residential area run-off. Source: From [130] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
3 Frequency table for height data from Example 3 on page 263. Source: Mock data produced by author . . . . . . 268
4 Frequency table for the phosphorus concentration data given in Problem 3 on page 266 Source: From [130] . . . 269
5 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
6 Solution to Problem 1 on page 282 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
7 Calculating expectation in the two-dice multiplication problem. Source: Author . . . . . . . . . . . . . . . . . . 306
8 Two-Way Frequency Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
9 Shark Attacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
10 Two-Way Frequency Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
11 Inference Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
12 Breast Cancer Screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
13 Radioactivity Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
14 Methane Contamination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
15 Prosecutor’s Fallacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
16 Biased Coin Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
17 Biased Coin Problem: Posterior Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
18 Biased Coin Problem After New Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
19 Biased Coin Problem: New Posterior Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333

1 Dilithium Power Plant Analysis: Present Values of Gain and Loss . . . . . . . . . . . . . . . . . . . . . . . . . . 355
2 Payoff Matrix for “Rock, Paper, Scissors” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
3 Payoffs for Software Engineers Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
4 Example Payoff Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
5 Payoff Matrix for Polluted Lakes Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
6 Table of Strictly Best Responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
7 Payoff Matrix for Shared Pollution Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
8 Should I Buy a Cow? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
9 Should I Buy a Cow? (We Both Have c Cows!) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
10 Should I Buy a Cow? (We Both Have 1 Cow) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
11 Should I buy a cow? (We both have 5 cows) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
12 Should I Buy a Cow? (We Both Have 6 Cows!) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
13 Should I Buy a Cow? (We Both Have 7 Cows!) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
14 Milk Production Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
15 Should I Buy a Cow? (We Both Have c Cows!) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
16 Should I buy a cow? (We Both Have c Cows and Tax is Payable!) . . . . . . . . . . . . . . . . . . . . . . . . . . 385

1 Rhetorical Appeals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
2 U.S. Wildfire Data, 2005–2016 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
3 U.S. Wildfire Data 2005–2016, Sorted by Acres Burned . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
4 Summary Statistics, 2005–2016 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
5 Summary Statistics, 1960–2016 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
6 Keetch–Byram Drought Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
7 Greenhouse Gas Benefits Associated with Recovery of Materials (2014) . . . . . . . . . . . . . . . . . . . . . . 431
8 Natural Radiation Exposures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
9 Thermal EROEI for Liquid Fuels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
10 EROEI for Electricity Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
500 List of Tables

1 Temperature Conversions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
2 Metric Prefixes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
3 Energy Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
4 Estimated Global Energy Reserves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
5 Fossil Fuel Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
6 Facts About “Standard” Human Adults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
Bibliography
Whenever possible we have used a permanent digital object identifier (doi) to provide an active link to material referenced in
the bibliography. To visit a source online using the doi listed, use the web address http://dx.doi.org/<doi>. For
example, the article “Impact of anthropogenic climate change on wildfire across western U.S. forests” can be accessed at
http://dx.doi.org/10.1073/pnas.1607171113. If you are using the ebook version of the textbook, clicking on a doi
below should take you directly to that source. For each online reference without a DOI, we have provided a URL. Whenever a
URL would span more than one line of text, we have opted to use the URL -shortening service http://tinyurl.com.
[1] Birth Control in Niger: Runaway Birth Rates are a Disaster. The Economist, August 2014.
URL http://tinyurl.com/jbpezy5. Accessed 02-01-2018.
[2] John T. Abatzoglou and A. Park Williams. Impact of anthropogenic climate change on wildfire across western u.s.
forests. Proceedings of the National Academy of Sciences, 113(42):11770–11775, 2016.
doi:10.1073/pnas.1607171113.
[3] Diane Ackerman. The Human Age: The World Shaped by Us. W. W. Norton & Company, New York, 1st edition,
September 2014. ISBN 978-0-393-24074-0.
[4] Ansel Adams. Photograph No. 79-AAB-1. Photograph of the Boulder Dam from Across the Colorado River, 1941.
URL http://catalog.archives.gov/id/519837. Records of the National Park Service, 1785–2006. Record
Group 79; National Archives at College Park, College Park, MD. Accessed 10-05-2017.
[5] EIA (U.S. Energy Information Administration). Residential Energy Consumption Survey, 2009.
URL http://www.eia.gov/consumption/residential/data/2009/. Accessed 11-04-2017.
[6] EIA (U.S. Energy Information Administration). EIA FAQ: How much gasoline does the United States consume? 2015.
URL http://tinyurl.com/k582ttr. Accessed 02-03-2017.
[7] EIA (U.S. Energy Information Administration). Electric Power Annual 2015, 2016.
URL http://www.eia.gov/electricity/annual/. Accessed 11-03-2017.
[8] EIA (U.S. Energy Information Administration). U.S. Dry Natural Gas Production (MMcf), 2016.
URL http://www.eia.gov/dnav/ng/hist/n9070us2m.htm. Accessed 02-02-2018.
[9] EIA (U.S. Energy Information Administration). FAQ: How is electricity used in U.S. homes? 2017.
URL http://tinyurl.com/yc8j99oe. Accessed 06-29-2017.
[10] EIA (U.S. Energy Information Administration). Annual U.S. Field Production of Crude Oil, 2017.
URL http://tinyurl.com/y97nexgz. Accessed 03-24-2017.
[11] EIA (U.S. Energy Information Administration). Total Energy, 2017. URL http://www.eia.gov/totalenergy/.
Accessed 05-15-2017.
[12] EPA (U.S. Environmental Protection Agency). EPA Technical Fact Sheet: Final Rule for Arsenic in Drinking Water,
2001. URL http://tinyurl.com/ya4s5htq. Accessed 02-02-2017.
[13] EPA (U.S. Environmental Protection Agency). Persistent Organic Pollutants: A Global Issue, a Global Response, 2009.
URL http://tinyurl.com/y78yd3gn. Accessed 02-03-2018.

https://doi.org/10.1007/978-3-319-76660-7
502 BIBLIOGRAPHY
[14] EPA (U.S. Environmental Protection Agency). Ecovative, U.S. EPA SBIR Award Recipient: Innovation with
Mushroom Insulation, 2014. URL http://tinyurl.com/y9vsmk36. EPA Small Business Innovation Research
(SBIR) Program. Accessed 01-29-2018.
[15] EPA (U.S. Environmental Protection Agency). Advancing Sustainable Materials Management: 2014 Fact Sheet, 2016.
URL http://tinyurl.com/y7afbp4f. Accessed 05-03-2017.
[16] EPA (U.S. Environmental Protection Agency). A Citizen’s Guide to Radon: The Guide to Protecting Yourself and Your
Family from Radon, 2016. URL http://tinyurl.com/y83s98la. Accessed 02-27-2017.
[17] EPA (U.S. Environmental Protection Agency). Top 10 Tips for Renters, 2017.
URL http://www.energystar.gov/products/top_10_tips_renters. Accessed 03-12-2017.
[18] EPA (U.S. Environmental Protection Agency). Mortality Risk Valuation, 2017.
URL http://www.epa.gov/environmental-economics/mortality-risk-valuation. Accessed 11-08-2017.
[19] EPA (U.S. Environmental Protection Agency). Climate Change Indicators: Climate Forcing, 2017.
URL http://www.epa.gov/climate-indicators/climate-change-indicators-climate-forcing.
[20] EPA (U.S. Environmental Protection Agency). National Primary Drinking Water Regulations, 2017.
URL http://tinyurl.com/ycpb3css. Accessed 12-15-2017.
[21] EPA (U.S. Environmental Protection Agency). Climate Impacts in the Southeast, 2017.
URL http://tinyurl.com/yajldfz7. Accessed 11-04-2017.
[22] Steve Ahillen. Panel starts review of Chimney Tops 2 fire. Knoxville News Sentinel, Feb 2017.
URL http://tinyurl.com/y8mc9mhd. Accessed 04-04-2017.
[23] Richard B. Alley. Earth: The Operators’ Manual. W. W. Norton & Company, New York, April 2011. ISBN
978-0-393-08109-1.
[24] Anthony L. Andrady and Mike A. Neal. Applications and societal benefits of plastics. Philosophical Transactions of
the Royal Society B: Biological Sciences, 364(1526):1977–1984, 2009. ISSN 0962-8436. doi:10.1098/rstb.2008.0304.
[25] K. R. N. Anthony, D. I. Kline, G. Diaz-Pulido, S. Dove, and O. Hoegh-Guldberg. Ocean acidification causes bleaching
and productivity loss in coral reef builders. Proceedings of the National Academy of Sciences, 105(45):17442–17446,
November 2008. ISSN 0027-8424, 1091-6490. doi:10.1073/pnas.0804478105.
[26] Edward Archer, Gregory Hand, and Stephen Blair. Validity of U.S. Nutritional Surveillance: National Health and
Nutrition Examination Survey Caloric Energy Intake Data, 1971-2010. PLOS One, 8:e76632, 2013.
doi:10.1371/journal.pone.0076632.
[27] Aristotle. The Nicomachean Ethics. Oxford World’s Classics. Oxford University Press, Oxford; New York, revised
edition, 2009. ISBN 978-0-19-921361-0. Lesley Brown (ed.).
[28] Hal R. Arkes and Wolfgang Gaissmaier. Psychological Research and the Prostate-Cancer Screening Controversy.
Psychological Science, 23(6):547–553, June 2012. ISSN 0956-7976. doi:10.1177/0956797612437428.
[29] Various authors. Copy of an Act of Parliament for enclosing 5 open fields at Shifnal, 1793.
URL http://tinyurl.com/ybn9r3rb. Shropshire Archives. Identity Document Reference: 539/1/5/3. Accessed
10-08-2017.
[30] Various authors. Growth chart: Fetal length and weight, week by week, 2015.
URL http://tinyurl.com/y7nwzfq6. Accessed 02-03-2018.
[31] Various authors. Signal to Noise Special Edition: GMOs and Our Food, 2015.
URL http://tinyurl.com/qbfmg4j. Accessed 05-03-2017.
[32] Various authors. Heavy-Duty Onroad Engines: Summary of Emission Limits, 2016.
URL http://www.dieselnet.com/standards/us/hd.php. DieselNet. Accessed 10-08-2017.
[33] Various authors. Genome Editing: An Ethical Review. Nuffield Council on Bioethics, 2016.
URL http://tinyurl.com/yas5327x. Accessed 05-03-2017.
BIBLIOGRAPHY 503
[34] Various authors. Climate Leadership Council, 2017. URL http://www.clcouncil.org/. Accessed 10-17-2017.
[35] Various authors. Household Income Percentile Calculator, 2017. URL http://tinyurl.com/ycez7yzh. Accessed
04-21-2017.
[36] Various authors. Total Wildland Fires and Acres (1960–2016), 2017.
URL http://www.nifc.gov/fireInfo/fireInfo_stats_totalFires.html. National Interagency Fire Center.
[37] Various authors. World Energy Council, 2017. URL http://www.worldenergy.org/. Accessed 02-29-2017.
[38] Patrick Bahls. Student Writing in the Quantitative Disciplines: A Guide for College Faculty. Jossey-Bass, 1st edition,
January 2012.
[39] Bang Goes the Theory: Human Power Shower. Clare Mottershead and Rachel Coughlan (producers), 2009.
URL http://www.youtube.com/watch?v=C93cL_zDVIM. BBC. Originally aired Dec. 3, 2009. Accessed
02-28-2017.
[40] The World Bank. World Bank Open Data, 2018. URL http://data.worldbank.org/.
[41] C J Barnhart and S M Benson. On the importance of reducing the energetic and material demands of electrical energy
storage. Energy and Environmental Science, 6:1083, 2013. doi:10.1039/C3EE24040A.
[42] Anthony D. Barnosky, Elizabeth A. Hadly, et al.. Approaching a state shift in Earth’s biosphere. Nature, 486(7401):
52–58, June 2012. doi:10.1038/nature11018. Accessed 02-26-2015.
[43] John M. Barry. The Great Influenza: The Story of the Deadliest Pandemic in History. Penguin Books, New York,
revised edition, October 2005. ISBN 978-0-14-303649-4.
[44] Al Bartlett. Arithmetic, Population and Energy, 2008. URL http://www.youtube.com/watch?v=sI1C9DyIi_8.

Video presentation. Accessed 02-03-2018.
[45] We Believers. Saltwater Brewery Edible Six Pack Rings, 2016.

URL http://www.youtube.com/watch?time_continue=49&v=-YG9gUJMGyw. Accessed 11-11-2017.
[46] Julia Belluz. Americans are eating way more fat. But it’s not butter.. Vox, Feb 2017.
URL http://tinyurl.com/gw7wa3j. Accessed 03-12-2017.
[47] Daniel Benjamin. Recycling Myths Revisited. The Property and Environment Research Center, 2010.
URL http://www.perc.org/articles/recycling-myths-revisited. Accessed 05-03-2017.
[48] Jeanine Bentley. U.S. Trends in Food Availability and a Dietary Assessment of Loss-Adjusted Food Availability,
1970–2014. Number 166 in EIB. U.S. Department of Agriculture Economic Research Service, 2017.
URL http://tinyurl.com/y7bdyvow. Accessed 11-05-2017.
[49] Khagendra P. Bhandari, Jennifer M. Collier, Randy J. Ellingson, and Defne S. Apul. Energy payback time (EPBT) and
energy return on energy invested (EROI) of solar photovoltaic systems: A systematic review and meta-analysis.
Renewable and Sustainable Energy Reviews, 47:133–141, July 2015. ISSN 1364-0321. doi:10.1016/j.rser.2015.02.057.
[50] Deborah Blum. Looney Gas and Lead Poisoning: A Short, Sad History. Wired, 2013.
URL http://www.wired.com/2013/01/looney-gas-and-lead-poisoning-a-short-sad-history/.
[51] Egbert Boeker and Rienk van Grondelle. Environmental Physics: Sustainable Energy and Climate Change. Wiley, 3rd
edition, September 2011. ISBN 0-470-66676-5.
[52] Steven J. Brams. Game Theory and the Humanities: Bridging Two Worlds. The MIT Press, Cambridge, Mass.;
London, August 2012. ISBN 978-0-262-51825-3.
[53] Adam R. Brandt, Jacob Englander, and Sharad Bharadwaj. The energy efficiency of oil sands extraction: Energy return
ratios from 1970 to 2010. Energy, 55:693–702, June 2013. ISSN 0360-5442. doi:10.1016/j.energy.2013.03.080.
[54] William McDonough an Michael Braungart. The NEXT Industrial Revolution. The Atlantic, 1998. ISSN 1072-7825.
URL http://tinyurl.com/y8ltkx63. Accessed 05-03-2017.
504 BIBLIOGRAPHY
[55] Stephen B. Broomell and Patrick Bodilly Kane. Public perception and communication of scientific uncertainty. Journal
of Experimental Psychology: General, 146(2):286–304, 2017. ISSN 1939-2222 0096-3445. doi:10.1037/xge0000260.
[56] Sally Brown. Greenhouse gas accounting for landfill diversion of food scraps and yard waste. Compost Science and
Utilization, 24(1), 2016. doi:10.1080/1065657X.2015.1026005.
[57] President George W. Bush. Remarks on Signing the Energy Independence and Security Act of 2007, 2007.
URL http://georgewbush-whitehouse.archives.gov/news/releases/2007/12/20071219-6.html.
[58] S. R. Carpenter, D. Ludwig, and W. A. Brock. Management of Eutrophication for Lakes Subject to Potentially
Irreversible Change. Ecological Applications, 9(3):751–771, August 1999. doi:10.2307/2641327.
[59] Sean B. Carroll. The Serengeti Rules: The Quest to Discover How Life Works and Why It Matters. Princeton University
Press, 1st edition, 2016. ISBN 978-0-691-17568-3.
[60] Jimmy Carter. Living Faith. Crown/Archetype, August 2011. ISBN 978-0-307-76470-6.
[61] Cary Funk and Brian Kennedy, Pew Research Center. The Politics of Climate, October 2016.
URL http://www.pewinternet.org/2016/10/04/the-politics-of-climate/. Accessed 11-05-2017.
[62] Nicolas Casey and Josh Haner. Climate Change Claims a Lake, and an Identity. The New York Times, jul 2016.
URL http://tinyurl.com/ya8bh96q. Accessed 02-03-2018.
[63] Andrew T. Chamberlain. Demography in Archaeology. Cambridge University Press, 2006.
[64] F. Stuart III Chapin and Erica Fernandez. Proactive ecology for the Anthropocene. Elem Sci Anth, 1(0), December
2013. ISSN 2325-1026. doi:10.12952/journal.elementa.000013.
[65] F.S. Chapin, M.C. Chapin, G.P. Kofinas, and C. Folke. Principles of Ecosystem Stewardship: Resilience-Based Natural
Resource Management in a Changing World. Springer-Verlag New York, 1st edition, 2014. ISBN 978-0-387-73032-5.
doi:10.1007/978-0-387-73033-2.
[66] Dan Charles. In a Grain of Golden Rice, a World of Controversy over GMO Foods, March 2013.
URL http://tinyurl.com/pbvz35y. NPR (National Public Radio) Morning Edition (transcript). Accessed
04-30-2017.
[67] EC (European Commission). EU Policy for a sustainable use of pesticides: The story behind the strategy. European
Commission, Luxemburg, 2007. ISBN 92-79-03221-6.
URL http://ec.europa.eu/environment/archives/ppps/pdf/pesticides_en.pdf.
[68] Barry Commoner. The Closing Circle: Nature, Man, and Technology. Random House Inc, New York, October 1971.
ISBN 978-0-394-42350-0.
[69] Barry Commoner. A Bulletin Dialogue on “The Closing Circle,” Response. Bulletin of the Atomic Scientists, May
1972.
[70] Wikipedia contributors. Lenape (potato) — Wikipedia, The Free Encyclopedia, 2017.
URL http://en.wikipedia.org/wiki/Lenape_(potato). Accessed 04-30-2017.
[71] Wikipedia contributors. Sunrise Dam Gold Mine — Wikipedia, The Free Encyclopedia, 2017.
URL http://en.wikipedia.org/wiki/Sunrise_Dam_Gold_Mine. Accessed 11-11-2017.
[72] Wikipedia contributors. Volkswagen emissions scandal — Wikipedia, The Free Encyclopedia, January 2017.
URL http://en.wikipedia.org/wiki/Volkswagen_emissions_scandal. Accessed 11-05-2017.
[73] Wikipedia contributors. Cradle-to-Cradle Design — Wikipedia, The Free Encyclopedia, 2017.
URL http://en.wikipedia.org/wiki/Cradle-to-cradle_design. Accessed 11-04-2017.
[74] Wikipedia contributors. Corporate average fuel economy — Wikipedia, The Free Encyclopedia, 2017.
URL http://en.wikipedia.org/wiki/Corporate_average_fuel_economy. Accessed 01-24-2018.
[75] Wikipedia contributors. Contraction and Convergence — Wikipedia, The Free Encyclopedia, 2017.
URL http://en.wikipedia.org/wiki/Contraction_and_Convergence. Accessed 11-08-2017.
BIBLIOGRAPHY 505
[76] Wikipedia contributors. Demographic Transition — Wikipedia, The Free Encyclopedia, 2017.
URL http://en.wikipedia.org/wiki/Demographic_transition. Accessed 04-23-2017.
[77] Wikipedia contributors. List of Countries by Electricity Consumption — Wikipedia, The Free Encyclopedia, 2017.
URL http://en.wikipedia.org/wiki/List_of_countries_by_electricity_consumption. Accessed
11-04-2017.
[78] Wikipedia contributors. Homeowner Flood Insurance Affordability Act of 2013 — Wikipedia, The Free Encyclopedia,
2017. URL http://en.wikipedia.org/wiki/Homeowner_Flood_Insurance_Affordability_Act_of_2013.
[79] Wikipedia contributors. Hybrid electric vehicles in the United States — Wikipedia, The Free Encyclopedia, 2017.
URL http://en.wikipedia.org/wiki/Hybrid_electric_vehicles_in_the_United_States. Data
summarized from Alternative Fuels and Advanced Vehicles Data Center (U.S. Department of Energy), except
2010–2013 which are from HybridCars.com. Accessed 02-06-2017.
[80] Wikipedia contributors. Transistor count — Wikipedia, The Free Encyclopedia, 2017.
URL http://en.wikipedia.org/wiki/Transistor_count. Accessed 03-23-2017.
[81] Wikipedia contributors. Solid waste policy in the United States — Wikipedia, The Free Encyclopedia, 2018.
URL http://en.wikipedia.org/wiki/Solid_waste_policy_in_the_United_States. Accessed
02-03-2018.
[82] Robert Costanza et al.. Twenty years of ecosystem services: How far have we come and how far do we still need to
go?. Ecosystem Services, 28:1–16, September 2017. doi:10.1016/j.ecoser.2017.09.008.
[83] U.S. National Security Council. National Security Study Memorandum 200: Implications of Worldwide Population
Growth for U.S. Security and Overseas Interests (The Kissinger Report), Dec 1974.
URL http://pdf.usaid.gov/pdf_docs/PCAAB500.pdf. Accessed 04-23-2017.
[84] Countable. Countable: Your Government Made Simple, 2017. URL http://www.countable.us/. Accessed
03-07-2017.
[85] Michael Dale and Sally M. Benson. Energy Balance of the Global Photovoltaic (PV) Industry - Is the PV Industry a
Net Electricity Producer?. Environmental Science & Technology, 47(7):3482–3489, April 2013. ISSN 0013-936X.
doi:10.1021/es3038824.
[86] Herman Daly. Beyond Growth: The Economics of Sustainable Development. Beacon, 2nd edition, 1996. ISBN
978-978-080-704-7.
[87] Herman E. Daly and Joshua Farley. Ecological Economics: Principles and Applications. Island Press, 1st edition,
November 2003. ISBN 1-55963-312-3.
[88] Herman E Daly, John B Cobb, and Clifford W Cobb. For the Common Good: Redirecting the Economy Toward
Community, the Environment, and a Sustainable Future. Beacon Press, Boston, 1994. ISBN 0-8070-4705-8
978-0-8070-4705-7.
[89] John Darley and Bibb Latane. Bystander Intervention in Emergencies: Diffusion of Responsibility. Journal of
Personality and Social Psychology, 8(4, Pt.1):377–383, 1968. doi:10.1037/h0025589.
[90] Data360. U.S. Death Rates by Age Group, 2006.

URL http://www.data360.org/dsg.aspx?Data_Set_Group_Id=587. Accessed 11-08-2017.
[91] Coral Davenport and Campbell Robertson. Resettling the First American “Climate Refugees.” The New York Times,
may 2016. URL http://tinyurl.com/jxalwy4. Accessed 02-03-2018.
[92] Gerard Debreu. Theory of Value: An Axiomatic Analysis of Economic Equilibrium. Yale University Press, 1959.
[93] William B. DeOreo, Peter Mayer, Benedykt Dziegielewski, and Jack Kiefer. Residential End Uses of Water, Version 2.
Water Research Foundation, 2016. URL http://www.waterrf.org/PublicReportLibrary/4309A.pdf.
[94] Desmos. Hudson’s Bay Company Pelts of Snowshoe Hare and Lynx 1845-1937, 2017.
URL http://www.desmos.com/calculator/txcjhlv60i. Accessed 11-08-2017.
506 BIBLIOGRAPHY
[95] John Dewey. Democracy and Educational Administration. In Jo Ann Boydston, editor, John Dewey: The later works,
1925–1953, Volume II: 1935–1937. Southern Illinois University Press, 1987.
[96] Jared Diamond. Collapse: How Societies Choose to Fail or Succeed. Penguin Books, New York, revised edition,
January 2011. ISBN 978-0-14-311700-1.
[97] Peter Ditto, Jesse Graham, Jonathan Haidt, Ravi Iyer, Spassena Koleva, Matt Motyl, Gary Sherman, and Sean Wojcik.
YourMorals.org, 2015. URL http://www.yourmorals.org. Accessed 05-17-2017.
[98] David Dornfeld. Green Manufacturing: Fundamentals and Applications. Springer Science & Business Media, 2012.
ISBN 978-1-4419-6016-0. doi:10.1007/978-1-4419-6016-0.
[99] John S. Dryzek, Richard B. Norgaard, and David Schlosberg. The Oxford Handbook of Climate Change and Society.
Oxford University Press, Oxford, 2011. ISBN 978-0-19-956660-0. doi:10.1093/oxfordhb/9780199566600.001.0001.
[100] David Easley and Jon Kleinberg. Networks, Crowds, and Markets: Reasoning About a Highly Connected World.
Cambridge University Press, July 2010. ISBN 0-521-19533-0.
URL http://www.cs.cornell.edu/home/kleinber/networks-book/networks-book.pdf. Accessed
08-12-2015.
[101] Paul R. Ehrlich. The Population Bomb. Ballantine Books, 1968.
[102] Paul R. Ehrlich and John P. Holdren. A Bulletin Dialogue on “The Closing Circle,” Critique: One Dimensional.
Bulletin of the Atomic Scientists, 1972. Available at http://tinyurl.com/y92lgczp. Accessed 11-08-2017.
[103] EIA (U.S. Energy Information Administration). International Energy Outlook 2016, 2016.
URL http://www.eia.gov/outlooks/ieo/pdf/0484(2016).pdf. Accessed 06-28-2017.
[104] EIA (U.S. Energy Information Administration). FAQ: How much carbon dioxide is produced when different fuels are
burned? 2017. URL http://tinyurl.com/y8oa2md8. Accessed 06-28-2017.
[105] Dwight D. Eisenhower. Atoms for Peace, Address Delivered by the President of the United States Before the General
Assembly of the United Nations. URL http://tinyurl.com/y9t2x2sh. New York City, Dec. 8, 1953.
[106] Aaron M. Ellison. An Introduction to Bayesian Inference for Ecological Research and Environmental
Decision-Making. Ecological Applications, 6(4):1036–1046, November 1996. doi:10.2307/2269588.
[107] Charles Elton and Mary Nicholson. The Ten-Year Cycle in Numbers of the Lynx in Canada. Journal of Animal
Ecology, 11(2):215–244, November 1942. doi:10.2307/1358.
[108] EPA (U.S. Environmental Protection Agency) emissions standards. Oil and Natural Gas Sector: Emission Standards
for New, Reconstructed, and Modified Sources. 81 Federal Register 35824–35942, June 2016.
URL http://tinyurl.com/ybyxr96y. (to be codified at 40 CFR Part 60). Accessed 03-23-2017.
[109] FAOSTAT. FAO Stastistical Database, 2016. URL http://www.fao.org/faostat/en/#home. Accessed

11-04-2017.
[110] FAOSTAT. Pesticide use data, 2017. URL http://www.fao.org/faostat/en/#data/RP/visualize. Accessed

05-03-2017.
[111] Federal Highway Administration, U.S. Department of Transportation. Our Nation’s Highways 2011 Chapter 4:
Drivers. United States Office of Highway Policy Information, 2011.
URL http://www.fhwa.dot.gov/policyinformation/pubs/hf/pl11028/chapter4.cfm. Accessed
02-03-2017.
[112] Martin S. Feldstein, Ted Halstead, and N. Gregory Mankiw. A Conservative Case for Climate Action. The New York
Times, February 2017. ISSN 0362-4331. URL http://tinyurl.com/gmxwl3p. Accessed 06-08-2017.
[113] Enrico Fermi. Thermodynamics. Dover Publications, New York, new edition, June 1956. ISBN 978-0-486-60361-2.
[114] Christoper Flavelle. The Nightmare Scenario for Florida’s Coastal Homeowners. Bloomberg, April 2017.
URL http://tinyurl.com/n5ay57c. Accessed 11-08-2017.
[115] FAO (United Nations Food and Agriculture Organization). Global Forest Resources Assessments, 2016.
URL http://www.fao.org/forest-resources-assessment/en/. Accessed 11-04-2017.
BIBLIOGRAPHY 507
[116] Eunice Newton Foote. Circumstances affecting the Heat of the Sun’s Rays. The American Journal of Science and Arts,
22:382–383, November 1856. URL http://books.google.co.uk/books?id=6xhFAQAAMAAJ&pg=PA382.
[117] CDC (Centers for Disease Control and Prevention). National Vital Statistics Reports, 2017.
URL http://www.cdc.gov/nchs/products/nvsr.htm. Accessed 11-05-2017.
[118] The European Organization for Packaging and the Environment. Extended Producer Responsibility, 2015.
URL http://www.europen-packaging.eu/policy/9-extended-producer-responsibility.html.
[119] GISS (NASA Goddard Institute for Space Studies). GISS Surface Temperature Analysis (GISTEMP), 2017.
URL http://data.giss.nasa.gov/gistemp/. Accessed 11-05-2017.
[120] Association for the Study of Peak Oil and Gas. Newsletter # 38, February 2004.
URL http://aspoireland.files.wordpress.com/2009/12/newsletter38_200402.pdf. Accessed
10-07-2017.
[121] Earl Ford and William Dietz. Trends in energy intake among adults in the United States: findings from NHANES.
American Journal of Clinical Nutrition, 97:848–853, 2013. doi:10.3945/ajcn.112.052662.
[122] Lara Fowler, Ryan Baxter, et al.. Flood Mitigation for Pennsylvania’s Rural Communities: Community-Scale Impacts
of Federal Policies. Technical report, The Center for Rural Pennsylvania, 2017.
URL http://www.rural.palegislature.us/documents/reports/Flood-Mitigation-2017.pdf. Accessed
01-30-2018.
[123] Pope Francis. Laudato Si – On Care for Our Common Home. Our Sunday Vistor, Huntington, IN, July 2015. ISBN
978-1-61278-386-4. URL http://tinyurl.com/ptcm9bz. Accessed 08-12-2015.
[124] Salifou Gado. The Energy Sector of Niger: Pespectives and Opportunities, 2015.
URL http://tinyurl.com/y7evwrbt.
[125] Shawna D. Ganley. Federal “Green” Product Procurement Policy in the United States, 2013.
URL http://tinyurl.com/yclopwhk. Sabine Center for Climate Change Law. Accessed 11-05-2017.
[126] Jianxi Gao, Baruch Barzel, and Albert-László Barabási. Universal resilience patterns in complex networks. Nature,
530(7590):307–312, February 2016. ISSN 0028-0836. doi:10.1038/nature16948.
[127] John Gibson, Sandra Barns, Michael Cameron, Steven Lim, Frank Scrimgeour, and John Tressler. The Value of
Statistical Life and the Economics of Landmine Clearance in Developing Countries. World Development, 35(3):
512–531, March 2007. ISSN 0305-750X. doi:10.1016/j.worlddev.2006.05.003.
[128] Gerd Gigerenzer. Calculated Risks: How to Know When Numbers Deceive You. Simon & Schuster, New York, 1st
edition, March 2003. ISBN 978-0-7432-5423-6.
[129] Natasha Gilbert. Case studies: A hard look at GM crops. Nature, 497(7447):24, 2013. doi:10.1038/497024a.
[130] Robert J. Gilliom. Estimation of Nonpoint Source Loadings of Phosphorus for Lakes in the Puget Sound Region,
Washington, 1983. URL http://pubs.usgs.gov/wsp/2240/report.pdf. U.S. Geological Survey Water Supply
Paper 2240. Accessed 10-20-2015.
[131] Malcolm Gladwell. The Tipping Point: How Little Things Can Make a Big Difference. Back Bay Books, Boston, MA,
January 2002. ISBN 978-0-316-34662-7.
[132] Russell Gold. The Boom: How Fracking Ignited the American Energy Revolution and Changed the World. Simon &
Schuster, New York, reprint edition, April 2015. ISBN 978-1-4516-9229-7.
[133] Jian Gong, Seth B. Darling, and Fengqi You. Perovskite photovoltaics: life-cycle assessment of energy and
environmental impacts. Energy & Environmental Science, 8(7):1953–1968, July 2015. ISSN 1754-5706.
doi:10.1039/C5EE00615E.
[134] Alan Grafen and Rosie Hails. Modern Statistics for the Life Sciences. Oxford University Press, Oxford ; New York, 1st
edition, May 2002. ISBN 978-0-19-925231-2.
[135] Jesse Graham, Jonathan Haidt, Sena Koleva, Matt Motyl, Ravi Iyer, Sean P. Wojcik, and Peter H. Ditto. Moral
Foundations Theory: The Pragmatic Validity of Moral Pluralism. Advances in Experimental Social Psychology, 47:
55–130, nov 2013.
508 BIBLIOGRAPHY
[136] C. Grainger, T. Clarke, et al.. Methane Emissions from Dairy Cows Measured Using the Sulfur Hexafluoride (SF6)
Tracer and Chamber Techniques. Journal of Dairy Science, 90(6):2755–2766, June 2007. ISSN 0022-0302.
doi:10.3168/jds.2006-697.
[137] Mark Granovetter. Getting a Job: A Study of Contacts and Careers. The University Of Chicago Press, Chicago, 2nd
edition, March 1995. ISBN 978-0-226-30581-3.
[138] Saul Griffith. Energy Flows in the United States, 2017. URL http://energyliteracy.com/. Accessed 02-03-2018.
[139] Fred Grimm. Massive fish kill makes Florida water emergency difficult to ignore. Miami Herald, apr 2016.
URL http://tinyurl.com/ydypbeqt. Accesed 02-03-2018.
[140] B. Groombridge and M.D. Jenkins. World Atlas of Biodiversity. University of California Press, 2002. ISBN
978-0520236684.
[141] Liza Gross. Switching Drugs for Livestock May Help Save Critically Endangered Asian Vultures. PLOS Biol, 4(3):
e61, January 2006. doi:10.1371/journal.pbio.0040061. Accessed 07-02-2016.
[142] Terry Gross. “Times” Reporters Describe How A Paper Trail Helped Break The Weinstein Story, 2017.
URL http://tinyurl.com/y9w7h4t5. Fresh Air. National Public Radio. Accessed 1-20-2018.
[143] David P. Gushee. Righteous Gentiles of the Holocaust: Genocide and Moral Obligation. Paragon House, St. Paul,
Minn, 2nd revised edition, November 2003. ISBN 978-1-55778-821-4.
[144] Mads Haahr. Random.org: True Random Number Service, 2017. URL http://www.random.org/. Accessed
11-05-2017.
[145] Helmut Haberl et al.. Quantifying and mapping the human appropriation of net primary production in Earth’s terrestrial
ecosystems. Proceedings of the National Academy of Sciences, 105:12942–12947, 2007.
doi:10.1073/pnas.0704243104.
[146] Charles Hadlock. Six Sources of Collapse: A Mathematician’s Perspective on How Things Can Fall Apart in the Blink
of an Eye. Mathematical Association of America, 2012. ISBN 978-0-88385-579-9.
[147] Jonathan Haidt. The Righteous Mind: Why Good People Are Divided by Politics and Religion. Vintage, New York,
reprint edition, February 2013. ISBN 978-0-307-45577-2. URL http://righteousmind.com.
[148] Danny Hakim. Doubts About the Promised Bounty of Genetically Modified Crops. The New York Times, 2016. ISSN
0362-4331. URL http://www.nytimes.com/2016/10/30/business/gmo-promise-falls-short.html.
[149] Daniel Halper. Over 100 Million Now Receiving Federal Welfare. The Weekly Standard, August 2012. Accessed
09-23-2017.
[150] Tegan Hanlon. Massive landslide crashes onto glacier in Southeast Alaska. Alaska Dispatch News, 2016.
URL http://tinyurl.com/yapvfx4z. Jul. 2, 2016. Accessed 02-06-2017.
[151] James Hansen, Makiko Sato, Reto Ruedy, Andrew Lacis, and Valdar Oinas. Global warming in the twenty-first
century: An alternative scenario. Proceedings of the National Academy of Sciences, 97(18):9875–9880, August 2000.
ISSN 0027-8424, 1091-6490. doi:10.1073/pnas.170278997.
[152] Frank Harary. On the measurement of structural balance. Behavioral Science, 4(4):316–323, October 1959. ISSN
1099-1743. doi:10.1002/bs.3830040405.
[153] Garrett Hardin. The Tragedy of the Commons. Science, 162(3859):1243–1248, December 1968.
doi:10.1126/science.162.3859.1243.
[154] Johnathan M. Harris, Brian Roach, and Anne-Marie Codur. The Economics of Global Climate Change. Global
Development and Environment Institute, 2017. URL http://tinyurl.com/kbpd3vf. Accessed 11-08-2017.
[155] Jonathan Harris and Brian Roach. Environmental and Natural Resource Economics: A Contemporary Approach.
Routledge, 3rd edition, 2014.
[156] John Harte. Consider a Spherical Cow. University Science Books, June 1988. ISBN 0-935702-58-X.
BIBLIOGRAPHY 509
[157] John Harte. Consider a Cylindrical Cow: More Adventures in Environmental Problem Solving. University Science
Books, February 2001. ISBN 1-891389-17-3.
[158] Harvard College Writing Center. Strategies for Essay Writing, 1998.
URL http://writingcenter.fas.harvard.edu/pages/strategies-essay-writing. Accessed 03-07-2017.
[159] Ran He, Daniel Kraemer, et al.. Achieving high power factor and output power density in p-type half-Heuslers
Nb1−x Tix FeSb. Proceedings of the National Academy of Sciences, 113(48):13576–13581, November 2016. ISSN
0027-8424, 1091-6490. doi:10.1073/pnas.1617663113.
[160] Frank Hobbs and Nicole Stoops. Demographic Trends in the 20th Century. U.S. Census Bureau, Census 2000 Special
Reports, Series CENSR-4. U.S. Government Printing Office, Washington, DC, 2002.
URL http://www.census.gov/prod/2002pubs/censr-4.pdf. Accessed 11-04-2017.
[161] Heather Holleman. How to Write with Flair. CreateSpace Independent Publishing Platform, April 2011. ISBN
978-1-4537-2120-9.
[162] M. King Hubbert. Nuclear Energy and the Fossil Fuels. Technical Report 95, American Petroleum Institute, San
Antonio, Texas, 1956. URL http://www.hubbertpeak.com/hubbert/1956/1956.pdf. Accessed 10-06-2015.
[163] Mike Hulme. Why We Disagree About Climate Change: Understanding Controversy, Inaction and Opportunity.
Cambridge University Press, Cambridge, UK ; New York, 4th edition, May 2009. ISBN 978-0-521-72732-7.
[164] Mason Inman. The true cost of fossil fuels. Scientifc American, pages 59–61, April 2013.
[165] Product Stewardship Institute. Map of State EPR Laws, April 2017.
URL http://www.productstewardship.us/?State_EPR_Laws_Map. Accessed 05-03-2017.
[166] Institute for Health Metrics and Evaluation. Global Burden of Disease (GBD), 2015.
URL http://www.healthdata.org/gbd. Accessed 05-04-2017.
[167] Greenpeace International. The trash vortex, 2012. URL http://tinyurl.com/o9mvtdq. Accessed 08-09-2015.
[168] Juha V. A. Itkonen. Social ties and concern for global warming. Climatic Change, 132(2):173–192, May 2015.
doi:10.1007/s10584-015-1424-0.
[169] Tim Jackson. Prosperity Without Growth: Foundations for the Economy of Tomorrow. Routledge, Abingdon, Oxon ;
New York, NY, 2nd edition, December 2016. ISBN 978-1-138-93541-9.
[170] Jenna R. Jambeck, Roland Geyer, Chris Wilcox, Theodore R. Siegler, Miriam Perryman, Anthony Andrady, Ramani
Narayan, and Kara Lavender Law. Plastic waste inputs from land into the ocean. Science, 347(6223):768–771, 2015.
ISSN 0036-8075, 1095-9203. doi:10.1126/science.1260352.
[171] Dale Jamieson. Reason in a Dark Time: Why the Struggle Against Climate Change Failed – and What It Means for
Our Future. Oxford University Press, Oxford ; New York, 1st edition, April 2014. ISBN 978-0-19-933766-8.
[172] Jeremy Singer-Vine, John R. Emshwiller, Neil Parmar, and Charity Scott. Waste Lands: America’s forgotten nuclear
legacy., 2014. URL http://projects.wsj.com/waste-lands. Report and interactive database. The Wall Stree
Journal. Accessed 2015-05-12.
[173] William Stanley Jevons. The Coal Question: An Inquiry Concerning the Progress of the Nation, and the Probable
Exhaustion of Our Coal-Mines. Classic reprint series. Forgotten Books, 2017. Reprinted from 2nd edition, originally
published in 1866.
[174] Gavin Jones. Bearing the Consequences of Population Policy in Thailand. East Asia Forum, 2011.
URL http://tinyurl.com/y7sy6t97.
[175] Dan Kahan. Making Climate-Science Communication Evidence-Based All the Way Down. In M. Boykoff and
D. Crow, editors, Culture, Politics and Climate Change, pages 203–220. Routledge, 2014.
[176] Jodi Kantor and Megan Twohey. Harvey Weinstein Paid Off Sexual Harassment Accusers for Decades. The New York
Times, oct 2017. URL http://tinyurl.com/y988ut9z. Accessed 1-20-2018.
[177] Hans Kaper and Hans Engler. Mathematics and Climate. Society for Industrial & Applied Mathematics,U.S.,
Philadelphia, November 2013. ISBN 978-1-61197-260-3.
510 BIBLIOGRAPHY
[178] Ashifa Kassam et al.. Europe needs many more babies to avert a population disaster. The Guardian, August 2015.
URL http://tinyurl.com/gu9bd2q.
[179] Carl Kaysen. The Computer That Printed Out W*O*L*F*. Foreign Affairs, 50(4):660–668, July 1972.
doi:10.2307/20037939.
[180] Kristi Keck. Big Tobacco: A history of its decline. CNN, 2009.
URL http://edition.cnn.com/2009/POLITICS/06/19/tobacco.decline/. Accessed 04-30-2017.
[181] Stephen King. On Writing: 10th Anniversary Edition: A Memoir of the Craft. Scribner, New York, July 2010. ISBN
978-1-4391-5681-0.
[182] Naomi Klein. This Changes Everything: Capitalism vs. the Climate. Simon & Schuster, reprint edition, August 2015.
ISBN 978-1-4516-9739-1.
[183] Judith S. Kleinfeld. The small world problem. Society, 39(2):61–66, 2002. doi:10.1007/BF02717530.
[184] Rakesh Kochhar. How Americans compare with the global middle class, July 2015.
URL http://tinyurl.com/qb4gt8b. Pew Research Center. Accessed 05-11-2017.
[185] Charles Komanoff. Carbon Tax Polling Milestone: 2/3 Support if Revenue Neutral, 2015.
URL http://tinyurl.com/yclmrn68. Accessed 11-04-2017.
[186] Johannes Kromdijk and Stephen P. Long. One crop breeding cycle from starvation? how engineering crop
photosynthesis for rising CO2 and temperature could be one important route to alleviation. Proc. R. Soc. B, 283(1826):
20152578, 2016. ISSN 0962-8452, 1471-2954. doi:10.1098/rspb.2015.2578.
[187] Christopher Landsea. What is my chance of being struck by a tropical storm or hurricane? 2014.
URL http://www.aoml.noaa.gov/hrd/tcfaq/G11.html. NOAA (National Oceanic and Atmospheric
Administration) Hurricane Research Division. Accessed 06-28-2017.
[188] Jill Lawless. Oldest human footprints outside Africa found in England. The Associated Press (as printed by CBC
News), 2014. URL http://tinyurl.com/y9dpx84s. Feb. 7, 2014. Accessed 08-07-2015.
[189] Debbie A. Lawlor, George Davey Smith, and Shah Ebrahim. Commentary: The hormone replacement–coronary heart
disease conundrum: is this the death of observational epidemiology?. International Journal of Epidemiology, 33(3):
464–467, June 2004. ISSN 0300-5771. doi:10.1093/ije/dyh124.
[190] Lawrence Berkeley National Laboratory. Standby Power Summary Table, 2012.
URL http://standby.lbl.gov/summary-table.html. Accessed 03-01-2017.
[191] Lawrence Livermore National Laboratory. Energy flow charts: charting the complex relationships among energy, water
and carbon, 2016. URL http://flowcharts.llnl.gov/. U.S. Department of Energy. Accessed 02-28-2017.
[192] Timothy Lenton. Earth System Tipping Points. In Improving the Assessment and Valuation of Climate Change Impacts
for Policy and Regulatory Analysis. U.S. Environmental Protection Agency, U.S. Department of Energy, ICF
International, June 2011. URL http://tinyurl.com/yb7s5rkl. Accessed 11-05-2017.
[193] Kelly Levin, Benjamin Cashore, Steven Bernstein, and Graeme Auld. Overcoming the tragedy of super wicked
problems: constraining our future selves to ameliorate global climate change. Policy Sciences, 45(2):123–152, June
2012. ISSN 0032-2687, 1573-0891. doi:10.1007/s11077-012-9151-0.
[194] Tom Di Liberto. How can we call something a thousand-year storm if we don’t have a thousand years of climate
observations? September 2016. URL http://tinyurl.com/yce62bjf. NOAA (National Oceanic and Atmospheric
Administration) Climate.gov. Accessed 09-20-2017.
[195] Joe Lindsey. Guide to Power Meter Metrics. Bicycling, July 2013.
URL http://www.bicycling.com/training/fitness/guide-power-meter-metrics. Accessed 03-01-2017.
[196] Jane Braxton Little. The Ogallala Aquifer: Saving a Vital U.S. Water Source. Scientific American, mar 2009.
URL http://www.scientificamerican.com/article/the-ogallala-aquifer/. Accessed 02-28-2017.
[197] Y. Liu, J. Stanturf, and S. Goodrick. Trends in global wildfire potential in a changing climate. Forest Ecology and
Management, 259(4):685–697, Feb 2010. doi:10.1016/j.foreco.2009.09.002.
BIBLIOGRAPHY 511
[198] Yongqiang Liu, Scott L. Goodrick, and John A. Stanturf. Future U.S. wildfire potential trends projected using a
dynamically downscaled climate change scenario. Forest Ecology and Management, 294:120–135, April 2013.
doi:10.1016/j.foreco.2012.06.049.
[199] Robin Lloyd. Metric mishap caused loss of NASA orbiter. CNN, September 1999.
URL http://www.cnn.com/TECH/space/9909/30/mars.metric.02/. Accessed 08-07-2015.
[200] Duncan Alexander Mac Lulich. Fluctuations in the numbers of varying Hare (Lepus americanus), volume 43 of
University of Toronto Studies: Biological Series. The University of Toronto Press, 1937.
[201] Matthew B. Lythe and David G. Vaughan. BEDMAP: A new ice thickness and subglacial topographic model of
Antarctica. Journal of Geophysical Research: Solid Earth, 106(B6):11335–11351, June 2001. ISSN 2156-2202.
doi:10.1029/2000JB900449.
[202] David JC MacKay. Sustainable Energy - Without the Hot Air. UIT Cambridge Ltd., Cambridge, England, 1st edition,
February 2009. ISBN 978-0-9544529-3-3. URL http://www.withouthotair.com. Accessed 08-12-2015.
[203] Doug Mackie. CO2 emissions change our atmosphere for centuries, 2015.
URL http://www.skepticalscience.com/co2-residence-time.htm. Skeptical Science: Getting skeptical
about global warming skepticism. Accessed 12-20-2017.
[204] Paul Mackun and Steven Wilson. Population Distribution and Change: 2000 to 2010: 2010 Census Briefs, March 2011.
URL http://www.census.gov/prod/cen2010/briefs/c2010br-01.pdf. Accessed 11-04-2017.
[205] Karl-Göran Mäler, Anastasios Xepapadeas, and Aart de Zeeuw. The Economics of Shallow Lakes. Environmental and
Resource Economics, 26(4):603–624, December 2003. ISSN 0924-6460, 1573-1502.
doi:10.1023/B:EARE.0000007351.99227.42.
[206] Thomas Malthus. An Essay on the Principle of Population. J. Johnson, in St. Paul’s Church-yard, London, 1798.
URL http://www.econlib.org/library/Malthus/malPlong.html.
[207] George Marshall. Don’t Even Think About It: Why Our Brains Are Wired to Ignore Climate Change. Bloomsbury
USA, London Oxford New York New Delhi Sydney, August 2015. ISBN 978-1-63286-102-3.
[208] John Mason. Understanding the long-term carbon cycle: weathering of rocks—a vitally important carbon sink, 2013.
URL http://www.skepticalscience.com/weathering.html. Skeptical Science: Getting skepitical about global
warming skepticism. Accessed 1-16-2018.
[209] Nathaniel Massey. Energy Storage Hits the Rails Out West. Scientific American, March 2014.
URL http://www.scientificamerican.com/article/energy-storage-hits-the-rails-out-west/.
Accessed 11-08-17.
[210] Anna Mazarakis and Alyson Shontell. How reporters Jodi Kantor, Irin Carmon, and Kim Masters exposed sexual
harassment and started a national reckoning. Business Insider, dec 2017. URL http://tinyurl.com/y835nfp8.
Accessed 1-20-2018.
[211] Douglas McCauley. Selling out to Nature. Nature, 443:27–28, September 2006. doi:10.1038/443027a.
[212] Robert L. McConnell and Daniel C. Abel. Environmental Issues: Looking Towards a Sustainable Future. Pearson
Learning Solutions, S.l., 4th edition, December 2012. ISBN 978-1-256-93309-0.
[213] Bill McKibben. Global Warming’s Terrifying New Math, July 2012. URL http://tinyurl.com/ydxl9jcd.
[214] Donella Meadows. Thinking in Systems: A Primer. Chelsea Green Publishing, White River Junction, VT, December
2008. ISBN 978-1-60358-055-7.
[215] Donella H. Meadows, Jorgen Randers, and Dennis L. Meadows. Limits to Growth: The 30-Year Update. Chelsea
Green Publishing, White River Junction, VT, 3rd edition, June 2004. ISBN 978-1-931498-58-6. Scan of original
edition is available at http://tinyurl.com/qxohw4h.
[216] Men standing with pile of buffalo skulls. Michigan Carbon Works, 1892. URL http://tinyurl.com/y7emgdpe.
Burton Historical Collection. Detroit Public Library. Accessed 10-07-2017.
[217] Ryan Menezes. Fermi Questions, 2015. URL http://www.fermiquestions.com/tutorial. Accessed

11-05-2017.
512 BIBLIOGRAPHY
[218] Jane E. Miller. The Chicago Guide to Writing About Numbers. University Of Chicago Press, Chicago; London, 2nd
edition, April 2015. ISBN 978-0-226-18577-4.
[219] MIT Joint Program on the Science and Policy of Global Change. Greenhouse Gamble, 2010.
URL http://globalchange.mit.edu/research/research-tools/risk-analysis/greenhouse-gamble.
[220] Michael Mitchell. A clever way to estimate enormous numbers, 2012.

URL http://www.youtube.com/watch?v=0YzvupOX8Is. TED-Ed Lessons Worth Sharing. Video. Accessed
02-03-2017.
[221] Matthew Moersbaecher and John W. Day. Ultra-Deepwater Gulf of Mexico Oil and Gas: Energy Return on Financial
Investment and a Preliminary Assessment of Energy Return on Energy Investment. Sustainability, 3:2009–2026, 2011.
[222] Lisa J. Molofsky, John A. Connor, Albert S. Wylie, Tom Wagner, and Shahla K. Farhat. Evaluation of Methane Sources
in Groundwater in Northeastern Pennsylvania. Groundwater, 51(3):333–349, May 2013. doi:10.1111/gwat.12056.
[223] George Monbiot. The Gift of Death, 2012. URL http://www.monbiot.com/2012/12/10/the-gift-of-death/.

[224] Rob Monroe. The History of the Keeling Curve, 2013. URL http://tinyurl.com/y99z3p9e. The Keeling Curve.
Scripps Institution of Oceanography. Accessed 12-10-2017.
[225] Chris Mooney and Brady Dennis. The world is about to install 700 million air conditioners. Here’s what that means for
the climate. The Washington Post, may 2016. URL https://tinyurl.com/ht6t459. Accessed 02-03-2018.
[226] Peter Moore. Young Americans most worried about vaccines, 2015.
URL http://today.yougov.com/news/2015/01/30/young-americans-worried-vaccines/. YouGov US.
[227] David J. Morin. Probability: For the Enthusiastic Beginner. CreateSpace Independent Publishing Platform, 1st edition,
April 2016. ISBN 978-1-5233-1867-4.
[228] Jeffrey Morris. Comparative LCAs for Curbside Recycling Versus Either Landfilling or Incineration with Energy
Recovery. The International Journal of Life Cycle Assessment, 10(4):273–284, 2005. ISSN 0948-3349, 1614-7502.
doi:10.1065/lca2004.09.180.10.
[229] John Muir. My First Summer in the Sierra. Dover Publications, Mineola, N.Y, November 2004. ISBN
978-0-486-43735-4.
[230] David J. Murphy and Charles A. S. Hall. Year in review—EROI or energy return on (energy) invested. Annals of the
New York Academy of Sciences, 1185(1):102–118, January 2010. ISSN 1749-6632.
doi:10.1111/j.1749-6632.2009.05282.x.
[231] Tom Murphy. The Energy Trap, 2012. URL http://dothemath.ucsd.edu/2011/10/the-energy-trap/. Do

the Math Blog. Accessed 06-27-2017.
[232] NASA. Earth Fact Sheet, 2016.

URL http://nssdc.gsfc.nasa.gov/planetary/factsheet/earthfact.html. Accessed 06-28-2017.
[233] National Academies of Science, Engineering, and Medicine (U.S.). Committee on Genetically Engineered Crops: Past
Experience and Future Prospects, Board on Agriculture and Natural Resources, Division on Earth and Life Sciences.
Genetically Engineered Crops: Experiences and Prospects. The National Academies Press, 2016.
URL http://nas-sites.org/ge-crops/2016/05/17/report/. Accessed 04-28-2017.
[234] National Renewable Energy Laboratory. PV FAQs: What is the energy payback for PV? 2004.
URL http://www.nrel.gov/docs/fy04osti/35489.pdf. Accessed 06-27-2017.
[235] Natural Resources Conservation Service. Balancing your animals with your forage: small scale solutions for your farm,
2009. URL http://tinyurl.com/yd49ah9p. Accessed 05-01-2017.
[236] R.L. Neilsen. Historical Corn Grain Yields for Indiana and the U.S.. Corny News Network (Purdue University), August
2012. URL http://www.agry.purdue.edu/ext/corn/news/timeless/yieldtrends.html. Accessed
04-30-2017.
BIBLIOGRAPHY 513
[237] Global Footprint Network. Footprint Calculator, 2017.

URL http://www.footprintnetwork.org/resources/footprint-calculator/. Accessed 04-14-2017.
[238] NOAA (National Oceanic and Atmospheric Administration) National Ocean Service. What is a kelp forest? 2017.
URL http://oceanservice.noaa.gov/facts/kelp.html. U.S. National Ocean Service. Accessed 03-27-2017.
[239] William Nordhaus. A Review of the Stern Review on the Economics of Climate Change. Journal of Economic
Literature, 45:686–702, 2007. doi:10.1257/jel.45.3.686.
[240] Hannah Northey. GAO: Death of Yucca Mountain Caused by Political Maneuvering. The New York Times, May 2011.
ISSN 0362-4331. URL http://tinyurl.com/3fhebcz. Accessed 07-10-2017.
[241] Nuclear Energy Agency and International Atomic Energy Agency. Uranium 2016: Resources, Production and Demand,
2016. URL http://www.oecd-nea.org/ndd/pubs/2016/7301-uranium-2016.pdf. Accessed 07-10-2017.
[242] J. Lindsay Oaks, Martin Gilbert, et al.. Diclofenac residues as the cause of vulture population decline in Pakistan.
Nature, 427(6975):630–633, February 2004. doi:10.1038/nature02317.
[243] Barack Obama. Remarks by the President on Climate Change, 2013. URL http://tinyurl.com/ychllgu2.
Georgetow University. June 25, 2013. Accessed 05-22-2017.
[244] Barack Obama. Remarks by the President at the GLACIER Conference—Anchorage, AK, 2015.
URL http://tinyurl.com/yc3an4jx. The White House Office of the Press Secretary. Accessed 02-06-2017.
[245] @ObamaWhiteHouse. Good news: America’s high school graduation rate has increased to an all-time high.. Twitter,
December 2015. URL http://tinyurl.com/ycau28gy. Accessed: 09-23-2017.
[246] NOAA (National Oceanic and Atmospheric Administration). Atlantic Cod: Overview, 2014.
URL http://tinyurl.com/ycyk28k8. Page last updated Sep. 26, 2014. Accessed through internet archive on
11-05-2017.
[247] NOAA (National Oceanic and Atmospheric Administration). NOAA’s Annual Greenhouse Gas Index (An
Introduciton), 2016. URL http://www.esrl.noaa.gov/gmd/aggi/. Accessed 10-06-2017.
[248] USDA (U. S. Department of Agriculture) Economic Research Service. Recent Trends in GE Adoption, 2016.
URL http://tinyurl.com/he936xm. Accessed 04-30-2017.
[249] Association of American Colleges & Universities. Quantitative Literacy VALUE Rubric, 2009.
URL http://www.aacu.org/value/rubrics/quantitative-literacy. Accessed 11-08-2017.
[250] NIH (National Institutes of Health). Intellectual Property and Genomics, 2017.
URL http://www.genome.gov/19016590/Intellectual-Property. National Human Genome Research
Institute. Accessed 05-03-2017.
[251] U.S. Department of the Interior Division of Fire and Aviation. Chimney Tops 2 Fire Review: Individual Fire Review
Report, 2017. URL http://wildfiretoday.com/documents/ChimneyTops2Report.pdf. Accessed
01-28-2018.
[252] Working Group on California Earthquake Probabilities. Earthquake Probabilities in the San Francisco Bay Region:
2002-2031, 2003. URL http://pubs.usgs.gov/of/2003/of03-214/OFR-03-214_FullText.pdf. Accessed
06-28-2017.
[253] IPCC (Intergovernmental Panel on Climate Change). Climate Change 2014: Synthesis Report. Contribution of
Working Groups I, II, and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change [Core
Writing Team, R.K. Pachauri and L.A. Meyer (eds.)], 2014. URL http://www.ipcc.ch/report/ar5. IPCC,
Geneva, Switzerland, 151 pp. Accessed 09-29-2015.
[254] IPCC (Intergovernmental Panel on Climate Change). IPCC Organizational History, 2016.
URL http://www.ipcc.ch/organization/organization_history.shtml. Accessed 03-06-2017.
[255] WCED (United Nations World Commission on Environment and Development). Our Common Future (Brundtland
Report). Oxford University Press, Oxford, 1987. URL http://www.un-documents.net/wced-ocf.htm.
[256] Cathy O’Neil. Weapons of Math Destruction. Brown Publishing, New York, 2016.
514 BIBLIOGRAPHY
[257] Naomi Oreskes and Erik M. Conway. Merchants of Doubt: How a Handful of Scientists Obscured the Truth on Issues
from Tobacco Smoke to Global Warming. Bloomsbury Press, New York, NY, reprint edition, May 2011. ISBN
978-1-60819-394-3.
[258] WHO (World Health Organization). Burning Opportunity: Clean Household Energy for Health, Sustainable
Development, and Wellbeing of Women and Children, 2016.
URL http://www.who.int/indoorair/publications/burning-opportunities/en/. Accessed 02-03-2018.
[259] Stephen G. Osborn, Avner Vengosh, Nathaniel R. Warner, and Robert B. Jackson. Methane contamination of drinking
water accompanying gas-well drilling and hydraulic fracturing. Proceedings of the National Academy of Sciences, 108
(20):8172–8176, May 2011. doi:10.1073/pnas.1100682108.
[260] Ignacio Palacios-Huerta. Professionals Play Minimax. The Review of Economic Studies, 70(2):395–415, April 2003.
ISSN 0034-6527. doi:10.1111/1467-937X.00249.
[261] Sumit Passery. U.S. Population Grows to 320.09 Million. California Still Reigns as Most Populous U.S. State. Tech
Times, Dec 2014. URL http://tinyurl.com/y8pc4h4y. Accessed 03-12-2017.
[262] Fred Pearce. El Niño and Climate Change: Wild Weather May Get Wilder. Yale Environment 360, 2016.
URL http://e360.yale.edu/features/el_nino_and_climate_change_wild_weather_may_get_wilder.
[263] Alan Penman and William Johnson. The changing shape of the body mass index distribution curve in the population:
implications for public health policy to reduce the prevalence of adult obesity.. Preventing Chronic Disease, 3(3), July
2006. URL http://www.cdc.gov/pcd/issues/2006/jul/05_0232.htm. Accessed 10-23-2015.
[264] Howard Perlman. The USGS Water Science School: How much water is there on, in, and above the Earth? 2016.
URL http://water.usgs.gov/edu/earthhowmuch.html. Accessed 06-30-2017.
[265] Raymond Pierrehumbert. Principles of planetary climate. Cambridge University Press, 2010. ISBN 9780521865.
[266] Lauren C. Ponisio, Leithen K. M’Gonigle, Kevi C. Mace, Jenny Palomino, Perry de Valpine, and Claire Kremen.
Diversification practices reduce organic to conventional yield gap. Proc. R. Soc. B, 282(1799):20141396, 2015. ISSN
0962-8452, 1471-2954. doi:10.1098/rspb.2014.1396.
[267] Proved reserves. BusinessDictionary.com. WebFinance, Inc., 2017.

URL http://www.businessdictionary.com/definition/proved-reserves.html. Accessed 11-04-2017.
[268] Jane Qiu. Genetically modified crops pass benefits to weeds. Nature, August 2013. doi:10.1038/nature.2013.13517.
[269] George Raine. SFO harnesses solar power for Terminal 3. San Francisco Chronicle, September 20, 2007.
URL http://tinyurl.com/y9m8wopv. Accessed 02-03-2017.
[270] Rhetoric. The Oxford American College Dictionary. Oxford University Press, Oxford, 2002. ISBN 9780399144158.
[271] Holli Riebeek and Robert Simmon. The Carbon Cycle, June 2011.
URL http://earthobservatory.nasa.gov/Features/CarbonCycle/. NASA Earth Observatory. Accessed
09-25-2017.
[272] Pete Riley. On the probability of occurrence of extreme space weather events. Space Weather, 10(2):S02012, February
2012. ISSN 1542-7390. doi:10.1029/2011SW000734.
[273] Gerard H. Roe and Marcia B. Baker. Why Is Climate Sensitivity So Unpredictable?. Science, 318(5850):629–632,
October 2007. doi:10.1126/science.1144735.
[274] Pamela C. Ronald. Lab to Farm: Applying Research on Plant Genetics and Genomics to Crop Improvement. PLOS
Biology, 12(6):e1001878, 2014. ISSN 1545-7885. doi:10.1371/journal.pbio.1001878.
[275] David Rose. Global warming stopped 16 years ago, reveals Met Office report quietly released. . . and here is the chart to
prove it. The Daily Mail, 2012. URL http://tinyurl.com/8c56ces. Accessed 12-10-2017.
[276] Max Roser. Global Economic Inequality, 2017.

URL http://ourworldindata.org/global-economic-inequality. Published online at OurWorldInData.org.
BIBLIOGRAPHY 515
[277] Max Roser. Child Mortality, 2017. URL http://ourworldindata.org/child-mortality. Published online at
OurWorldinData.org. Accessed 04-23-2017.
[278] Max Roser. Fertilty Rate, 2017. URL http://ourworldindata.org/fertility. Published online at
OurWorldInData.org. Accessed 11-08-2017.
[279] Max Roser and Esteban Ortiz-Ospina. World Population Growth, 2017.
URL http://ourworldindata.org/world-population-growth. Published online at OurWorldInData.org.
[280] Roswitha Krautgartner and others. Oilseeds and Products Annual Report, 2016.
URL http://tinyurl.com/y8ukuj7l.
[281] Jessica Roy. Microbrewery’s edible six-pack rings create eco-friendly alternative to plastic. Los Angeles Times, May
2016. ISSN 0458-3035.
URL http://www.latimes.com/food/la-dd-edible-six-pack-rings-20160524-snap-story.html.
[282] C. Ford Runge and Benjamin Senauer. How Biofuels Could Starve the Poor. Foreign Affairs, May 2007. ISSN
0015-7120. URL http://www.jstor.org/stable/20032348. Accessed 07-05-2017.
[283] Carl Sagan. Cosmos. Random House, 1980.
[284] Michael J. Sandel. What Money Can’t Buy: The Moral Limits of Markets. Farrar, Straus and Giroux, New York, NY,
reprint edition, April 2013. ISBN 978-0-374-53365-6.
[285] Robert Sanders. Report warns of of climate change “tipping points” within our lifetime. Berkeley News, Dec 2013.
URL http://tinyurl.com/jan65g8. Accessed 03-21-2017.
[286] George Saunders. George Saunders: what writers really do when they write. The Guardian, March 2017. ISSN
0261-3077. URL http://tinyurl.com/gnqf2gl. Accessed 03-05-2017.
[287] Ian Savage. Comparing the fatality risks in United States transportation across modes and over time. Research in
Transportation Economics, 43(1):9 – 22, 2013. ISSN 0739-8859. doi:10.1016/j.retrec.2012.12.011.
[288] Bruce Schneier. Liars and Outliers: Enabling the Trust that Society Needs to Thrive. John Wiley & Sons, 2012.
[289] Leila Schneps and Coralie Colmez. Math on Trial: How Numbers Get Used and Abused in the Courtroom. Basic
Books, New York, March 2013. ISBN 978-0-465-03292-1.
[290] Barry Schwartz. The Paradox of Choice: Why More Is Less. Harper Perennial, New York, NY, January 2005. ISBN
978-0-06-000569-6.
[291] Peter Scott, Jennifer Thomson, David Grzywacz, Serge Savary, Richard Strange, Jean B. Ristaino, and Lise Korsten.
Genetic modification for disease resistance: a position paper. Food Security, 8(4):865–870, 2016. ISSN 1876-4517,
1876-4525. doi:10.1007/s12571-016-0591-9.
[292] Roy Scranton. Learning to Die in the Anthropocene: Reflections on the End of a Civilization. City Lights Publishers,
San Franciso, CA, October 2015. ISBN 978-0-87286-669-0.
[293] SEIA (Solar Energy Industries Association) and GTM Research. U.S. Solar Market Insight — Executive Summary,
2017. URL http://www.seia.org/research-resources/solar-market-insight-report-2017-q2.
Source: Wood Mackenzie, Limited/SEIA U.S. Solar Market Insight. Accessed 11-04-2017.
[294] Senate Budget Committee Republican Staff, ranking member Jeff Sessions. Over 100 Million People in U.S. Now
Receiviing Some Form of Federal Welfare, 2012. Graphic released to public August 8, 2012 and accompanying a press
release from Senator Sessions’s office on August 21, 2012. URL http://tinyurl.com/yd75n3v5. Accessed via
Weekly Standard [149]. Accessed 09-23-2017.
[295] U.S. National Park Service. Great Smoky Mountains National Park Weather, 2016.
URL http://www.nps.gov/grsm/planyourvisit/weather.htm. Accessed 04-04-2017.
[296] Shark Research Institute. Global Shark Attack File, 2015. URL http://www.sharkattackfile.net/. Accessed
05-12-2017.
516 BIBLIOGRAPHY
[297] Leslie Sharma and Robert K. D. Peterson. The Benefits and Risks of Producing Pharmaceutical Proteins in Plants. Risk
Management Matters, 2(4):28–33, 2004. URL http://agbiosafety.unl.edu/biopharm.shtml. Accessed
04-30-2017.
[298] Donald E. Simanek. Perpetual futility: a short history of the quest for perpetual motion, 2015.
URL http://lockhaven.edu/~dsimanek/museum/people/people.htm. The Museum of Unworkable Devices.
[299] Julian Lincoln Simon. The Ultimate Resource 2. Princeton University Press, revised edition, July 1998. ISBN
978-0-691-00381-8.
[300] Adam Smith. The Wealth of Nations. Simon & Brown, September 2016. ISBN 978-1-61382-931-8.
[301] Stuart J. Smyth, Peter W.B. Phillips, and David Castle. Benefits of genetically modified herbicide tolerant canola in
Western Canada. International Journal of Biotechnology, 13(4):181–197, 2014. ISSN 0963-6048.
doi:10.1504/IJBT.2014.068928.
[302] Kelly Soderland. Take two minutes to brush. ADA News, August 2012. URL http://tinyurl.com/yaeq5vna.
[303] J. H. Song, R. J. Murphy, R. Narayan, and G. B. H. Davies. Biodegradable and compostable alternatives to
conventional plastics. Philosophical Transactions of the Royal Society B: Biological Sciences, 364(1526):2127–2139,
2009. ISSN 0962-8436, 1471-2970. doi:10.1098/rstb.2008.0289.
[304] Susan Sontag. WRITERS ON WRITING; Directions: Write, Read, Rewrite. Repeat Steps 2 and 3 as Needed.. The
New York Times, December 2000. ISSN 0362-4331. URL http://tinyurl.com/ybopdoou. Accessed 09-12-2017.
[305] James Gustave Speth. The Bridge at the Edge of the World: Capitalism, the Environment, and Crossing from Crisis to
Sustainability. Yale University Press, New Haven, Conn., March 2009. ISBN 978-0-300-15115-2.
[306] Jenny Stanton. Japan draws up urgent measures to increase birth rate after its population shrinks by a million in five
years. The Daily Mail, 2016. URL http://tinyurl.com/y7ysyz9n.
[307] Lynn Arthur Steen, editor. Mathematics and Democracy: The Case for Quantitative Literacy. Woodrow Wilson
National Foundation, Princeton, N.J., 1st edition, December 2001. ISBN 978-0-9709547-0-1.
[308] Will Steffen. Global Change and the Earth System: A Planet Under Pressure. Springer, Berlin ; New York, September
2005. ISBN 978-3-540-26594-8.
[309] Nicholas Stern. The Stern Review: The Economics of Climate Change, Executive Summary. United Kingdom
National Archives, 2006. URL http://tinyurl.com/y8cn4gqu.
[310] Nicholas Stern. Why Are We Waiting? The Logic, Urgency and Promise of Tackling Climate Change. MIT Press, 2016.
ISBN ISBN: 978-0-262-52998-3.
[311] Joshua Stevens and Kathryn Hansen. Bloom in Lake Okeechobee, July 2016.
URL http://earthobservatory.nasa.gov/NaturalHazards/view.php?id=88311. NASA Earth
Observatory. Accessed 02-03-2018.
[312] Lewis L. Strauss. Remarks For Delivery At The Founders’ Day Dinner, National Association of Science Writers.
URL http://www.nrc.gov/docs/ML1613/ML16131A120.pdf. Atomic Energy Commission. New York City, Sept.
16, 1954. Accessed 01-30-2018.
[313] Steven Strogatz. Visualizing Vastness. The New York Times, October 2012.
URL http://opinionator.blogs.nytimes.com/2012/10/15/visualizing-vastness/. Accessed
08-07-2015.
[314] Meera Subramanian. Global health: Deadly dinners. Nature, 509(7502):548, May 2014. doi:10.1038/509548a.
[315] U.S. Geological Survey. Water Use in the United States: Domestic Water Use. U.S. Geological Survey, 2010.
URL http://water.usgs.gov/watuse/wudo.html. Acccessed 11-08-17.
[316] Ana Swanson. One chart captures why China’s letting families have more kids. The Washington Post, October 2015.
URL http://tinyurl.com/ycstxh4t.
BIBLIOGRAPHY 517
[317] U.S. Forest Service Wildland Fire Assessment System. Keetch-Byram Drought Index.
URL http://www.wfas.net/index.php/keetch-byram-index-moisture--drought-49. Accessed
04-04-2017.
[318] T. M. Lenton and V. N. Livina and others. Early warning of climate tipping points from critical slowing down:
comparing methods to improve robustness. Philosophical Transactions of the Royal Society of London A:
Mathematical, Physical and Engineering Sciences, 370(1962):1185–1204, March 2012. doi:10.1098/rsta.2011.0304.
[319] Stephanie R. Taylor. Data Analysis and Visualization: Good and Bad Examples (course website), January 2014.
URL http://cs.colby.edu/courses/S14/cs251/goodbad.php. Accessed 09-23-2017.
[320] Dhiraj Thakare, Jianwei Zhang, Rod A. Wing, Peter J. Cotty, and Monica A. Schmidt. Aflatoxin-free transgenic maize
using host-induced gene silencing. Science Advances, 3(3):e1602382, 2017. ISSN 2375-2548.
doi:10.1126/sciadv.1602382.
[321] John Tierney. Recycling Is Garbage. The New York Times, 1996. ISSN 0362-4331.
URL http://www.nytimes.com/1996/06/30/magazine/recycling-is-garbage.html. Accessed
05-03-2017.
[322] John Tierney. The Reign of Recycling. The New York Times, 2015. ISSN 0362-4331.
URL http://www.nytimes.com/2015/10/04/opinion/sunday/the-reign-of-recycling.html. Accessed
05-03-2017.
[323] J. R. R. Tolkien. The Lord of the Rings: 50th Anniversary, One Vol. Edition. Mariner Books, Boston, October 2005.
ISBN 978-0-618-64015-7.
[324] Jeffrey Travers and Stanley Milgram. An Experimental Study of the Small World Problem. Sociometry, 32(4):
425–443, 1969. doi:10.2307/2786545.
[325] Edward R. Tufte. The Visual Display of Quantitative Information. Princeton University Press, 1986.
URL http://www.edwardtufte.com/tufte/books_vdqi.
[326] Graham Turner and Cathy Alex. Limits to Growth was right. New research shows we’re nearing collapse. The
Guardian, September 2014. URL http://tinyurl.com/oyv3jo4. Accessed 09-29-2014.
[327] Daniel Kahneman; Amos Tversky. Prospect theory: an analysis of decision under risk. Econometrica, pages 263–291,
1979. doi:10.2307/1914185.
[328] Christopher Uhl and Dana L. Stuchul. Teaching as If Life Matters: The Promise of a New Education Culture. Johns
Hopkins University Press, Baltimore, Md, 1st edition, March 2011. ISBN 978-1-4214-0039-6.
[329] United Nations, Department of Economic and Social Affairs, Population Division. The World Population Prospects:
2015 Revision, Key Findings and Advance Tables, 2015. Working Paper No. ESA/P/WP.241. Accessed 11-08-2017.
[330] U.S. Army Corps of Engineers. Tonawanda Landfill Vicinity Property Report, 2015.
URL http://www.lrb.usace.army.mil/Missions/HTRW/FUSRAP/Tonawanda-Landfill/. Accessed
05-12-2017.
[331] U.S. Department of Energy. Where the Energy Goes, 2017.

URL http://www.fueleconomy.gov/feg/atv.shtml. Accessed 03-01-2017.
[332] Egbert H. van Nen and Marten Scheffer. Slow Recovery from Perturbations as a Generic Indicator of a Nearby
Catastrophic Shift.. The American Naturalist, 169(6):738–747, June 2007. doi:10.1086/516845.
[333] Various. Alliance of Small Island States (website), 2016. URL http://aosis.org/. Accessed 05-30-2017.
[334] Various. Welcome to the Anthropocene, 2017. URL http://anthropocene.info. Stockholm Resilience Centre.
[335] Various authors. Insight Maker, 2017. URL http://insightmaker.com/. Accessed 03-14-2017.
[336] Various authors. Skeptical Science: Getting skeptical about global warming skepticism, 2017.
URL http://www.skepticalscience.com/. Accessed 3-10-2017.
[337] Mathis Wackernagel and William Rees. Our Ecological Footprint: Reducing Human Impact on the Earth. New
Society Publishers, Gabriola Island, British Columbia, Canada, 1996.
518 BIBLIOGRAPHY
[338] Paul E. Waggoner and Jesse H. Ausubel. A framework for sustainability science: A renovated IPAT equation.
Proceedings of the National Academy of Sciences, 99(12):7860–7865, jun 2002. doi:10.1073/pnas.122235999.
[339] Robert M. Wah. Physicians agree: Vaccination crucial to public health. AMA Wire, February 2015.
URL http://wire.ama-assn.org/ama-news/physicians-agree-vaccination-crucial-public-health.
[340] Martin E. Walter. Mathematics for the Environment. Chapman & Hall/CRC Press, London; Boca Raton, FL, 2011.
ISBN 978-1-4398-3472-5.
[341] Peter Wason. Reasoning about a rule. Quarterly Journal of Experimental Psychology, 20:273–281, 1968.
doi:10.1080/14640746808400161.
[342] Robert Watson, A.H. Zakri, et al.. Ecosystems and Human Well-being (The Millenium Ecosystem Assessment). Island
Press, Washington DC, 2005. URL http://www.millenniumassessment.org/en/Framework.html.
[343] Alvin M. Weinberg. Social Institutions and Nuclear Energy. Science, 177(4043), jul 1972.
doi:10.1126/science.177.4043.27.
[344] Zach Weinersmith. The Prisoner’s Dilemma, June 2010.

URL http://www.smbc-comics.com/comic/2010-06-05. Saturday Morning Breakfast Cereal. Web Comic.
[345] Lawrence Weinstein and John Adam. Guesstimation: Solving the World’s Problems on the Back of a Cocktail Napkin.
Princeton University Press, Princeton, N.J, April 2008. ISBN 978-0-691-12949-5.
[346] Charles Weiss. Expressing scientific uncertainty. Law, Probability and Risk, 2:25–46, 2003. doi:10.1093/lpr/2.1.25.
[347] D. Weissbach, G. Ruprecht, A. Huke, K. Czerski, S. Gottlieb, and A. Hussein. Energy intensities, EROIs (energy
returned on invested), and energy payback times of electricity generating power plants. Energy, 52:210–221, April
2013. ISSN 0360-5442. doi:10.1016/j.energy.2013.01.029.
[348] Michael Westendorf. Stall Waste Production and Management. Rutgers University Extension, 2015.
URL http://articles.extension.org/pages/18868/stall-waste-production-and-management.
[349] James White and others. Abrupt Impacts of Climate Change: Anticipating Surprises. National Academies Press, 2013.
URL http://tinyurl.com/y7pxlch5. Accessed 02-21-2015.
[350] Lynn White. The Historical Roots of Our Ecologic Crisis. Science, 155(3767):1203–1207, March 1967.
doi:10.1126/science.155.3767.1203.
[351] William B. Whitman, David C. Coleman, and William J. Wiebe. Prokaryotes: The unseen majority. Proceedings of the
National Academy of Sciences, 95(12):6578–6583, June 1998. ISSN 0027-8424, 1091-6490.
URL http://www.pnas.org/content/95/12/6578. Accessed 06-30-2017.
[352] WHO (World Health Organization) and FAO (Food and Agriculture Organization of the United Nations). Diet,
Nutrition and the Prevention of Chronic Diseases, chapter “Global and regional food consumption patterns and trends.”
WHO Techical Report Seriers 916. World Health Organization, Geneva, 2003.
URL http://www.who.int/dietphysicalactivity/publications/trs916/en/. Accessed 06-28-2017.
[353] Chris Woodyard. Icons: Elon Musk doesn’t let up at Tesla, SpaceX. USA Today, April 2013.
URL http://tinyurl.com/crcw9le. Accessed 10-10-2017.
[354] Xiaoyu Yan. Energy demand and greenhouse gas emissions during the production of a passenger car in China. Energy
Conversion and Management, 50(12):2964–2966, December 2009. doi:10.1016/j.enconman.2009.07.014.
[355] Dennis Tao Yang. China’s Agricultural Crisis and Famine of 1959–1961: A Survey and Comparison to Soviet Famines.
Comparative Economic Studies, March 2008. doi:10.1057/ces.2008.4.
[356] Nancy Zumoff, Christopher Schaufele, et al.. Earth Algebra, College Algebra with Applications to Environmental
Issues. Addison Wesley Publishing Company, Reading, Mass, 2nd edition, August 1998. ISBN 978-0-321-01597-6.
Index
absolute size, 19 reverse, 463

agent, 349, 361 central limit theorem, 275, 297, 314
albedo, 111 change, 183
allocation, 349 relative v. absolute, 186
amplifying feedback, see feedback chaos, 162
Anthropocene, 394–396, 400, 450 charts, see graphs
antilogarithm, 188 cherry-picking, 56
appeal to authority fallacy, 30 chicken game, 369
approximately equal, 38 citing sources, 418
area class interval, 267, 333
and radiant heat, 109 climate change, 73, 107, 141, 379
forested, 185 anthropogenic, 73, 204, 356, 379
in heat loss calculation, 100 climate sensitivity, 203, 237, 260
units of, 9, 484 clustering coefficient, 135
Arrhenius, 114, 202 codistill, 97
atmosphere, 15, 50, 72, 110, 374, 486 commons, 373, 392
tragedy, see tragedy of the commons
bar chart complement, 287
column, 269 complete graph, 149
basic demographic model, 159, 170, 195 component, 132
Bayes’ theorem, 325 compound interest, 165
best response, 366 concave, 311
blackbody radiation law, see Stefan’s law concentration, 15, 97, 313
border leakage, 386 conditional probability, 318
box plot, 263, 265 rule, 320
breeder reactor, 453, 487 conduction, 99
Brundtland Report, xi, 393 control
burping v. farting, 28 arrow, 154, 171
bystander effect, 142, 360 linear, 222
nonlinear, 212, 227
cap and trade, 381 strength of, 228
capital, 195 sign of, 156, 171, 221
carbon budget, 219, 406 strength, 221–232
carbon cycle, 72 control arrow, 95
carbon dioxide, 15, 50, 73, 79, 106, 114, 202 convection, 99
and carbon, by mass, 73 conversion factor, 7–15, 86, 483–485
and the atmosphere as a commons, 374 cooperate v. defect, see Prisoners’ Dilemma
cap and trade, 381 core values, 390
flows, 72 correlation v. causation, 323, 344, 417
revenue-neutral taxation, 383 critical slowing down, 235, 248, 249
carbon fee and dividend, 384 critical thinking, 79, 80
carbon intensity, 459
carbon neutral, 442 data, 261–275, 330
Carnot limit, 91, 461 and politics, 146

https://doi.org/10.1007/978-3-319-76660-7
520 INDEX
communicating, 44–57 eutrophic, 227, 248, 369

precision of, 34 event, 280
set, 261 excludable, 380
decimal notation, 25 expectation, 303
degree, 130, 135 expected payoff, 370
demographer, x expected utility rule, 308
demographic, see basic demographic model expected value, 303
profile, 74, 437 exponent, 24
transition, 434 exponential
depreciation, 195 and growth rates, 206
descriptive statistic, 270 and logarithms, 192
directed network, 133 decay, 165, 174
discount rate, 354 discounting, 354
disjoint, 289 growth, x , 165, 166, 169, 172
distribution, 262, 281 law of, 168
skewed, 274 reserve index, 201
doubling time, 172 exponential model, 164
approximation, 173 offset, 165
exact rule, 194 externality, 350, 380
drinking water regulations, 16 extrapolation, 216
extreme value, 264
Earth, xi, 19, 55, 107, 151, 210, 246
physical data, 486 fallacy of multiplication, 56
Summit (Rio, 1992), 398 false negative error, 322
surface temperature, 108 false positive error, 322
ecological footprint, 209, 440 feedback
ecosystem services, xiii , 196, 351 amplifying, 155
criticism of, 351 delay, 159
efficiency, 90, 459 ice-albedo, 242
theoretical maximum, see Carnot limit loop, 153
electromagnetic radiation, 108 negative, 155
emissions trading, 393 positive, 155
energy, 20, 49, 81–94, 141, 354, 470 rule of signs, 156
and recycling, 429 stabilizing, 155
and thermodynamics, 88 Fermi problems, 42
centrality of, 84 fertility rate, 436
conservation of, 88 fissile, 451
consumption, 93, 167, 209 flow, 69, 70, 74, 77, 79, 80
definition, 82 dependence on stock, see control arrow
different forms of, 83–84 flow arrows, 95
from biomass, 416 foot-pounds, 82
from food, 61 fossil fuels, 162
from fossil fuels, see fossil fuels carbon intensity of, 488
investments and returns, 470 energy content of, 488
nuclear, 84, 450–452 framing, 414
radiant, 84, 108–110 frequency, 268
units of, 86 frequency table, 267, 268
use in home, 99, 153, 458–465 friend network, 130
useful data, 485, 487
energy balance (in climate modeling), 112, 240–247 game, 361
energy trap, 468 game theory, 360, 361
epoch, 394 genetic modification, 444
equally likely probability rule, 282 geoengineering, 254, 407
equilibrium, 69, 72–73, 80, 112, 151, 368, 376 geometric progression, 197
calculating, 95–105 glaciers, 18
Nash, see Nash equilibrium globe, 77
sensitivity of, 232 GMO, see genetic modification
stable v. unstable, 225–227 goods, 349
static v. dynamic, 73 grade (of heat energy), 82, 90
sudden change in, see tipping point graphs
EROEI, 470, 487 bar charts, 48
INDEX 521
line graph, 50 kelvin, 91

misleading, 51, 55
pie charts, 49 latent heat of evaporation, 92, 486
scatter plots, 50 law of large numbers, 296
use in communicating quantitative information, 48 length, 8
use in determining equilibria, 97, 112, 226, 233, units of, 8
240–247 link (in network), 129
use in exponential models, 191 directed, 133
use in logistic models, 215 strength of, 136, 137
use in modeling dynamic behavior of a system, loan formula, 199
152, 161 logarithmic approximation, 190
gravitational constant, 86, 120, 486 logarithmic change, 190
Great Law of Peace, xii logarithmic transformation, 191
Great Pacific Garbage Patch, 46, 69 logarithms, 188
greenhouse effect, 113 lower quartile, see quartile
glass ball model, 114 lumen, 92, 459
gross domestic product, 356 luminous coefficient, 93, 459
growth rate, 207
Malthus, Thomas, 433
half-life, 174 market, 360
approximation, 174 market failure, 350, 356
exact rule, 194 market paradigm, 349
Harary’s theorem, 149 mass
heat, 84 units of, 12
high-grade, 88 mean, 270
low-grade, 90 measure of spread, 273
heat engine, 91 measurement claim, 34
heat pump, 462 median, 263
histogram, 268 meta-game, 403
homophily, 147 micron, 62
human power, 87 mixed strategy, 370
human terms, 9, 17, 45, 416 model, 22, 77, 80
measurement rules, 17 and scientific method, see scientific method
hypothesis testing, 321 dynamic, xvi
stock-flow, 69–80
independent, 291, 307, 320 money, 349
Indian vulture crisis, 132 mutually exclusive, 289
individual
in data set, 261 naked planet model, 110
individualism, 391 Napier, 188
inference, 279, 321, 326, 417 Nash equilibrium, 368, 377
numerical, 328 natural capital, 196
inflow, 69, 70, 80 neighbors, 130
information cascade, 143, 360 nepers, 190
information deficit, 141 net flow, 69, 73, 80, 164
integrated assessment modeling, 356 net metering, 309
intellectual capital, 196 network, 129, 360
Intergovernmental Panel on Climate Change, 106, 141, connected, 131
204, 237, 298 degree of node, 130
interquartile range, 266, 421 distance between nodes, 131
intersection of events, 289 heterogeneity, 138
investment, 195 strength of ties, 136
invisible hand, 348 node (in network), 129
IPAT equation, 441 nonrenewable resource, 71
Iroquois Confederacy, xii nonsubstitutable, 85
nonuniform probability rule, 285
Jevons paradox, 94, 200, 458 normal distribution, 275–276, 333
joule, 86 nuclear power, 260, 450–457
Keeling curve, 50, 52, 201 observation, 261

Kelvin, 14 odds, 287
522 INDEX
offset exponential model, 165 quantitative reasoning, 411, 413–418

Ogallala aquifer, 71, 119, 373, 379 quartile, 264
order of magnitude, 39–43
order-of-magnitude estimation, 41 R-value, 100
oscillation, 159 radiant energy, 108
outcome, 278 radiation, 99
outflow, 69, 70, 80 radiative forcing, 202
outlier, 266 random process, 278, 370
overshoot and collapse, 152, 162, 211 random variable, 301
oversimplification, 74, 116, 184, 259 range, 264
rate constant, 164, 165
Pandora’s Promise, 457 ratio, 15
parameter, 102, 314 regulatory capture, 383
Pareto efficiency, 375 relative frequency, 294
Pareto optimal, 349, 360, 377 relative size, 19
parts per billion, 15 rescaling, 20
parts per million, 15 example of, 44
path, 130 reserve index
payback time, 309, 466 exponential, see exponential
energy, 467 static, 201
payoff, 361 residence time, 69, 75, 76, 80, 152
payoff matrix, 361, 362 resilience, xii, 134
peak oil, 217 measure of, 137–139
percentage, 14 revealed preferences, 357
perfect information, 361, 365 revenue-neutral tax, 383
perpetual motion machine, 88 rhetoric, 413
Planck’s law, 110 risk aversion, 311–313
planet walk, 21 risk seeking, 312
Poiseuille’s law, 60
pollution, 350 Saint Matthew Island, 162
air, 61, 62, 417 sample space, 280, 300, 370
dilute and disperse, 98 satisficing, 399
phosphorus, 159, 226, 247 savings formula, 198
plastic, 17, 46 scientific method, 69, 77–78, 80, 279, 329–330
population, x scientific notation, 24
population momentum, 437 calculations with, 27
portfolio of sustainability choices, 400 calculator, 31
posterior distribution, 331 conversion, 26
power, 83, 86 standard form, 25, 26
precautionary principle, 398 semilogarithmic, 192
precision, 34, 483 set point, 155, 165
in tables, 48 significand, 24
of measurement , 333 significant figures, 34–39, 483
order of magnitude, 39 calculations, 37
significant figures, 36 sink, 71, 374
v. accuracy, 34 slippery we, 345–347
prediction, 78 small-world, 131
present value, 354, 472 social capital, 196
prior distribution, 329 social optimum, 376
prisoners’ dilemma, 363, 375, 378 socially optimal, 377
privatization, 380 solar constant, 111
probability, 281 source, 71, 374
probability distribution, 204 square root heuristic, 268
probability model, 281 stabilizing feedback, see feedback
prosecutor’s fallacy, 327 stable equilibrium, 244
proxy, 97 standard deviation, 273
public goods, xiv standard form, 24
pure number, 14–15, 20, 97 static reserve index, 452
pure time preference, 355, 357 steady state, 358
Stefan’s law, 109, 485
quantitative literacy, xviii stock, 67, 69–71, 74, 76, 77, 79, 80
INDEX 523
computing equilibrium levels, 95–105 uniform probability, 282

in exponential process, see exponential union of events, 288
in logistic model, see logistic unit-factor method, 5–7, 19, 76
in tragedy of commons, 378 units
nonequilibrium behavior, 151 control strength, 222–223
of heat conversion, 6, 12, 87, 483–485
in a home, 100, 154 energy, 86
on Earth, 110, 240 flow, 231
stock-flow system, 70 heat loss, 101
Stockholm Convention, 406 human scale, 17, 46, 416
strategy, 361 length, 8
dominant, 366 mass, 12
pure, 370 of flow, 70
strictly dominant, 363, 365, 376 power, 86
strictly best response, 364 radiation exposure, 455
substitutable, 85 temperature, 13
substrate, 97 time, 7
surface temperature volume, 10
of Earth, 55, 108, 240–246 weight, 12
of Mars, 113, 123 unstable equilibrium, 244
of Mercury, 59 upper quartile, see quartile
of Sun, 109 utility, 308
of Venus, 113, 123
sustainability, x–xiii, 73, 85, 106 vaccination, 140
and decision-making, 399–400 value of a statistical life, 352
Brundtland Commission definition of, xi variable, 261
question, xi, 80, 85, 345, 355, 392 categorical, 262
numerical, 262
sustainable, xi, 162, 379, 397, 452
Venn diagram, 318
system diagram, 69, 70, 77
virtue ethics, 392
volume, 10
tables of water on Earth, 18
best practices in use of, 47 units of, 10
temperature, 13, 91, 100, 112, 203
absolute zero of, 14 Wason 4-card task, 78
and climate change, 107, 242 watt, 86
and offset exponential model, 165 wavelength, 108
anomaly, 64 weight
global mean, 55 units of, 12
probability distribution, 204, 285 work, 82
conversion, 14, 485
surface relative to upper atmosphere, 116 zero sum, 348, 362, 363
units of, 13
test, 321
thermal imaging, 99
thermal radiation, 109
thermal resistance, 100
thermodynamics, 88–93
laws of, 89
thermostat, 100
thesis, 414
time
units of, 7
time step, 164, 165
tipping point, 239, 246
climate system, 250
sensitivity, 248
tragedy of the commons, 378, 435
transmission, 458
triadic closure, 135
two-way frequency table, 323

Roe Et Al., 2018. Mathematics For Sustainability

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Roe Et Al., 2018. Mathematics For Sustainability

Uploaded by

Copyright:

Available Formats

Texts for Quantitative Critical Thinking

More information about this series at http://www.springer.com/series/15949

Mathematics for Sustainability

ISSN 2523-8647 ISSN 2523-8655 (electronic)

Library of Congress Control Number: 2018935919

© Springer International Publishing AG, part of Springer Nature 2018

Printed on acid-free paper

4.3 Measuring Feedback Strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

II Case Studies 409

7 Case Studies 411

III Resources 481

8 Resources for Student and Instructor 483

List of Figures 491

List of Tables 498

0.1 To the Student

• In a democracy, as many people as possible need to participate in well-informed discussion of

Sustainability—The Key Idea

Figure 1: Some measures of the “size” of humanity (from [308]).

The “old world” that Speth describes is a world in which

definition of “sustainability” itself.

Think about it. . .

Sustainability and Resilience

the definition of a sustainable society implicitly involves

that are available to each generation and also limits on Economy

Think about it. . .

About This Book

Summary of Ideas: To the Student

• Sustainability refers to the ability of a social or economic system to keep functioning

0.2 To the Instructor

Quantitative Literacy (QL) is a “habit of mind,” competency, and comfort working

Think about this quotation from a BBC news broadcast:

We know 1 million years is a long time, but “a long time” is

© Springer International Publishing AG, part of Springer Nature 2018 3

1.1 Units and Measurement

1.1.1 Number Part and Unit Part

1.1.2 The Unit-Factor Method

1 cowyr 9,500 lb 9,500 × 20

1.1.3 Standard Units

Table 1: Units of Time

Table 2: Units of Length (U.S.)

Unit (abbr) Definition Example

Table 3: Units of Length (Metric)

Unit (abbr) Definition Example

Figure 3: Illustrating the formation of units of area.

Units of Area and Volume

(12 in)2 144 in2

Problem 6: Express 15 square centimeters in square meters.

Solution: Working as in the problem above,

The answer is 0.0015 square meters.

Table 4: U.S. Units of Area

The answer is approximately 0.7 cubic feet.

Problem 8: How many gallons are in a cubic foot?

Figure 4: Illustrating the formation of units of volume.

Table 5: Units of Volume

Problem 9: Approximately how many cubic centimeters make a teaspoon?

Units of Weight or Mass

Table 6: Units of Mass (U.S.)

Table 7: Units of Mass (Metric)

Table 8: Unit Conversions

Table 9: Temperature Scale Conversions

K K = C + 273 K = 59 F + 255 K=K

1.1.4 Percentages and Other Pure Numbers

0.12 = 0.12 × 100% = 12%,

1 cowyr 9,500 lb 9,500 × 20