(eBook PDF) Mind on Statistics:

Australian & New Zealand 2nd

Visit to download the full and correct content document:
https://ebooksecure.com/download/ebook-pdf-mind-on-statistics-australian-new-zeala
nd-2nd/
MIND ON STATISTICS
VIII CONTENTS

6.4 Expected values, standard deviations, medians, quartiles and percentiles of

random variables ..................................................................................................................... 212
6.5 Parameters and estimates ...................................................................................................... 217
6.6 Three special distributions ...................................................................................................... 218
Case study 6.2 Does caffeine enhance the taste of cola? 222
6.7 Normal probability plots.......................................................................................................... 230
CHAPTER 7 Estimating proportions with confidence 247
7.1 Percentages and proportions abound .................................................................................... 248
7.2 Confidence intervals for proportions ...................................................................................... 253
Case study 7.1 How many passersby would notice this? 257
7.3 Confidence intervals for the difference in two proportions .................................................. 260
7.4 Sample size to estimate a proportion..................................................................................... 265
7.5 Background for confidence intervals for proportions ........................................................... 268
7.6 Confidence intervals and decisions ........................................................................................ 274
CHAPTER 8 Analysis of variance: categorical predictors, continuous
response 285
8.1 Examples of data investigations that include continuous responses and
possible categorical explanatory variables ............................................................................ 286
8.2 One-way ANOVA ....................................................................................................................... 287
8.3 Assumptions and diagnostics for ANOVA............................................................................... 294
8.4 Multiple comparisons .............................................................................................................. 300
8.5 Two-way ANOVA ....................................................................................................................... 303
8.6 More on continuous response and categorical explanatory variables ................................. 312
8.7 Other methods for investigating effects of a categorical variable on a
continuous variable .................................................................................................................. 314
8.8 Models, notation and calculations for ANOVA........................................................................ 318
CHAPTER 9 Regression: investigating relationships between
quantitative variables 351
9.1 Some examples of data investigations that include continuous responses and
quantitative explanatory variables .......................................................................................... 352
9.2 Simple linear regression ......................................................................................................... 354
9.3 Messages from residuals ........................................................................................................ 366
 CONTENTS IX

9.4 Multiple regression .................................................................................................................. 375

Case study 9.1 Weights and lengths of fish (Example 9.4) 380
Case study 9.2 Experience counts 381
Case study 9.3 Can we predict secondhand car prices? 383
9.5 Some formulae in regression.................................................................................................. 386
CHAPTER 10 Interval estimation and quantitative variables 409
10.1 Sample statistics as estimates ............................................................................................... 410
10.2 Confidence interval for the mean of a quantitative variable ................................................. 419
10.3 Interval estimates for the median........................................................................................... 423
10.4 Confidence interval for difference between two means........................................................ 426
Case study 10.1 Confidence interval for relative risk 438
10.5 Tolerance intervals for individual values................................................................................ 438
10.6 Confidence interval for a standard deviation ......................................................................... 440
10.7 Sample size required to estimate a mean with a desired precision .................................... 442
CHAPTER 11 Testing hypotheses in one and two samples 455
11.1 Overview of statistical hypothesis testing............................................................................... 456
11.2 Testing hypotheses about a proportion .................................................................................. 465
11.3 Testing hypotheses about the difference in two proportions................................................ 471
11.4 Connection with testing independence in contingency tables .............................................. 476
11.5 Testing hypotheses about one mean ...................................................................................... 478
11.6 Testing hypotheses about the mean of paired differences ................................................... 482
11.7 Testing hypotheses about the difference between two means ............................................. 483
11.8 Non-parametric tests and medians........................................................................................ 487
11.9 Tests for one or two standard deviations ............................................................................... 494
11.10 The relationship between tests and confidence intervals ................................................... 497
11.11 t-tests and ANOVA: correspondences and pitfalls of t-testing in real investigations ....... 499
Case study 11.1 Dissolving times for soluble aspirin 501
11.12 The rejection region approach to hypothesis testing........................................................... 501
11.13 Sample size, statistical significance, practical importance and effect size....................... 507
CHAPTER 12 More on probability, random variables and distributions 529
12.1 Foundations for probability.................................................................................................... 530
12.2 Probability rules ..................................................................................................................... 534
X CONTENTS

12.3 Independence and conditional probability ............................................................................ 538

12.4 Using conditional probabilities .............................................................................................. 543
12.5 Bayes’ theorem....................................................................................................................... 546
Case study 12.1 The Punjabi drug addict report 549
12.6 Continuous distributions ........................................................................................................ 550
12.7 A very special process: the Poisson...................................................................................... 556
CHAPTER 13 Sums and differences of random variables 569
13.1 Examples for which this chapter is needed.......................................................................... 570
13.2 Sums and differences of two random variables................................................................... 571
13.3 Means and variances of linear combinations of random variables .................................... 575
Case study 13.1 Variation in a process 577
13.4 Linear combinations of normal random variables............................................................... 578
13.5 Sums of some other independent random variables........................................................... 581
13.6 The sample mean and the central limit theorem ................................................................ 582
13.7 Combining ‘errors’ ................................................................................................................. 585
CHAPTER 14 Some further data situations 593
14.1 Binary logistic regression ...................................................................................................... 594
14.2 Failure and survival data........................................................................................................ 599
Appendix A: Appendix of tables 620
Appendix B: Answers to selected questions 629
Appendix C: References 636
Index 638
PREFACE
From students – for students
Many thousands of students from across many disciplines have contributed to this book. Their
questions, writings, comments, feedback, assessment responses and reports have helped shape
the way it is organised, the approaches to learning, the styles of explanations and the design of the
examples and exercises. Many of the examples and rich datasets have come from students’ own
data investigations planned and carried out by groups on freely chosen topics or issues of interest
to them. Because the vignettes, examples and exercises use real data and/or real contexts, all the
vignettes and almost all examples are essentially case studies. The vignettes and the returns to
vignettes at the end of each chapter are particularly designed not only to bring statistics to life for
students but also to bring together the key concepts, messages and procedures of each chapter in
real contexts of interest to all students and across a variety of disciplines.
Questions and feedback from past students, graduates in diverse workplaces and postgraduate
students in disciplines ranging across science, engineering, health, information technology,
education, social sciences and business have also influenced the structure, language,
explanations, examples and exercises. Last, but not least, the approaches throughout the book are
permeated by students’ reactions as they plan, explore, analyse, problem-solve and discover that
statistics is accessible, powerful, and even exciting, as well as important for them personally now
and in their future.

From staff – for staff

The book is also the product of working with many staff as a team member, leader or mentor,
particularly staff who work at the coalface with students, and of extensive discussion with
colleagues of all levels of experience across Australia, New Zealand and the world. Comments,
suggestions and requests from a wide range of reviewers have been carefully studied and
integrated in a balanced and systematic way.
Feedback from students across disciplines from the least to the most quantitative, including
statistical graduates and postgraduates, is that the approaches of this book gave them a lasting
and enabling foundation in core statistical thinking. Courses or symposia for postgraduates from
MBA’s to researchers in all other disciplines have successfully used the strategies of this book.
Feedback from staff is that the approaches are enablers for their teaching, helping them to reduce
the pressures often associated with teaching statistics, to connect with their students and make
the essence of statistics live and meaningful. The emphases are on learning by doing; on developing
understanding of concepts and processes through examples in contexts of familiarity or relevance;
on step-wise progression but at the instructor’s choice; on flow within, and connections across,
material; on straightforward language and exposition but without trivialisation; and on flexibility for
both students and instructors to cater for a variety of students, courses, practical constraints and
learning preferences.
XII PREFACE

Using this book

The book is structured to cater for different choices in sequencing and extent of material, and for
different types of cohorts and different amounts of time available for a course. It is expected that a
single course would select from the topics in the book, but the intent is to provide choice and
coverage for introductory courses across a variety of types of cohorts and disciplines, with
sufficient content for revisiting or use beyond a first course. Although topics, examples, case
studies and exercises are cross-referenced, the book is sufficiently modularised to facilitate various
sequencing of material with good flow within and between chapters. In addition, material within
chapters is arranged to facilitate choices in the extent of material to be covered, with sections on
core topics followed by sections that can be omitted or partially selected by an instructor, or read
or revisited by students, to broaden or develop a topic.
Chapter 1 consists of real examples that illustrate the roles of statistics in tackling complex
problems in complicated contexts; it can be used as optional background reading or to lead into
discussion of key points.
Sections 2.1 to 2.3 are of core importance for statistical thinking reflecting the practice of
statistics. Sections 2.4 to 2.7 may be selectively chosen to receive more emphasis or even extension
in some courses, or to be background reading. Most introductory statistics courses would include
all of Chapter 3, and Sections 4.1 to 4.4 of Chapter 4, with an emphasis on revision and
consolidation of statistics from today’s school curricula. Sections 4.5 and 4.6 are new in this
edition, providing an optional introduction to the core inferential concepts of interval estimation
and hypothesis testing. Only the motivation for, and interpretation of, interval estimates are
presented, and simple randomisation tests illustrate the essential concepts of statistical
hypothesis testing – the emphasis on turning research questions into statistical questions
pervades the book.
It is after Chapter 4 that this book differs considerably from conventional approaches, not only
in the sequence as presented in the book, but in the variety of choices and options open to readers
and courses. The sequence as presented in Chapters 5–11 has been developed and successfully
implemented in courses built around hands-on learning of statistical data investigations, for large
student cohorts diverse in interests and attitudes, catering for quantitative inclinations ranging
from the mathematically averse to the mathematically able. But the book is written to optimise
freedom of choice in both order of progression and depth of coverage. To facilitate user
preferences and course selections, core concepts in data investigations, interval estimation and
statistical hypothesis testing are included in different contexts, providing consolidation no matter
what sequencing is preferred. Chapters 12 and 13 may be included in courses for more
quantitative cohorts or left for future courses. Similarly, it may be desired to include the further
procedures of either of Sections 14.1 and 14.2 in more extensive courses for particular disciplines.
Just some examples of the many other choices of sequences, with inclusion or not of chapters
and sections, are:

• Chapters 6, 7, 10, 11, 5, with the choice of omitting some of the later sections of some chapters,
and then continuing to Chapters 8 and/or 9 in either order.

• The above could be varied by moving Chapter 5 to before Chapter 10 or after Chapters 8 and/or
9 if desired.
 PREFACE XIII

• Chapter 12 could be included after Chapter 6 in either the original sequence or in the sequences
above.

• Chapters 6 and 9, preceded by either (Sections 4.5 and 4.6), or (Sections 10.1–10.4 and Sections
11.1, 11.5–11.7), could be followed by Section 14.1.

• Section 14.2 could be preceded by Chapter 6, and either (Sections 4.5 and 4.6), or (Sections 10.1–
10.4 and Sections 11.1, 11.5–11.7), and Chapter 12.

It should be noted that introducing statistical inference via categorical data, with hypothesis
testing as in Chapter 5, and interval estimation as in Chapter 7, has proven remarkably successful
for all disciplines and quantitative inclinations or non-inclinations.

New to this edition

New features in this edition include:

• Each chapter starts with a rich vignette of universal interest, which is revisited at the end of the
chapter as a case study illustrating the material of the chapter.
• The Keep in mind feature captures core concepts, results and procedures.

• The Mind your step feature provides cautions and pointers to common mistakes or
misunderstandings.

• The Signposts at the beginning of each chapter outline what will be covered by each section.
• An extensive Key terms glossary has been added to each chapter, to maximise student
understanding of the key terms as they appear within each chapter of the book.
• Sections 4.5 and 4.6 have been added as described above, and are also intended to optimise
flexibility.
• The number of available datasets has been considerably increased.

• As well as the addition of vignettes, examples have been updated to include new scenarios and
data.
• The Bringing minds together feature replaces ‘For discussion’.

• Correspondence of exercises to sections is indicated.

In addition, all the text has been carefully examined and modified in the light of reviewers’
comments and to enhance reader-friendly exposition, accessibility of concepts and language, as
well as progression and flexibility. For example, the former Section 4.5 has (mostly) moved to
become Section 6.5. The alternative procedures in Chapter 7 are now mentioned only at the
chapter’s end. In Chapter 10, Section 10.1 discusses sample statistics generally, and non-
parametric procedures have their own section. More of the sections which provide background or
may be chosen to be omitted are moved further to the end of chapters. The large number of
exercises has been retained.
ACKNOWLEDGEMENTS
Thanks to all the colleagues in Australia, New Zealand, UK, South Africa, Canada, USA and
elsewhere internationally, who have used or commented on my materials and strategies, and with
whom I have shared many discussions on the never-ending challenges, fascinations and
richnesses of teaching, learning and students in statistics. Thanks to the Australian and New
Zealand statistical education communities for their initiatives and dedication in statistics
education. Thanks to all lecturing and tutoring staff who have worked with me at a number of
universities and who care passionately about helping students’ learning in statistics.
Thanks to Jessica Utts and Robert Heckard for the spirit of their book and the opportunity to
build on that spirit.
Thanks to all the reviewers and colleagues who have provided valuable feedback and comments –
both general and detailed – on the book. Your efforts and dedication are much appreciated.
Thanks to the Australian Learning and Teaching Council (ALTC) for the many and far-reaching
opportunities given to me by the award of a national Senior Teaching Fellowship, and the
subsequent support as an ongoing Fellow.
Thanks to the thousands of students who worked with commitment in their statistics courses,
and who developed statistical understanding, confidence and appreciation of the importance and
power of statistical thinking beyond their expectations. Their enthusiasm and teamwork in their
data investigations and problem-solving and their willingness to share their thoughts with me
have contributed to my ongoing learning in understanding students’ learning in statistics.
Thanks to the judges of the Australian Educational Publishing Awards for their comments in
awarding the first edition a joint winner in the 2011 Tertiary Education Awards – Teaching and
Learning category (Adaptation).
My sincere appreciation and gratitude to Fiona Hammond, Emily Spurr, Kylie McInnes,
Michaela Skelly, Greg Alford and the staff of Cengage, without whom this book could not, and
would not, have been written. Finally, my thanks and gratitude to my family for their ongoing
support, encouragement and patience, especially Bernie, Bryony and Jen.
The authors and Cengage Learning would also like to thank the following reviewers for their
incisive and helpful feedback:

• Patricia van den Nieuwenhuijzen, ANU

• Raymond Summit, Deakin University

• Irene David, University of Canterbury

• Judi McWhirter, Waikato University

• Di Warren, University of Sydney

• Jason Hay, Griffith University

• Fiona Kate Barlow, University of Queensland

• Gareth Evans, The University of Queensland

• Bronwen Whiting, ANU
 ACKNOWLEDGEMENTS XV

• Carmel Coady, University of Western Sydney

• Maureen Townley-Jones, University of Newcastle

• Peter Petocz, Macquarie University

• Phillip Rayment, Monash University

• Sharon Gunn, University of Melbourne and several other anonymous reviewers.

ABOUT THE AUTHORS

Helen MacGillivray
Helen’s university teaching and curriculum design experience extends across many areas of
statistical sciences and their applications, across all levels of subjects, all class sizes and most
disciplines. Her work has received support through many national or university grants, including a
national (ALTC) leadership grant and one of the first national (ALTC) Senior Fellowships. She has
published widely, including textbooks, book chapters and more than 80 refereed, keynote or
invited papers, and delivered approximately 100 local, national or international presentations and
workshops on learning, teaching and assessment in statistics and quantitative learning support.
Helen was the first female President and the first female Honorary Life Member of the
Statistical Society of Australia Inc. (SSAI), and is now a Vice-President of the International
Statistical Institute. She is a past president of the International Association for Statistical
Education, a Fellow of the Royal Statistical Society, and has also been President of the Australian
Mathematical Sciences Council, Board member of the Federation of Australian Scientific and
Technological Sciences and a member of the Institutional Grants Committee of the ARC. She is
joint chair and editor of OZCOTS, the Australian Conference on Teaching Statistics, and member
of the International Programme Executives for both the 8th and 9th International Conferences on
Teaching Statistics. She has been a member of the organising or editorial committees for many
conferences, including World Statistics Conferences, Australian Statistics Conferences, Southern
Hemisphere Conferences on Undergraduate Mathematics and Statistics Teaching and Learning,
and Australasian Engineering Education Conferences. She has chaired reviews of university
departments and centres, and worked as a consultant on teaching statistics in Australian
universities, and with the Royal Statistical Society Centre in Statistical Education and the UK
Learning and Teaching Support Network for Mathematics, Statistics and OR. She is also currently
Australian representative on the editorial board of the journal Teaching Statistics.
Helen has played key roles in mathematics and statistics school education with the Queensland
Studies Authority, the Australian national curriculum and the Australian Mathematical Sciences
Institute’s TIMES project. This has included work on syllabus committees, state panels, core skills
scrutineering and as a statistical adviser in research and moderation. She has given many
professional development workshops for teachers, and a variety of successful extension and
enrichment programs in mathematics and statistics for high school students.

Jessica M Utts
Jessica Utts is a Professor of Statistics at the University of California at Irvine, previously at Davis
where she joined the faculty in 1978. She received her BA in Math and Psychology at SUNY
Binghamton, and her MA and PhD in Statistics at Penn State University. She is the author of Seeing
Through Statistics (3rd edition, 2005) and the co-author with Robert Heckard of Statistical Ideas and
Methods (1st edition, 2006) both published by Duxbury Press, an imprint of Cengage Learning. She is
also the Editor-in-Chief of CYBERSTATS, an interactive online introductory statistics course.
Jessica has been active in the statistics education community at the high school and college level.
 ABOUT THE AUTHORS XVII

She served as a member and then chaired the Advanced Placement Statistics Development
Committee for six years, and was a member of the American Statistical Association task force that
produced the GAISE (Guidelines for Assessment and Instruction in Statistics Education)
recommendations for Elementary Statistics courses. She is the recipient of the Academic Senate
Distinguished Teaching Award and the Magnar Ronning Award for Teaching Excellence, both at
the University of California at Davis. She is also a Fellow of the American Statistical Association,
the Institute of Mathematical Statistics and the American Association for the Advancement of
Science. Beyond statistics education, Jessica’s major contributions have been in applying statistics
to a variety of disciplines, most notably to parapsychology, the laboratory study of psychic
phenomena.

Robert F Heckard
Robert Heckard is a Senior Lecturer in Statistics at the Pennsylvania State University where he has
taught for over 30 years. He has taught introductory and intermediate applied statistics to more
than 15 000 college students. Bob has been awarded several grants to develop multimedia and
web-based instructional materials for teaching statistical concepts. He is the co-author of Statistical
Ideas and Methods (1st edition, 2006) and is a co-author of CYBERSTATS, a web-based introductory
course. As a consultant, he is active in the statistical analysis and design of highway safety
research and has frequently been a consultant in cancer treatment clinical trials.
RESOURCES GUIDE

As you read this text you will find a number of features to enhance your study of
statistics and help you understand its applications. We have added a number of new
features, based on your feedback, to help you navigate through the text and find what
is most important. These features are indicated by the new to this edition! icon. NEW TO TH
EDITIO IS
N!

CHAPTER 2 GATHERING AND PREPARING USEFUL DATA 55

Fairfax/Steven Siewert

Using the past as a source of data

If at all possible, prospective studies should be used. That is not always possible. For example,
researchers who first considered the potential causes of AIDS or toxic-shock syndrome had to start
with those who were ill and try to find common factors from their pasts. If possible, retrospective
studies should use authoritative sources such as medical records rather than relying on memory.
However, as with other examples, the key is to be aware of the possibilities of the context. Often
this may mean including other variables in the study.

WHERE DID THE CANCER COUNCIL’S DATA COME FROM?

In November 2009, the Cancer Council of Australia marked the start of a National Skin Cancer Action week with a video and
pictures that graphically depicted the dangers of tanning (http://www.youtube.com/watch?v¼QlcTDHvbvIs&noredirect¼1).
The campaign used a picture of 1700 beach towels stretched out on Sydney’s Bondi beach, each with a chalk outline of a
‘victim’ to represent the 1700 Australians who, according to the Cancer Council, die of skin cancer each year in Australia. The
data prompting this campaign, and a range of news and web items since, came from the 2006–2007 National Sun Protection
Survey (http://annualreview.cancer.org.au/2008-09/ReducingTheImpact/ReducingRisk/ReducingSkinCancer/SunSurveyResults.htm).
This second National Sun Protection Survey aimed to investigate people’s behaviour and attitudes to sun protection, including
whether people get sunburnt or attempt to tan. New Zealand’s SunSmart organisation also runs a National Sun Protection Survey,
and has been doing so since 1994. New Zealand’s Sunsmart website (http://sunsmart.org.nz/) says: ‘New Zealand has the
highest rate of melanoma skin cancer in the world. But the good news is it’s easy to reduce your chances of developing skin
cancer – you only have to be SunSmart – Slip, Slop, Slap and Wrap!‘ Compare this with Australia’s slogan since 2007 ‘Slip, Slop,
Slap, Seek, Slide’.
The difficulties in asking questions relating to people’s behaviour are that behaviour tends to vary from day to day and
depends on many individual influences. How does the Cancer Council collect their data so that they can defend the statistics
they quote? Interpretations of questions can also cause problems. For example, a question such as ‘do you wear a hat when
outside in summer?’ is too vague to enable consistent, comparable replies across respondents. Is there evidence that people
tend to heed the messages in the Cancer Council’s campaigns? Are adults different to teenagers?
EXAMPLE 2.10
Do left-handers die young?
Some years ago, a highly publicised study pronounced that left-handed people did not live as long as right-
handed people (Coren and Halpern, 1991). In one part of the study, the researchers had sent letters to next of
kin for a random sample of recently deceased individuals, asking which hand the deceased had used for writing,
drawing and throwing a ball. They found that the average age of death for those who had been left-handed was
66, while for those who had been right-handed, it was 75.
What the researchers failed to take into account was that in the early part of the 20th century, many
children were forced to write with their right hands, even if their natural inclination was to be left-handed.
Therefore, people who died in their 70s and 80s during the time of this study were more likely to be right-
handed than were those who died in their 50s and 60s. The confounding factor of how long ago one learned to
write was not taken into account. A prospective study, following current left- and right-handers to see which
group survived longer, would have enormous practical difficulties; the participants could well outlive the
researchers!
WHERE DID THE CANCER COUNCIL’S DATA COME FROM?

This example is revisited

The beginning of this chapter describes some of the
work of the Australian Cancer Council and New
Zealand’s Sunsmart in providing advice, conducting
campaigns and collecting data on attitudes to
protection from the harmful effects of the sun.
Clearly the type of investigation to obtain such data
is a survey, so how do they conduct it and what
difficulties are encountered?
The National Sun Protection Survey investigates
people’s sun-related knowledge, attitudes and
behaviours. The challenge in this general issue or
topic is to investigate people’s behaviour and
attitudes to sun protection, including whether
people get sunburnt or attempt to tan. The
difficulties in asking questions relating to people’s
behaviour are that behaviour tends to vary from day
to day and to depend on many individual influences.
Interpretations of questions can also cause
problems.
NEW
The Australian National Sun Protection Survey
of 2006–2007 was the second such survey; the first
was conducted in 2003–2004. The study was funded
TO TH
EDITIO IS
in the return to vignette
N!
by the Cancer Council Australia and the Australian
Government through Cancer Australia. The 2006–
2007 survey was carried out by telephone interviews
at the end of the chapter.
on Monday and Tuesday evenings over summer and
asked about weekend behaviour. The interviews
also recorded whether the person was an
These two features act as
adolescent or an adult, and in which state they
lived. cc
‘bookends’ to demonstrate
and illustrate the key concepts
in the chapter.

Chapter opening vignettes

NEW provide you with a contextual
TO TH
EDITIO IS
N!
example to illustrate the key
concepts of each chapter.

5
Investigating categorical variables and their
relationships

In this chapter you will see:

CHAPTER OBJECTIVES
• more about creating and managing categories of categorical variables

• more about planning collection of categorical data

• more about presenting and interpreting presentations of categorical data

• row and column percentages for two categorical variables and their interpretation in context

•
Chapter objectives give you a clear
the possible effects of other variables on the relationships between two categorical variables

• concepts and principles of testing statistical hypotheses, test statistics and the interpretation of p-values

• how to test a model of probabilities/proportions for one categorical variable

sense of what each chapter will cover •

•
how to test if two categorical variables are statistically dependent on each other
measurement and understanding of risk and relative risk.

and what you should be able to do after Section 5.1 summarises key aspects of collecting, presenting and summarising data on categorical variables and poses
SIGNPOSTS

reading the chapter.

questions about them, using data from some real and contemporary investigations as illustrations. This section
thus covers relevant points about planning and collecting categorical data as discussed in Chapter 2, and about
graphs and summaries of categorical data as in Section 3.1, including bar charts, contingency tables, row and
column percentages, and introducing the concepts of conditional percentages.

Section 5.2 looks at exploration and presentation of data from more than two categorical variables, and the importance of
considering the effect(s) of possible confounding or lurking variables, including the phenomenon known as
Simpson’s paradox.
Section 5.3 introduces the chi-square test for the statistical hypothesis specifying a set of proportions for one categorical

Chapter signposts give you a Section 5.4

variable.

uses the test procedure of Section 5.3 to discuss p-values and the general principles of statistical hypothesis

NEW list of the major sections to be

testing. This section thus links with, and builds on, the concepts introduced in Section 4.6, contributing to the
development of understanding of statistical hypothesis testing.
TO TH Section 5.5 covers the chi-square test of independence of two categorical variables – a test that is much used across all
EDITIO IS disciplines. The section also includes some comments on research questions versus statistical questions, and
N!
covered in the chapter. Section 5.6
statistical versus practical significance.
defines and discusses risk and relative risk with examples illustrating the care required in interpreting risk
statistics. The odds ratio is also defined.

Key terms are highlighted in bold and
defined at the end of the chapter.
Mind your step margin icons
NEW highlight things for you to be
TO TH
EDITIO IS
N!
aware of, as well as common
mistakes you can learn to
identify and avoid.
Keep in mind margin icons
NEW
TO TH
EDITIO IS
and highlighted text has
N!
been selected by the author
to draw attention to critical
concepts and knowledge that
you ‘need to know’.
Bringing minds together
NEW
TO TH
EDITIO IS
boxes appear throughout
N!
each chapter and encourage
discussion and questioning
about statistical ideas.
RESOURCES GUIDE XIX
252 MIND ON STATISTICS
Not only is this a voluntary online survey, but the total number of respondents is not given and there is no
disclaimer as in Example 7.1 above. On an episode (24 October 2012) of the Gruen Planet (ABC TV), the Guardian
was praised for its many ways of interacting with, and involving its readers online. How do you think the above
poll could be improved while retaining interaction with readers?
EXAMPLE 7.8
‘Pension savings gender gap widens’
Relevant examples form the basis for
A report ‘Pension savings gender gap widens’ in the UK’s Guardian by Hilary Osborn, 22 October 2012 (http://
www.guardian.co.uk/money/2012/oct/22/pension-savings-gender-gap-widens), includes the following: discussion in each chapter and walk
The report, which was based on interviews with 5,200 adults, found the number of women saving
nothing at all increased year on year from 23% in 2011 to 26% today. The proportion of men not
saving for retirement stands at 19%.
Women who are saving put aside £203.21 a month on average, down from £227 in previous years,
you step-by-step through real-life uses
of statistical concepts.
and 29% are saving on a regular basis.
5 200 is a lot of observations, but the report does not give the breakup into men and women which is critical
information as the emphasis is on comparing men and women. Also it is not clear further in the article if the
29% who are saving on a regular basis is a percentage of the women who are saving or of all the women in the
survey.
In all of the above reports, percentages obtained from data are an integral part of the story, and
all are being used, explicitly or implicitly, as estimates of proportions or probabilities in more
general situations or populations that the data are being assumed to randomly represent. Such
estimates are called point estimates because they provide a single value with no information about
the error of the estimate. Other names that are used are sample estimates or sample statistics. The
quantities they are being used to estimate are called parameters. A parameter is a quantity that is a
measure of some feature or characteristic of a general situation or population.
In considering how appropriate and how good these estimates are, we need to know how the
LINK ME
data were collected so that we can judge for which situation or population the estimates could be
Chapter 2
applied, and we need sufficient information to be able to quantify the variability there could be in
these estimates.
LINK ME
Section 4.5
The fundamental rule for using data for statistical inference is that available data can be used
to make inferences about a more general situation or a much larger group if the data can be
considered to be randomly representative with regard to the question(s) of interest.
MIND YOUR STEP This section has focused on the abundance of proportions quoted in reports, the
1 If you are reporting data, always
report how it was collected and
numbers of observations.
importance of reporting how the data are collected and of including full
information in reports. The remainder of this chapter focuses on procedures to
provide intervals to estimate proportions – intervals that allow for sampling
2 If you are reading reports with variation and that we are confident include the appropriate parameter proportion
percentages, remember that you need
desired to be estimated. Such intervals are called confidence intervals.
to know how the data were collected,
including sufficient information about However, we must never lose sight of the need to know how the data are collected.
the numbers of observations.
262 MIND ON STATISTICS
^2 ¼ sample proportion for the sample from the second situation or population.
p
The point estimate of the difference between the proportion parameters is the sample statistic
^1  p
p ^2 ¼ the difference in sample proportions.
LINK ME You will have noticed in Section 7.2 that the value 1.96 used in the 95% confidence intervals
Section 7.5 comes from the standard normal, and that the interval (1.96, 1.96) has probability 0.95 lying in it
for the standard normal. Section 7.5 explains that we can use this because for large values of n, the
sample proportion has an approximately normal distribution. Similarly, the distribution of the
difference of proportions in two independent samples is approximately normal for sufficiently
large values of the sample sizes n1 and n2, provided the samples are independent and randomly
selected from the two situations or populations of interest.
An approximate 95% confidence interval for p1 – p2 is
0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1
^1 ð1  p
p ^1 Þ p ^ ð1  p ^2 Þ ^1 ð1  p
p ^1 Þ p ^ ð1  p ^2 Þ
KEEP IN MIND @p
^1  p
^2  1:96 þ 2 ^1  p
,p ^2 þ 1:96 þ 2 A
n1 n2 n1 n2
If a different confidence level is wanted, the appropriate standard normal multiplier can be
used instead of 1.96.
You can see the similarities between this confidence interval and the traditional approximate
method for a single proportion; these are discussed in Section 7.5.
The conditions that must be met in order to use the above approximate confidence interval are:
1 Independent sample proportions are available. These could be based on two independent,
randomly selected samples on a categorical variable, or from a randomly selected sample on
two categorical variables, one of which has two categories that we wish to compare.
2 The sample sizes must be reasonably large and the proportions not too close to 0 or 1. One ‘rule
of thumb’ that is sometimes used to help ensure that the approximations are reasonable is that
^1 , n1 ð1  p
all of the quantities n1 p ^2 , and n2 ð1  p
^1 Þ, n2 p ^2 Þ are at least 10. These quantities
represent the counts observed in the category of interest, and not in that category, respectively,
Link me margin icons identify
for the two samples.
EXAMPLE 7.11
(CONTINUED)
NEW where chapter concepts link
Lift or stairs and time of day TO TH
Consider the data in Table 7.2 to compare proportions of people who use the lift to go up in the morning peak EDITIO IS
N!
time with those in the evening peak time. First let us consider the condition we need that we have independent
samples. Do we? We have no way of knowing how many people in the morning sample and in the evening sample
are the same and, if so, is it random if they use the lift or the stairs? The morning and evening samples might be
to concepts in other chapters,
to help you connect the
taken in times or in places in the bus station such that we can assume independence. The bus station is clearly a
very busy place as it was easy to obtain large samples. If we have context knowledge of the bus station and
exactly where and when the samples were taken, we would be able to better comment on this. This illustrates
the importance of reporting as much as possible about the context of the data and how they were collected.
Here all we can do with the information given is to state that we are assuming that the two sample
proportions of morning and evening commuters who use the lift when going up are independent. There is
nothing wrong with stating this assumption. Stating assumptions is very important in all statistical work and all
theory and concepts in a
statistical reports.
larger context.
CHAPTER 2 GATHERING AND PREPARING USEFUL DATA 25
first challenge was to define what they meant by ‘dissolved’ in order to have a consistent measure of time to
dissolve. They decided to define the tablet as ‘dissolved’ when the form of tablet first disappeared, that is, when it
first could not be discerned that there was a tablet. They decided not to stir the liquid to avoid the difficulty of
ensuring the same stirring conditions each time. They chose their experimental conditions to be water temperature
(two temperatures, one approximately room temperature and one cool), pH of water (neutral and a selected
slightly acidic one), and water type (normal and slightly salt water at a selected concentration). Five brands of
soluble aspirin were tested and three tablets from each were tested for each of the eight combinations of
temperature, pH and water type. The same volume of water was used each time. Randomisation came from using a
random order of testing conditions for each brand and choosing the tablets at random for each test.
Bringing minds together 2.2
For chemistry fans: is there any other information about the design of this experiment that should be
reported?
EXAMPLE 2.3
Are brown eggs stronger than white?
This question motivated an investigation into measuring the
Data set margin icons identify where
A. Warr and S. Yu, QUT
strength of eggs – that is, of their shells. The home-made
apparatus held the egg steady on its side under a container
that was gradually filled with water until the shell first
cracked. It had been intended to distinguish between load
Dataset:
Egg crush
relevant datasets are available online
until first crack and load until egg collapsed but in practice
once the egg cracked it collapsed. The measure of strength
was the weight of water at which an egg collapsed. The
explanatory variables were colour (white and non-white),
on the CourseMate website. A list of
examples and cases that relate to
brand of egg, housing (caged or free range), length and width
of egg. The experimenters did not expect the dimensions of
the egg to matter, but recorded them in case they did, as such
information certainly cannot be obtained after testing to
destruction. Other conditions were kept the same, with the
same person adding the water and the same person observing
the eggs.
these datasets can also be found on
Is this an experiment or an observational study? Certainly
an experiment was designed and carried out, but note that colour and dimensions of the eggs are observed not
controlled. Many real investigations are a mixture of experiment and observation. Randomisation in this
experiment is due to random selection of eggs from brands and random order of eggs in carrying out the
the website.
experiment.
XX RESOURCES GUIDE

72 MIND ON STATISTICS CHAPTER 7 ESTIMATING PROPORTIONS WITH CONFIDENCE 257

manufacturing process, it is likely that the company would want to be more precise in a 95%
picture focuses on the comparison, that is, on the parts relative to the whole and this comparison is the same. confidence interval for the proportion of defectives. A range of just below 2% up to just over 7%
The category ‘other’ is very small and could be combined with one of the other categories. Which category would may not be sufficiently precise in a situation where hundreds of thousands of cells are being sold
you choose to combine it with? This depends on the context. In this case, looking at the speeds of the users in by the company. We will see in Section 7.5 how many observations need to be collected in order to
the ‘other’ category of transport shows that they are close in speed to joggers, so they could be combined with
give a more precise estimate for this company. In that section, we will also see that it takes many
the joggers. An alternative procedure could be to omit them from the dataset because there are so few, but the
more observations to reliably estimate a proportion close to 0.5 than it does to
usual procedure is not to throw data away without good reason, and in the case of categorical variables, we
often combine categories in real data investigations. estimate a proportion close to 0 or 1.

Software tip boxes offer details on the

MIND YOUR STEP
The other aspect to notice in this example is that the difference between the
t 3FNFNCFS UIBU UIF GPSNVMB HJWFO
traditional approximate method and the accurate method is not much in actual IFSF JT BQQSPYJNBUF BOE SFRVJSFT n
values but would be important in a situation like this where reliable information is MBSHF BOE p OPU UPP DMPTF UP  PS 

Table and graphs for one categorical variable

• To determine how many and what Simple to graph one variable. Then, in the
use of statistical software technology, needed about defectives in a manufacturing process.

percentage fall into the categories of a ‘Categorical variables’ box, specify the
including Minitab, Excel, SPSS, HOW MANY PASSERSBY WOULD NOTICE THIS?

CASE STUDY 7.1

single categorical variable, use
Stat>Tables>Tally Individual Variables. In
the dialog box, specify a column containing
the raw data for a categorical variable.
column containing the raw data for the
variable. To graph percentages rather
than counts, use the Bar Chart Options A group of university students decided to
button.
R and Genstat. investigate how curious people were about
objects in a public place. Conducting such an

Source: A. Davidson, S. Palmer, S. Ross and

Click on any desired options for counts and
percentages under ‘Display’.
• To draw a bar graph, use Graph>Bar
Minitab tip Chart. In the resulting display, select
Table and graphs for one categorical variable
• To draw a pie chart, use Graph>Pie Chart.
In the pie options, the default is to combine experiment took a lot of thought and more than
two (or more) slices containing fewer than one pilot study, as well as consultation with
0.6% of the observations. security staff and their permission. How to
measure human curiosity was a major challenge
of this investigation, but this example considers
only the categorical measure of degree of notice
Case studies apply statistical ideas to

C. Watterson, QUT
or attention, with categories ‘no attention’, ‘some
• To create a frequency table for one • To create a bar chart or pie chart for one attention’ and ‘considerable attention’. A large
categorical variable, use categorical variable, continue from box was placed in a public thoroughfare for

SPSS tip
Analyze>Descriptive
Statistics>Frequencies.
creating frequencies by selecting Charts
and selecting the desired chart. interesting real-world scenarios, to help 1 hour on three days with sufficient separation
to be able to assume no overlap of people or
sufficient separation of the experience. On the first day, the box was plain. On the second day, the box had the

The history of the pie chart is very interesting. It was developed by William Playfair in 1801.
Florence Nightingale used a form of pie chart with great effect in the presentation of her data on
develop your understanding of concepts visual stimuli as shown in the accompanying photograph. On the third day the box had no visual stimuli but a
mobile phone was constantly ringing inside the box. On each day, observations were taken on 51 randomly
selected ‘groups’ where the number in the group was also recorded (taking values of 1, 2,. . .). The numbers that
Dataset:
Human Curiosity

causes of mortality in military hospitals. Pie charts can be used only for one categorical variable so
have limited use. Unfortunately fancy versions have also been invented that often distort the very
feature that makes a pie chart useful and representative of the data – namely the areas of the
covered in the text. fell in the category of ‘no attention’ were 35 for the plain box, 14 for the visually decorated box, and 36 for the
box with the phone constantly ringing.
From these data, an approximate 95% confidence interval for the probability that no notice is taken of a
pieces of pie which give the relative frequencies of the categories. Three-dimensional pie charts plain box is (0.56, 0.81). And a 95% confidence interval for the probability that no notice is taken of a box with
and doughnut pie charts are poor graphs because neither the third dimension nor the doughnut considerable visual stimuli is (0.15, 0.4).
‘hole’ represents any information, and the essential information of the accurate representation of In the first case, the accurate 95% confidence interval from software is (0.17, 0.41). Notice that the difference
the relative frequencies is hidden or distorted. Similarly, three-dimensional bar charts are poor between the traditional 95% approximate confidence interval and the Jeffrey 95% accurate confidence interval is
graphs because the third dimension does not represent any information and distorts the cc
presentation of the information, which is the height of the bars. Section 3.7 comments on some
reasons for graphs not doing the job they are supposed to do. The various ‘innovations’ in pie
charts seem to be some of the worst in this respect.

At the end of each chapter you’ll find several tools to help you to review the chapter and key learning
concepts, and also to help extend your learning.
The summary section recaps the keys points from each section of the chapter, giving you a snapshot of the
important concepts covered.
Exercises are found at the end of each chapter. They include questions designed for practice and review, as well
as conceptual and data analysis exercises.

Answers to selected exercises are Exercises marked with # have related Key terms are bolded when first
indicated by blue question numbers in datasets that can be found on introduced in the text and are
the exercises section. These questions the CourseMate website. A list of listed with a definition at the end of
have solutions in the back of the text examples and cases that relate to each chapter. You will also find them
for checking your answers and guiding these datasets can also be found on the defined in the full glossary on the
your thinking on similar problems. CourseMate website. CourseMate site.

102 MIND ON STATISTICS CHAPTER 4 DATA SUMMARIES AND INFERENTIAL CONCEPTS 139 122 MIND ON STATISTICS

Summary
3.1 Exploratory data analysis (EDA)
Exploratory data analysis (EDA) refers to procedures to present
data in an informative way, using graphical, pictorial and
summary methods.
3.2 Categorical data
Categorical data fall into categories and are reported in terms of
frequencies or relative frequencies of observations in categories,
We have already seen the interquartile range and, to a lesser extent, the overall range
Exercises
(maximum – minimum) of the data in commenting on how spread out or how concentrated the
# Denotes dataset is available on the website but is not required to solve the exercise. data are.
of the data. Each type of plot has its advantages and
Blue-numbered exercises have answers in the back of the text and on the website.
disadvantages.
3.4 Continuous and categorical data Sections 4.1–4.4
Boxplot of price ($)
4.1 # Student textbooks (refer Example 3.5): EXAMPLE 4.3 (CONTINUED)
It is often desired to compare continuous data across categories
Data were collected on all textbooks from the university 200 *
of categorical variables. Any of the plots of Section 3.3 may be * Referring to the fish lengths of Table 4.1, we can find the quartiles. For the 57 observations on fish lengths
bookshop with staff permission. Course notes were not * * (without the shark):
used provided the same scale is used. Dotplots and boxplots tend
included, nor were general reference books. To qualify, a 150
to be the most useful because they facilitate comparisons.
book was required to have at least five copies on the shelf. t UIF NFEJBO JT UIF UI PCTFSWBUJPO ¼  NN
Price ($)

either of a single variable or the joint categories of two variables.
Tables summarise these, and visual displays can be pie charts
and bar charts. Pie charts can be used only for single variables
but bar charts can present data from two-way tables for two (or
more) categorical variables. Row or column percentages in two-
way tables can indicate or illustrate association between two
categorical variables.
3.3 Graphs and plots for one continuous
variable
A dotplot is a plot of the individual observations of raw data of a
continuous variable with a dot for each observation, or for a
fixed number of observations in very large datasets. A histogram
groups continuous data into ‘bins’ based on intervals chosen by
the investigator or by computer software. If equal-sized intervals
are chosen the heights of the boxes represent frequencies and
relative frequencies of the intervals. A histogram is not a bar
chart and the bins must abut each other. A stem-and-leaf plot is
like a histogram on its side but it retains the original observed
values to a certain number of significant figures, with repeated
digits in the ‘leaves’ representing the frequency of the
corresponding observed value. A boxplot divides the
observations from smallest to largest into four equally
sized groups, with identification of the median, the quartiles
and individual observations if they are away from the bulk
Exercises
# Denotes dataset is available on the website but is not required to solve the exercise.
Blue-numbered exercises have answers in the back of the text and on the website.
Section 3.2
3.1 In a pilot study of vehicles travelling on residential streets,
Where multiple editions of a book were available, only the
3.5 More than one continuous variable most recent was included. The books were classified
Scatterplots are essential tools in exploring relationships according to discipline area, whether the cover was hard or
between continuous variables. The relationships may not fit soft, and if it came with a CD. The level of colour used was
simple representations, may be affected by other variables and considered to be either full (F), some (Y) or none (N). Each
may involve much variation across the whole range or for book was weighed, its thickness measured and its price
particular parts of the range. Scatterplots facilitate and guide (in $) and year of publication noted.
a The stem-and-leaf plot below is of the prices of
exploration and later analysis. Continuous data collected over
textbooks in Law. Use this plot to answer the following
time and dependent on previous observations in time are usually
questions.
explored through time series plots.
Stem-and-leaf of Law price N ¼ 72
3.6 Outlying observations
Leaf Unit ¼ 1.0
Many datasets have extreme observations which may be part of
4 1 1799
the natural variability or may need individual investigation to
7 2 369
check if they correspond to different conditions or are due to
10 3 137
error. No observation should be omitted without good reason. 23 4 2355677777899
Graphs and plots can help identify observations that may need 29 5 223666
checking of circumstances. Sometimes such observations (8) 6 26667799
provide valuable information. 35 7 257
32 8 1145556
3.7 Good graphs and bad graphs 25 9 13999
20 10 334667
Good graphs summarise information in the data in pictures that
14 11 224557799
provide insight and with clear representation. Bad graphs distort
5 12 0239
or misrepresent information, through unnecessary third 1 13 4
dimensions, poor choice of scaling or optical deception.
i Obtain the median of the price for Law texts.
ii Find the lower quartile of the price for Law texts.
4.2
iii From the data, estimate the probability that the
price for a Law textbook is more than $50.
b The boxplots below are of the prices in A$, classified by
colour and type of cover (H ¼ hard, S ¼ soft). For each of
the statements following, decide whether the statement
a Which plot would you choose to represent these data? is an appropriate one to make based only on the boxplots
b Either row or column percentages could be useful for below.
100
* t UIF MPXFS RVBSUJMF JT UIFO IBMGXBZ CFUXFFO UIF UI BOE UI PCTFSWBUJPOT GSPN UIF MPXFS FOE UIJT HJWFT
50 * **
* *
*
0
Cover H S H S H S
Colour F N
i Hard cover books are generally more expensive
than soft cover books.
ii The prices of hard cover books are generally more
variable than those of soft cover books.
iii The average price of a soft cover textbook with no
colour is $75.
iv The standard deviation of the prices of soft cover
textbooks with no colour is approximately $50.
v Half of the soft cover books with some colour are
more expensive than three-quarters of the soft
cover textbooks with no colour.
vi The prices of the soft cover books with some colour
are skew to the left.
vii Some of the observations should be discarded.
(Data source: Textbooks on the website.)
Real-estate data:
The data are of 280 houses sold in four regions from 2000–
2003. Townhouses and duplexes were omitted from the
study.
a The boxplots below are of the selling price per unit land
area, across four regions. For each of the statements
below, decide whether the statement is an appropriate
one to make based only on the boxplots on the next page.
t FBDI PG UIF UXP AIBMWFT PG UIF EBUB IBWF  PCTFSWBUJPOT JO UIFN
 NN
*
*
*
t UIF VQQFS RVBSUJMF JT IBMGXBZ CFUXFFO UIF UI BOE UI PCTFSWBUJPOT GSPN UIF VQQFS FOE UIJT HJWFT  NN
5IF JOUFSRVBSUJMF EJTUBODF JT UIFO  NN
Y
Another measure of spread is the data standard deviation, usually called the sample standard
deviation. We find this by first finding the sample variance. You can think of the sample variance as
roughly the average squared distance that values fall from the mean. Put another way, it measures
variability by summarising how far individual data values are from the mean.
In words, the sample variance is the total of the squared distances from the sample average,
divided by (the number of observations – 1). We will not do an example of finding this ‘by hand’ as
all technology aids provide it – from calculators to statistical software. If you use a calculator,
check that the function you use has the divisor (n – 1) where n is the number of observations.
General notation
In general notation, with n observations with values x1, x2 . . ., xn, and denoting the data or sample
average by x, the sample variance can be written as
!
Xn
ðxi  xÞ2 =ðn  1Þ
i¼1
This is usually denoted by s2. The data or sample standard deviation is then the square root of this
and is denoted by s.
The formula above should not be used for calculations by hand. Any spreadsheet or calculator
with basic statistical functions will calculate s for you. If you need to calculate s by hand, use
! !
Xn

these two variables. If you were commenting on data xi 2  nx2 =ðn  1Þ

the following table gives the numbers by gender and by
i¼1
type of vehicle. for these two variables, give an example of a question
for which you would quote row percentages, and an Any statistical software, and many other software packages of all sorts and varieties will give you
Type example of a question for which you would quote .
what are called descriptive or summary statistics (even if their statistical plotting capabilities are non-
Gender Sedan Family 4WD Sports Total column percentages.
existent or negligible!), including the data median, quartiles, minimum, maximum, average and
Female 39 14 19 8 80 c Do you think the classification of vehicle type is clear by
standard deviation.
Male 34 3 8 5 50 the names? How would you describe ‘family’ and
Total 73 17 27 13 130 ‘sports’?
 RESOURCES GUIDE XXI

Online resources
Visit http://login.cengagebrain.com and login using the code card in the front of
this text for 12 months’ access to the CourseMate website. You’ll find an eBook,
interactive self-assessments, datasets, technical manuals, glossary, flashcards,
crosswords, case questions and more tools to help you excel in your studies.

Enhanced Web Assign

Online homework has never been easier! Enhanced Web Assign is the perfect
solution for your homework management needs, with assignable questions and
an online gradebook. Once set up by instructors, students can access online
review questions related to the text.
Enhanced Web Assign is an extensive online program for statistics to encourage
the practice that’s so critical for concept mastery. The meticulously crafted
pedagogy and exercises in our texts become even more effective in Enhanced
WebAssign, supplemented by multimedia tutorial support and immediate
feedback as students complete their assignments. Key features include: all of
the end-of-chapter problems, conceptual questions, master its, watch its, active
figures and quick quizzes.

For the instructor

Cengage Learning is pleased to provide you with a selection of resources that will help you prepare for your lectures.
These teaching tools are available on the companion website accessible via http://login.cengage.com.

Data sets Technical manuals ExamView® test bank

Provided in CSV format, you can import
these datasets into any statistics
software program. A grid showing the
cases, examples and exercises related to
each dataset is provided for instructors
on the companion website.
Instructor’s manual
The instructor’s manual
provides you with content to
help set up and administer
your subject. It includes
learning objectives,
teaching notes for the ‘bringing minds
together’ discussion questions in the
text, case projects and more.
Solutions manual
The solutions manual provides you
with solutions to all the end-of-chapter
exercises in the text.
The technical manuals provide you with
information on some of the commonly
used statistics technology packages,
and how they can be used with this text.
PowerPointTM presentations
C h a p t e r- b y - c h a p t e r P o w e r P o i n t
presentations cover the main concepts
addressed within the text and can be
edited to suit your own requirements.
Use these slides to enhance your lecture
presentations and to reinforce the
key principles of your subject, or for
student handouts.
Test bank
The question bank for each chapter
addresses learning objectives and key
topics. It can be exported into your
learning management system so you
can easily create, customise and deliver
tests, both in print and online.
ExamView helps you
to create, customise
and deliver tests
in minutes for both print and online
applications. The ‘quick test wizard’
and ‘online test wizard’ guide you
step-by-step through the test-creation
process. With ExamView’s complete
word-processing abilities, you can add
an unlimited number of new questions
to the bank, edit existing questions
and build tests of up to 250 questions
using up to 12 question types. You can
also export the files into Blackboard
or WebCT.
Artwork
These digital files of graphs, tables,
pictures and flow charts from
the text can be used in a variety of
media. Add them into your course
management system, use them within
student handouts or copy them into
lecture presentations.
 shutterstock.com/Claudio Zaccherini
HOW CAN BUSINESSES AND GOVERNMENTS FIND OUT ABOUT TOURISTS?

Tourism is an increasingly important industry for many countries, including Australia and New Zealand. Information is vital
for the tourist industry – for governments and for the many and wide-ranging businesses that depend on tourists. Desirable
information can range from numbers of tourists to their opinions on their experiences and intentions of potential future
tourists. How can governments and the tourist industry find the information they want? And what are some of the problems
with information that is available?
 1
Thinking statistically

In this chapter you will see:

CHAPTER OBJECTIVES
• a little of what statistics is and why it is important for so many disciplines and across all of modern society

• some statistical ‘stories’ illustrating just a little of the roles, importance and usefulness of statistics

• some of the statistical thinking needed even before a data investigation and data collection are done

• a little about key statistical aspects in the discovery of new knowledge.

Section 1.1 provides some initial thoughts about the nature of the discipline of statistics and why it matters across disciplines

SIGNPOSTS
and across society.
Section 1.2 uses a selection of real contexts as case studies to showcase the interest and importance of statistics.

Section 1.3 gives just a few examples of how statistical thinking must start much earlier than might be thought to avoid
problems before data investigations begin. These examples and the case studies of Section 1.2 also illustrate the
richness of statistical thinking.
Section 1.4 gives a few basic steps in the discovery of knowledge in which statistics plays a vital role.
4 MIND ON STATISTICS

1.1 What is statistics?

When you hear the word statistics you probably think of numbers. You might think of sporting
statistics, or percentages and numbers quoted in reports from government or industry or in the
media. This is because information comes from data, so statistics is very much involved in
obtaining, understanding, analysing and interpreting data. But because statistics is concerned with
systems and processes involving variation and uncertainty, it is also the science of probability and
randomness.
The word ‘science’ does not restrict statistics to any particular area of science, such as the
physical, life, health or social sciences, but rather has its general meaning of ‘discipline’,
‘knowledge’, ‘skills’ or ‘art’, and refers to the objectivity and evidence-based principles and
procedures of statistics. Certainly, statistics is very important in all areas of science, as well as in
business, engineering, law and any discipline that uses quantitative information and evidence-based
reasoning.
Thus statistics is the discipline that helps in obtaining, understanding and using information in
all real and complex systems and processes that involve uncertainty and variation. It is not a set
of procedures that provide answers by pushing buttons. Statistics provides ways of tackling
uncertainty and variation that are objective and powerful but require understanding and
interpretation. Hence statistics is demanding but exciting in how it can help solve an amazing
variety of problems.
Uncertainty and variation are tricky phenomena to tackle, but they are everywhere in real life.
Real systems and processes also tend to be complex. So it should be no surprise to learn that
statistics is a big and complex discipline and that statistics and statisticians are involved in
tackling all sorts of problems, large and small, across an amazingly broad set of disciplines. But
statistics also tackles problems and interprets results in context, and so is a highly collaborative
discipline. This means that statistical understanding must be shared – statisticians have to find
out about contexts in order to help solve problems in them, and people whose problems need
statistical assistance need at least some understanding of what statistics is about and how to
think statistically. In addition, because variation and uncertainty permeate our everyday lives,
everyone needs to understand enough about variation and uncertainty for ‘citizenship fitness’. So
no matter what you are studying or what careers you follow or what types of workplaces lie in
your future, you are likely to benefit from a foundation in how statistics tackles problems.
Statistics is also like other quantitative disciplines in that mathematics and mathematical thinking
provide invaluable tools and underpin its procedures and skills, but it has its own principles and
thinking. So those who like mathematics are able to find and pursue mathematics in statistics.
Those who do not like mathematics can focus on the thinking, application and interpretation of
statistics in real contexts, remembering that mathematics is a servant to statistics in the same
way that it serves other disciplines.
Because obtaining, understanding and interpreting data tend to be the statistical focus in all
disciplines needing statistics, most of this book aims to develop a common foundation in the
principles and procedures for gathering and analysing data. But there is much more to this than
being handed some numbers, applying a formula or pushing a button and recording the result.
The case studies and examples of this chapter aim not only to give you glimpses into the powerful
and exciting world of statistics, but also to introduce some of the richness and challenge of thinking
statistically that is needed in learning and using statistical principles and procedures.
 CHAPTER 1 THINKING STATISTICALLY 5

1.2 Some statistical stories in real and complex problems

The best way to gain an understanding of some of the concepts and methods in statistics is to see
them in action. Each of the case studies presented in this section gives some idea of what can be
involved in statistics-in-action. Although the case studies here are complex, with some requiring
very advanced statistical methodology, they illustrate aspects of statistical problem-tackling that
appear throughout the book. These case studies will give you some examples of why statistics is
such an important and valuable discipline.

CASE STUDY 1.1

TREATING PARKINSON’S DISEASE WITH DEEP BRAIN STIMULATION
Parkinson’s disease is a complex neurodegenerative disorder, affecting an estimated 1–2% of people aged 60
years and over worldwide. There is substantial variation in symptom profiles between affected patients, arising
from the presence and severity of many different symptoms. St Andrew’s Hospital in Queensland specialises in a
treatment called deep brain stimulation, and researchers and neurologists there are constantly searching for ways
to understand and improve patient outcomes.
Statisticians are working with medical researchers and practitioners on a number of aspects. The first
problem was to improve the classification and summary system for symptoms to assess patients before
treatment and three months after treatment. The data for each patient were scored on a paper chart, and
occupied almost 80 lines, consisting of 42 different symptoms, with a number having sub-symptoms such as
left and right characteristics, with a rating of severity increasing from 0 to 5. These scores were added over
broad categories; previous methods had often been restricted only to presence or absence of symptoms rather
than ratings. The statistical challenge was to find the best way to group the symptoms and to combine the
ratings of the symptoms to produce useful and validated profile measures with which to rate patients’
conditions before and after treatment. Entering all the data was the least of the challenge! The result is a
much improved system for characterising patients’ conditions and for evaluating effects of treatments
(White et al., 2012).
Another research challenge for statistics is to develop methods to characterise and analyse signal data from
the brain during the treatment. Deep brain stimulation requires extremely accurate pinpointing of a target in
the brain. Signals from the brain not only indicate correct positioning of the stimulation but also monitor the
effect of the treatment in creating neurological connections. As you would expect, such signals are highly
variable and have a number of characteristics. The statistical research seeks measures that capture the key
features of the signals, and that describe patterns that correspond to different stimuli. These will help
understand neural behaviour and response to stimuli, thus assisting in measuring effects of treatment and in
improving deep brain stimulation.

Finding ways to represent complex phenomena with lots of variation is a statistical challenge
in many applications. In Case study 1.1, finding, scaling, choosing, classifying and combining
measures are just some parts of the problem. Detecting, defining, characterising and summarising
the important properties of a varying process or signal are others. You will see in this book
examples of simpler investigations in which the same types of questions arise. Some of these
questions are: What can we observe or measure? Can we measure what we want? What else
Another random document with
no related content on Scribd:
order of Asteroidea to which Asterias belongs, the adambulacrals
themselves do not project much, but in all other cases they form
prominent mouth-angles, so that the opening of the mouth becomes
star-shaped (Fig. 211, p. 483).

Except in the case of the ambulacral and adambulacral plates little

regular arrangement is to be detected in the ossicles of the skeleton
which, as has already been mentioned, form a mesh-work. If,
however, the arm be cut open and viewed from the inside it will be
seen that the edge is strengthened above and below by very thick,
powerful, rod-like plates. These are called the "supero-marginal" and
"infero-marginal" ossicles; they are not visible from the outside, since
they are covered by a thick layer of the body-wall containing other
smaller plates (Fig. 190, marg). In many genera, however, they are
exposed, and form a conspicuous edging to the arm above and
below. In many genera, also, there are three conspicuous series of
plates on the back of each arm, viz. a median row, called "carinals"
(car., Fig. 191), and two lateral rows, termed "dorso-laterals" (d.lat.,
Fig. 191). These three rows, with the two rows of marginals, one of
ambulacrals, and one of adambulacrals on each side (11 rows in all),
constitute the primitive skeleton of the arm, and appear first in
development.

The structure of all these elements of the skeleton is the same. They
may be described as scaffoldings of carbonate of lime,
interpenetrated by a mesh-work of cells fused with one another, by
which the carbonate of lime has been deposited. The matrix in which
the ossicles lie is a jelly-like substance traversed by a few bands of
fibres which connect the various rods with one another. This jelly is
almost fluid in the fresh state, but when heated forms a hard
compound, possibly allied to mucin, which will turn the edge of a
razor.

When the covering of the back is dissected off the coelom is

opened. This is a spacious cavity which apparently surrounds the
alimentary canal and extends into the arms. It has, however, its own
proper wall, which is called the "peritoneum," both on the outer side,
where it abuts on the skin, and on the inner side, where it comes in
contact with the wall of the alimentary canal. The outer wall is called
the "somatic peritoneum," and it is possible to dissect off the rest of
the body-wall and leave it intact; the inner wall, from its close
association with the alimentary canal, is termed the "splanchnic
peritoneum." This wall can only be distinguished in microscopic
sections from that of the alimentary canal, to which it is closely
applied.

The coelom is filled with a fluid, which is practically sea water with a
little albuminous matter in solution. Through the thin walls of the
papulae oxygen passes into this fluid, whence it easily reaches the
inner organs, since they are all in contact with some part of the
coelomic wall. Similarly CO2 is absorbed by the coelomic fluid from
all parts of the body, and diffuses through the papulae to the
surrounding water.

The Starfish possesses no definite kidney for getting rid of

nitrogenous waste. In most of the higher animals with a well-
developed coelom it has been proved that the kidney is simply a
specialised portion of the coelom, and in many cases some parts of
the coelomic wall still retain their excretory functions, which
apparently the whole originally possessed. In the Starfish and in
Echinodermata generally this primitive state of affairs is still retained.
From the cells forming the coelomic wall, cells are budded off into
the fluid, where they swim about. These cells from their movements
are called amoebocytes. If a substance such as indigo-carmine,
which when introduced into the tissues of the higher animals is
eliminated by the kidney, is injected into the Starfish, it is found soon
after to be vigorously absorbed by the amoebocytes. These later
accumulate in the dermal branchiae, through the thin walls of which
they make their way[442] to the outside, where they degenerate.

The coelom is indented by five folds, which project inwards from the
interradii. These folds are called the "interradial septa"; they are
stiffened by a calcareous deposit, which is not, however, sufficiently
dense to constitute a plate. In one of the septa the axial sinus and
stone-canal (see below) are embedded. These septa are to be
regarded as areas of lateral adhesion between the arms.

Fig. 188.—View of upper half of a specimen of Asterias rubens, which has been
split horizontally into two halves. ax.c, Axial sinus; g.d, genital duct; oe, cut
end of the oesophagus, the narrow neck of the stomach; py, pyloric sac;
py.c, pyloric caeca; r, rectum; r.c, rectal caeca; sept, interradial septum; st.c,
stomach lobe.

The alimentary canal consists of several distinct portions. The

mouth leads by a narrow neck called the "oesophagus" into a
voluminous baggy sac termed the "stomach," which is produced into
ten short pouches, two projecting into each arm. The stomach leads
in turn by a wide opening into a pentagonal flattened sac, the "pyloric
sac," which lies above it. Each angle of the pyloric sac is prolonged
into a tube—the so-called "pyloric duct"—running out into the arm,
where it immediately bifurcates into two forks, each beset by a large
number of small pouches and attached to the dorsal wall of the
coelom by suspensory bands of membrane called mesenteries.
These ten forks are called "pyloric caeca"; they are of a deep green
colour owing to the pigment in their wall. Beyond the pyloric sac the
alimentary canal is continued as the slender "rectum" to the anus.
The rectum gives off two small branched pouches of a brown colour
called "rectal caeca." This comparatively complicated form of
alimentary canal is related to the nature of the food of the animal and
the method it employs to capture its prey.

Fig. 189.—View of a Starfish (Echinaster) devouring a Mussel. 1. The

madreporite.

The favourite food[443] of Asterias consists of the common bivalves

of the coast, notably of the Mussel (Mytilus edulis). There is,
however, no animal which it will not attack if it is fortunate enough to
be able to catch it. The Starfish seizes its prey by the tube-feet, and
places it directly under its mouth, folding its arms down over it in
umbrella fashion. The muscles which run around the arms and disc
in the body-wall contract, and the pressure thus brought to bear on
the incompressible fluid contained in the coelom, forces out the thin
membranous peristome and partially turns the stomach inside out.
The everted edge of the stomach is wrapped round the prey.

Soon the bivalve is forced to relax its muscles and allow the valves
to gape. The edge of the stomach is then inserted between the
valves and applied directly to the soft parts of the prey which is thus
completely digested. When the Starfish moves away nothing but the
cleaned shell is left behind. If the bivalve is small it may be
completely taken into the stomach, and the empty shell later rejected
through the mouth.
It was for a long time a puzzle in what way the bivalve was forced to
open. Schiemenz[444] has, however, shown that when the Starfish
folds itself in umbrella-like form over the prey it holds on to the
substratum by means of the tube-feet of the distal portions of the
arms, whilst, by means of the tube-feet belonging to the central
portions, it drags apart the valves by main force. He has shown
experimentally: (1) that whilst a bivalve may be able to resist a
sudden pull of 4000 grammes it will yield to a pull of 900 grammes
long continued; (2) that a Starfish can exert a pull of 1350 grammes;
(3) that a Starfish is unable to open a bivalve unless it be allowed to
raise itself into a hump, so that the pull of the central tube-feet is at
right angles to the prey. A Starfish confined between two glass plates
walked about all day carrying with it a bivalve which it was unable to
open.

The lining of the stomach is found to consist very largely of mucus-

forming cells, which are swollen with large drops of mucus or some
similar substance. It used to be supposed that this substance had
some poisonous action on the prey and paralysed it, but the
researches of Schiemenz show that this is incorrect. If when an
Asterias is devouring a bivalve another be offered to it, it will open it,
but will not digest it, and the victim shows no sign of injury but soon
recovers. The cells forming the walls of the pyloric sac and its
appendages are tall narrow cylindrical cells crowded with granules
which appear to be of the nature of digestive ferment. This
substance flows into the stomach and digests the captured prey.

A very small amount of matter passes into the rectum and escapes
by the anus, as the digestive powers of the Starfish are very
complete. The rectal caeca are lined by cells which secrete from the
coelomic fluid a brown material, in all probability an excretion, which
is got rid of by the anus.

When the meal is finished the stomach is restored to its former place
by the action of five pairs of retractor muscles, one pair of which
originates from the upper surface of the ambulacral ossicles in each
arm and extends to the wall of the stomach, where they are inserted
(Fig. 190, ret).

The tube-feet, which are at once the locomotor and the principal
sensory organs of the Starfish, are appendages of that peculiar
system of tubes known as the water-vascular system, which is
derived from a part of the coelom cut off from the rest during the
development of the animal. This system, as already mentioned,
consists of (1) a narrow "ring-canal," encircling the mouth and lying
on the inner surface of the membranous peristome; (2) a radial canal
leaving the ring-canal and running along the under surface of each
arm just above the ambulacral groove; (3) a vertical stone-canal
running from the madreporite downwards to open into the ring-canal
in the interspace between two arms. The madreporite is covered
externally by grooves lined with long cilia, and is pierced with narrow
canals of excessively fine calibre, the walls of which are also lined by
powerful cilia. Most of these narrow canals open below into a main
collecting canal, the stone-canal, but some open into a division of the
coelom termed the axial sinus, with which also the stone-canal
communicates by a lateral opening. The cavity of the stone-canal is
reduced by the outgrowth from its walls of a peculiar Y-shaped
projection, the ends being rolled on themselves in a complicated way
(Fig. 190, B). The walls of the canal consist of a layer of very long
narrow cells, which carry powerful flagella, and outside this of a crust
of calcareous deposit, which gives rigidity to the walls and has
suggested the name stone-canal.

The tube-feet are covered externally by ectoderm, inside which is a

tube in connexion with the radial water-vascular canal. This latter is
lined by flattened cells, which in the very young Starfish are
prolonged into muscular tails; in the older animal these tails are
separated off as a distinct muscular layer lying between the
ectoderm and the cells lining the cavity of the tube. The tube-foot is
prolonged inwards into a bulb termed the "ampulla," which projects
into the coelom of the arm and in consequence is covered outside by
somatic peritoneum. Just where the ampulla passes into the tube-
foot proper the organ passes downwards between two of the
powerful ambulacral ossicles which support the ambulacral groove,
and a little below this spot a short transverse canal connects the
tube-foot with the radial canal which lies beneath these ossicles (Fig.
191).

Fig. 190.—A, view of the under half of a specimen of Asterias rubens, which has
been horizontally divided into two halves. B, enlarged view of the axial
sinus, stone-canal and genital stolon cut across. amb.oss, Ambulacral
ossicle; amp. ampullae of the tube-feet; ax.s, axial sinus; gon, gonad; g.stol,
genital stolon; marg, marginal ossicle; nerv.circ, nerve ring; oe, cut end of
oesophagus; pst, peristome; ret, retractor muscle of the stomach; sept,
interradial septum; stone c, stone-canal; T, Tiedemann's body; w.v.r, water-
vascular ring-canal.

The tube-feet are, therefore, really a double row of lateral branches

of the radial canal. The appearance of being arranged in four rows is
due to the fact that the transverse canals connecting them with the
radial canal are alternately longer and shorter so as to give room for
more tube-feet in a given length of the arm. Each tube-foot ends in a
round disc with a slightly thickened edge. The radial canal terminates
in a finger-shaped appendage, called the median tentacle, at the
base of which is the eye.
The manner in which this complicated system acts is as follows:—
When the tube-foot is to be stretched out the ampulla contracts and
drives the fluid downwards. The contraction of the ampulla is brought
about by muscles running circularly around it. The tube-foot is thus
distended and its broad flattened end is brought in contact with the
surface of the stone over which it is moving and is pressed close
against it. The muscles of the tube-foot itself, which are arranged
longitudinally, now commence to act, and the pressure of the water
preventing the tearing away of the sucker from the object to which it
adheres, the Starfish is slowly drawn forward, whilst the fluid in the
tube-foot flows back into the ampulla.

Fig. 191.—Diagrammatic cross-section of the arm of a Starfish. adamb,

Adambulacral ossicle; amb, ambulacral ossicle; amp, ampulla of tube-foot;
branch, papula; car, carinal plate; d.lat, dorso-lateral plate; inf.marg, infero-
marginal plate; p.br, peribranchial space; ped, pedicellaria; s.marg, supero-
marginal plate. The nervous ridge between the bases of the tube-feet and
the two perihaemal canals above this ridge are shown in the figure but not
lettered.

If each tube-foot were practically water-tight, then each would be

entirely independent of all the rest, and it would not be easy to
suggest a reason for the presence of the complicated system of
radial canals and stone-canal. Just at the spot, however, where the
transverse canal leading from the radial canal enters the tube-foot
there is a pair of valves which open inwards and allow fluid to pass
from the radial canal into the tube-foot but prevent any passing
outwards in the reverse direction. The presence of these valves
renders it probable that the tube-foot is not quite water-tight; that
when it is distended under the pressure produced by the contraction
of the muscles of the ampulla, some fluid escapes through the
permeable walls; and that the loss thus suffered is made up by the
entry of fresh fluid from the radial canal. The radial canal in turn
draws from the ring-canal, and this last is supplied by the stone-
canal, the cilia of which keep up a constant inward current.

In the fluid contained in the water-vascular system, as in the

coelomic fluid, there are amoebocytes floating about. These are
produced in short pouches of the ring-canal, nine in number, which
are called after their discoverer "Tiedemann's bodies" (Fig. 190, T).
From the cells lining these the amoebocytes are budded off.

The nervous system of the Starfish is in a very interesting

condition. The essential characteristic of all nervous systems is the
presence of the "neuron," a cell primitively belonging to an
epithelium but which generally has sunk below the level of the others
and lies amongst their bases. This type of cell possesses a round
body produced in one direction into a long straight process, the
"axon," whilst in the other it may have several root-like processes, or
"dendrites," which may spring from a common stem, in which case
the neuron is said to be "bipolar." The axon is often distinguished as
a "nerve-fibre" from the round body which is termed the "nerve-cell."
This is due to the fact that for a long time it was not recognised that
these two structures are parts of a whole.

Now at the base of the ectoderm all over the body of the Starfish
there is to be found a very fine tangle of fibrils; these are to be found
partly in connexion with small bipolar neurons lying amongst them
and partly with isolated sense-cells scattered amongst the ordinary
ectoderm cells. This nervous layer is, however, very much thickened
in certain places, so as to cause the ectoderm to project as a ridge.
One such ridge is found at the summit of each ambulacral groove
running along the whole under surface of the arm and terminating in
a cushion at the base of the median tentacle of the water-vascular
system. This ridge is called the radial nerve-cord. The five radial
nerve-cords are united by a circular cord, the nerve-ring, which
appears as a thickening on the peristome surrounding the mouth.
The sense-organs of the Starfish are chiefly the discs of the tube-
feet. Round the edges of these there is a special aggregation of
sense-cells; elsewhere, as in the skin of the back, only isolated
sense-cells are found, and it becomes impossible to speak of a
sense-organ.

A prolongation of the radial nerve-cord extends outwards along one

side of each tube-foot. This is often spoken of as the "pedal nerve,"
but the term nerve is properly retained for a mere bundle of axons
such as we find in the higher animals, whereas the structure referred
to contains the bodies of nerve-cells as well as their outgrowths or
cell-fibres and is therefore a prolongation of the nerve-cord.

Fig. 192.—Diagrammatic longitudinal section through a young Asteroid passing

through the tip of one arm and the middle of the opposite interradius. This
diagram is generalised from a section of Asterina gibbosa. ab, Aboral sinus;
ax, axial sinus; ax1, basal extension of axial sinus forming the inner
perihaemal ring-canal; br, branchia = gill = papula; g.r, genital rachis; mp,
madreporite; musc.tr, muscle uniting a pair of ambulacral ossicles; nerv.circ,
nerve-ring; n.r, radial nerve-cord; oc, eye-pit; oss, ossicles in skin; p.br,
peribranchial sinus; p.c, pore canal; perih (on the right), perihaemal radial
canal, (on the left), outer perihaemal ring-canal; py, pyloric caecum; rect,
rectum; rect.caec, rectal caeca; sp, spines; st.c, stone-canal; t, median
tentacle terminating radial canal; w.v.r, water-vascular radial canal. The
genital stolon (not marked by a reference line) is seen as an irregular band
accompanying the stone-canal, its upper end projects into a small closed
sac, also unmarked, which is the right hydrocoele or madreporic vesicle.

At the base of the terminal tentacle the radial nerve-cord ends in a

cushion. This cushion is called the "eye," for it is beset with a large
number of cup-shaped pockets of the ectoderm. Each pocket is lined
partly by cells containing a bright orange pigment and partly by
visual cells each of which ends in a small clear rod projecting into the
cavity of the pit (Fig. 193, A, vis.r). The pit is apparently closed by a
thin sheet of cuticle secreted by the most superficial cells.

An exposed nervous system and simple sense-organs such as the

Starfish possesses lend themselves admirably to the purposes of
physiological experiment, and so Starfish have been favourite
"corpora vilia" with many physiologists.

Fig. 193.—A, longitudinal section of a single eye-pit of Asterias. s.n, Nucleus of

supporting cell; vis.n, nucleus of visual cell; vis.r, visual rod. B, view of the
terminal tentacle showing the eye-pits scattered over it. (After Pfeffer.)

The light-perceiving function of the eye is easily demonstrated. If a

number of Starfish be put into a dark tank which is illuminated only
by a narrow beam of light they will be found after an interval to have
collected in the space reached by the beam of light.[445] If all the
median tentacles but one be removed this will still be the case; if,
however, they are all removed the Starfish will exhibit indifference to
the light.

If the under surface of a Starfish be irritated by an electric shock or a

hot needle, or a drop of acid, the tube-feet of the affected area will
be strongly retracted, and this irritation will be carried by the pedal
nerves to the radial nerve-cord, with the result that finally all the
tube-feet in the groove will be retracted and the groove closed by the
action of the transverse muscle connecting each ambulacral ossicle
with its fellow. If, on the other hand, the back of a Starfish be irritated
this may produce a contraction of the tube-feet if the irritation be
strong, but this will be followed by active alternate expansions and
contractions, in a word, by endeavours to move. Preyer[446] by
suspending a Starfish ventral surface upward, by means of a small
zinc plate to which a string was attached which passed through a
hole bored in the back and through the mouth, caused movements of
this description which lasted for hours. Irritation of the back causes
also activity of the local pedicellariae, which open their valves widely
and then close them with a snap in the endeavour to seize the
aggressor.

The uninjured Starfish in moving pursues a definite direction, one

arm being generally directed forwards, but this may be any one of
the five. The tube-feet of this arm are directed forwards when they
are stretched out, by the slightly unequal contraction of the
longitudinal muscles of opposite sides of the foot, which persists
even when the circular muscles of the ampulla are contracting. They
thus may be said to swing parallel to the long axis of the arm. The
tube-feet of the other arms assist in the movement, and hence swing
obliquely with reference to the long axis of the arm to which they
belong, although they move parallel to the general direction in which
the Starfish is moving. A change in the direction of the swing of the
tube-feet will bring about a change in the direction of the movement
of the animal as a whole. If now the connexion of each radial nerve-
cord with the nerve-ring be cut through, each arm will act as a
separate Starfish and will move its tube-feet without reference to the
movement of those in the other arms, so that the animal is pulled
first one way and then another according as the influence first of one
arm and then of another predominates. Similarly, when a Starfish is
placed on its back, it rights itself by the combined action of the tube-
feet of all the arms, extending them all as widely as possible, those
which first catch hold being used as the pivot for the turning
movement. If, however, the radial nerve-cords are cut through, each
arm tries to right itself and it is only by chance that the efforts of one
so predominate as to turn the whole animal over. From these
experiments it is clear that the nerve-ring acts as co-ordinator of the
movements of the Starfish, that is to say as its brain.

If a section be taken across the arm of a Starfish (Fig. 191), it will be

seen that between the V-shaped ridge constituting the radial nerve-
cord and the radial water-vascular canal there are two canals lying
side by side and separated from one another by a vertical septum.
These canals are not mere splits in the substance of the body-wall,
but have a well-defined wall of flattened cells. They are termed, for
reasons which will be explained subsequently, perihaemal canals,
and they open into a circular canal called the "outer perihaemal ring,"
situated just beneath the water-vascular ring-canal (Fig. 192, perih).
These canals originate as outgrowths from the coelom. From their
upper walls are developed the muscles which connect the pairs of
ambulacral ossicles and close the groove, and also those which
connect each ossicle with its successor and predecessor and help to
elevate or depress the tip of the arm.

In most of the higher animals the processes of many of the ganglion-

cells are connected together in bundles called "motor nerves," which
can be traced into contact with the muscles, and thus the path along
which the stimulus travels in order to evoke movement can clearly be
seen. No such well-defined nerves can be made out in the case of
the Starfish, and it is therefore interesting when exceptionally the
paths along which stimuli travel to the muscles can be traced. This
can be done in the case of the muscles mentioned above. Whereas
they originate from the dorsal walls of the perihaemal canals,
ganglion-cells develop from the ventral walls of these canals, which
are in close contact with the nerve-cord, so that the nervous system
of the Starfish is partly ectodermic and partly coelomic in origin.
Stimuli reaching the ectodermic ganglion-cells are transmitted by
them to the nervous part of the wall of the perihaemal canal and from
that to the muscular portion of the same layer of cells.

Besides the radial perihaemal canals and their connecting outer

perihaemal ring there are several other tubular extensions of the
coelom found in the body-wall. These are:—

(1) The "inner perihaemal canal," a circular canal in close contact

with the inner side of the outer perihaemal canal (Fig. 192, ax1).

(2) The "axial sinus" (ax) a wide vertical canal embedded in the
body-wall outside the stone-canal. This canal opens into the inner
perihaemal canal below; above it opens into several of the pore-
canals and into the stone-canal. The separation of the axial sinus
from the rest of the coelom is the remains of a feebly marked
metamerism in the larva.

(3) The "madreporic vesicle," a closed sac embedded in the dorsal

body-wall just under the madreporite. This sac by its history in the
larva appears to be a rudimentary counterpart of the water-vascular
system, since this organ in correspondence with the general bilateral
symmetry of the larva is at first paired. Into this a special process of
the genital stolon projects.

(4) The "aboral sinus" (Fig. 192, ab), a tube embedded in the dorsal
body-wall running horizontally round the disc. The aboral sinus
surrounds the genital rachis (see p. 452) and gives off into each arm
two branches, the ends of which swell so as to surround the genital
organs. It has no connexion with the axial sinus though the contrary
has often been stated by Ludwig.[447]

(5) The "peribranchial spaces," circular spaces which surround the

basal parts of the papulae (Fig. 192, p.br).

Besides these, large irregular spaces have been described as

existing in the body-wall by Hamann[448] and other authors, but for
various reasons and especially because they possess no definite
wall they appear to be nothing more than rents caused by the
escape of CO2 gas during the process of decalcifying, to which the
tissues of the Starfish must be subjected before it is easy to cut
sections of them.

The question as to whether or not there is a blood system in the

Starfish has an interesting history. It must be remembered that the
examination of the structure of Echinodermata was first undertaken
by human anatomists, who approached the subject imbued with the
idea that representatives of all the systems of organs found in the
human subject would be found in the lower animals also. So the
perihaemal canals were originally described as blood-vessels. Later,
Ludwig[449] discovered a strand of strongly staining material running
in each septum which separates the two perihaemal canals of the
arm. Each of these radial strands could be traced into connexion
with a circular strand interposed between the outer and the inner
perihaemal ring-canals. This circular strand again came into
connexion with a brown, lobed organ, lying in the wall of the axial
sinus, and this in turn joined at its upper end a circular cord of
pigmented material adhering to the dorsal wall of the coelom (lying in
fact within the aboral sinus), from which branches could be traced to
the generative organs. Ludwig concluded that he had at last
discovered the true blood-vessels, though the facts that the radial
strands and the oral circular strand absorbed neutral carmine
strongly and that the vertical and aboral strands were pigmented,
constituted a very slender basis on which to found such a
conclusion. The colour apparently appealed to the imagination, and it
is undoubtedly true that the "plasma" or blood-fluid of other animals
often absorbs stain strongly.

The strands were accordingly named "radial blood-vessels," "oral

blood-ring," "aboral blood-ring"; and the brown vertical strand was
called the "heart," although no circulation or pulsations had ever
been observed. When later investigations revealed the fact that the
so-called heart was practically solid, the term "central blood-plexus"
was substituted for heart, although it was still regarded as the central
organ of the system. The name "perihaemal" was given to the
spaces so called because they surrounded the supposed blood-
vessels.

In order to come to a satisfactory conclusion on the matter some

general idea as to the fundamental nature and function of the blood-
vessels in general must be arrived at. Investigations made on
various groups of animals, such as Annelida, Mollusca, Crustacea,
Vertebrata, show that at an early period of development a
considerable space intervenes between the alimentary canal and the
ectoderm, which is filled with a more or less fluid jelly. Into this cavity,
the so-called "primary body-cavity" or "archicoel," amoebocytes,
budded from the ectoderm or endoderm or both, penetrate. In this
jelly with its contained amoebocytes is to be found the common
rudiment both of the connective tissue and of the blood system. The
resemblance of the archicoele and its contents to the jelly of a
Medusa is too obvious to require special insistence on, and therefore
in the Coelenterata it may be stated that there is to be found a tissue
which is neither blood system nor connective tissue but is the
forerunner of both.

In the higher animals as development proceeds the jelly undergoes

differentiation, for some of the amoebocytes become stationary and
connected with their pseudopodia so as to form a protoplasmic
network. A portion of this network becomes altered into tough fibres,
but a portion of each strand remains living, and in this way the
connective tissue is formed. In the interstices of the network of fibres
a semi-fluid substance (the unaltered jelly) is found, and this is
traversed by free, wandering amoebocytes. In other places the jelly
becomes more fluid and forms the plasma, or liquid of the blood,
whilst the amoebocytes form the blood corpuscles. The blood
system thus arises from regions of the archicoel where fibres are not
precipitated.

Now in the Starfish the whole substance of the body-wall intervening

between the ectoderm and the coelomic epithelium really represents
the archicoel. The formation of fibres has, it is true, proceeded to a
certain extent, since there are interlacing bundles of these, but there
are left wide meshes in which amoebocytes can still move freely.
Apart from the skeleton, therefore, the tissues of the body-wall of the
Starfish do not exhibit much advance on those of a Jellyfish. If
anything is to be compared to the blood system of the higher animals
it must be these meshes in the connective tissue. From observations
made on other Echinoderms it appears probable that the colour of
the skin is due to amoebocytes loaded with pigment wandering
outwards through the jelly of the body-wall and disintegrating there.
The strands regarded as blood-vessels by Ludwig are specially
modified tracts of connective tissue in which fibres are sparse, and in
which there are large quantities of amoebocytes and in which the
"jelly" stains easily. Cuénot[450] suggests that they are placed where
new amoebocytes are formed; this is quite possible, and in this case
they ought to be compared to the spleen and other lymphatic organs
of Vertebrates, and not to the blood-vessels.[451]

The organ regarded as the heart, however, belongs to a different

category: it is really the original seat of the genital cells and should
be termed the "genital stolon." Careful sections show that at its
upper end it is continuous with a strand of primitive germ-cells which
lies inside the so-called aboral blood-vessel, and is termed the
"genital rachis" (Fig. 192, g.r). The germ-cells are distinguished by
their large nuclei and their granular protoplasm. The genital organs
are only local swellings of the genital rachis, and from the shape of
some of the germ-cells it is regarded as highly probable that the
primitive germ-cells wander along the rachis and accumulate in the
genital organs. The genital rachis itself is an outgrowth from the
genital stolon, and this latter originates as a pocket-like ingrowth of
the coelom into the wall separating it from the axial sinus; when fully
formed it projects into and is apparently contained in this latter
space.

Not all the cells forming the genital stolon become sexual cells. Many
degenerate and become pigment-cells, a circumstance to which the
organ owes its brown colour. In very many species of Starfish many
of the cells of the genital rachis undergo a similar degeneration, and
hence is produced the apparent aboral blood-vessel. Further, the
rachis is embedded in connective tissue which has undergone what
we may call the "lymphatic" modification, and this for want of a better
name we call the "aboral" blood-ring.

The size of the genital organs varies with the season of the year;
they are feather-shaped, and attached to the genital rachis by their
bases, but project freely into the coelom of the arm. From their great
variation in size and also from the shape of some of the cells in the
genital rachis, Hamann concludes that as each period of maturity
approaches fresh germ-cells are formed in the rachis and wander
into the genital organ and grow there in size. It is probable that the
aboral end of the genital stolon is the seat of the formation of new
germ-cells.

In the Starfish, therefore, as in other animals with a well-defined

coelom, the genital cells ultimately originate from the coelomic wall.

The genital ducts are formed by the burrowing outwards of the germ-
cells. When it is remembered that the fundamental substance of the
body-wall is semi-fluid jelly, this process will be better understood.

When the ova and spermatozoa are ripe, they are simply shed out
into the sea and fertilisation occurs there. The development is
described in Chapter XXI. The free-swimming larval period lasts
about six weeks.

Having described a single species with some degree of fulness, we

must now give some account of the range of variation of structure
met with in the group.

Number of Arms.—In the overwhelming majority of Starfish the

number of arms is 5, but deviations from this rule are met with not
only as individual variations, but as the characteristics of species,
genera, and even families.

The number 5 is rarely diminished, but amongst a large collection of

specimens of Asterina gibbosa, belonging to the author, some 4-
rayed individuals are met with. One species of Culcita, C. tetragona,
is normally 4-rayed.

On the other hand the number 5 is often exceeded. The families

Heliasteridae and Brisingidae are characterised by possessing
numerous (19-25) arms. In the normally 5-rayed family Asteriidae
Pycnopodia has 22 arms; and in the Solasteridae the genera
Rhipidaster and Solaster are characterised by possessing 8 and 11-
15 arms respectively; whilst Korethraster and Peribolaster have only
5. The common Starfish of the Gulf of St. Lawrence, Asterias polaris,
is 6-rayed, whilst most of the other species of the same genus are 5-
rayed, though 6 rays are often met with as a variation.

In some species the fact that the number of arms exceeds 5 seems
to be connected with the power of multiplication by transverse
fission. Thus Ludwig[452] has shown that in Asterias tenuispina the
number of arms is usually 7, but sometimes 5, 6, or 8, and that in
most cases the arms are arranged in two groups—one consisting of
small arms, the other of large.

Shape.—Apart from the varying number of arms, differences in the

shape of the Starfish are due to two circumstances:—

(1) The proportion of breadth to length of arm; and

(2) The amount of adhesion between adjacent arms.

The adhesion can go so far that the animal acquires the shape of a
pentagonal disc. This is the case for instance in Culcita. The fact that
the body of this animal is really composed of adherent arms is at
once made clear when the coelom is opened. This space is found to
be divided up by inwardly projecting folds called interradial septa,
which are stiffened by calcareous deposits and represent the
conjoined adjacent walls of two arms.

In the family Heliasteridae the mutual adhesion between the arms

has gone on merely to a slight extent, for the interradial septa are
still double.

Skeleton.—Most of the schemes of classification have been

founded on the skeleton, largely because the greater number of
species have only been examined in the dried condition, and little is
known of their internal anatomy or habits. There is, however, this
justification for this procedure, that the habits and food of the species
(with the exception of the Paxillosa) which have been observed in
the living condition appear to be very uniform, and that it is with
regard to the skeleton that Asteroidea seem to have split into
divergent groups through adopting different means of protecting
themselves from their foes.

The description of the various elements of the skeleton will be

arranged under the following heads:—(a) Main framework; (b)
Spines; (c) Pedicellariae; (d) Ambulacral skeleton.

(a) Main Framework.—The type of skeleton which supports the

body-wall of Asterias is called reticulate. As already indicated it
consists of a series of rods bound together by bundles of connective-
tissue fibres so as to form a mesh-work. This is a very common type
of aboral skeleton, but in a large number of Starfish a different type
occurs, consisting of a series of plates which may fit edge to edge,
leaving between them only narrow interstices, as in the
Zoroasteridae, or which may be placed obliquely (as in Asterina) so
that they imbricate or overlap one another. In a very large number of
Asteroidea the supero- and infero-marginal ossicles are represented
by squarish plates even when the rest of the skeleton is reticulate;