You are on page 1of 16

Water Resources and Economics 4 (2013) 22–37

Contents lists available at ScienceDirect

Water Resources and Economics

journal homepage: www.elsevier.com/locate/wre

Household behavior related to water conservation$


Diane P. Dupont, Steven Renzetti n
Department of Economics, Brock University, 500 Glenridge Avenue, St Catharines, Ontario, Canada, L2S 3A1

a r t i c l e in f o a b s t r a c t

Article history: This research investigates the role of the level, structure, and rate of
Received 25 July 2012 change of water prices, as well as non-price policies and socio-
Received in revised form demographic characteristics on households0 decision-making regard-
4 December 2013
ing indoor and outdoor water conservation practices. In this paper,
Accepted 7 December 2013
indoor water conservation choices refer to the presence/absence of
low volume toilets and low flow showers in the home. Outdoor water
Keywords: conservation choices refer to the frequency of lawn and garden
Indoor and outdoor water conservation watering by the household during summer months. The former is
choices
estimated with a bivariate probit model, while the latter requires an
Water pricing structure
ordered bi-variate probit model. The data come from Statistics
Ordered probit
Endogeneity Canada0 s 2006 Households and the Environment Survey. It provides
Simultaneous equations a national cross-sectional sample of household-level observations on
indoor and outdoor water choices. These data are linked to municipal
level information on residential water prices, non-price water
conservation policies, and weather information assembled by the
authors. Potential endogeneity arising from the presence of price and
non-price water policy variables is investigated using a simultaneous
equations approach. Estimation results provide several insights. First,
water prices and household characteristics play important roles in
shaping household decision-making regarding both indoor and
outdoor water conservation. Second, there is little evidence that
non-price water conservation policies influence either indoor or
outdoor decision-making. Third, there is evidence of correlation for
the two indoor water-conserving choices and also for the frequency of
lawn and garden watering decisions. Finally, there is no evidence of
endogeneity of water prices and other policy variables.
& 2013 Elsevier B.V. All rights reserved.


The authors thank the staff of the McMaster University Research Data Centre for computing assistance and Neil Arnold,
Yang Tang, Arlo Matisz, Josh Macdonald, and Patricia Hennigar for research assistance. Funding support from the Canadian
Water Network is gratefully acknowledged. All errors and omissions are our own.
n
Tel.: þ 1 905 688 5550; fax: þ 1 905 688 6388.
E-mail addresses: ddupont@brocku.ca (D.P. Dupont), srenzetti@brocku.ca (S. Renzetti).

2212-4284/$ - see front matter & 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.wre.2013.12.003
D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37 23

1. Introduction

Recent reports summarizing the findings of Statistics Canada0 s Households and the Environment
Survey [21,22] provide a number of interesting observations about Canadian households0 water
conserving behavior. In 2005 fifty-four percent of Canadian households reported having a water-
saving showerhead and 34% a water-saving toilet. However, when similar questions were asked in
1994, only 42% had low-flow showerheads while 15% of households had a low-volume toilet. Outdoor
water use practices – which can add 50% to a household0 s overall water consumption – have also
declined. In 2009, 43% of respondents (with a lawn) said that they watered their lawn at some point
over the summer while 73% of those with a garden indicated that they had watered their garden. In
2005, comparable numbers were 54% for lawns and 81% for gardens.
There are a number of potential explanations for the observed changes in willingness to conserve
on both indoor and outdoor water use through either incorporation of new types of water-saving
technologies by homeowners and/or alterations in water-using behavior. One possible reason for
indoor conservation relates to government efforts to promote the purchase of water-saving appliances
and fixtures. A second reason that may affect both indoor and outdoor water conservation efforts
stems from increases in water prices and/or the shift by municipalities towards volumetric water
rates, thereby, raising the cost of being profligate with water.
A major constraint in the development of efficacious water conservation policies is the lack of
quantitative knowledge regarding households0 sensitivity to these different types of policy
instruments. This research seeks to address this information gap by investigating the factors
influencing households0 decision-making regarding frequency of outdoor water use, along with their
choices regarding the presence of indoor water-saving appliances.1 Specifically, we investigate the
separate roles of the level, the structure, and the rate of change of water prices, as well as water utility
non-price conservation policies, climate variables, and socio-demographic characteristics on house-
holds0 decision-making regarding their indoor and outdoor water conservation choices. Our bi-variate
choice models allow us to test for jointness of decision-making between indoor water saving
appliances, on the one hand, and, separately, between frequency of lawn and garden watering, on the
other hand. To our knowledge, this is the first effort to examine both indoor and outdoor water
conserving decisions in response to both the structure of water prices and non-price policies, as well
as climate variables when appropriate, as determinants of these decisions. An issue in modeling
households0 ’ responses to water prices and municipal water conservation policy measures is that the
structure of water prices and the features of water conservation programs chosen by municipalities
may be endogenously determined (see the discussion in [17]). This implies that some observable and
unobservable characteristics of local communities that determine pricing and policy design choices
may also influence residential water conservation decisions. For example, a community facing rising
water demands may be induced to use an increasing block rate pricing which may, in turn, influence
water conservation decisions. Our modeling of households0 decision-making and investigating the
role of prices and other policy measures takes into account the potential endogeneity of the pricing
and conservation policy choices.
We combine data from a number of sources to create a cross sectional household-level set of
observations containing information on both indoor and outdoor water conservation efforts, water
prices/policies, socio-economic variables, and climate variables. The sources include Statistics
Canada0 s 2006 Households and the Environment Survey (HES), as well as Environment Canada0 s
water pricing data set and climate variables [3,12]. Indoor conservation choices are examined with
bivariate probit models that incorporate potential endogeneity of water prices and non-price
conservation policies. Outdoor conservation behavior focuses upon frequency and is modeled using
bivariate ordered probit models. The next section of the paper reviews the literature concerned with
household choices as they pertain to water conservation. The third section sets out the econometric

1
It is possible that the adoption of water conserving devices may not, in fact, lead to reductions in consumption [23]. We
do not observe water consumption directly and, thus, are unable to test whether the observed behavioral changes result in
decreased water use.
24 D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37

models. The fourth section details the data sets employed and variable construction. The fifth section
reports the results of the estimation. The final section concludes.

2. Literature review

A small number of papers seek to extend residential water demand models by explicitly
considering households0 decision-making regarding technology adoption. Cameron and Wright [1] are
motivated by a concern that most residential water demand studies estimate price and income
elasticities for a given stock of water-using capital without examining whether changes in prices
might have a two-fold effect: change water use for a given level of capital and then change the level of
capital installed. Thus, the authors0 research is concerned with the latter issue: what factors influence
retrofit decisions? The authors estimate a series of multinomial logit models to determine the factors
influencing households in their decision to install water-conserving devices in the shower and toilet.
Data come from a cross sectional survey of households in the Los Angeles area in 1983. The estimated
models indicate that the decision to install shower flow restrictors is influenced primarily by potential
to save on energy costs. Toilet retrofits, however, are more influenced by “General Conservation
Mindedness” since the cost savings from reducing water use are still quite small.
Californian water utilities0 policy responses to the droughts experienced during the 1990s provide
an opportunity to examine household decision-making and to assess the effectiveness of alternative
demand management options. Renwick and Archibald [16] use a two-stage procedure to jointly
estimate the adoption of water conserving technologies (low-flow toilets and showerheads, new
irrigation methods) and the structure of household water demands. The estimated equations are then
used to assess the relative efficacy of price and non-price measures that are aimed to induce
residential water conservation. Data are household observations on water use, prices, adoption of
water conservation strategies and characteristics from two Californian cities that experienced severe
droughts. The estimation results suggest that both price and non-price measures reduce household
water use although their relative impact is a function of household structural features (size of lot, etc.)
and characteristics (income).
Hurd [7] considers the choice of landscape and turf type as one form of households0 efforts to
conserve water. Using a mail survey on choices related to the type of lawn coverage, the estimated
discrete choice models indicated that households0 choices regarding landscape type were sensitive to
water prices, public utilities0 education programs and households0 awareness of water conservation
responsibilities. In particular, households0 price elasticity of landscape choice ranges from  2.8 to
0.6 as landscape type changes from all turfgrass to no turfgrass.
Two recent papers make use of a large-scale survey of households in OECD countries in order to
investigate water use behavior and the factors that influence households0 decisions. Millock and
Nauges [11] are concerned with identifying the factors that influence the presence of water-
conserving appliances such as: water-efficient washing machine, low volume toilet, low flow
showerhead, and rainwater collecting tank. The survey also collects information on each respondent0 s
characteristics (e.g. income, education), attitudes and actions and features of the dwelling. One
important limitation of the survey, however, is that it only asks whether the household0 s water use is
metered and whether payment for water is tied to the volume of water consumed. The authors
estimate a series of individual probit equations to separately explain whether a household has each of
the four water-saving appliances. Explanatory variables include age, education, income of respondent;
ownership status of dwelling; characteristics of dwelling (such as age, size), attitudinal variables and
policy variables (binary for each of water metering and volumetric pricing). The equations are
estimated using the entire dataset with country-specific dummies and also separately for each
country0 s observations. The estimation models identify several factors that have significant influences
on the presence of appliances making including ownership status, presence of water metering and
certain attitudinal variables (e.g. membership in environmental group). All of these are positively
correlated with ownership of most water-saving appliances, however, no effort is made to address
issues of potentially endogenous explanatory variables.
D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37 25

Grafton et al. [5] employs a subset of the same OECD data for which respondents answered
optional questions related to household water use and expenditures. With this additional information,
the authors are able to estimate two models. The first is a water demand equation relating water use
to price (calculated as expenditures divided by quantity), income, the presence of water-saving
appliances and household characteristics. The second model is an ordered probit of the frequency of a
number of different types of water-conserving activities. The presence of a volumetric water charge is
found to increase the frequency of a number of self-reported activities including turning off the water
while brushing teeth, taking a shower instead of a bath, collecting rainwater and recycling waste-
water.
A number of papers from economics and other disciplines2 have considered the potential impacts
of government conservation programs aimed at influencing water use and appliance adoption.
Nieswiadomy [13] estimates residential water demand models and tests for the influence of
conservation and public education programs. Data are a cross section of aggregate water use in major
U.S. cities in 1984. Demands are found to be price and income inelastic while conservation programs
do not have statistically significant impacts on demands. Public education programs, however, do
reduce demand in some regions. Corral et al. [2] also consider the impacts of price and non-price
policy instruments. A model of aggregate residential water demand is estimated to study the
influence of pricing and non-price conservation programs on consumption and conservation behavior
in the San Francisco Bay Area, over a 10-year period that includes both drought and normal years.
Empirical results show that pricing can be effective in reducing water consumption, particularly
during the annual dry season. Non-price conservation such as seasonal water-use restrictions and
landscaping audits are particularly effective in inducing outdoor conservation during periods of
drought.

3. Data

The data are drawn from several sources with the key one being individual household level data
from Statistics Canada0 s 2006 Households and the Environment Survey (HES) [20]; these data were
originally collected for the Canadian Environment Sustainability Indicators project. Households
included in the Labor Force Survey (LFS) conducted in 2006 were asked to voluntarily answer a
questionnaire pertaining to how they had dealt over the last year with a number of environmental
issues including: water quality, consumption and conservation of water, consumption and
conservation of energy, recycling and composting practices and attitudes, as well as other
environmental issues. Statistics Canada [19] notes that respondents were generally happy to answer
the survey and that “the majority of refusals for the HES were also refusals and refusal conversions to
the LFS” (p. 24). As a result of this, selection bias is not likely to be present in the HES data. The final
sample of households in the HES numbered 28,334, a 77.8% response rate. Interviewing was done
using computer-assisted telephone (CATI) methods. The HES data provide household-specific
observations on the following: income, number of household members, composition of household,
highest level of education of the respondent, the city or town in which the household is located, the
availability of waste recycling programs, and a number of variables relating to both indoor and
outdoor water conservation choices.
With respect to indoor conservation, the survey provides information on whether the house
has a low-volume toilet and/or a low-flow showerhead but asks no further information about the
equipment. We construct two discrete choice variables. The first identifies households that
report the presence in the home of a low volume toilet and the second does the same for a low
flow shower. Outdoor water conservation choices are explored in more detail in the survey. The
HES asks respondents about the presence of a lawn and/or garden and the frequency with which
each is watered. Responses to these questions allow us to construct two dependent variables to

2
A number of studies have employed models from psychology to gain understanding of why households do and do not
undertake actions to conserve water. Van den Bergh [23] provides a survey of the economic and psychological factors
influencing households0 environmental decision-making.
26 D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37

reflect the frequency of lawn and garden watering, respectively. 3 Each variable takes on a value of
0, 1, 2, 3 if this is the number of times per week the respondent reports watering his/her lawn or
garden, respectively. A value of 4 is assigned for 4 or more waterings per week. A household
without a garden or lawn is coded as a 0. In our sample, 90% of the respondents report that they
have a lawn.
Explanatory variables used to explain the indoor and outdoor conservation choices are those
suggested by economic theory and/or have been identified as significant in previous research. There
are several categories of explanatory variables: those related to prices facing the household,
characteristics of the household, non-price conservation policy variables, provincial indicators, and
climate variables. These data are obtained from other sources and combined with the HES data via
information on the respondent0 s geographical location.
The first category of explanatory variables relates to the water prices faced by respondents. This
information is not collected in the HES; nor was this type of information collected by the OECD survey data
used by Millock and Nauges [11] and Grafton et al. [5], as discussed in the literature survey. In order to
combine the HES data with pricing data, we restrict our attention to observations for which Statistics
Canada provides specific CMA or Census Metropolitan Area information. These are large urban areas with
populations greater than 100,000. Water prices are collected every few years from municipal water
providers from across Canada and collated by Environment Canada in the Municipal Water and Wastewater
Survey [3]. All municipalities in Canada (except those on federal lands) with more than 1000 residents are
surveyed, as well as a representative sample of 660 communities with fewer than 1000 residents. The
closest year of data available are for 2004, so the price observations contained in MWWS are inflated to
2005 values (using province-level Consumer Price Indexes) to correspond with the data collection period
for the 2006 HES. (Respondents were asked about the previous year, i.e., 2005.)
In order to examine the potential role that the price of water plays in explaining the likelihood
and/or frequency of indoor and outdoor water conservation choices, a number of price-related
explanatory variables are constructed. First, we use data on the type of water pricing structure faced
by households in order to create a binary variable reflecting whether households are charged on an
increasing volumetric basis or not. This takes on a value of 1 if the household faces an increasing block
rate structure and 0 otherwise. Second, we construct a variable reflecting the level of the water price.
This is defined as the marginal price of a cubic meter of water at an average monthly consumption
level of 25 m3. Since the HES does not provide information on actual household water consumption in
order to assign a price, we assume that each household is at the average consumption level. Third, we
construct a variable that allows us to investigate whether households are influenced by the recent rate
of change of water prices. The primary motivation for including this variable is that water prices have
historically been quite low in Canada but some jurisdictions have seen increasing rates recently. In
order to calculate the rate of change of real water prices we use prices from the Environment Canada
water price survey prior to 2004. The earlier survey provides data for 2001. After expressing the 2001
marginal price in real terms we calculate a simple percentage change in the water price over the
3-year period. One of the surprises is the observation of a mix of positive and negative real water price
changes over this period. In particular, while Table 1 shows the mean percentage price change to be
positive, the median percentage price change over the period in real terms is actually negative.
Moreover, the real percentage price change is also negative at the 75th percentile of the data.
The second set of explanatory variables relate to characteristics of the household. Questions
contained in the HES allow us to construct variables for the following: the number of persons in the
household, the level of after tax household income, a binary variable reflecting completion of
university or college, and a binary variable to indicate the presence of children in the household.
Unfortunately, the HES does not ask any further questions (such as age of house and/or lot size) that
might assist in better understanding household water conserving choices.
The next explanatory variable relates to municipal utilities0 efforts to promote water conservation
through a variety of non-price policies. The MWWS provides information on different types of water

3
Survey respondents are not asked to provide the number of minutes/hours used each time they water, merely the number
of times in a week.
D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37 27

Table 1
Descriptive statistics.

Variable Units Mean Mina Maxa Stan. dev.

Dummy low flow shower 0.604 0 1 0.489


Dummy low volume toilet 0.402 0 1 0.49
Frequency lawn watering 1.245 0 4 1.325
Frequency garden watering 1.766 0 4 1.601
Marginal price of water $/m3 0.875 0 2.15 0.711
Dummy increasing volumetric price schedule 0.209 0 1 0.407
Dummy access to recycling program 0.972 0 1 0.164
Dummy non-price conservation policy 0.571 0 1 0.495
Household income $ 79,831 15,000 150,000 81,193
Dummy university education 0.806 0 1 0.396
Dummy children in household 0.359 0 1 0.48
Rainfall mm 77.88 23.9 130 22.5
Degree days 67.89 3.3 116.7 36.45
Average summer daily temperature 1C 18.11 14.20 21.6 1.79
Size of household 2.825 1 5 1.394
Dummy Atlantic provinces 0.046 0 1 0.209
Dummy Ontario 0.432 0 1 0.495
Dummy Manitoba 0.032 0 1 0.176
Dummy Saskatchewan 0.0252 0 1 0.157
Dummy Alberta 0.114 0 1 0.318
Dummy British Columbia 0.137 0 1 0.457
ΔPrice of water % 12.519  12.03 44.61 111.445

a
In the case of household income, Statistics Canada disclosure rules require that we report at no less than the 5% and no
greater than the 95% percentile of responses.

conservation programs that are in place within a municipality. Using this information, we construct a
binary variable that takes a value of one if water utilities employ any measures relating to non-price
conservation measures. In the MWWS, these are described as the following: “Lead by example –
efficient municipal/company facilities”, “Media”, “Outdoor advertising (billboards, buses, etc.), “School
curriculum programs”, “Water use bylaws for lawn watering”, and “voluntary measures/restrictions”.
The final set of explanatory variables controls for the influence of climate on outdoor water
conservation choices only. Climate-related data are obtained from Environment Canada [4]. The first
climate-related variable measures total rainfall in the area in which the household lives during the
months of June, July and August. The second climate-related variable (degree days) measures the
number of days during June, July and August when the average daytime temperature is above 18 1C.
The third climate-related variable (temp) measures the average summer daytime temperature in the
area in which the household lives.
Since we are interested in examining the behavior of households whose water is municipally
supplied and for whom water conservation decisions may respond to either price or non-price
incentives we include only those responses from households living in a house obtaining water from a
municipal water supply system. After removing respondents living in apartments, as well as those
who self supply their water through wells, do not live in census metropolitan areas, and those for
whom we do not have complete information (including income and price and non-price water utility
information), our final sample size is 9479 households. Table 1 provides sample characteristics for the
variables included in the analysis.

4. Econometric models

In this section we set out the econometric models used to explain a household0 s indoor and/or
outdoor water conservation choices as observed from the HES dataset. According to the data we only
know the following about indoor water conservation choices: whether households have a low flow
28 D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37

toilet in their home and whether they use a low flow shower. Information from the HES provides
more information on outdoor water conservation choices, specifically, the weekly frequency of lawn
and garden watering, separately.

4.1. Indoor water conservation choices

Consider the decision made by a household i to have a single type of indoor water conserving
appliance such as a low-volume toilet. Assume that the household compares the utility obtained from
both having the appliance in place and not having it and aims to maximize utility. While the
researcher does not observe these utilities, she does observe the specific choice that a particular
household has made, conditional upon its circumstances. Eq. (1) explains the presence (Tn1 ¼1) or
absence (Tn1 ¼ 0) of the appliance, conditional upon observable explanatory variables described by the
vector, X. Elements in the vector include the structure of water prices and presence of non-price water
conservation policies, as well as household specific variables.
T n1 ¼ X0 β þ ε ð1Þ

Since we do not observe all characteristics, the error component captures unobserved influences upon
choices. This is assumed to be distributed as a standard normal. The probability that a household with
characteristics, X, will choose to have a low flow toilet is given in Eq. (2).
Z X0 β
Prob½T 1 ¼ 1jX ¼ ϕðtÞ dt ¼ ΦðX0 βÞ ð2Þ
1

Eq. (2) describes a (single) probit model, where ϕ and Φ are, respectively, the univariate standard
normal density and the cumulative density function. A log-likelihood function that incorporates both
types of choices by households (adopt or not) can be estimated using maximum likelihood methods to
obtain consistent and efficient estimates of the parameters. Eq. (2) can be defined for each separate
behavioral change or technology adoption decision that a household might make. This is the modeling
choice made by Millock and Nauges [11] but it has the shortcoming that it assumes that a household0 s
indoor water conservation choices are made independently. One can posit reasons why decisions
might be correlated. First, they are different ways of achieving the same end. Thus, a household owner
may make several behavioral adjustments together to reduce water use. Second, budget and time
constraints might imply that if one action is taken, the household has insufficient funds or time to
undertake others.
In these circumstances, the researcher can estimate a bivariate probit to explain two jointly
determined decisions in which there is a possible non-zero correlation in the random disturbances
associated with a particular household0 s choices [6]. Define variable, T1, as above to describe the
presence or absence of a low flush toilet and a similar variable T2 to describe the presence/absence of a
low flow shower. In order to account for possible jointness in the two decisions, we make the
following assumptions in Eq. (3):
E½εi jX ¼ 0; i ¼ 1; 2
Var½εi jX ¼ 1; i ¼ 1; 2
Cov½εi ; εj jX ¼ ρij ia j ð3Þ

The resulting bivariate probit model describes the joint probability associated with the two water
conserving choices, Eq. (4). In this equation Φ2 is the bivariate normal cumulative density function and
ρ is the matrix of correlations between the random disturbances.
Prob½T i ¼ 1jX ¼ Φ2 ðX0 β; ρÞ i ¼ 1; 2 ð4Þ

The associated log-likelihood function includes four possible segments: a household uses both types
of water conservation appliances, it uses only one or the other, and it uses neither. Maximization of
this log-likelihood function provides estimates of both the β coefficients and the correlation between
the choices (ρ). A test for the significance of ρ indicates whether the two choices are joint. This is
estimated with the biprobit routine in STATA.
D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37 29

The equations described above include the price of water and other policy variables. Given the
nature of municipal water pricing structures, specifically those that employ some form of volumetric
prices, there is a concern that these variables may be endogenous [17,14]. Similar concerns arise in
relation to potential endogeneity associated with the adoption of non-price water conservation
policies by water utilities experiencing high levels of water demands or those situated in regions of
the country that obtain lower amounts of rainfall/experience more frequent dry conditions. Since our
observational units are households across Canada served by a number of different local water utilities,
we address these issues of potential endogeneity by estimating our model of indoor water
conservation twice: first as a bivariate probit (subsequently, we refer to this as Model #1) and second
as a bivariate probit with instrumental variables (Model #2). In the second approach, we undertake a
joint estimation of instruments for those potentially endogenous variables along with the original
equations of interest [8,10]. As set out in Table A1, we jointly estimate the bivariate probit model with
three additional equations designed to obtain instruments for three water price/non-price policy
variables that might be endogenous. Our choices for instruments were guided by the goal of finding
variables that are correlated with the potentially endogenous variables while being uncorrelated with
the household choice variables. The first equation is modeled as a probit to explain the likelihood of a
household facing an increasing block rate structure. The explanatory variables are electricity prices,
degree-days, and a dummy representing access to recycling programs. We expect each of these
variables to positively influence a community0 s use of an increasing block rate structure. The second
equation explains the marginal price of water that a household faces as a function of electricity prices,
degree days, and a set of provincial dummies that allow for potential differences across Canada to be
incorporated. The excluded province is Quebec. Finally, the third equation employs a probit to explain
the likelihood that a household faces non-price conservation measures, as described above. This is
explained by electricity prices, rainfall, a dummy representing access to recycling programs, a dummy
signaling the presence of water metering, and the instrumented value of the marginal water price.
The water price is instrumented in the second equation but is included as an explanatory variable
in the third equation. The instrumented values of the marginal price, rate type variable and
conservation policy variable are included as explanatory variables in Eqs. (4) and (5). This makes for a
recursive system of five equations (three instrumental variable equations and two indoor water
conservation choices). This is estimated with the cmp (conditional mixed process) routine in STATA
[18]. Table A1 sets out, the system of three instrumental equations and two behavioral equations. The
errors in the five equations are assumed to share a multivariate normal distribution and the equations
are estimated simultaneously as a system of seemingly unrelated equations (SUR) with a recursive
structure using a maximum likelihood estimation procedure. In order to obtain parameter estimates
in the case of 5-dimensional normal integrals, the cmp routine employs the commonly used
Geweke–Hajivassiliou–Kean (GHK) algorithm, a simulated maximum likelihood procedure. We use
100 Halton draws for the simulation. Output includes parameter estimates for the partial correlations
between pairs of errors. This allows us to test for statistically significant correlations for errors in pairs
of equations as a means of identifying both whether indoor conservation choices are jointly
determined and whether the policy variables are jointly determined with indoor conservation
choices.

4.2. Outdoor water conservation choices

While probit models are appropriate for examining indoor water conserving behavior, the implicit
all-or-nothing format may not be the best approach for describing outdoor water conserving choices.
Instead of the choice between do not water and water one0 s lawn, a household may decide on the
relative frequency with which watering is done (e.g., never, once a week, twice a week, or more
frequently). An ordered probit model can be used to describe this.
Suppose there are K ascending but finite choices to describe the frequency with which one water0 s
the lawn, where k ¼0 describes no watering and k ¼K is the maximum amount of watering. Again,
assuming that a household0 s decision depends upon a comparison of the utility to be obtained from
choosing a particular frequency level, we can define an indicator variable (yn) to help explain a
30 D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37

particular household0 s choice, conditional upon its circumstances (Eq. (5)), where X, β, and ε are as
defined above.
yn ¼ X0 β þε ð5Þ
n
The researcher does not observe y directly, however, she does observe the following possible K
discrete outcomes for a given household (Eq. (6)).
y ¼ 0 if yn r0
y ¼ 1 if 0 o yn rμ1
y ¼ 2 if μ1 oyn r μ2


y ¼ K if μK  1 r yn ð6Þ
The parameters identified as μ1, μ2, etc. are unknown threshold parameters that define the regions
into which yn might fall. They are estimated with the β parameters. Assuming that ε is normally
distributed we have an ordered probit model with associated probabilities for the different discrete
choices (Eq. (7)).
Prob½y ¼ 0jX ¼ Φ ð X0 βÞ
Prob½y ¼ 1jX ¼ Φ ðμ1  X0 βÞ  Φð  X0 βÞ
Prob½y ¼ 2jX ¼ Φ ðμ2  X0 βÞ  Φðμ1 X0 βÞ


Prob½y ¼ KjX ¼ 1 Φ ðμK  1  X0 βÞ ð7Þ
The log-likelihood function in this case allows for each of the K segments to be represented. Again,
maximization of the log-likelihood function provides estimates of the parameters of interest, both the
β0 s and the μ0 s.
In order to explore potential correlation between the frequency with which an individual
waters both his/her lawn and his/her garden, we define a bi-variate ordered probit. Defining a
second indicator variable, z, with M possible discrete outcomes and ∂M unknown threshold
parameters, we jointly estimate this two equation system with maximum likelihood, assuming a
jointly normal distribution. As with the bivariate probit model used for indoor water conservation
choices, we can examine the correlation between the two equations for evidence of jointness in
decision-making. Estimating such a model, however, does not address the potential endogeneity of
the water price/policy-related variables
Thus, in parallel to what we do for indoor water conservation choices, we estimate two models for
outdoor water choices. The first is the bi-variate ordered probit (Model #3) and the second is the
bi-variate ordered probit with instrumental variables (Model #4). In the case of the latter, we again
have a system of five equations in order to account for potential price and non-price policy
endogeneity. We instrument three variables – a dummy for non-price measures, a dummy for
increasing volumetric rate type, and the marginal price of water – in a way similar to that described
above for indoor conservation choices. Table A2 sets out the structure of the system of equations
relating to outdoor water conservation decisions. The key difference from the equation system in
Table A1 is that degree days and rainfall are no longer used as instruments since we include them as
explanatory variables in the outdoor water conservation decision equations. Instead for the first
equation in the system (volumetric rate type) we include a variable representing average summer
daytime temperature as an instrument. The other two instrumented equations (water price and
presence of non-price water conservation measures) are estimated with the same explanatory
variables as described above. Finally, the equations describing the ordered probits for lawn and garden
water frequency use the same explanatory variables as those for indoor conservation choices with the
inclusion of two climactic variables specific to outdoor water use, namely, rainfall, and degree-days.
We estimate Model #3 with the bioprobit routine in STATA and Model #4 with the cmp routine
in STATA because of the recursive structure of the five equations [18]. As above, we can test the
significance of correlation coefficients for each pair of equations. In each case, if the null hypothesis
D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37 31

of independence is rejected for the last two equations (lawn and garden watering frequency), then the
data provide support for the hypothesis that these outdoor water choices decisions are made jointly.

5. Results

5.1. Indoor water conservation

Table 2 provides estimated coefficient and standard errors for the two models used to identify the
roles that water prices and non-price policies, as well as socio-demographic factors, play in jointly
explaining two indoor water conservation choices: presence of a low-flow shower and presence of a
low volume toilet. The first model (Model #1) provides estimates of a bi-variate probit model that
allows for non-zero correlation between the two indoor water-related decisions. The second model
(Model #2) is the same as the first but it instruments the price and policy variables and, as a result,
attempts to account for their potential endogeneity. The structure of Model #20 s recursive system of
equations is set out in Table A1 and the estimated coefficients and model statistics for the instrument
variable equations are reported in Table A3.
The estimates provide a number of insights into the factors influencing households0 decision-
making regarding the adoption of indoor water conservation devices. Looking at the estimates from
Model #1, we first consider the three variables related to water prices (the marginal price of water,

Table 2
Bivariate probits for indoor water conservation choices.

Variable Model #1 Model #2 (with instrumented water


price/policy equations)

Low volume Low flow Low volume toilet Low Flow


toilet showerhead Showerhead

Marginal price of water 0.1605*** (0.0199)a  0.1088*** (0.0198) 0.1615* (0.0866)  0.1105 (0.0933)
Dummy: increasing volumetric  0.0006 (0.05) 0.2366*** (0.0517) 0.0006 (0.0683) 0.2354*** (0.0707)
price
ΔPrice of water  1.61E  05 5.02E  05 (0.0001) 1.15E-05 (0.0001) 4.72E  05 (0.0002)
(0.0001)
Dummy: Non-price Conservation 0.0433 (0.0277)  0.0016 (0.0277) 0.0437 (0.3647)  0.0022 (0.3929)
Household Income 1.15E 06*** 7.15E  07*** 1.15E 06*** 7.68E  07***
(1.82E  07) (1.92E  07) (1.85E  07) (1.98E  07)
Dummy: university education 0.0533 (0.0339) 0.0357 (0.0336) 0.0534 (0.034) 0.0326 (0.0337)
Dummy: children in household  0.1732*** (0.0366)  0.1419 *** (0.0369)  0.1715*** (0.0368)  0.142*** (0.037)
Size of household 0.067*** (0.0136) 0.0931 *** (0.0137) 0.0665*** (0.0137) 0.0923*** (0.0138)
Constant  0.6507*** (0.0429) 0.0486 (0.0424)  0.6546*** (0.1175) 0.0485 (0.1251)

Rho (toilet/shower) 0.3817*** (0.015) 0.2376 *** (0.0163)


Rho (toilet/volumetric price N/A 0.0067 (0.2467)
dummy)
Rho (shower/volumetric price N/A 0.0228 (0.2667)
dummy)
Rho (toilet/water price) N/A 0.0124 (0.0513)
Rho (shower/water price) N/A 0.0063 (0.0539)
Rho (toilet/non-price conservation N/A  0.012 (0.0487)
policy)
Rho (shower/non-price N/A 0.0415 (0.0524)
conservation policy)
LLF  12,340.907  22,631.96

a
Estimated standard errors are in parentheses.
nnn
1% Significance.
nn
5% Significance.
n
10% Significance.
32 D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37

a dummy indicating the household faces an increasing volumetric price, and the percentage change in
water price over the last three years). In two cases the coefficients of the price-related variables are
statistically significant; specifically, the presence of higher marginal prices increases the likelihood of
the presence of a low-volume toilet while the presence of an increasing block rate structure increases
the probability of observing a low-flow showerhead. Conversely, higher marginal water prices appear
to decrease the likelihood of low-flow showerheads. This may be because householders believe they
can control their water use in the shower by altering the flow of water with their existing
conventional showerhead or through controlling the duration of their shower. On the other hand, a
recent increase in the price of water does not affect the likelihood of a household having either a low
flow shower or a low flush toilet. The variable representing non-price water conservation polices is
also not a statistically significant factor affecting the likelihood of either indoor water conservation
choice. In contrast, household income, education and size of household (potentially a proxy for square
footage of the house) each have positive signs for both water conservation choices and are mostly
statistically significant. The presence of children, however, has a statistically significant negative sign,
appearing to lower the likelihood of a household having either appliance. Lastly, the two indoor
conservation choices are positively and significantly correlated (ρ¼0.3817 with p-value ¼0.000). Thus,
the null hypothesis of independence with respect to these two choices is rejected.

Table 3
Bivariate ordered probits for outdoor watering choices.

Variable Model #3 Model #4 (with instrumented water


price/policy equations)

Lawn wateringa Garden watering Lawn watering Garden watering

*** b ***
Marginal price of water 0.0579 (0.0182)  0.0500 (0.0181) 0.0575 (0.0701)  0.0540 (0.0618)
Dummy: increasing 0.2147*** (0.0493)  0.0386 (0.0499) 0.2167** (0.0988)  0.0372 (0.0994)
volumetric price
ΔPrice of water 0.0005*** (0.0001) 0.0001 (0.0001) 0.0005*** (0.0001) 0.0001 (0.0001)
Dummy: non-price  0.0055(0.0245)  0.0706*** (0.0243)  0.0029 (0.3022)  0.0672 (0.2575)
conservation
Household income 1.54E  06*** 1.64E  06*** 1.54E  06*** 1.75E  06***
(1.56E  07) (1.77E 07) (1.56E  07) (1.80E  07)
Dummy: university 0.0532* (0.0298) 0.1391*** (0.0298) 0.0484 (0.0298) 0.1346*** (0.0299)
education
Dummy: children in  0.1257*** (0.0318)  0.1205*** (0.0318)  0.1256*** (0.0318)  0.1180***
household (0.0318)
Size of household 0.0982*** (0.0118) 0.0906*** (0.0119) 0.0978*** (0.0119) 0.0873*** (0.0119)
Rainfall  0.0076*** (0.0005)  0.0017*** (0.0005)  0.0076*** (0.0005)  0.0017***
(0.0005)
Degree days 0.0024*** (0.0005) 0.0038*** (0.0005) 0.0024*** (0.0006) 0.0038*** (0.0006)

Threshold 1  0.1001 (0.055) 0.1484*** (0.0549)  0.1058 (0.1183) 0.1557 (0.1044)


Threshold 2 0.3347*** (0.0551) 0.3437*** (0.055) 0.3306*** (0.1187) 0.3517*** (0.1046)
Threshold 3 0.9391*** (0.0554) 0.7545*** (0.0552) 0.9341*** (0.119) 0.7606*** (0.1048)
Threshold 4 1.6057*** (0.057) 1.2799*** (0.0558) 1.5991*** (0.119) 1.2843*** (0.1051)

Rho (lawn/garden) 0.4114*** (0.011) 0.3352*** (0.0122)


Rho (lawn/nonprice policy) N/A  0.0134 (0.2002)
Rho (garden/nonprice policy) N/A 0.0057 (0.17)
Rho (lawn/volumetric price) N/A 0.021 (0.0572)
Rho (garden/volumetric price) N/A 0.0027 (0.0601)
LLF  26,143.225  36,356.343

a
Ordering is defined as increasing frequency of lawn or garden watering.
b
Estimated standard errors are in parentheses.
nnn
1% Significance.
nn
5% Significance.
n
10% Significance.
D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37 33

Table 2 also contains the comparable estimated parameters relating to indoor conservation choices
from Model #2 that instruments the water price and conservation policy variables. The most
important difference from the previous set of results is that, for a number of the coefficients related to
the price of water, the estimated values have changed very little but their respective standard errors
have increased. Thus, instrumenting the explanatory variables suspected of being endogenous has not
changed the magnitudes of these coefficients noticeably but has led to less efficient estimates. This
indicates that there is no evidence of endogeneity between indoor water appliance choices and the
price and conservation policy variables. This conclusion is confirmed by an examination of the
correlation coefficients between individual pairs of equations of instrumented variables and
the behavioral choices which reveals no statistically significant relationships. There is, however, a
statistically significant and positive relationship between the behavioral choice variables. The latter
may be interpreted as evidence of the jointness in the decision-making.

5.2. Outdoor water conservation

Table 3 presents coefficient and estimated standard errors from the two models that examine
outdoor water conservation choices. In order to interpret these results of the outdoor water
conservation equations, it is necessary to note that an increasing value in the dependent variable
implies a likelihood of greater frequency of lawn or garden watering, and hence, less effort to conserve

Table A1
System of recursive equations for indoor water conservation choices (Model #2).

Equation # and dependent variable Explanatory variables Equation type

1. Y1 X1, X2, X11 Probit


2. Y2 X1, X2, X5, X6, X7, X8, X9, X10 Continuous
3. Y3 X1, X11, X12, X13, Y^ 2 Probit
4. Y4 X3, X4, X14, X15, X16, Y^ 1 ; Y^ 2 ; Y^ 3 Probit
5. Y5 X3, X4, X14, X15, X16, Y^ 1 ; Y^ 2 ; Y^ 3 Probit

Y1 dummy for increasing volumetric price schedule (^ denotes instrumented value)


Y2 marginal price of water (^ denotes instrumented value)
Y3 dummy for use of non-price conservation policy (^ denotes instrumented value)
Y4 dummy for low volume toilet
Y5 dummy for low flow showerhead
X1 electricity price
X2 degree days
X3 household income
X4 household size
X5 dummy for Atlantic province resident
X6 dummy for Ontario resident
X7 dummy for Manitoba resident
X8 dummy for Saskatchewan resident
X9 dummy for Alberta resident
X10 dummy for British Columbia resident
X11 dummy for access to glass/plastic/paper recycling program
X12 rainfall
X13 dummy for water meter
X14 dummy for university education
X15 dummy for kids
X16 real percentage change in water price
X17 temperature
34 D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37

on outdoor water use. Model #3 presents the bioprobit results while the structure of Model 40 s
recursive system of equations is set out in Table A2 and the estimated coefficients and model statistics
for the instrumental variable equations are reported in Table A3.
Beginning with Model #3, we first observe that the three price variables have quite different
impacts in the two choice equations. In the case of the frequency of garden watering, a higher
marginal price and the presence of an increasing block rate structure are associated with the greater
likelihood of less frequent garden watering although only the marginal price coefficient is significant.
An unexpected finding, however, is that all of the price-related variables are positively (and
significantly) associated with a greater likelihood of more frequent lawn watering. While such a
finding seems contrary to expectations, there is a plausible explanation. Householders may be
responding to higher prices and the presence of increasing block rates by reducing the overall
quantity of water applied to their lawns but compensating for this reduced quantity by watering more
frequently but for shorter time periods. As stated earlier, however, we do not have data on household
water use and, as a result, this remains a hypothesis that must be tested in future work. Further, with
respect to the positive coefficient on the percentage change in real water prices, recall that the median
value of this variable is negative. Thus, over half of the households in our sample have faced recent
decreases in real water prices and may have responded by increasing their frequency of lawn
watering.
Secondly, the coefficient on the non-price conservation dummy is negative for both lawn and
garden watering frequency and significant for garden watering, suggesting that municipal non-price

Table A2
System of recursive equations for outdoor watering choices (Model #4).

Equation # and dependent variable Explanatory variables Equation type

1. Y1 X1, X11, X17 Probit


2. Y2 X1, X5, X6, X7, X8, X9, X10 Continuous
3. Y3 X1, X11, X13, Y^ 2 Probit
4. Y6 X2, X3, X4, X12, X14, X15, X16, Y^ 1 ; Y^ 2 ; Y^ 3 Ordered probit
5. Y7 X2, X3, X4, X12, X14, X15, X16, Y^ 1 ; Y^ 2 ; Y^ 3 Ordered probit

Y1 dummy for increasing volumetric price schedule (^ denotes instrumented value)


Y2 marginal price of water (^ denotes instrumented value)
Y3 dummy for use of non-price conservation policy (^ denotes instrumented value)
Y6 frequency of lawn watering (categorical)
Y7 frequency of garden watering (categorical)
X1 electricity price
X2 degree days
X3 household income
X4 household size
X5 dummy for Atlantic province resident
X6 dummy for Ontario resident
X7 dummy for Manitoba resident
X8 dummy for Saskatchewan resident
X9 dummy for Alberta resident
X10 dummy for British Columbia resident
X11 dummy for access to glass/plastic/paper recycling program
X12 rainfall
X13 dummy for water meter
X14 dummy for university education
X15 dummy for kids
X16 real percentage change in water price
X17 temperature
D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37 35

water management campaigns do appear to have a beneficial impact on outdoor water conservation
efforts. Thirdly, the socio-demographic variables are all statistically significant. Higher incomes and
household size lead to greater likelihood of weekly watering frequencies, while the presence of
children acts in the opposite way. Rainfall and degree-days have opposite signs, as expected, and are
consistent with a priori expectations and statistically significant. Finally, the estimated correlation
coefficient for the two decisions is positive and statistically different from zero, thereby, implying
jointness in these outdoor watering choices.
The second outdoor model (Model #4) is the same as the first except it instruments the price and
non-price water conservation policy variables. The most significant difference between the two
models of outdoor watering frequency is that the coefficients for some of the price-related and
conservation policy variables are no longer statistically significant. Just as in the case of Models #1 and
#2, we see that instrumenting changes these coefficients by small relative amounts but raises their
standard errors substantially. Thus, we again find no evidence to support the hypothesis of
endogeneity of the price and conservation policy variables.

6. Conclusions

The goal of this research was to investigate factors influencing households0 decision-making
regarding indoor and outdoor water conserving behavior. Despite the interest in this topic expressed
by government, this has been a rather understudied facet of residential water demand studies. Indeed,
the recent studies by Millock and Nauges [11] and Grafton et al. [5] and our study are the first to

Table A3
Equations for instrumental variables for indoor and outdoor conservation choices models (Models #2 and #4).

Dependent variable Explanatory Variable Estimated coefficients (standard errors)

Indoor Outdoor

***
Increasing volumetric price Electricity price 0.5536 (0.0238) 0.5234*** (0.0221)
Degree days 0.0518*** (0.0017) N/A
Access to recycling program 1.2155*** (0.3859) 1.2011*** (0.3891)
Temperature N/A 1.2125*** (0.0443)
Constant  11.9842*** (0.5331)  31.8517*** (1.1389)
Pseudo-R2 0.5853 0.5939

Marginal price of water Electricity price 0.0678*** (0.0033) 0.069*** (0.0034)


Degree days 0.0014*** (0.0002) N/A
Atlantic provinces  0.1561*** (0.024)  0.1888*** (0.0235)
Ontario 0.8717*** (0.0167) 0.9098*** (0.0162)
Manitoba 1.8391*** (0.0229) 1.8433*** (0.0229)
Saskatchewan 0.7159*** (0.0281) 0.6963*** (0.027)
Alberta 1.286*** (0.0222) 1.2413*** (0.0214)
British Columbia 0.2981*** (0.0155) 0.2587*** (0.0142)
Constant  0.3942*** (0.0217)  0.3269*** (0.0206)
R2Adj 0.7304 0.7282

Non-price conservation policies Electricity price 0.1778*** (0.0075) 0.1805*** (0.0124)


Dummy: access to recycling program 0.1803** (0.0844) 0.1641* (0.0886)
Rainfall  0.0052*** (0.0008) N/A
Dummy: water meter  0.1898*** (0.0483)  0.1251*** (0.0421)
Marginal price of water 0.2727*** (0.0328) 0.2744*** (0.0322)
Constant  1.4405*** (0.1141)  1.8934*** (0.1121)
Pseudo-R2 0.1047 0.0981

nnn
1% Significance.
nn
5% Significance.
n
10% Significance.
36 D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37

employ Canadian household level observations to investigate this phenomenon. Their studies benefit
from employing a survey that included a number of attitudinal questions that could be used to
examine the role of environmental attitudes on water conservation decisions. Our work does not have
access to these attitudinal questions but we have augmented the household observations with data on
the level and form of water prices. In addition, we estimate bivariate probit and ordered bivariate
probit models, both with and without instrumenting the price and conservation policy variables, to
investigate the relationships amongst decisions that pertain to indoor water conservation choices and
outdoor water conservation decisions.
Our findings largely are largely consistent with economic theory. The level and form of water
prices each plays a role in influencing the adoption of water-conservation behaviors by households.
Generally, however, prices appear to play a more significant role at influencing decision-making
regarding indoor water conservation than outdoor water conservation choices.
Perhaps equally important, we find that the presence of non-price water conservation measures by
local water agencies does not have a strong effect on watering frequency undertaken by households
with the exception of garden watering. Household characteristics such as income, education level and
family size, however, are significant factors. In addition, there is strong evidence in support of the
view that decisions relating to indoor and outdoor water conserving efforts are being taken jointly.
Conversely, we do not find evidence of endogeneity between prices, policy variables and household
characteristics and behaviour.

Appendix A

See Tables A1–A3.

References

[1] T. Cameron, M. Wright, Determinants of household water conservation retrofit activity: a discrete choice model using
survey data, Water Resour. Res. 26 (2) (1990) 179–188.
[2] L. Corral, A.C. Fisher, N. Hatch, Price and Non-Price Influences on Water Conservation: An Econometric Model of Aggregate
Demand under Nonlinear Budget Constraint, Department of Agricultural & Resource Economics, University of California,
Berkeley, 1999 (CUDARE Working Papers, Paper 881).
[3] Environment Canada, Municipal Water Pricing Report: 2004 Statistics, , 2008 (Catalogue En14-6/2008E).
[4] Environment Canada, National Climate Data and Information Archive, 2010.
[5] R.Q. Grafton, M.B. Ward, H. To, T. Kompas, Determinants of residential water consumption: Evidence and analysis from a
10-country household survey, Water Resour. Res. 47 (2011) W08537, http://dx.doi.org/10.1029/2010WR009685.
[6] W. Greene, Econometric Analysis, 7th edition, Prentice Hall, New York, 2008.
[7] B. Hurd, Water conservation and residential landscapes: household preferences, household choices, J. Agric. Resour. Econ.
31 (2) (2006) 173–192.
[8] Y. Konishi, K. Adachi, A framework for estimating willingness -to- pay to avoid endogenous environmental risks, Resour.
Energy Econ. 33 (1) (2011) 130–154.
[10] N. Lyssenko, R. Martínez-Espiñeira, ‘Been There Done That’: disentangling option value effects from user heterogeneity:
when valuing natural resources with a use component, Environ. Manage. 50 (2012) 819–836.
[11] K. Millock, C. Nauges, Household adoption of water-efficient equipment: the role of socio-economic factors, environmental
attitudes and policy, Environ. Resour. Econ. 46 (2) (2010) 539–565.
[12] National Climate Data and Information Archive Canadian Climate Normals or Averages 1971–2000, 2010, website: 〈http://
climate.weatheroffice.gc.ca/climate_normals/index_e.html〉 (accessed 15.06.10).
[13] M. Nieswiadomy, Estimating urban residential water demand: effects of price structure, conservation, and education,
Water Resour. Res. 28 (3) (1992) 609–615.
[14] S. Olmstead, W.M. Hanemann, R. Stavins, Water demand under alternative price structures, J. Environ. Econ. Manage. 54 (2)
(2007) 181–198.
[16] M. Renwick, S. Archibald, Demand side management policies for residential water use: who bears the conservation
burden?, Land Econ. 74 (3) (1998) 343–359.
[17] A. Reynaud, S. Renzetti, M. Villeneuve, Pricing structure choices and residential water demand in Canada, Water Resour.
Res. 41 (2005) 11.
[18] D. Roodman, CMP: Stata Module to Implement Conditional (Recursive) Mixed Process Estimator, Statistical Software
Components, Boston College Department of EconomicsS456882.
[19] Statistics Canada, Microdata User Guide: Households and the Environment 2006 Survey, Environment Accounts and
Statistics Division, Ottawa, Canada, 2006.
[20] Statistics Canada, Households and the Environment 2006 Survey, Catalogue 11-526-X, Environment Accounts and
Statistics Division, 2006.
D.P. Dupont, S. Renzetti / Water Resources and Economics 4 (2013) 22–37 37

[21] Statistics Canada, Canadian lawns and gardens: where are they the ‘greenest’? EnviroStats, 2007 (Catalogue number 16-
002-XWE), Ottawa, Canada.
[22] Statistics Canada, Households and the Environment 2009 Survey, Environment Accounts and Statistics Division, 2012
(Catalogue no. 11-526-X).
[23] J.C.J.M. Van den Bergh, Environmental regulation of households: an empirical review of economic and noneconomic
factors, Ecol. Econ. 66 (4) (2008) 559–574.

You might also like