Urgent a high level analysis

Applied Economics, 2006, 38, 383–393

Effect of price information on

residential water demand

S. Gaudin

Department of Economics, Oberlin College, Oberlin, OH 44074, USA

E-mail: [email protected]

Microeconomic theory predicts that people decrease consumption when

price increases, the magnitude of the effect depending on price elasticity.

The law of demand, however, implicitly assumes that consumers know

prices, an assumption that is not always satisfied in markets with ex post

billing. When prices are not transparent, elasticity estimates are potentially

lower than their full information potential. Evidence of low price elasticity

abounds in residential water demand studies, limiting the effectiveness and

desirability of using price signals as a conservation tool. It is hypothesized

that resident’s sluggish response to price is partly due to the absence

of price information on water bills. Differences in the informational

content of bills are documented for the first time on the basis of sample

bills collected from 383 utilities across the USA. A standard aggregate

water demand model is augmented with qualitative variables describing

differences in billing information, allowing such variables to affect the

intensity with which consumers respond to price signals. No evidence

is found that non-price information items affect price elasticity but there is

a statistically significant effect in the case of price-related information;

in our sample, price elasticity increases by 30% or more when price

information is given on the bill.

I. Introduction

The power of price signals in motivating demand

responses is put into question when price elasticity

is low and when people have imperfect information

about prices. When it comes to residential energy and

water use, evidence of low price elasticity of demand

abounds. 1

Recent empirical analyses have also

found evidence that, within price ranges observed,

there is a large amount of water use under which

demand is insensitive to price (Gaudin et al., 2001;

Martı́nez-Espiñeira and Nauges, 2004). The focus

of most studies has been to estimate intrinsic

parameters of water demand in order to assess the

relative merits of price versus non-price demand

reduction policies. However, existing research has

not considered the possibility that price elasticity

itself could be influenced by policy. We hypothesize

1 The average estimate of price elasticity from 18 studies of annual residential water demand reported in Hanemann (1997)

is 0.46 (absolute value) with a mean elasticity of 0.36 for winter demand and 0.70 for summer demand (computed from Hanemann, 1997, Table 2.5, pp. 67–72). Espey et al. (1997), in a meta-analysis of residential water demand studies, report a median short run price elasticity of 0.38 and a median long-run price elasticity of 0.64.

Applied Economics ISSN 0003–6846 print/ISSN 1466–4283 online � 2006 Taylor & Francis 383 http://www.tandf.co.uk/journals DOI: 10.1080/00036840500397499

that price elasticity results are not exogenous to the level of information provided to consumers. The issue of price transparency has been advanced in past research to explain demand responses and price perception in utility demand (Foster and Beattie, 1981; Shin, 1985; Nieswiadomy and Molina, 1991), but variation in information across utilities has not been documented. While water utilities are legally required to provide price information, there are no specific regulations about the way the information is to be provided, creating potential differences across communities in the cost of acquiring price information. In particular, although information costs become minimal to consumers when they can find the information they need on their utility bill, water utilities are not required to give price informa- tion on bills. Would price elasticity increase with better price information on water bills?

Historically, growing cities in the USA have relied on expanding supply to satisfy rapidly increasing water demands. As new supply sources become more and more difficult to secure, measures to encourage conservation have become an important part of water management and planning in areas subject to drought and/or experiencing high population growth. The effectiveness and desirability of price instru- ments in demand management, however, depends on the magnitude of the price elasticity of demand. The lower the price elasticity, the greater the price increase needed to absorb a given shortage. In practice, large price increases are unlikely because of distributional implications and political pressures. Higher price elasticity would increase the effect of price signals and, consequently, the attractiveness of price-based policies. If weak sensitivity to price stems from the intrinsic nature of the good, then no economic policy instrument can change it; e.g. if the water we use is a necessity to life, as is often implicitly assumed, price elasticity cannot be changed, short of finding substitutes for water. However, very little of the water used by residential consumers can be considered a necessity.

2 Basic economic theory points to at

least two other reasons why consumers would not be responsive to price in their decision to consume water: water bills constitute a small portion of their budgets, and price information is imperfect.

This paper is concerned with the latter. In particular, we hypothesize that residents’ sluggish response to price is partly due to the fact that the information necessary to make informed decisions is not con- veniently available to them. If this hypothesis is valid, including clear price information on water bills should increase price elasticity.

The Environmental Protection Agency (EPA) guidelines for water conservation recommend the use of an ‘understandable’ and ‘informative’ water bill.

3 Best Management Practices produced by state

agencies started to include the layout and information content of bills as a tool to foster conservation (e.g. Envision Utah, 2002). However, cross sectional variations in billing practices and, in particular differences in the informational content of bills have not been considered in empirical studies of residential water demand. The experience of gas and electric utilities gives some indication that the content and presentation of bills matter. For example, Fast (1990) found a significantly larger price elasticity of resi- dential electricity demand after the 1985 change to ‘plain language’ bills in the state of New York. Fast’s study provided evidence of the overall effect of the new bill but could not identify the specific effect of price information. Other evidence in energy demand relates to consumption feedback rather than price information (Egan et al., 1996; Matsukawa, 2004).

4

Growing cities, especially those in the West and Southwest of the USA but also in water scarce areas around the world, would benefit from understanding whether a slight modification on their water bills could reinforce the effectiveness of price signals, allowing utilities to rely more on prices to reflect changes in scarcity rather than rationing methods, thereby reducing the welfare impact of shortages. This study provides the first quantitative measure of the impact of price information on household’s sensitivity to price variation in utility demand analyses.

Billing information was collected directly from utilities across the USA to document the variation in billing practices and to test the impact of price information. In our sample, 17% of the utilities clearly indicated marginal prices next to unit

2 Baumann and Boland (1997) dismiss the ‘water is a necessity’ argument as a water management myth (p. 21).

3 The Safe Drinking Water Act Amendments of 1996 mandated EPA to publish guidelines for water conservation to be used

by public systems. These guidelines are purely informative and available on the EPA website. ‘Understandable’ refers to the inclusion of price and quantity information on the bill; ‘informative’ refers to additional quantity information such as comparison with previous usage and the inclusion of conservation tips. 4 Egan et al. (1996) compare the effectiveness of different graphical displays to provide feedback on own energy use and find

that consumers react differently to different displays. Matsukawa (2004) finds that the use of monitors providing continuous feedback to customers on their energy consumption promotes energy conservation, and that the more the monitor is used, the greater the price elasticity of demand.

384 S. Gaudin

consumed on their water bill while 78% gave no price information other than total amount due. We used this data along with utility and community level data from secondary sources in a standard water demand model where features of the bill may affect price elasticity. The analysis indicates that the presence of marginal price information on the bill has a statistically significant impact on price elasticity while other types of information are not found to affect price elasticity.

II. Water Demand Model and Data

Model specification

Basic aggregate annual residential water demand

model. We use an aggregate water demand model to be estimated using cross-sectional community level data. The use of aggregate cross-sectional data is motivated by the nature of our quest. Indeed, features of water bills scarcely change over time and are identical for all consumers in the service area of a given utility, therefore this research would not gain from using microeconomic or pooled cross-sectional data.

5 Our choice of demand variables, functional

form, and estimation procedure are guided by insights provided by a large number of published studies, data availability, and interpretability of the coefficients of interest to this study.

6 We base our

analysis on the following aggregate water demand model:

Q ¼ f ðAP, I, H, D, AAP, T 90Þ

where Q is per capita annual water consumption, AP is average price, I is income, H is average household size, D is density, AAP is average annual precipitation, and T90 is the number of days when temperature exceeded 90�F. While AAP is a 30-year average and as such will capture structural differences between communities mostly related to outdoor water use and practices, T90 is specific to the survey year and will capture both structural and temporary features of water demand. It is likely that a high level of T90 will not only increase outdoor water use due to higher evaporation but also possibly increase indoor water use due to more frequent washing and

showering, both effects that cannot be captured by AAP.

The choice of an average price specification is driven by specificities of our data set. Several studies show that consumers tend to respond to average prices for water and electricity demand rather than marginal prices (Foster and Beattie, 1979, 1981; Shin, 1985; van Helden et al., 1987; Griffin and Chang, 1990). Others find that neither an average price specification, nor a marginal price specification can be rejected in favour of the other (Williams and Suh, 1986). A few studies have shown that, in some circumstances, marginal prices are more appropriate (Nieswiadomy and Molina, 1991; Taylor et al., 2004).

7 When using marginal prices,

however, one must be aware of the effective price structure used, as people are likely to respond differently to marginal prices depending on whether price schedules are increasing, decreasing or uniform (Nieswiadomy and Molina, 1991; Olmstead et al., 2003). Unfortunately, although we know the type of rate structure that utilities used in 1995/6 – as reported in the American Water Works Association (AWWA) database – we do not know the full price schedule and cannot identify the amount of fixed and variable charges for each utility.

8 Without

knowledge about fixed fees and free allowances, we cannot distinguish between cases when the marginal price is less, equal to, or greater than the average price therefore, using a marginal price specification with heterogeneous price structures would create additional estimation and interpretation problems. Although the use of marginal prices rather than average prices may affect elasticity estimates (Espey et al., 1997), the magnitude of the effect of informa- tion on price responses should not be affected. Finally, although it is often argued that billing information may affect the appropriateness of a given price specification (Foster and Beattie, 1981; Shin, 1985), the limited information on marginal prices does not allow us to test for the appropriate price specification; we leave this enquiry for further research and focus our contribution on the impact of information on price elasticity.

Data constraints also motivate the exclusion of other prices in the demand equation. While most aggregate residential water demand models in the

5 Espey et al. (1997) found no significant differences between long-run elasticity estimates calculated using single period cross

sectional data and other types of data. 6 See Renzetti (2002) for a review of the literature.

7 Nieswiadomy and Molina (1991) found evidence that, for a true increasing block rate structure – i.e. without large quantity

allowances included in the fixed fee – the role of marginal prices dominated average price; a recent article by Taylor et al. (2004) attributes the response to average price to the magnitude of fixed fees. 8 We do know the cost to consumers of 3750 gallons and 7500 gallons per month but a pseudo-marginal price calculated from

this information is not likely to be the appropriate marginal price for the level of water consumed.

Effect of price information on residential water demand 385

literature ignore prices of other goods, sewer charges

are often included in the average price specification (Renzetti, 2002). While we do not know sewer prices, we do know whether the utility charged for sewer, electricity, or other utilities on the same bill and we

are able to test whether the inclusion of such charges on the bill affect price elasticity in the transformed model.

Modified demand model. Price and quantity-related billing information is assumed to enter the demand equation through its effect on price elasticity. The presence of such information is incorporated in the estimation using qualitative dichotomous variables

interacted with the price variable (slope dummies). Other type of information, such as messages aimed at sensitizing consumers to the importance of conserva-

tion, may affect demand independently of prices through their potential effect on consumer’s prefer- ences, and are therefore included in the estimable

equation as intercept dummy variables. Our choice of a log–log functional form as opposed to a linear form is motivated by previous research where it was shown that forcing price elasticity to decrease along

the demand curve is not an appropriate specification for water demand (Gaudin et al., 2001).

9 As opposed

to other functional forms, the log–log functional

form facilitates interpretation of the coefficient as elasticity estimates and allows direct comparison with the existing literature.

10 The following relationship

is assumed:

lnðQÞ ¼ �0 þ�0 lnðAPÞþ �ið�i lnðAPÞXiÞþ�1 lnðI Þ

þ�2 lnðDÞþ�3ðHÞþ�4 lnðAAPÞþ�5T 90

þ �j�jZj, i ¼ 1, . . . , x; j ¼ 1, . . . , z

where the Xi and Zj are qualitative dichotomous

variables representing the different features of the

water bill: the Xs represent billing features that do not

directly affect consumer’s utility but are likely to

change consumers’ sensitivity to price; the Zs are

variables likely to directly affect consumers prefer-

ences for water. Price elasticity in a standard low

information water bill is �0 while price elasticity on a bill that includes information item i is �0þ�i.

Data

Primary source data on water bills. The American Water Works Association (AWWA) periodically

collects data on member utilities across the USA.

In 1996, AWWA surveyed a total of 3200 utilities

(the smallest utilities were excluded) and received

898 responses with 501 utilities serving residential

customers. 11

Utilities with un-metered (flat fee) or

seasonal rates were dropped, leaving us with

495 utilities to locate and contact by phone during

the summer of 2003. The goal was to find out what

information was given on a residential water bill that

could potentially affect water use. Since most utilities

used December 1995 as the end date for the yearly

data they provided to the AWWA, we asked about

bills as they were in 1995. 12

We obtained usable

information from 383 utilities of which 130 reported

charging decreasing block rates, 104 increasing block

rates, and 149 uniform per unit rates. 13

Although

the sample may not be considered representative

of US utilities, sample selection cannot be related

in any systematic manner to variables in the demand

equation and therefore may be considered random

for our analysis. 14

9 A better specification in cases when researchers are interested in changes in price elasticity along the demand curve would

require functional forms such as a Stone-Geary form or flexible forms that allow decreasing price elasticity with decreasing quantity and increasing price (Gaudin et al., 2001). In terms of demand elasticities at the mean, however, Espey et al. (1997) find that the choice of functional form does not systematically affect price elasticity estimates. 10 Coefficients on household size and number of hot days are left in their raw form to be conveniently interpreted as semi-

elasticities. A similar regression with the two variables in their log form reduced the overall fit but did not affect the level and significance of other parameter estimates. 11 AWWA membership in 1996 consisted of about 4000 utilities. Although there are approximately 56 000 utilities in the USA,

most of them are small utilities with a customer base lower than 500. Over two-thirds of the 500 largest utilities are represented in the AWWA data. Although another survey was conducted in 1999, it did not include enough information for our analysis. 12 When a copy of a 1995 bill could not be found but someone with an accurate recollection of the information could be

located, we asked the utility to answer a simple questionnaire. If there was any doubt about the accuracy of the information, the observation was dropped. 13 The price structure is as reported by the utility in the AWWA survey. As indicated in the previous section, we do not know

the details of the price schedule. In our phone interview we asked utilities that still had a decreasing block structure the quantity level that would push consumers into the lower priced second block and found that in most cases, the second block was too high to be reached by any single-family household. Such a price structure is effectively similar to the uniform rate for the household. 14 Sixty utilities had either changed ownership (most due to the American Water Company merger in 2000) or could not be

located; 24 did not have records or recollection of their 1995 bill; 5 had bills changed in the 1995–96 period; 3 were removed for miscellaneous reasons, such as a utility serving only summer homes. Twenty utilities refused to participate.

386 S. Gaudin

The kind of information given on water bills from

city to city varied along several dimensions. All bills

included meter readings, quantity used, and total

amount due for water (separate from other charges).

There was significant variation on whether the bill

included marginal price information and history

of use. Table 1 summarizes our survey results on

information variables that were found to differ

significantly across utilities. Only 17% of utilities in our sample indicated price

per unit next to consumption and an additional 3%

indicated the price schedule somewhere on the bill.

For history of use, a simple comparison to the same

period last year was the most common (23% of

utilities overall), while only 6% gave more extensive

historical data (multiple months, generally presented

with a graph). Although price information and

consumption history are positively correlated

(�¼0.27, p-value¼0.00), many utilities with price information did not include history and vice versa.

15

There is no evidence of correlation between price

information and conservation messages on the bill

(�¼�0.03, p-value¼0.53).16 A break down of the sample by regions and by rate structure revealed that

a disproportionate share of utilities in the West

included different types of information on the bill.

The same is true for utilities using increasing block

rates. The empirical analysis will take account of

these features of the data. Another feature of bills that may affect water

use although not related to their informational

content is whether they include utility charges other

than water. The effect on water use could be two-fold:

on the one hand, including other utilities reduces the

consumer’s ability to understand the cause of changes

in the total bill, especially if the customer only looks

at the bottom line and there is little breakdown,

possibly reducing price elasticity; on the other hand,

consumers are more likely to pay attention to a larger

total bill and price elasticity tends to be higher for

goods that constitute a larger portion of income.

Other utilities were included in about three-quarters

of the bills (286 included sewer charges and 59

included energy charges, of which 55 included both

sewer and energy). Bills with sewer and/or energy

charges were evenly distributed among the different

regions and across rate structures. Finally, billing frequency may influence water

consumption through price perception. Arbues et al.

(2003) recommend that billing frequency be included

as a relevant variable in residential water demand

models. Two opposite forces could be at play: on the

Table 1. Prevalence of information on water bills by region and by rate structure (1995)

Region a

Rate structure b

Full sample MW NE SE SW W DB IB UR

Number of collected bills 383 109 64 77 31 102 130 104 149

Percent of bills with Price per unit consumed

c 17.2 11.0 23.4 3.9 19.4 29.4 8.5 26.9 18.1

Price schedule d

2.9 2.8 3.1 5.2 0 1.2 2.3 2.9 3.4 Consumption history I

e 22.7 11.0 7.8 15.6 28.6 50.0 10.8 39.4 21.5

Consumption history II f

5.7 1.8 0 3.9 16.1 11.7 2.3 14.4 2.7 Conservation messages 9.7 4.6 3.1 6.5 6.5 22.6 3.0 18.3 9.4 Other non-price info

g 10.2 3.7 7.8 6.5 9.7 21.6 3.9 17.3 10.7

Notes: a US Census definition.

b DB¼declining block rate; IB¼ increasing block rate; UR¼uniform rate.

c Rate per unit indicated next to units consumed. With block pricing, the price per unit is next to quantity consumed for all relevant consumption blocks. d Bills that include information about the rate schedule on the bill but not next to consumption.

e Consumption in the same period in the previous year is included for comparison with current period.

f Consumption for all periods in the previous year is included (the information is either presented in a table or a graph). Note that these bills are a subset of the bills reported to include Consumption history I. g Includes features used by a small number of utilities that provide information on quantity including: benchmark

comparisons, daily consumption, and % age changes in consumption.

15 Among utilities that gave price information on the bill, 32% gave simple history and 13% gave more detailed history

(compared to 23 and 6% in the full sample), while 40% of all bills with consumption history included price information (compared to 20% in the full sample). 16 Eight per cent of the bills with price information gave conservation messages (compared to 10% for the full sample) while

16% of the bills with messages gave price information (compared to 20% for the full sample).

Effect of price information on residential water demand 387

one hand, frequent bills are a reminder that water is not free and may create a better understanding on the part of consumers of the price structure and the relation between consumption and cost, increasing price elasticity; on the other hand, more frequent billing causes smaller overall bills, which would dampen price elasticity.

17 Over half of the utilities

in our sample used monthly billing. 18

Qualitative (dummy) variables were created to record the presence of the different types of informa- tion. To avoid low sample problems, we do not create dummy variables for information items used by fewer than 30 utilities. We are most interested in the presence of price information, theoretically most likely to increase price elasticity. Quantity informa- tion may also affect price elasticity by attracting the attention of individuals to prices in an effort to figure out whether changes in total bills are due to quantity or price movements.

19 The presence

of conservation messages is more likely to affect consumers’ preferences independently of prices and is therefore allowed to enter directly in the demand function.

Qualitative variables used in the model are separated into two groups. One group includes billing features that may affect demand through price responses (Xis); the other includes billing feature that may affect demand by altering preferences (Zj s). A single variable, message, is included in the latter group and set equal to 1 if the utility commonly included conservation related messages on bills. The first group of qualitative variables includes three categories: price information variables, quantity information variables, and variables related to other aspects of billing. For price information we test the model with two measures: mpinfo and priceinfo. The former is set equal to 1 if marginal price is indicated clearly next to units consumed; the latter includes all the utilities with mpinfo¼1 plus the ones with full information about the price schedule somewhere else on the bill. To describe quantity information we use the quantityinfo variable, equal to 1 when the bills include simple or advanced consumption history or other detailed quantity information such as

daily average use. 20

Other variables related to the size of the bill and the rate structure are: CombinedBilling¼1 if sewer charges, electricity charges, and/or gas charges were included together with water charges on the same bill; MonthlyBill¼1 if water was billed monthly; and IB¼1 if the utility reported using an increasing block rate structure.

21

Secondary data sources. All variables pertaining to price and quantities and other characteristics of the utilities come from the 1996 AWWA survey. Per capita quantity was calculated as the ratio of volume of residential sales to retail population served and average price as the ratio of total revenue to total volume of residential water sales. Density was calculated as size of the service area divided by the retail population served. Census data on average income, median income and household size was aggregated to match the utilities’ service areas.

22

Finally, the Annual Climatological Summary of the National Oceanic and Atmospheric Administration (NOAA) was used to locate the closest weather station with 1995 data on temperatures and normal rainfall. Table 2 presents summary statistics and description for the variables used in the estimation.

III. Estimation and Results

Endogeneity issues

Concerns about endogeneity arise when an explana- tory variable is potentially correlated with the error component of the dependent variable, thereby violating a basic assumption of ordinary least squares (OLS) and creating bias in coefficient estimates. Two potential sources of endogeneity need to be addressed in regards to the estimation of our water demand model. One source is related to the specifica- tion of the price variable; the other, recently high- lighted in the water demand literature, comes from the possibility that there are unobserved community- related components of per capita consumption that may be correlated with community-related

17 Stevens et al. (1992) find that higher frequency billing decreases price elasticity while Kulshreshtha (1996) did not find

conclusive results on billing frequency (see Arbués et al., 2003). 18 Billing frequency in our sample varied from twice a year (only two utilities) to 12 times a year (200 utilities); 92 utilities

billed four times a year and 88 billed six times a year. There were no clear differences in billing frequency by rate structure. 19 However, the presence of detailed quantity information on the bill could also complicate the bill and obscure price

information, leading to a reverse effect. 20 Variables were also created for simple history, advanced history, and others quantity related information to test individual

effects. 21 Additional dummy variables were created to test the individual effects of sewer charges and energy charges as well as

different billing frequencies. 22 The 2000 US Census was used because the 1990 Census did not allow us to match a good number of service areas because

of changes in zip codes or names of places.

388 S. Gaudin

components of price or, in our case, to the choice of billing information in a specific community (Nauges and Thomas, 2000).

The existing literature on residential water demand recognizes different reasons for believing that the price variable in a residential water demand model may be endogenous (Renzetti, 2002).

23 In particular,

endogeneity may arise when using an average price specification in a sample where some utilities have non-linear price structures. The use of aggregate data mitigates the problem generated by the simultaneous determination of price and quantity in consumer choices, thus reducing the likelihood of a simultaneity bias (Shin, 1985). Models of water demand that have estimated price elasticity using different price specifications with aggregate data have not found significant differences (Griffin and Chang, 1989; Martı́nez-Espiñeira, 2003). The endogeneity in this model is therefore more likely to come from the fact that marginal price is usually not equal to average price. If marginal price would be a better specification in some communities, measurement errors may result in biased OLS estimates. Following Hausman (1978), we test for endogeneity of the price variable by running both OLS and two-stage least squares (2SLS) on the same demand model and comparing the coefficient estimates, assuming that in the absence of endogeneity, 2SLS estimates are consistent but inefficient. We perform the test using the standard (base) demand model where average price is the only variable that needs to be instrumented. In the two-stage procedure, the log values of the total charges for 3750-gallon and 7500-gallon monthly bills are used in addition to the other exogenous variables to instrument price.

24 Testing for the equality of OLS

and 2SLS coefficient estimates, we find no evidence

of a systematic bias in the OLS coefficients with a �2 value of 3.46 (p-value¼0.75).

The second source of endogeneity is more prob- lematic in the absence of instruments for price information variables. If the quantity of water use per capita motivates utility managers to include more information on their bills (i.e. if they believe it can foster conservation), the coefficients on information variables will be biased. It is also possible that community characteristics – e.g. environmental con- sciousness of the population or familiarity with water scarcity issues – motivate both lower water use and informative billing. In the absence of good instruments for billing information, this potential source of bias cannot be tested formally. Political data about environmental consciousness of the population could be obtained for some communities but the number of utilities for which we could obtain data was too small to carry a meaningful analysis. Several factors, however, lead us to believe that OLS results are appropriate to make meaningful conclusions. First, correlations between informational features of the bill are low or insignificant as reported earlier. For example many utilities that include conservation messages on the bill do not provide detailed price information. The EPA guidelines about the impor- tance of understandable and informative bills were not published until the late 1990s while the bills we collected date back to 1995. We found no evidence while collecting the data that bill formats were chosen with water conservation strategies in mind. When asked about the motivation for including or not including different features on the bills the utility staff that provided us with the information on billing features either did not know or made reference to software capabilities, mailing costs, and attempt to

Table 2. Summary statistics

Variables Description Mean Std. dev. Min. Max. N a

Q Per capita residential consumption in thousand gal./yr 32.48 15.94 7.47 112.6 378 AP Average price of water in US$ per thousand gal. 2.33 1.04 0.13 6.30 380 Income Per capita income in thousand US$ 22201 8584 9480 98643 383 HHsize Avg. number of household members in housing units 2.56 0.37 1.71 4.91 383 Density Population density in persons per square mile 2246 2033 8.07 11923 353 AAP 30-year average annual precipitation in inches 35.08 13.28 1.66 79.49 383 T90 Number of days with temperature >90�F. 41.00 33.40 0 172 383

a Missing entries in the AWWA database reduce the sample size for Q, AP, and Density.

23 These reasons are somewhat different from usual simultaneity issues in demand estimation since the supply side of the

market is regulated to balance revenues and expenses each period. In effect, this means that a higher price cannot be interpreted as an incentive for utilities to supply more water. 24 To gauge the appropriateness of the instruments, we use them to obtain predicted values for the average price variable.

Correlation between the predicted and the original average price variable is 0.65. A linear regression of the predicted value of AP on all the predetermined endogenous variables of the demand equation produces residuals that do not significantly explain the variation in Q (p-value of 0.57).

Effect of price information on residential water demand 389

reduce consumer enquiries. More importantly, water bills are rarely changed. Most utilities included in our survey were still using the same bill as in 1995 in 2003. In cases when the bill format was changed since 1995, all utilities included indicated that the previous bill had been in effect for as long as they could remember. Since the format of water bills cannot be changed easily in response to changes in current quantity demanded, it is more likely to be correlated with structural variables already included in the model (such as average annual precipitation or density) than to the dependent variable.

25 While we cannot quanti-

tatively assess the extent to which omitted, community related variables bias coefficient estimates on price and price information variables, the same type of bias is expected for price and price information (lower quantity use and higher price with greater price information) in ‘water aware’ communities. However, a simple calculation of the correlation coefficient between average price and the price information dummy did not reveal any significant relationship between price and price information (�¼0.04, 95% confidence interval from�0.06 to 0.1).

Results and discussion

Results of the OLS regressions are given in Table 3. The first column gives results of the simplest model with variables most commonly used in the literature. In order to clarify the role of billing information relative to other billing features added to the model as determinants of price elasticity, another base model is estimated with the same variables as in the standard model plus non-information-related billing variables (Base Modelþ). In the information models, we use two definitions of price information: INFO I is the more restrictive model where only bills with unit price given next to units consumed are included in the price information variable (mpinfo); INFO II includes in the price information variable all bills with the full price information on them, whether next to consumption or somewhere else on the bill (priceinfo). To address problems highlighted in the data section concerning the higher prevalence of information for utilities in the West and Southwest and those using increasing block price schedules, we run INFO II on sub-samples of the data excluding (a) all utilities in the West (column 5), (b) all utilities in the West and Southwest (column 6), and (c) all utilities with increasing block (IB) pricing (last column).

26

Results from the base model (standard aggregate

model with no information variables) compare

favorably to the literature. We find a price elasticity of �0.37 and an income elasticity of 0.30.

27 The

coefficients on average rainfall and high temperatures

are of expected sign and significant. Density, used as a proxy for describing the housing stock and

size of yards, has the expected negative effect on per capita water use. The significant and positive

effect of household size on per capita consumption

is more surprising but the result is likely a feature of using aggregate data: household consumption is

clearly positively related to household size and in

our sample, per capita and per household usage are highly correlated (�¼0.93). Overall, the adjusted R-squared of 0.44 is similar to other studies of water

demand that use aggregate cross-sectional data. In the extended Base Model in the second column of

Table 3, neither billing frequency nor combination

billing is found to have a significant effect on price elasticity in our sample. We also ran the model (and

subsequent regressions) with sewer and power as

separate variables and included lower frequency billing in addition to monthly billing. In all cases,

we found no significant individual effects. Results from the models with information variables

indicate that price information has a significant

positive impact on price elasticity. The presence

of marginal price information on the bill next to quantity consumed increases price elasticity by a

factor of 1.4 (price elasticity is �0.36 for areas that do

not include the information and �0.51 for areas that do). Assuming constant elasticity, this means that

for any given quantity reduction target, required price

increase can be close to 30% lower with price information on the bill; for example, a 10% decrease

in quantity requires a price increase of approximately

20% when price information is on the bill, compared to 29% otherwise. Accounting for the presence of

price information anywhere on the bill (Info II) yields a larger coefficient (although not significantly so)

and lower standard error. We interpret this effect

as further evidence that individuals find it costly to seek simple price information outside of the bill

but do react more strongly when the price system

is transparent. We cannot reject the hypothesis that all other types of information have no effect on the

price elasticity with probability values equal to 0.5

for messages and 0.4 for quantity related information

25 For example, it is likely that consumers in areas with low annual precipitation are more sensitized to water conservation

than those in water abundant areas. 26 Note that the sample size decreases significantly, thus affecting standard errors and test statistics.

27 The estimation was run with median income instead; other parameter estimates were not affected significantly and the

income elasticity was 0.24.

390 S. Gaudin

interacted with average price. We also ran the model

with history and other information separately with

no significant individual effects. The joint significance

of non-price billing features was tested using different

linear combinations and the null hypothesis could

not be rejected in all cases. 28

Given such high

probabilities that the coefficients may be zero,

we cannot make inferences from their signs. As in

the Base Modelþ results, the evidence on the impact

of larger bills (through the inclusion of sewer and/or

power on the same bill) is also inconclusive.

Similarly, there is no evidence that, controlling for

information on the bill, the use of increasing

block rates has a significant impact on price

elasticity. 29

Despite the low significance levels of all

billing and price structure dummies, inclusion of

Table 3. Estimations results

Dependent variable ln(Q)

Base model N¼349

Base modelþ N¼349

Info I N¼349

Info II n¼349

West excluded n¼254

W & SW excluded n¼226

IB excluded n¼252

ln(AP) �0.37*** �0.38*** �0.36*** �0.35*** �0.28*** �0.28*** �0.37*** (0.039) (0.053) (0.056) (0.055) (0.066) (0.071) (0.066)

ln(Income) 0.30*** 0.30*** 0.31*** 0.31*** 0.30*** 0.25*** 0.30*** (0.057) (0.060) (0.060) (0.060) (0.075) (0.084) (0.073)

HHsize 0.25*** 0.25*** 0.27*** 0.27*** 0.24*** 0.34*** 0.33*** (0.048) (0.049) (0.050) (0.050) (0.092) (0.11) (0.059)

ln(Density) �0.048*** �0.047*** �0.044*** �0.042*** �0.026 �0.019 �0.036* (0.016) (0.016) (0.016) (0.016) (0.020) (0.022) (0.020)

ln(AAP) �0.23*** �0.23*** �0.23*** �0.22*** �0.23*** �0.14 �0.19*** (0.035) (0.035) (0.036) (0.035) (0.063) (0.10) (0.044)

dt90 0.0015*** 0.0015*** 0.0014*** 0.0015*** 0.0023*** 0.0018* 0.0023*** (0.0005) (0.0005) (0.0005) (0.0005) (0.0007) (0.0010) (0.0007)

ln(AP)�mpinfo �0.15*** (0.055)

ln(AP)�priceinfo �0.16*** �0.21*** �0.22*** �0.14** (0.051) (0.060) (0.065) (0.066)

ln(AP)�QuantityInfo 0.045 0.043 �0.025 0.018 0.018 (0.049) (0.048) (0.060) (0.065) (0.060)

ln(AP)� �0.0014 �0.013 �0.009 0.0068 0.014 0.038 CombinationBilling (0.045) (0.045) (0.045) (0.052) (0.057) (0.054)

ln(AP)� 0.0041 0.0015 �0.002 �0.036 �0.039 0.013 HighFrequencyBill (0.044) (0.044) (0.044) (0.050) (0.054) (0.050)

ln(AP)�IB 0.048 0.057 0.064 0.028 0.055 (0.044) (0.046) (0.046) (0.060) (0.072)

Message 0.043 0.040 �0.054 �0.052 �0.023 (0.058) (0.058) (0.090) (0.10) (0.083)

Intercept 1.07* 1.13* 0.93 0.89 0.81 0.82 0.60 (0.59) (0.63) (0.63) (0.63) (0.76) (0.86) (0.77)

Adjusted R-squared 0.441 0.439 0.446 0.452 0.314 0.256 0.427 F-test 46.84 31.19 24.35 24.89 10.67 7.46 18.02 Hausman test �2 3.46 (w/instruments for AP) (�¼0.75)

Notes: *** statistically significant at the 1% level or better; ** significant at the 5% level; * significant at the 0% level. Standard errors are in parentheses.

28 For all models with information we test the joint significance of (1) all non-price variables interacted with ln(AP) with and

without IB, (2) history and otherinfo interacted with ln(AP) with and without IB, (3) power and sewer interacted with ln(AP), (4) all non-price information variables interacted with ln(AP) plus the message. We found no evidence of joint significance as F-test values ranged between 0.01 and 0.92 with probability values from 0.99 to 0.43. 29 Recall that utilities that report using and increasing block rate structure may have free allowances that would effectively

create a decreasing rate structure for lower consumption levels. Again, the structure of the AWWA data does not allow us to identify such nuances in rate structures.

Effect of price information on residential water demand 391

these variables was important to make sure that price information results were not capturing other features of the bill or utility.

30

The magnitude and significance of our results appear robust to sample selection and not solely driven by the fact that we included utilities from the West and Southwest – where attitudes toward water are likely to be different than in the rest of the country – or utilities that opted for increasing block prices that may have opted for the price structure to promote conservation. The regression without Western states reveals a larger increase in elasticity with price information (although the coefficients are not statistically different from each other). While total price elasticity is approximately the same (�0.49), elasticity for utilities without price information is only �0.28. Such numbers imply that the percentage price increase required to obtain any size reduction in quantity is 40% less than what it would need to be without the price information (for example, a 10% decrease in quantity requires a 20% increase in price instead of 35%). Excluding the Southwest in addition to the West reduces the magnitudes and significance levels of some of the coefficients, but results on price elasticity and information variables are virtually unchanged. Finally, excluding utilities across the USA that reported using increasing block rates slightly reduces the estimated effect of price information on price elasticity but not significantly so.

IV. Conclusions

While issues of price information have been suggested before as possible factor contributing to low price elasticity in water and electricity demand, the existing literature did not provide any quantitative analysis of the impact of price information on household’s sensitivity to price variation. This paper constitutes a first attempt at quantifying the effect of provid- ing price information on water bills. We started by asking the question whether the demand for residential water could be made more elastic by providing consumers with more informative water bills. In order to assess differences in billing informa- tion across the US, we surveyed all utilities with residential water sales included in the American

Water Works Association’s 1996 database. We found enough variation in the content of bills (whether they included price information, history of use, and/or conservation messages) to test our hypothesis. We merged the data thus collected with existing data sets to estimate parameters of a simple residential water demand model where price and quantity information on bills is allowed to affect price elasticity and where the presence of conservation messages may affect preferences independently of prices.

Our results provide evidence in support of the hypothesis that price information increases the price elasticity of demand but the inclusion of additional information aside from simple price information is not found to significantly affect water demand. The magnitude and statistical significance of the price information effect is large enough to merit notice from researchers and utility managers. All other factors held constant, a utility that gives marginal price information on the water bill can attain the same level of conservation with a 30 to 40% lower rate increase.

The significance of the price information variable should be taken into account when interpret- ing elasticity results from the literature. Elasticity estimates calculated using microeconomic data are conditional on billing formats. Results from cross sectional studies that do not take account of the existence of price information on the bill are likely to be biased downward for utilities that provide such information. Data limitation did not allow us to test the effect of billing information on price perception. For future research, data on billing information needs to be collected along with full price schedules for a cross section of utilities that use identical price structures. With this new data set, we can investigate whether Shin’s price perception parameter (Shin, 1985) decreases when consumers receive more infor- mative bills, providing a direct test of whether the price perception parameter is indeed a function of information costs.

31 Finally, this study constrains the

price information to have the same impact on price elasticity at all price levels. Further research using a more flexible functional form and possibly a large microeconomic data set that spans utilities with different billing practices could test whether people pay more or less attention to the information on the bill depending on the size of the bill.

30 At the suggestion of a referee, we ran the same model taking account of the possibility that price elasticity was not constant

for all income levels. We created dummy variables for four income levels around the mean. All results were largely statistically insignificant except for communities in the highest 10% of the income distribution for which the price elasticity was reduced by 0.12 (with a standard error of 0.07 and a p-value of 0.083). 31 In Shin’s article (1985) individuals respond to a perceived price P

� ¼MP� (AP/MP)

k , where MP is marginal price and

AP is average price. The price perception parameter k is between 0 and 1. A decrease in k indicates that individuals’ price perception is closer to marginal price than average price.

392 S. Gaudin

No one likes the prospect of water rationing and no one likes exorbitant price increases; policies to increase the price elasticity in residential water demand have the potential to reduce reliance on rationing by reducing the level of price increases necessary to absorb shortages. We hope the results of this research will encourage water utilities in the USA and other countries to use bills to their full potential by incorporating information that is most likely to make a difference for their community by increasing price sensitivity and foster water conserva- tion at low cost. More generally, our results provide evidence in support of the intuitive but often ignored fact that the predictions of economic theory (that people respond to price signals) are necessarily weaker when prices are imperfectly observed.

Acknowledgements

I wish to thank Adam Greeney for invaluable assistance in the data collection process and the Mellon Foundation for financial support. I also thank Barbara Craig, Hirschel Kasper, Kenneth Kuttner, Celine Nauges, Steven Renzetti, John Swinton, seminar participants at McMaster University and the Lawrence Berkeley National Laboratory, as well as two anonymous referees for comments on earlier drafts. All remaining errors are mine.

References

Arbués, F., Garcia-Valiñas, M. A. and Martinez- Espiñeira, R. (2003) Estimation of residential water demand: a state-of-the-art review, The Journal of Socio-Economics, 32, 81–102.

Baumann, D. A. and Boland, J. (1997) The case for managing urban water, in Urban Water Demand Management and Planning (Eds) D. Baumann, J. Boland and W. M. Hanemann, McGraw Hill, New York, pp. 1–30.

Egan, C., Kempton, W., Eide, A., Lord, D. and Payne, C. (1996) How consumers interpret and use comparative graphics of their energy use, in Proceedings of the 1996 ACEEE Summer Study on Energy Efficiency in Buildings 8, American Council for an Energy Efficient Economy, Washington, DC and Berkeley, CA, pp. 39–45.

Envision Utah (2002) Urban Planning Tools for Quality Growth, Ch. 5, p. 130. http://www.envisionutah.org/ Chapter_5.pdf

Espey, M., Espey, J. and Shaw, W. D. (1997) Price elasticity of residential demand for water: a meta-analysis, Water Resources Research, 33, 1369–74.

Fast, J. E. (1990) The effect of plain language billing procedures on residential energy consumption, in Enhancing Consumer Choice: Proceedings of the

Second International Conference on Research in the Consumer Interest (Ed.) R. N. Mayer, American Council of Consumer Interests, Columbia, MO.

Foster, H. S. and Beattie, B. R. (1979) Urban residential demand for water in the United States, Land Economics, 55, 43–58.

Foster, H. S. and Beattie, B. R. (1981) On the specification of price in studies of consumer demand under block price scheduling, Land Economics, 57, 624–9.

Gaudin, S., Griffin, R. and Sickles, R. (2001) Demand specification for municipal water management: evaluation of the Stone-Geary form, Land Economics, 77, 399–422.

Griffin, R. C. and Chang, C. (1990) Pretest analyses of water demand in thirty communities, Water Resources Research, 26, 2251–5.

Hanemann, W. M. (1997) Determinants of urban water use, in Urban Water Demand Management and Planning (Eds) D. Baumann, J. Boland and W. M. Hanemann, McGraw Hill, New York, pp. 1–75.

Hausman, J. A. (1978) Specification tests in econometrics, Econometrica, 46, 1583–8.

Kulshreshtha, S. N. (1996) Residential water demand in Saskatchewan communities: role played by block pricing system in water conservation, Canadian Water Resources Journal, 21, 139–55.

Matsukawa, I. (2004) The effects of information on residential demand for electricity, The Energy Journal, 25, 1–17.

Nauges, C. and Thomas, A. (2000) Privately operated utilities, municipal price negotiation, and the estima- tion of residential water demand: the case of France, Land Economics, 76, 68–86.

Martı́nez-Espiñeira, R. (2003) Estimating water demand under increasing-block tariffs using aggregate data and proportions of users per block, Environment and Resource Economics, 26, 5–23.

Martı́nez-Espiñeira, R. and Nauges, C. (2004) Is all domestic water consumption sensitive to price control?, Applied Economics, 36, 1697–703.

Nieswiadomy, M. L. and Molina, D. J. (1991) A note on price perception in water demand models, Land Economics, 67, 352–9.

Olmstead, S. M., Hanemann, W. M. and Stavins, R. N. (2003) Does price structure matter? Household water demand under increasing-block and uniform prices, mimeo, Presented at NBER 2003.

Renzetti, S. (2002) The Economics of Water Demands, Series in Natural Resource Management and Policy, Kluwer Academic Publishers, Boston, MA.

Shin, J. (1985) Perception of price when price information is costly: evidence from residential electricity demand, The Review of Economics and Statistics, 67, 591–8.

Stevens, T. H., Miller, J. and Willis, C. (1992) Effect of price structure on residential water demand, Water Resources Bulletin, 28, 681–5.

Taylor, R. G., McKean, J. R. and Young, R. A. (2004) Alternate price specifications for estimating residential water demand with fixed fees, Land Economics, 80, 463–75.

van Helden, G., Leeflang, P. and Sterken, E. (1987) Estimation of the demand for electricity, Applied Economics, 19, 69–82.

Williams, M. and Suh, B. (1986) The demand for urban water by customer class, Applied Economics, 18, 1275.

Effect of price information on residential water demand 393