Professional Documents
Culture Documents
Campbell - Policy On Water Conservation
Campbell - Policy On Water Conservation
with
Abstract
Water conservation policy is a natural laboratory for addressing general issues of relevance to policy
analysis and public administration, particularly the relative effectiveness of policies in changing human
behavior. Using multivariate regression analysis of a dataset comprised of more than 200,000 monthly
observations of more than 19,000 household accounts over six years, this paper makes three main con-
tributions. First, it contributes to the substantive area of water conservation policy by analyzing what
works. Second, it contributes to our understanding of generic policy instruments. Third, it raises an
important issue regarding standard assumptions about selection bias. The results support earlier work
warning that offsetting behavior can negate engineering solutions to policy problems, but further indi-
cate that adding communication to engineering solutions can create cooperation that overcomes offset-
ting behavior. They also provide evidence that appropriate regulation can be effective, and that pricing
can be effective even when price elasticities of demand are low.
Introduction
Since the United States uses more than three times as much water per person
per day as average European countries, which themselves use much more than
developing countries (Graves, 1993), it is reasonable to claim that United States
water policies are relevant to the world. This is particularly important because
many fear that water scarcity will soon become a worldwide problem (Rothfeder,
2001). As various regions face drought, water pollution, and increasing demand,
understanding how to stimulate water conservation is one key. Further, it seems
that understanding how to induce water conservation should help in understand-
ing how to induce other conservation-oriented behaviors, such as recycling and
energy conservation. The latter is poignant as California and other regions of the
United States and South America face energy problems.
In Arizona, water conservation is of continuing policy importance, an issue that
administrators grapple with on an ongoing basis: water has been central to the
development of the state; the city of Tucson has already caused land damage (sub-
sidence) through groundwater depletion; and growth rates are so high in the city
of Phoenix that projections indicate that, absent droughts or interventions, in some
areas demand will exceed supply by 2025. For these reasons, the Arizona Depart-
ment of Water Resources (ADWR) and the City of Phoenix sought analysis of the
effectiveness of actual, fielded, water conservation techniques. The City of Phoenix
is an excellent policy laboratory because it has an active water conservation depart-
ment that has tried many conservation techniques over many years, a large and
diverse population, and high-quality data resources.2
Phoenix conservation measures include several types of policy interventions that
cover a gamut of generic policy types. They have tried what may be considered
engineering interventions; for example, they have directly provided water-saving
devices, such as flow restrictors for showerheads, to people. They have tried reg-
ulation: ordinances were passed by the city council that all new and replacement
fixtures must meet low-flow requirements. They have provided information—and
attempted to reduce information asymmetry—through educational methods such
as the distribution of self-audit kits and brochures. They have also tried methods
that are not in the generic policy toolkit, such as infusing a more human touch
through programs that involve senior citizens helping other seniors and students
helping neighbors. And, they have altered price structures.
The empirical evidence suggests that price interventions and regulation can be
very successful, that pure engineering solutions are almost entirely offset, but that
adding a human touch greatly improves the effectiveness of engineering solutions.
In addition, there is room for cautious optimism that some educational efforts can
be effective.
Engineering solutions to policy problems have been tried in many different set-
tings, such as in industrial pollution where specific types of smokestack scrubbers
Prices, Devices, People, or Rules 639
have been required, and in automobile safety regulation where particular devices,
such as airbags, have been added to cars. However, offsetting behavior may be a
problem with engineering solutions. In 1975, Peltzman’s work on automobile safety
regulation raised the specter of offsetting behavior decreasing policy effectiveness—
and perhaps even leading to perverse outcomes—when people consume more of
a good, such as safety, than they normally would (Peltzman, 1975). In his study,
when drivers were safer in their cars, they drove more recklessly, increasing danger
to pedestrians. Chirinko and Harper (1993) found further evidence of offsetting
behavior in this policy arena.
There is also evidence of offsetting behavioral responses to water conservation
policies. Geller, Erickson, and Buttram (1983) found evidence of offsetting behav-
ior in their experiment involving educational, behavioral, and engineering strate-
gies for inducing residential water conservation (and all combinations of the three
in their 2 ¥ 2 ¥ 2 design).3 They found that, even though their engineering devices
treatment did save water, it was significantly less than it should have been accord-
ing to the capacities of the devices themselves and laboratory data (p. 108). This
suggests offsetting behavior. Even more significant, they report evidence that in sit-
uations where people don’t know that the engineering retrofits are in place, such
engineering devices do succeed in conserving significant amounts of water (p. 109).
This strongly suggests that people engage in offsetting behavior when they know
devices are causing conservation. For example, if people know that their shower-
head is low-flow, they may feel free to take longer showers. The possibility of off-
setting behavior is of significant importance to policy analysts seeking to predict
the magnitude of outcomes from policy interventions, so these findings should
cause caution in adopting engineering device solutions in situations where behav-
ioral changes can alter effectiveness.
Both the city and ADWR were aware that laboratory estimates of water savings
generated by devices tended to be larger than actual savings. For this reason, they
were very interested in having an analysis of policy programs that field devices,
rather than of devices themselves. They were aware that human behavior was dif-
ferent from laboratory behavior, and wanted estimates of effectiveness incorporat-
ing actual behavior.
games is to permit the subjects to talk to one another.” Van de Kragt, Orbell, and
Dawes (1983) found that allowing subjects to communicate not only resulted in
provision of a public good—whereas without communication the good often was
not provided—but resulted in efficient provision. In Ostrom, Walker, and Roy’s
(1993) complex common pool resource experiments studying the relative effec-
tiveness of sanctions and communication, they found that communication alone
was more successful than sanctioning alone in inducing cooperation toward pro-
viding a common good (communicating and sanctioning together was the most
successful).
These findings and others (e.g., Anglin, 1995) tend to suggest that adding a
human dimension through direct communication may assist in the achievement
of policies, including conservation policies. As mentioned above, in some cases the
city of Phoenix combined engineering solutions with elements of direct, human
communication.
Prices
Rules—Regulation
Regulation implies use of the coercive force of law. Engineering solutions can be
regulatory, as are many mandatory automobile safety measures, but the majority
of city of Phoenix engineering solutions were not because (as discussed more fully
below), they were voluntary: people were given devices designed to save water, but
not required to use them. Pricing certainly can be regulatory if prices are inten-
tionally set in a regulatory process to achieve policy goals, but pricing still allows
an element of voluntary behavior as it allows people to make their own decisions
Prices, Devices, People, or Rules 641
Method
Review of water consumption literature (Campbell, Johnson, & Waits, 1996) and
discussion with experts in Arizona suggest that individual household water con-
sumption can be conceptualized as caused by the following categories of factors:
Functional Form
Water experts believe that the relationship between water consumption and most
causal factors is curvilinear: for example, while people with more wealth use more
water, water use increases at a decreasing rate at high levels of wealth; while people
use more water when it is hotter, the rate of increase decreases with high temper-
atures. Therefore, the majority of the model is log-log in structure to allow for the
change in rates of increase. This form is customary in water consumption models
(see, e.g., Chesnutt & McSpadden, 1995).
With the exception of the nonprice program variables, which are measured
dichotomously, most variables in this model are continuous. When using a loga-
rithmic transform, values of 0 must either be omitted—if they are not considered
meaningful—or they must be converted to some nonzero number that may be con-
sidered zero for purposes of analysis. In these data, zeros are meaningful numbers,
so they were converted as appropriate for those variables to undergo log transfor-
mation. Of course, the dichotomous variables were not transformed, so those ele-
ments of the model are semi-log in form (i.e., the dependent variable is transformed
using logs, but these independent variables are not).
Variables
The multivariate regression analysis includes forty-one variables to control for the
five broad causal categories described above. Of these, four measure attributes of
water pricing, and another thirteen measure nonprice conservation measures. Of
the additional variables, while some are policy relevant, others are simple controls;
their purpose is to increase confidence in estimates of the independent effects of
policy-relevant variables. The following sections briefly describe each variable used
in the analysis, with significant discussion only for those variables that are policy
relevant. For additional discussion of measurement of and rationale for control
variables, please see Appendix A.
Because of the directly behavioral concerns of the study, the most appropriate
dependent variable is individual household water consumption. This is measured
monthly at the single-family account level in “units” of water. A “unit” of water is
748 gallons, or 100 cubic feet. The average in the data is about twenty units per
month, with high variance. After screening, 233,928 monthly records were ana-
lyzed for 19,494 accounts.6
Water Pricing
As is common with public utility pricing, water prices for the city of Phoenix are
relatively complex. Therefore, four variables are used to measure water price.
Water Included in the Service Charge—Early in the study period, the city of Phoenix
charged per unit for all water used by residential customers. However, this policy
Prices, Devices, People, or Rules 643
was changed in mid-1990 so that the fixed monthly service charge included up to
six units (4,488 gallons) from October through May and ten units (7,480 gallons)
from June through September. After the change, only water use above these limits
was charged on a price-per-unit basis.
Economic theory suggests that water included in the service charge is consid-
ered free to consumers (the service charge is a sunk cost). Therefore, if all other
factors were held equal, including more units of water in the fixed monthly service
charge should increase water consumption. Please note that, though water demand
is seasonal, here “all else held equal” explicitly includes climatic factors of water
demand.
Sewer Charge (Constant Dollars)—In Phoenix, the residential sewer fee for the whole
year is based on average water usage during a three-month period. So, during
those three months it may seem to consumers as if the water charge is increased,
for lowering water consumption during those months will lower the sewerage com-
ponent of the water bill for the whole year. If consumers see this as an increase in
water price, then water consumption should decrease with real increases in the
sewer charge.
During the time under study, the city of Phoenix created two water conservation
policies that could be measured and can be considered command and control reg-
ulation.8 They instituted a low-flow fixtures and devices ordinance, and also a water
waste ordinance.
Low-Flow Fixtures and Devices Ordinance—The low-flow fixtures and devices ordi-
nance was implemented in three phases, from January of 1990 through January
of 1992. The ordinance required that all new and replacement fixtures meet low-
flow requirements, with different fixtures and levels of requirement phased in over
time. Three time-varying dichotomous variables control for this ordinance, one for
each phase. To the extent that the ordinance is effective, water consumption should
decrease (variable coefficients should be negative).9
644 Heather E. Campbell with Ryan M. Johnson and Elizabeth Hunt Larson
Water Waste Ordinance Enforcement—The City of Phoenix created a water waste ordi-
nance. Its goal was to focus on street flooding. We were able to measure, dichoto-
mously, whether the ordinance was being enforced or not. If effective, it should
decrease water consumption.
During the study period, the City of Phoenix tried several water conservation pro-
grams that involved giving people water-saving things, usually hardware. Whether
people installed the devices or left them on after installation was voluntary.
If people do not install the devices, the policies’ effects should be zero. If people
do install the devices, but then engage in offsetting behavior, water use should not
decline much and might even increase (estimated coefficients should be small in
magnitude or perhaps even positive). Negative effects on household water con-
sumption indicate that at least some devices are installed and offsetting behavior
(if any) does not overwhelm device benefits.
Retrofit Device Canvassing and Union Hills Plumbing Products Drop-Off—For both of
these programs, the city went door-to-door, leaving packets of water-saving fixtures
at people’s homes. It is, of course, unknown what percentage of households actu-
ally installed the dropped-off devices, but this is not a problem for this policy analy-
sis because the estimation of program effectiveness (rather than device effectiveness)
is the goal.
Depot Plumbing Products Pick-Up—In this program, water-saving hardware was also
given away. However, people had to come to a location (a depot) to pick up the
devices. Presumably, a higher proportion of these people would be inclined to
install the fixtures.
The differences between these two programs are of particular interest from a
social science perspective because participating household groups are different: in
the first two programs, selection for participation in the program was essentially
random; in the second, participants were self-selected, so could be inferred to be
people who are more interested in water-conserving fixtures.
Book of Seeds—Though not like the other programs in this category, this program
involved giving people things—here, seeds of water-conserving plants. This could
instead be considered an educational program (see next category) since the intent
was to let people know that they could grow nice plants that did not use a lot of
water.
Unsolicited Audit Kits—For two years of the study period, kits that explained how to
perform home water audits were sent, unsolicited, to households. If households
performed the audit, they could fill out and send in a card and then receive water-
conserving fixtures and devices. Which households sent in the card and received
fixtures could be directly observed. As with other fixture programs discussed above,
however, it cannot be known whether fixtures were installed.
Though this program was distributed to individual households, records did not
allow measuring which households received the kits, but only whether the kits were
in effect citywide. Therefore, estimated coefficients should be lower than the actual
house-by-house effects of the program (if any).
Seniors Helping Seniors—In this program, people who were elderly, at 125% of the
poverty level or below, or disabled were provided with hardware retrofit assistance
in their homes by trained senior citizens. During the study period, over 4,900
households had retrofit devices installed through this program.
Climate Factors
The literature and experts suggested a number of variables regarding the persons
living in households as possibly important to household water consumption. Some
factors are policy relevant, others are not, but are nonetheless important as con-
trols. This subsection describes those factors after a discussion of measurement
issues.
646 Heather E. Campbell with Ryan M. Johnson and Elizabeth Hunt Larson
Household Income and Poverty—The hypothesis with regard to poverty and water
consumption is opposite to that a simple use of demand theory would suggest.11
Experts in the Phoenix area expected that poorer residents actually used more water
Prices, Devices, People, or Rules 647
all else equal because they could not afford to repair leaks, replace old fixtures with
newer, more water-saving ones, and so on. The idea here is that it takes an invest-
ment of money to save water. To measure this, we used both average income and
percent of households below the poverty level. Holding constant average income
(and all other factors), it is expected that increases in the percent of households in
poverty would actually increase water consumption. Additionally, holding constant
all other factors (such as other attributes of wealth, discussed below), increases in
tract average income should decrease water consumption.
These measures were directly policy relevant to the city of Phoenix, which was
beginning a pilot project that involved going into poor households and directly
fixing leaks and replacing toilets, and so on. In general, support for these hypothe-
ses would suggest that water conservation policies may need to be quite different
for poor households.
Western Conservation Ethic—Some water experts and literature suggest that those
who have lived in the (often arid) Western states have a conservation ethic. To
control for this effect, we also measured the percentage of households in the tract
that had lived in another house but in a Western state (including Arizona). If there
is a Western conservation ethic, this element controls for it and increases the like-
lihood that General Education Messages is measuring the effect of general educa-
tion messages in the Phoenix area itself. If there is a Western conservation ethic,
then when the likelihood that a household has lived in the West increases, then
water consumption should decrease.
Households with Children—Many conservation experts felt that one of the most effec-
tive conservation tactics was youth education programs (as an example, see Anton,
1995). We wanted to try to capture this effect, but youth education programs
were not specific city of Phoenix programs during the time analyzed (though
Phoenix children could receive water education provided by other organizations).
The best measure we could come up with was percent of households in the
tract with children present. If youth-oriented conservation programs are effective,
then children may not only conserve water themselves, but encourage their parents
to do so.
648 Heather E. Campbell with Ryan M. Johnson and Elizabeth Hunt Larson
We realize this measure is tenuous. It only very indirectly measures what we care
about and, further, there might be other reasons that households with children
might conserve water. For example, parents might be more careful to conserve
water because they want to be good role models. Or, parents might have a more
future-oriented perspective because they are aware their children will live in the
future. Still, the purpose of including this measure was to try to capture the effect
of youth-oriented water conservation education, and all of this reasoning predicts
that households with children should use less water, ceteris paribus. Though the
measure is weak, the concept is policy relevant.
Many attributes of the physical infrastructure of houses and their lots should affect
water consumption: for example, lot size and type of landscaping (e.g., low-
water-use native plants versus high-water-use turf). However, good measures of
such attributes are not always available. The following were used to control for
such features:
• Number of Bathrooms (houses with more bathrooms should use more water),
• House Value (more valuable homes are expected to have more water-using
features),
• House Age and House-Age-Squared (new homes should be more water con-
serving, though the difference may decline as older homes replace fixtures),
and
• Landscape Flood Irrigation (because flood irrigation is off-meter, homes
with flood irrigation should use less metered water, though not less water
overall).
Estimation Issues
As mentioned, a multivariate regression analysis was performed on more than
200,000 monthly observations of household water use by more than 19,000 indi-
Prices, Devices, People, or Rules 649
vidual account households during the years 1990 through 1996. A log transform
was taken of the dependent variable and of most independent variables. Dichoto-
mous variables (these include all nonprice conservation program variables) were
not transformed. In addition, home age, home-age-squared, and income were not
transformed, resulting in a semi-log functional form for the relationship between
those variables and the dependent variable. Home age and home-age-squared were
not transformed for a purely instrumental reason: when transformed, extreme col-
inearity prevented estimation of coefficients for these variables. The reason income
is not transformed is discussed below.
Growth in the Error over Time—Another possible error process arises from the use of
census data. Used as they are here, it must be expected that the size of the error
grows over time: that is, the census data, which are fixed in time, will measure
household attributes with greater error over time. Fortuitously, using a logarith-
mic transformation of the data, as was necessitated by empirical understanding of
water consumption patterns, is a recommended correction for this type of error
process (Pindyck & Rubinfeld, 1981).
Simultaneity—One factor that might strike readers is that the dependent variable is
consumption (quantity demanded) and one independent variable is price. Gener-
ally, estimates of quantity demanded require simultaneous-equations models
because price and quantity are simultaneously determined by the forces of supply
and demand. However, the price for public utilities such as water is an adminis-
tered price, not a price set in markets. Thus, price is exogenous to the consump-
tion decision, and there is no simultaneity.
Results
Table One presents the estimation results. In interpreting the results, the follow-
ing should be kept in mind. First, for cases where both the dependent variable and
650 Heather E. Campbell with Ryan M. Johnson and Elizabeth Hunt Larson
Seniors Helping Seniors is -0.064. This means that using the Seniors program
(going from 0 to 1) is expected to result in a 6% decrease in household water
consumption.12
Policy Implications—Prices
Four variables are used to measure the impact of pricing decisions on household
water consumption. Three of the four are of the sign expected. “Free” Service
Charge Water is measured in units, taking on 0, 6, or 10 in the data. The estimated
coefficient, 0.198, indicates that a 1% increase in the amount of free water included
will result in approximately a 0.2% increase in water consumption. However, the
increase from 6 units (nonsummer) to 10 units (summer) is a 66% increase, which
is expected to result in an increase of 13.1% in water consumption. For a house-
hold average of about 20 units per month, this would be an increase of about 2.6
units (almost 2,000 gallons), all else constant. Including some water in the monthly
service charge is often justified on equity grounds, but these results strongly indi-
cate that it is a bad idea for water conservation. According to the estimates, water
prices would have to be raised by almost 50% in order to make up for the water
loss caused by the increase in free water included in the service charge between
nonsummer and summer.
Because it can be difficult to understand the policy meanings of the estimated
coefficients, Table Two shows predicted policy magnitudes for this variable and
others of particular policy interest.
As discussed above, the estimated price elasticity of demand for Real Volume
Price is -0.271. As typically found in the literature, this estimate suggests that
household water demand is quite price inelastic. The estimated elasticity for Real
Sewer Price is even lower, at -0.04. However, though the price elasticities are small,
their effectiveness can nonetheless be high. For example, in nominal terms the
highest price in the data was $1.45 per unit. A 10% increase in this low sum would
only be $0.14 per unit, but is estimated to result in a 2.7% reduction in water con-
sumption, or about 0.54 units per household per month. Because this would apply
to every household—and at low administrative cost—it can have large aggregate
effects. Assuming 278,000 single-family residences, such a tiny price increase could
result in water savings of more than1.8 million units (over 1 billion gallons) per
year. It is hard to believe that a price increase of 14 cents for every 748 gallons
used is politically salient. Again assuming an average use of 20 units per house-
hold, this would cost only $1.96 per non-summer month, on average (even less in
the summer) assuming that households do not change their behavior. But, of course, the
purpose of pricing is to induce behavior change, so actual budgetary increases will
be lower.
The estimated coefficient for Real Environmental Charge is very small, but pos-
itive, completely contrary to economic theory. As mentioned above, this may be
because the environmental charge is separated from the water charge, so customers
may not understand that it is part of their water price. On the city of Phoenix bill,
per-unit water price is called “Water Use Fee,” while the environmental charge is
called “Environment Mandates,” with the phrase “Water Charge” below it as a
subset. Perhaps use of the word “Mandates” makes customers think their behavior
cannot affect it.
Taken as a group, these results suggest that, even though price elasticities of
demand for water are low, pricing should be seriously considered as a conserva-
tion measure. In addition to pricing itself, price structure—such as water included
within a service charge, without a specific price—should also be considered as an
element important to inducing conservation.
Policy Implications—Devices
Policy Implications—People
Policy Implications—Rules
Two rule-based regulatory efforts were also included in the analysis: Water-Waste
Ordinance Enforcement, and the Low-Flow Fixtures and Devices Ordinance, mea-
sured separately for each of its three phases. According to the estimate, Water-
Waste Enforcement is not effective. Its coefficient estimate is positive and fairly
large.13
Interpreting results of the phased-in hardware Ordinance is more complex.
Phase I’s coefficient is estimated as positive though small, Phase II’s as negative
and fairly large, and Phase III’s as negative and quite small. One way to make sense
of this change in signs between the phases may be to consider the possibility of off-
setting behavior of another sort. The fact that Phase I is estimated to cause an
increase in water use may indicate that people were motivated to purchase high-
flow devices while they could. In economic terms, a change in consumer expecta-
tions may have led to an increase in demand for nonsaving devices as people
realized that only saving ones would be available in the future.
654 Heather E. Campbell with Ryan M. Johnson and Elizabeth Hunt Larson
Taken together, these estimates do suggest that the ordinance saved significant
water. Adding the coefficients together indicates that the ordinance resulted in total
savings of 3.5%. Not only is this number fairly large in magnitude when compared
with those estimated for other policy interventions, but it—like price and unlike
the devices and people programs—applies to every household in the city.
Thus, using the same assumptions as those given for a 10% increase in price
(278,000 single-family residences using an average of 20 units per month), the ordi-
nance is estimated to save over 2.3 million units per year (more than 1.7 billion
gallons).
Though the results regarding regulation are not so convincing as those for prices
and people—after all, Water-Waste Enforcement is estimated to increase water con-
sumption—this section still indicates that regulation can be effective if appropri-
ately designed. Part of the reason for regulation’s effectiveness is its wide
applicability.
Policy Implications—Education
Other Educational Efforts—As discussed, two other measures of the effect of educa-
tional efforts were included. Percent of Households with Children was included to
try to control for the effect of child-targeted water conservation education, and the
Percent who lived in the Same City in 1985 was included to try to control for
general, citywide conservation information. Both of the coefficients estimated for
these variables are negative. Increasing the tract percent with children by 1% is
estimated to decrease water use by 0.31%, while increasing the tract percent who
Prices, Devices, People, or Rules 655
have lived in their houses since 1985 by 1% is predicted to decrease water use by
approximately 0.05%, all else equal.
To the extent that these variables indeed measure what they are intended to,
they suggest that child-targeted education and general water conservation messages
can decrease household water consumption. Though one may expect that it is
easier to alter children’s behavior than adults’, it is puzzling why educational mes-
sages delivered to homes would be less effective than general educational messages.
Perhaps it is the effect of the accretion of many messages, as compared to the effect
of a single message. As discussed in the footnote, people may not look at brochures,
but billboards and TV and radio messages may be more difficult to ignore.
Age—Some experts in the Phoenix area expect that young adults use more
water than all other groups. The estimates support this. For these estimates,
656 Heather E. Campbell with Ryan M. Johnson and Elizabeth Hunt Larson
children under the age of 17 are the reference group. Those aged 17–24 use
more water than this group, while adults between the ages of 25 and 64 use less.
A 1% increase in the percent of 17–24-year-olds in the tract is estimated to increase
water consumption by 0.18%. With the enormous Echo Boom coming, this may
suggest the efficacy of devising water conservation programs especially for this age
cohort.
Conclusions
City of Phoenix water conservation programs turn out to be a natural laboratory
that can provide insight into many questions of relevance to policy analysis and
even social science more broadly. Because the city’s water conservation adminis-
trators took a page from Franklin Delano Roosevelt’s book and decided to “above
all, try something,”15 over several years they put together a variety of specific
instances of generic policy interventions, including providing information, non-
market providing of water-saving devices, regulating, and pricing. In addition, they
tried adding person-to-person communication, a technique that is not in the
generic policy tool-kit. Because their population is large and diverse, their situa-
tion also allowed for testing the promise of demographically targeted policies that
had not been tried.
Consequently, the results of this study not only provide specific information
for those who seek to induce water conservation, but they address several issues
of general relevance to policy analysis. For example, they allow consideration of
the standard assumption regarding the effects of self-selection versus random
selection. Further, they bring into relief the issue of policies that are more certain
versus those that are not (Weimer & Vining, 1999). Prices apply with high certainty
to all households in a water provider’s area. So, even with small elasticities
of demand, small price increases can result in large water savings. Similarly, rule-
type regulations apply with high likelihood to all households; if the ordinances are
designed correctly, they can lead to large water savings. But the other edge of
this sword is that incorrectly designed regulations may also have big impacts.
Further, effective policies that do not apply to all households may have small total
effects.
As introduced at the beginning and discussed throughout, the possibility of off-
setting behavior should be kept firmly in the mind of the policy analyst and public
administrator. Evidence of offsetting behavior can be seen in the results: giving
people devices is estimated to cause very small decreases or fairly large increases
in water consumption, regardless of whether the people to whom they are given
are selected randomly or are self-selected volunteers—and regardless of engineer-
ing tests of device effectiveness. Further, Phase I of the Fixtures and Devices
Ordinance was estimated to cause an increase in water consumption.
One way to think about these results is to consider that engineering solutions
(devices) were tried in several different ways. The city tried to legislate the use of
water-saving devices, to randomly give people water-saving devices, to invite people
to pick up water-saving devices, and had individual people work with individual
households to install water-saving devices.
Prices, Devices, People, or Rules 657
These different settings help illuminate the conditions under which offsetting
behavior appears and what can be done to ameliorate it. One indication is that
ignorance prevents offsetting behavior (Geller, Erickson, & Buttram, 1983, p. 109),
though this has grave implications for citizenship and governance. One factor
that clearly seems to ameliorate offsetting behavior is the use of one-on-one
communication, which accords well with the cooperation literature, and fits much
better with notions of participatory democracy. Considering this all together,
perhaps policy analysts should add techniques for inducing cooperation to other
generic solutions, such as regulation, information, and pricing, to the standard
tool-kit.
The cooperation literature may also include a warning, as it indicates that
repeated communication works better than one-shot communication (Ostrom,
Walker, & Gardner, 1993, p. 140). Thus, the durability of effects may decline over
time. However, this is true of all policy interventions. Hardware devices themselves
wear out over time. Price increases are overtaken by inflation. Engineering-focused
regulations, such as those used here, become obsolete as technology improves, and
so on. Thus, all of these concerns just remind us that policies cannot be static—we
cannot expect to find the best one and then move on; we must revisit even poli-
cies that have been effective.
Given the findings, the effectiveness of educational information is unclear, with
evidence that printed educational information sent to homes is not effective,
though broad-based education to children and households may be. This may tend
to support the importance of repeated communication—a brochure is a one-shot
communication; billboards and radio messages may be heard many times. Depend-
ing on how the program is designed, education to children may combine personal
communication with multiple messages.
Certain demographic factors were hypothesized to be policy-relevant: age, eth-
nicity, and poverty. Though there is evidence of poverty and Hispanic ethnicity
increasing water consumption slightly, the effect of young adults is estimated to be
almost an order of magnitude greater. Given the imminent young-adulthood of
the Echo Boom generation, it may be worthwhile to design policies specifically for
older youths and young adults. If so, because prices and regulation are not easily
targeted in this way and engineering devices are not effective, cooperative and edu-
cational programs seem the most promising. This reminds us of something we
know. We have many generic policies in our toolkit partially because some may
work better in specific settings than others.
658 Heather E. Campbell with Ryan M. Johnson and Elizabeth Hunt Larson
In sum, in the trade-offs between prices, people, devices, and rules, prices and
rules are the most certain and, because they can be applied generally to all house-
holds, their cumulative effects can be great. At least for water conservation, simply
giving households engineered devices is not effective. Offsetting behavior is so
strong that effectiveness is swamped; here, nonmarket provision by government is
a poor choice, unless mediated by direct communication. However, programs that
must be administered one-on-one to individual households will tend to have a
smaller impact overall, even if they have a greater impact per affected household—
but these techniques do have the advantage of inducing cooperation, which is in
keeping with participatory democracy. Prices and rules are administratively more
inexpensive with high effectiveness; cooperative programs educate citizens but
require more resources and are less effective overall.
APPENDIX A
Additional Variable Details
Climate Factors
Numbers of Persons Living in the Household—As the number of people living in the
household rises, it is expected that household consumption of water will rise. This
variable is measured by the tract average number of persons per household.
Renters and Owners—It seems that renters who pay for water use will use less water
on the landscape than those who live in houses they own, while renters whose land-
lord pays for the water may use the same amount or even more, perceiving the
marginal price of water as zero. We had no information about which type of rental
situation predominates, but felt that we needed to include a measure of this as a
control variable. We used tract percent of owner-occupied housing.
However, if this is true, then female-headed households should use less water, ceteris
paribus, than either male-headed or dual-headed households. We measured percent
of tract households that were female-headed and percent that were male-headed
(the omitted group is dual-headed).
House Value—It is likely that more valuable homes have more water-using features
than do lower-value homes. For example, more expensive homes are likely to have
larger land—and therefore more landscaping. In our area, they may also have dec-
orative water-using features such as fountains. Value was measured using census
tract average home value.
House Age—New homes should be more water conserving for two reasons: first,
newer homes tend to have smaller yards in our area; second, newer homes use
more of the newer (often mandated) low-water-use technology. However, it is not
expected that this effect is linear since, once a house is old enough, fixtures will
tend to need to be replaced, and therefore these fixtures will be newer, water-saving
technology.
This variable is measured using actual age when available and census tract
average age when actual age is not available. Home-age-squared is also used to
allow for a changing effect of age. Using a variable in both the level and squared
allows the effect of age to increase water consumption up to a point, but then for
the effect to decline.
Landscape Flood Irrigation—In our area, some landscapes can be watered using
unmetered flood irrigation. Flood irrigation enables a household periodically to
flood the landscape with water. Because this water is unmetered, then metered water
use should be lower for households with available flood irrigation. This variable is
measured dichotomously; either households have available flood irrigation, or they
do not.
Notes
1 This work was partially funded by a grant from the Arizona Department of Water Resources
(ADWR). Special thanks to Tom Babcock, Shannon Autwell, Jeff De Witt, and Scott Willett of the
City of Phoenix for extraordinary information provision and data collection efforts. Special thanks
to Mary Jo Waits and the Morrison Institute for Public Policy for their grant administration efforts.
Special thanks to Barbara McCabe for insightful discussions and Carol Tschudi for library science.
Thanks also to attendees of the Association for Public Policy Analysis and Management Fall 2000
Research Conference. Ideas presented in this article are the authors’ and do not necessarily repre-
sent those of ADWR or the City of Phoenix.
660 Heather E. Campbell with Ryan M. Johnson and Elizabeth Hunt Larson
2 Some of the material presented in this paper was first presented in the research report provided to
ADWR, Campbell, Larson, Johnson, & Waits (May 1999).
3 In their experiment, Geller, Erickson, and Buttram (1983) cross “Education versus No Education”
with “Daily Consumption Feedback versus No Feedback” and “Low Cost Conservation Devices
versus No Devices” in all combinations. “Education” consisted of distributing a handbook discussing
problems of excess water consumption and techniques for reducing water consumption. “Feedback”
included smiling or frowning faces. Engineering devices were toilet dams, flow restrictors, aerators,
and shut-off shower controls installed by the researchers.
4 Depending on the study, these goods may be pure public goods, or common pool resources (CPRs)
(goods that exhibit rivalry in consumption, but not excludability).
5 Approaches with quick-response “reward” or “punishment,” such as those used in Geller, Erickson,
and Buttram (1983) (Daily Consumption Feedback) are also possible, but were not used by the City
of Phoenix during the study period.
6 The sample included both households that had participated in household-specific conservation pro-
grams and those that had not. About 1,200 nonparticipating households were selected for each year.
For each year and each program, participating households were either selected randomly from the
pool of participants, or all participants were used, based on the total number of available program
participants. For a given year and a given program, there were generally between 100 and 500 par-
ticipants observed. Campbell et al. (1999) gives additional information regarding the sample, sample
selection, and omissions.
7 All price data are adjusted using a metropolitan Phoenix consumer price index, base 1982–1984.
Index data were provided by the Center for Business Research, William Seidman Research Insti-
tute, College of Business, Arizona State University.
8 Phoenix also passed a water theft ordinance, but they were not confident of any possible measure
of this regulation.
9 It should be noted that the federal government passed a similar law in 1992. However, the federal
law was never funded nor enforced. For this reason, the federal law is not measured in the analy-
sis, so the effect of the 1992 phase-in could capture effects of the federal law, if any.
10 It should be noted that issues of alienation and culture are really quite different from whether house-
hold members speak English. Ideally, one would measure fluency in the dominant language
separately.
11 But do note that there are goods for which demand does not increase with increases in income.
These are called “inferior” goods.
12 There is some confusion with regard to how to interpret the coefficient of a dummy variable in a
model where the dependent variable is logarithmically transformed. Here I follow Kennedy (1985),
which explicitly says “When a regression linear in logarithms includes a dummy variable, for
example ln Y = a0 + a1 ln X + a2D + e, the coefficient a2 of the dummy variable D is interpreted as
the percentage impact on Y of the qualitative variable it represents” (p. 185). Since the vast major-
ity of this equation is log-log in form, this seems appropriate. However, some of the model is semi-
log in form, and the interpretation of a dummy variable coefficient is more complex in semi-log
models, as outlined in Halvorsen and Palmquist (1980) and Kennedy (1981). Regardless, the correct
interpretation for dummies in semi-log equations is “approximately equal” for “small values” to the
interpretation given here (Halvorsen & Palmquist, 1980, p. 475). As can be seen by examination of
Table One, all these coefficient estimates are quite small—smaller in magnitude than all but one of
the 13 examples Halvorsen and Palmquist provide (their smallest example is 0.00).
13 It is hard to know what to make of this estimate. It does not seem like a case where offsetting behav-
ior could be the culprit. Perhaps they enforced more when there was more of a known problem?
14 A. J. Pfister, former General Manager of SRP, one of the largest energy providers in the Phoenix
area, reports that SRP found that, though bill inserts are easy and inexpensive, they are not effec-
tive. Surveys revealed that only about 15% of recipients looked at them at all, and only about 1/2 of
those looked at them closely (personal communication, May 23, 2001).
15 Address at Oglethorpe University, Atlanta, Georgia, May 22, 1932. As quoted in John Bartlett (1980),
Bartlett’s Familiar Quotations, p. 779, Boston: Little, Brown.
versity of California, San Diego, and an M.Phil. and a Ph.D. in public policy analysis from
Carnegie Mellon’s Heinz School. Research interests are in public policy, emphasizing
telecommunication and environmental regulation, and the scholarship of teaching.
Ryan M. Johnson is Public Affairs Director for WorldatWork, a worldwide association based
in Arizona of human resource professionals who work in the compensation field. He holds
a B.A. in political science from U.C. Santa Barbara and an M.P.A. from Arizona State Uni-
versity. His research interests are broad public policy and compensation issues. He was for-
merly a research analyst at the Morrison Institute for Public Policy, Arizona State University.
Elizabeth Hunt Larson, M.B.A., is the Director of Institutional Research at Mesa Commu-
nity College. Her responsibilities include analyzing Student Outcomes Assessment Data and
assisting faculty in refining and developing instruments. She also provides data analysis
for institutional effectiveness studies. She has worked as a statistician for National Data
Corporation, as a policy analyst as the Morrison Institute for Public Policy at Arizona State
University, and has conducted educational research at the local and state levels.
References
Anglin, R. (1995). Constructing cooperation: Instituting a state plan for development and redevelopment.
Journal of Policy Analysis and Management, 14(3), 433–445.
Anton, W. F. (1995). Implementing ASCE water-conservation policy. Journal of Water Resources Planning, 121(1),
80–89.
Avalos, M., & De Young, T. (1995). Preferences for water policy in the Ogallala region of New Mexico:
Distributive vs. regulatory solutions. Policy Studies Journal, 23(4), 668–685.
Baumann, D. D., & Opitz, E. M. (1993). Measurement of conservation attitudes and behaviors: Trends in
Southern California. Proceedings of CONSERV 96 (pp. 1031–1045). Denver, CO: American Water Works
Association.
Briggs, R. K., Jr. (1989). Evaluating a water conservation surcharge program in Orange County, Florida.
Government Finance Review, April, 7–10.
Brown, P. W. (n.d.). AZMET evapotranspiration estimates: A tool for improving water management of turf-
grass. Retrieved from http://ag.arizona.edu/azmet/et1.htm.
Bruvold, W. H. (1990). Extending the model of residential water conservation. Proceedings of CONSERV 90
(pp. 5–9). Denver, CO: American Water Works Association.
Campbell, H. E., Johnson, R. M., & Waits, M. J. (1996). Developing a statistical model for evaluating water
conservation measures: A review of literature. Tempe, AZ: Morrison Institute for Public Policy.
Campbell, H. E., Larson, E. H., Johnson, R. M., & Waits, M. J. (1999, May). Some best bets in residential water
conservation: Results of multivariate regression analysis, City of Phoenix, 1990–1996. Tempe, AZ: Morrison
Institute for Public Policy.
Chesnutt, T. W., & McSpadden, C. N. (1995). Determinants of Phoenix water demand. Santa Monica, CA: A&N
Technical Services.
Chirinko, R. S., & Harper, E. P. Jr. (1993). Buckle up or slow down? New estimates of offsetting behavior and
their implications for automobile safety regulation. Journal of Policy Analysis and Management, 12(2),
270–296.
Geller, S. E., Erickson, J. B., & Buttram, B. A. (1983). Attempts to promote residential water conservation with
educational, behavioral and engineering strategies. Population and Environment, 6(2), 96–112.
Graves, W. (1993, November 1). Introduction, National Geographic Special Edition, p. 1.
Halvorsen, R., & Palmquist, R. (1980). The interpretation of dummy variables in semilogarithmic equations.
American Economic Review, 70(3), 474–475.
Kennedy, P. E. (1981). Estimation with correctly interpreted dummy variables in semilogarithmic equations.
American Economic Review, 71(4), 801.
Kennedy, P. (1985). A guide to econometrics (2nd ed.). Cambridge, MA: MIT Press.
Lee, C. (1987). Water in the region. In Water for millions: At what cost, a resource book on water demand and use in
the greater New York metropolitan region. Poughkeepsie, NY: Scenic Hudson.
Ostrom, E., Walker, J., & Gardner, R. (1993). Covenants with and without a sword: Self-governance is possi-
ble. In T. L. Anderson and R. T. Simmons (Eds.). The political economy of customs and culture: Informal solu-
tions to the commons problem (pp. 27–156). Lanham, MD: Rowman & Littlefield.
Peltzman, S. (1975). The effects of automobile safety regulation. Journal of Economic Issues, 11, 587–600.
Pindyck, R. S., & Rubinfeld, D. L. (1981). Econometric models and economic forecasts (2nd ed.). New York:
McGraw-Hill.
662 Heather E. Campbell with Ryan M. Johnson and Elizabeth Hunt Larson
Rothfeder, J. (2001). Every drop for sale: The desperate battle over water in a world about to run out. New York:
Tarcher.
SPSS Inc. (1998). SPSS Professional Statistics 8.0 for Windows. Chicago: SPSS
Thaler, R. H., & Dawes, R. M. (1992). Cooperation. In R. H. Thaler, The winner’s curse: Paradoxes and anomalies
of economic life (6–20). Princeton, NJ: Princeton University Press.
U.S. Drought Monitor (2002, November 12). Retrieved November 15, 2002, from http://www.
drought.unl.edu/dm/monitor.html.
van de Kragt, A. J. C., Orbell, J. M., & Dawes, R. M. (1983). The minimal contributing set as a solution to
public goods problems. The American Political Science Review, 77, 112–122.
Weimer, D. L., & Vining, A. R. (1999). Policy analysis concepts and practice. Upper Saddle River, NJ: Prentice
Hall.
Copyright of Review of Policy Research is the property of Wiley-Blackwell and its content may not be copied
or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission.
However, users may print, download, or email articles for individual use.