The URL can be used to link to this page

Home My WebLink About5 - Discussion of "Residential Water Demand Management: Lessons from Aurora, Colorado", Journal of American Water Resources Association paper CITY OF BOULDER WATER RESOURCES ADVISORY BOARD AGENDA ITEM MEETING DATE: June 15, 2009 AGENDA TITLE: Discussion of "Residential Water Demand Management: Lessons from Aurora, Colorado", Journal of American Water Resources Association paper PRESENTERS: Dr. Charles Howe, University of Colorado & WRAB member Dr. Douglas S. Kenney, University of Colorado Carol Ellinghouse, Water Resources Coordinator Joanna Crean, Public Works Administrator EXECUTIVE SUMMARY: At the July 20, 2009 WRAB meeting, board members will continue to discuss the Drought Response Plan update. In preparation, at the June 15, 2009 WRAB meeting, Dr. Charles Howe and Dr. Douglas Kenney, University of Colorado professors, will present the lessons learned on residential water demand management from a "Drought in Aurora, Colorado" case study. Drs. Howe and Kenney will discuss factors affecting residential water demand during drought focusing on price, water restrictions, and rebate programs. The principle of price elasticity and the revenue implications of different controls will be discussed. Following the presentation, WRAB will have an opportunity to comment and provide thoughts on the study and conclusions. RESIDENTIAL WATER DEMAND - AURORA, COLORADO CASE STUDY: Dr. Kenney and his colleagues completed a case study on residential water demand management in Aurora, CO. Attachment A is the paper based on the study published in the Journal of American Water Resources Association (JAWRA). A primary study focus is on the effectiveness of demand management using water pricing. Price elasticity of water demand measures the sensitivity of water use relative to changes in the price of water. Water is considered to be an "inelastic" commodity because a specific percent change in price results in a smaller percent change in demand. For example, if elasticity is -0.6, a 10 percent price increase in water results in a 6 percent reduction in water demand and an increase in revenue even with reduced demand. Overall, the findings from the study demonstrate that residential water demand is greatly influenced by price, water use restrictions, and weather. The case study documents that water use reductions achievable through pricing and reductions from outdoor water restrictions are not additive. The study also demonstrates that the effectiveness of pricing and water restrictions policies varies among different types of users (high versus low volume users) and based on weather conditions (pre-drought versus drought). Lastly, the study illustrates that real-time information about water consumption assists customers in reaching full use of allocated water budgets resulting in an increase in water use. AGENDA ITEM # 5 PAGE 1 USING THE CASE STUDY RESULTS IN BOULDER'S DROUGHT PLAN: Boulder's water budget rate structure allows pricing to be used for water use reduction even without an increase in the base rate for water because a decrease in water budget size reduces the size of Blocks 1 and 2. This effectively increases the cost of water if a water customer maintains the previous level of use by moving a portion of water use from Blocks 1 and 2 into higher rate blocks. In response, the water customer would be expected to reduce use of that portion of previous use moved to a higher rate block that is no longer included in the water budget in Blocks 1 and 2. However, because of the inelasticity associated with water use, the customer would likely not stop use of the full amount of that portion of previous use for which he is now being charged more. Therefore, the water budget must be decreased by more than the desired overall water use reduction goal. For example, if a customer has previously used only his budget amount and if the water budget is reduced by 10 percent, Blocks 1 and 2 are reduced by 10% in size and 10 percent of the previous use will move into Block 3. The cost of Block 3 is 100% greater than Block 2, so if elasticity is -0.6, the customer might be expected to decrease use of the water at the margin in Block 3 by 60%. The customer's use would be 104% of his current budgeted amount, but would be 94% of his previous use, satisfying a use reduction goal of 6%. Boulder can also use pricing mechanisms to reduce water use by applying drought surcharges to standard block rates. Water use can also be reduced using mandatory water use restrictions when a more immediate response with greater participation by wealthier customers is required, including methods such as limits on days or times for irrigation use or prohibition of sidewalk washing. If surcharges alone are used to achieve the desired reduction, overall water revenues will actually increase because of inelastic demand. Mandatory restrictions have a doubly negative effect on overall water revenues, first by reducing the total amount of water used and sold and, secondly, by bumping some users into a lower rate block. Information on the impact and the inter-relationship of each of these tools for water demand management, as outlined in the Aurora, CO study is being modified as appropriate for Boulder and incorporated into water budget scenarios. Specifically, the Aurora study information is being used to help analyze options for how much: • water budgets need to be reduced, • drought surcharges need to be added, and • mandatory watering restrictions need to be implemented in order to reach the water use reduction goals established for each drought stage through water system modeling. As a result, staff is drafting several options for WRAB to review at the July meeting that illustrate the effects of various drought management tools. For instance, how much does the water budget, and the corresponding size of Blocks 1 and 2, for a single-family residential user need to be reduced for each drought stage if surcharges and restrictions are not used? The scenarios allow comparison of a water budget reduction only option to an option that uses all three tools to meet the water use reduction goals. It is anticipated that all options will be included in the Drought Response Plan to provide a range of information to future drought responders since each drought will have different characteristics. NEXT STEPS: At the July 20, 2009 WRAB meeting, staff will present scenarios/options to meet the water use reduction goals for each drought stage. Feedback from WRAB will be incorporated into a AGENDA ITEM # 5 PAGE 2 revised version of the Drought Response Plan (Volume 1). It is anticipated that a final version of the Drought Response Plan will be provided to WRAB in the fall of 2009. Volume 2 of the Drought Plan (technical information) is anticipated to be updated in 2010 following the completion of new water system model runs and studies as recommended in the Source Water Master Plan. Staff will present a draft of Volume 2 to WRAB when complete. ATTACHMENT: A - "Residential Water Demand Management: Lessons from Aurora, Colorado", Journal of American Water Resources Association paper AGENDA ITEM # 5 PAGE 3 Atttachment A JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION Vol. 44, No. 1 AMERICAN WATER RESOURCES ASSOCIATION February 2008 RESIDENTIAL WATER DEMAND MANAGEMENT: LESSONS FROM AURORA, COLORADO' Douglas S. Kenney, Christopher Goemans, Roberta Klein, Jessica Lowrey, and Kevin Reidy2 ABSTRACT: Residential water demand is a function of several factors, some of which are within the control of water utilities (e.g., price, water restrictions, rebate programs) and some of which are not (e.g., climate and weather, demographic characteristics). In this study of Aurora, Colorado, factors influencing residential water demand are reviewed during a turbulent drought period (2000-2005). Findings expand the understanding of res- idential demand in at least three salient ways: first, by documenting that pricing and outdoor water restriction policies interact with each other ensuring that total water savings are not additive of each program operating independently; second, by showing that the effectiveness of pricing and restrictions policies varies among differ- ent classes of customers (i.e., low, middle, and high volume water users) and between predrought and drought periods; and third, in demonstrating that real-time information about consumptive use (via the Water Smart Reader) helps customers reach water-use targets. (KEY TERMS: water conservation; drought; residential water demand; water pricing.) Kenney, Douglas S., Christopher Goemans, Roberta Klein, Jessica Lowrey, and Kevin Reidy, 2008. Residential Water Demand Management: Lessons from Aurora, Colorado. Journal of the American Water Resources Associa- tion (JAWRA) 44(1):192-207. DOI: 10.1111/j.1752-1688.2007.00147.x INTRODUCTION developing after core cities have already appropriated the most abundant and reliable local supplies. In these settings, the majority of water demands are A century ago, most western water issues focused typically for single-family homes; consequently, one on the pursuit of federally funded (and constructed) of the strongest management needs is to better projects serving agricultural water demands through understand and predict how these household increased storage and conveyance facilities. Today, demands are likely to respond both to management the landscape is dramatically different, as municipali- interventions (such as price increases and outdoor ties have emerged as the focal point of most water water use restrictions) and exogenous factors (such issues and decision-making, and as the scope of water as weather and demographic changes). This informa- management has come to focus on demands as well tion is particularly valuable in the context of drought as supplies. In many cases, this municipal focus is on planning and mitigation. suburbs rather than core cities, as the suburbs often In the following pages, a discussion of residential face the strongest growth pressures coupled with water demand in a rapidly growing western city is the least robust supply systems - a consequence of presented, focusing primarily on the levels of demand 'Paper No. J07-0012 of the Journal of the American Water Resources Association (JAWRA). Received January 24, 2007; accepted July 5, 2007. © 2008 American Water Resources Association. Discussions are open until August 1, 2008. ZRespectively, Deputy Director, Research Associates, and Professional Research Assistant (members of the NOAA-sponsored Western Water Assessment), University of Colorado, UCB 401, Boulder, Colorado 80309-0401; and Water Conservation Supervisor for Aurora Water, Aurora, Colorado 80012 (E-Mail/Kenney: douglas.kenney@colorado.edu). JAWRA 192 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION RESIDENTIAL WATER DEMAND MANAGEMENT: LESSONS FROM AURORA, COLORADO management effectiveness associated with water pric- premium on demand management. In this respect, ing, outdoor water-use restrictions, and technology Aurora is similar to cities across Colorado's Front rebate programs. This discussion begins with back- Range and much of the southwestern United States ground information about the case study setting, fol- (Nichols and Kenney, 2003). lowed by a review of the relevant literature, a In 2002, water officials along the Front Range were description of the data and methodology employed, confronted with one of the worst drought years on and a discussion of results and conclusions. A brief record (Pielke et al., 2005), threatening the adequacy appendix is also included to provide additional details of Aurora's water supply. In response, Aurora Water about the study methodology. implemented a variety of short and long-term demand management programs over the next few years. Programs included: drought restrictions (i.e., Case Study: Drought in Aurora, Colorado limits on outdoor water use), incentive programs, introductions of new technologies, and multiple The investigation of residential water demand changes in billing structures and rates, culminating featured in this paper focuses on the city of Aurora, in the adoption of an increasing block rate (IBR) pric- Colorado, a rapidly growing Denver suburb of approx- ing structure with individualized (household-specific) imately 309,000 residents served exclusively by a sin- block widths (i.e., the volume of water priced at a gle municipal provider: Aurora Water. Based on our given rate level) based on water budgets adjusted analysis of billing records provided by Aurora Water, annually in response to consumption levels, water approximately 70-80% of deliveries in the utility's storage conditions, and revenue considerations. A service area are to residential customers, with single- timeline of the key management interventions - i.e., family homes accounting for the bulk of these deliver- the pricing and water restrictions policies - is pro- ies. Stretching supplies to meet demands in Aurora vided in Figure 1. Collectively, these water demand has been a growing challenge for several decades, as efforts were highly successful, reducing total annual rapid population growth, combined with limited deliveries in 2002 and 2003 by 8 and 26%, respec- opportunities to expand supply, have placed a tively, relative to average deliveries in the 2000-2001 $ S 6.12 8.95 9.20 6.68 4.08 5.90 6.25 5.01 2.04 FC = 2.87 2.68 FC -3.30 2.03 1.34 FC ° 3.79 FC-379 TH Gal T H Gat TIT Gal TIT Gal FC - 2.69 FC = 2.87 FC = 2.87 FC - 3.30 PI -191 P1=2.04 PI 2.04 PI 2.34 511102 716102 1011102 j 1/1/03 513/03 111/04 5/1104 12/31/04 5115 to 8/31; 9/1 to 10/14; 10/15 to 4/30; No 5-1 to 6-30; 9-10 to 10-31; 5-1 to 10-31; Lawn circle/square/diamond Outdoor watering outdoor watering Lawn watering Lawn watering watering 2 days per watering restrictions limited to 3 days (except trees) 2 days per 2 days per week, no time limit per week week, one hour week, no time per day limit 7-1 to 9-9; Lawn watering Single Rate 2 days per I 1-holds face a fixed service cost (FC) plus a uniform per unit week, two hours charge (PI); all prices are shown per thousand gallons (TI I Gal) per day Inc. Block Rate In addition to FC, households face an increasing block rate structure for all units consumed. Block widths are standardired across all households Inc. Block Rate based In addition to FC, households face an increasing block rate structure on Water Budgets for all units consumed. Block widths are specific to each household, based on average daily indoor consumption (ADIC) and an irrigation allowance (IA). Households receive a varying percentage of their ADIC and IA in each block depending on drought conditions. * Block widths in diagrams not to scale Rate structure type relleets the rate structure utilized during summer months Source: City of Aurora: Water Management Plan (2002-2004) and ratesall.txt provided by the City of Aurora Utilities Department. FIGURE 1. '15meline of Pricing and Restriction Policies. JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 193 JAWRA KENNEY, GOEMANS, KLEIN, LOWREY, AND REIDY period (Aurora Management Plan, 2005). The vast utilities and those that are not (a convention utilized majority of these cutbacks came from the single-fam- by Gegax et al., 1998). Most of our emphasis, accord- ily home sector and occurred during the summer irri- ingly, is on the former category; nonetheless, consid- gation season. ering the full spectrum of influences on water Enthusiasm regarding the success of the demand demand is necessary for understanding and project- management program was tempered somewhat by ing demand, and for assessing opportunities for the inability to easily assess which of the simulta- demand management. neously employed tools were responsible for the observed declines, and subsequently, which reduc- tions could (and could not) be relied upon in the Factors Under Utility Control future. Answering these questions is necessary to improve both long-term and short-term planning. To Pricing and Rate Structures. A consistent point investigate these questions, Aurora Water in the fall of emphasis in the literature is the attempt to quan- of 2005 entered into an ongoing research partnership tify price elasticity of water demand - i.e., the eco- with the Western Water Assessment (a NOAA-funded nomic measure of how demand for water moves in effort based at the University of Colorado's Coopera- response to price changes. This is a question of great tive Institute for Research in Environmental Sci- practical importance, as pricing provides an obvious ences) to explore influences and recent trends in mechanism for water utilities to strategically manipu- residential water demand. The timing for this late customer behavior. The tremendous experimen- research is ideal, as the extreme nature of the recent tation recently with new rate and pricing structures drought, combined with the aggressiveness and com- has provided many opportunities for this research, plexity of Aurora Water's drought response, provide with dozens of studies confirming the intuitive notion an unusually broad spectrum of factors against which that raising prices does in fact reduce demand, albeit to track demand patterns. Aurora Water was able to only modestly (i.e., demand is largely price inelastic). provide a panel database of monthly consumption Estimates of the price elasticity of residential water records over the study period tracking water demand demand vary widely; one summary of this literature at a household-by-household scale, which allowed us by Brookshire et al. (2002) suggests a fairly typical to investigate the impacts of different demand man- value to be -0.5 (meaning that a 10% increase in agement programs enacted at different times and price nets a 5% decrease in consumption). evaluate the behavior of different types of house- Nested within this general conclusion regarding holds. In contrast, most similar water demand studies price elasticity is a variety of subtle, but practically rely on aggregated, citywide data (Hewitt and Hane- important, uncertainties and research questions. mann, 1995; Arbues et al., 2003). Recent exceptions Chief among these is the notion that many individu- to this include Pint (1999), Hewitt and Hanemann als lack a clear understanding of their rate structure (1995), Renwick and Archibald (1998), and Renwick and water bill, raising difficult research issues about and Green (2000). Collectively, these qualities provide which price signals customers actually respond to a largely unprecedented opportunity to explore sev- (e.g., see Billings and Agthe, 1980; Shin, 1985; Jor- eral facets of residential water demand. Results from dan, 1999). In the modern era, more and more cus- Phase 1 of research are presented herein; a Phase 2 tomers throughout the Southwest face an IBR is under development. structure which means that water gets progressively more expensive as their level of use moves them into and through pricing tiers designed to discourage excessive use (Western Resource Advocates, 2003). LITERATURE REVIEW The rationale of this approach is based on the notion that consumers respond to marginal prices (i.e., the cost of the last unit purchased); however, there is The literature on residential water demand has reason to think that this viewpoint is too simplistic, expanded significantly in recent years in terms of as customers not only often lack an understanding of scope and sophistication, as quantitative, regression- their rate structure, but rarely have anything resem- based studies have illuminated many relationships bling real-time information about their current level while simultaneously identifying several new of consumption (Foster and Beattie, 1979; Arbues research questions (e.g., see Olmstead et al., 2003 et al., 2003; Carter and Milon, 2005). A further com- and Gaudin, 2006). Given our focus in this study on plication is identified by Olmstead et al. (2003), who informing real-world demand management, our sum- provided evidence that the mere existence of an mary in this and subsequent sections explicitly dis- increasing block structure can reduce demand tinguishes between factors under the control of water irrespective of the change in price. Still additional JAWRA 194 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION RESIDENTIAL WATER DEMAND MANAGEMENT: LESSONS FROM AURORA, COLORADO complications associated with calculating and utiliz- Research into the effectiveness of outdoor water- ing price elasticities derive from the observation that ing restrictions generally focuses on the comparison price elasticity can vary significantly among seasons, of voluntary vs. mandatory programs. The literature uses, regions, and various social/economic conditions, is consistent in showing significant (sometimes 30% and can be influenced by the existence of other or more) savings from mandatory restrictions; find- demand management strategies (e.g., public educa- ings regarding voluntary restrictions are much more tion and water-use restrictions) (e.g., see Howe and variable, but with savings estimates generally lag- Linaweaver, 1967; Renwick and Green, 2000; Cava- ging far behind the mandatory programs (e.g., see nagh et al., 2002). A more sophisticated understand- Lee, 1981; Lee and Warren, 1981; Shaw and Maid- ing of these influences is key to translating a general ment, 1987, 1988; Renwick and Green, 2000; Kenney understanding of price elasticity into effective et al., 2004). demand management policies. Part of the challenge in assessing the impact of restrictions programs is that they are usually com- Nonprice Strategies. Due perhaps to political bined with other price and nonprice efforts. Few stud- opposition, equity concerns, and legal limitations, ies have included both types of policies in their water utilities are frequently reluctant to rely solely analysis (e.g., Michelsen et al., 1999; Renwick and on price to allocate scarce supplies of water. Thus, in Green, 2000), and even among those studies which conjunction with price policies, utilities often imple- include both sets of policies, two important factors ment a variety of nonprice programs designed to pro- are typically omitted. First, aggregate responsiveness duce both temporary (drought-motivated) and to restrictions will depend heavily on the distribution permanent reductions in quantity demanded. of users (Goemans, 2006). For example, cities with a The range of nonprice strategies for managing relatively small number of large water users are water demand can generally be grouped into three likely to experience less reductions in response to categories: public education, technological improve- restrictions than those with a large number of these ments, and water restrictions. Research into the types of consumers. Second, as noted by Howe and first category, public education programs, generally Goemans (2002), the response of households to show them to be modestly beneficial, especially in changes in price is likely to differ when restrictions the short-term (Michelsen et al., 1999; Syme et al., are in place. 2000). However, most water demand studies, includ- ing this one, offer little quantitative analysis on this variable as it remains a challenge to (1) separate Factors Beyond the Control of the Water Utility the effect of education programs from other pricing and nonprice programs, (2) make meaningful dis- Weather. In addition to the various price and tinctions between the nearly infinite variety of edu- nonprice tools that utilities can utilize to manage cational efforts, and (3) assess the long-term value demand are a host of independent factors known to of public education in promoting a conservation influence residential water demand. Chief among ethic. Research seems to suggest that a certain these is weather. It is well documented that weather "critical mass" of educational programs is necessary can impact short-term water demand decisions (par- to generate significant benefits, but that utilities ticularly for landscape irrigation), and for this reason, soon reach a point of declining returns as additional weather variables are typically controlled for in efforts are implemented thereafter (Michelsen et al., regression-based studies focused on price and non- 1999). price tools (e.g., see Gutzler and Nims, 2005). But Somewhat more attention has been given to beyond the intuitive conclusion that hot-dry weather understanding the effectiveness of technological generates higher demands than cool-wet conditions, changes, especially indoor retrofitting of water-using the exact nature of the weather/water demand rela- devices such as toilets, showerheads, and washing tionship has several areas of uncertainty. For exam- machines. Studies with this focus are frequently ple, researchers continue to search for the best based on engineering assumptions of expected reduc- combination of weather variables to explain consump- tions (Michelsen et al., 1999). One notable exception tion patterns, often finding precipitation to be the is provided by Renwick and Archibald (1998), whose most useful predictive variable, but also finding value empirical research of household water demand in in measures of temperature, ET (evapotranspiration), Santa Barbara and Goleta, California, found that and in some cases, indices designed to measure the installing low flow toilets reduced consumption by unmet water needs of landscape plantings (e.g., see 10% (per toilet), low flow showerheads by 8% (per fix- Maidment and Miaou, 1986; Woodard and Horn, ture), and adoption of water efficient irrigation tech- 1988; Rhoades and Walski, 1991; Gutzler and Nims, nologies by 11%. 2005). JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 195 JAWRA KENNEY, GOEMANS, KLEIN, LOWREY, AND REIDY Exactly how to consider these variables is a chal- availability of household level data for many vari- lenging question; for example, what is more impor- ables (namely price and consumption), the extreme tant: total precipitation over a month, the number of drought conditions that characterized the study per- precipitation events, or the time between events? iod, and the aggressiveness and diversity of the man- Questions of this nature are difficult to answer for a agement interventions. Although the city of Aurora variety of reasons, including issues of microclimate provided consumption records for every customer (i.e., weather conditions in one neighborhood may not over the period 1997-2005, our analysis focuses on a match another), the existence of major outdoor water subset of single family residential customers. Specifi- uses other than for irrigation (e.g., the use of evapo- cally, the results presented below correspond to the rative coolers), and distinguishing the impact of sample of households for which we had a complete, weather from the broad spectrum of pricing and non- uninterrupted billing history between 1997 and 2005. price management tools that are most frequently This timeframe was utilized because it allowed us to (and/or aggressively) employed during the hottest categorize each household based on its water use hab- and driest seasons. The literature does not identify a its during the relatively normal years 1997-1999, preferred method for modeling weather variables. leaving observations occurring from 2000 to 2005 for Furthermore, research is frequently constrained by inclusion in the regression analysis. As discussed the fact that household-level consumption data are later, this approach allowed us to examine important only available at a monthly scale while weather vari- differences in behavior among subsets of the study ables change daily. population as well as during predrought and drought time periods. Demographic Considerations. Data limitations After cleaning the data, we are left with roughly are a common impediment to assessing the impact of 680,000 unique billing period observations from over demographic characteristics on residential water 10,000 household accounts. It is this subset of the city demand. Researchers rarely have datasets that allow of Aurora population that constitutes our study popu- them to match household level consumption data lation, and is referred to as "All Households" in sub- with demographic data about the people and house sequent discussions. Two important points are worth associated with a residential water account. Nonethe- noting regarding this population. First, the consump- less, research to date is sufficient to suggest that tion patterns for those households included in the household water demand is influenced by heterogene- study were not significantly different from those ity associated with differences in wealth (income), excluded due to the absence of a complete record. Sec- family size and age distribution, and household pref- ond, and more important, the coefficient estimates erences towards water use and conservation (Hanke obtained using the study sample (Column 1, Table 3) and de Mare, 1982; Jones and Morris, 1984; Lyman, were not significantly different from those obtained 1992; Renwick and Green, 2000; Syme et al., 2000; when using household data for all city residents. Var- Cavanagh et al., 2002). Similarly, housing character- iable definitions and source information are provided istics useful in explaining residential water demand in Table 1. can include the type of dwelling (e.g., single family home vs. apartment), age of house, size of house/lot, Price, Pricing Structures, and Consumption. At and the water-using technologies featured (Billings the heart of the research database are monthly and Day, 1989; Lyman, 1992; Mayer et al., 1999; Ren- billing records from Aurora Water keyed by a cus- wick and Green, 2000; Cavanagh et al., 2002). Con- tomer number and customer location which allowed sidering these influences is difficult not only due to us to track individual behavior while still preserving the aforementioned lack of the relevant house- the anonymity of specific customers. Billing records hold/account level data, but also given that many fea- provide two critically important types of information: tures of a home (e.g., size) are likely to be correlated consumption levels and the pricing structures (i.e., with household features, particularly income. the delineation of tiers and their associated rates) associated with the observed levels of consumption. As shown earlier in Figure 1, these pricing structures have changed significantly in recent years. In sum- DATA AND METHODOLOGY mer of 2002, Aurora transitioned from a flat rate to an IBR pricing structure, with all households subject to the same rates and block widths (i.e., quantity of Data water sold at each price). Soon thereafter, Aurora began to refine its IBR structure by tailoring the size As noted earlier, the dataset compiled for this of each block width on a household by household investigation is unusually strong, in part due to the basis, an approach known as individual water JAWRA 196 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION RESIDENTIAL WATER DEMAND MANAGEMENT: LESSONS FROM AURORA, COLORADO TABLE 1. Variable Definitions. Variable Definition Units Source Consum Household consumption per billing period TH Gallons Aurora Factors Under Utility Control cpilagap CPI adjusted average price paid per thousand gallons during the previous 1999 Dollars Aurora bill period restrict Indicator variable, equal to one if restrictions where in place at some 0-1 Aurora point during the current bill period blprddays Length of current bill period Days Aurora outdoorrebate Indicator variable, equal to one if household participated in outdoor 0-1 Aurora rebate program indoorrebate Indicator variable, equal to one if household participated in indoor rebate 0-1 Aurora program wsr Indicator variable, equal to one if household purchased a water smart 0-1 Aurora reader Factors Outside of Utility Control Seasonal/Weather related Irrigation Indicator variable, equal to one if any portion of the bill period occurred 0-1 during the irrigation season (May-October) Holiday Indicator variable, equal to one if Christmas or Thanksgiving occurred 0-1 during some portion of the current bill period avemaxt Average daily maximum temperature over the course of the current bill Fahrenheit NOAA period totprecip Total precipitation over the course of the current bill period Inches NOAA Economic-Demographic (block-level) hhinc Median household income 1999 Dollars 2000 Census medage Median age of homeowner Years 2000 Census pph Median size of household Persons 2000 Census houseowned Percentage of homes owner occupied Percentage 2000 Census newhome Percentage of homes built after 1991 Percentage 2000 Census oldhome Percentage of homes built prior to 1970 Percentage 2000 Census numbedrooms Median number of bedrooms # of bedrooms 2000 Census budgets. This was initially done (in 2003) by focusing among other things, to identify the marginal price of only on the width of the first block, based largely on water in the next month given a particular level of the customer's historic average winter consumption. use. None were able to do this correctly. Since 2004, the size of each account's second block We followed common convention by lagging this has also been determined on a household by house- price variable a month; i.e., water use in a given hold basis. month is assumed to be influenced by the magnitude Over the course of the study period, nominal rates of the water bill in the preceding month. Average ranged from a low of $1.91 per thousand gallons price from the previous bill is used because this is (under the uniform rate structure in place prior to the only pricing information available to consumers 2002) to $9.20 (in the highest (third) block in 2004). when making their current month's water use deci- Thus, the effective marginal price for a consumer sions. The database also includes a variable for num- using a large volume of water has increased by more ber of billing days in each cycle. than $7 per thousand gallons (almost a factor of five), by far the largest swing we have observed in the Restrictions. The dataset also tracks periods fea- literature. turing drought-inspired restrictions on outdoor water In this analysis, we chose to use the average cost of use, primarily focusing on the frequency and duration water as the price signal in the statistical analysis, a of lawn watering. Aurora Water, like most Colorado conclusion reached after reviewing the extensive liter- utilities, has recently employed restrictions as part of ature on the subject (e.g., see Michelsen et al., 1999; efforts to curb summer water demand (Kenney et al., Gaudin, 2006), and after an informal experiment 2004). In Aurora, mandatory outdoor water-use among our university colleagues confirmed our suspi- restrictions of various degrees of severity were in cion that most customers likely have difficulty inter- place between May 15, 2002 and October 31, 2003, preting their bill and billing structure beyond the then again between May 1 and October 31, 2004 (see general conclusion that charges increase with usage. Figure 1). (Note in the following discussion of meth- In our experiment, we provided several colleagues odology and results that the interaction of restrictions with copies of sample Aurora water bills, asking them, and price is given particular attention in this study.) JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 197 JAWRA KENNEY, GOEMANS, KLEIN, LOWREY, AND REIDY Rebates and Water Smart Readers. Another reviewing daily (system-wide) water delivery records, unusual quality of our dataset is our ability to iden- it was also decided to utilize a "holiday" parameter to tify and track households that have taken part in account for the noticeable spikes observed in the city-sponsored rebate programs for water efficient daily water deliveries seen in the late November technologies. Our analysis focuses on three different (Thanksgiving) and late December (Christmas) billing classes of programs: (1) those for indoor appliances, periods. such as toilet retrofits; (2) those for outdoor technolo- gies, such as sprinkler system upgrades; and (3) the Demographic Data. The billing data are supple- Water Smart Reader (WSR) program. A WSR is an mented with a variety of household-level demo- in-home device (similar in appearance to a pager) graphic data which are potentially useful in exploring that intercepts radio signals from an individual's how water demand varies among different types of water meter, displaying real-time information about families and houses. The U.S. Census data are levels of water consumption. Use of a WSR allows reported at the block level, so average or median individuals to track their water usage in relationship neighborhood values were assigned to the correspond- to their monthly water budget. ing individual records. Data included are: median Rebates offered for water efficient indoor appli- household income (1999 dollars), median age of home- ances range from $100 for one low-flow toilet to $400 owner, median size of household, percentage of homes for one water-efficient washer and two dual-flush toi- owner-occupied, percentage of homes built after 1991, lets. Aurora also offered rebates of 50%; of total cost percentage of homes built prior to 1970, and median up to a maximum of $200 for irrigation efficiency number of bedrooms. As noted below and in Appendix upgrades. Aurora Water customers wanting a WSR 1, while our model of demand includes these demo- are assessed a charge of $30 (roughly half the cost of graphic factors, our choice of statistical technique providing the product). cannot utilize data that are static over the study period, so our presentation of demographic data is Weather and Climate. The research dataset limited to descriptive statistics. utilizes daily weather data from the National Oceanic and Atmospheric Administration to construct average maximum daily temperature and total precipitation Methodology over the course of each billing period. As noted before, this dataset is unusual in its variability, as Model of Demand. Our model of household-level the study period contains several years of drought, water demand is conceptually similar to those found particularly 2002 which has been estimated by some in previous studies that assume water demand is climate researchers as having a return period of primarily a function of price, weather, house and roughly 400 years for some parts of Colorado's Front household characteristics, and any other notable (and Range (the urban corridor running between the observed) policy interventions taken during the study Wyoming border to the north and Pueblo to the period (e.g., restrictions) (e.g., see Hewitt and Hane- south, along the eastern edge of the Rocky Moun- mann, 1995; Olmstead et al., 2003; Gaudin, 2006). tains) (Pielke et al., 2005). This is highly significant, Specifically, we assume that total demand for water as previous studies of residential water demand typi- by household i during billing period t is defined as tally use climate variables from relatively normal follows: periods to estimate responses in drought conditions; in contrast, we have the data necessary to measure this response directly. Additionally, some of those flo + # i 1ri (aveprice i,,_ I)+ studies that have had extreme conditions as part of f2(ln(avepriee,,,_1) *restriet,)+ the study period have been limited by not having individual household data (e.g., Renwick and Green, f33restrictt + f34blocllrate+ 2000; Kenney et al., 2004). f35111 (blprdd(ysi,,) + f6owdoorrebi,,+ The climate in the study region is also considered (37indoorr-eb,,,+fl8wsr-,,,+f39lrrigationt+ by coding all billing period observations based on flioHoliday, + fat iavernuxtr + #12totptcc"' 'rpr+ whether they occurred during the irrigation season, defined with respect to the start and end dates ~i h~(hJzinc,) + ~zrnedage; + ~2pph;+ at which most households are believed to begin 03hoiiseowrredi + 0411etivhoniei+ and end lawn watering (May-October). (Including 05 oldhoniei + 00w?ibedroonisi + a;t dummy variables for each individual month was also _ originally done, but was found not to offer any bene- fits beyond the irrigation season approach.) After 1 JAWRA 198 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION RESIDENTIAL WATER DEMAND MANAGEMENT: LESSONS FROM AURORA, COLORADO ELt represents unobserved factors that influence 40 demand. This term is composed of two parts: pit 35 reflects random unobserved influences, where the 30 mean of pi, is assumed to be zero; rh reflects differ- g ences between households which are unobserved from " zs the analyst's perspective (e.g., lot size, irrigation tech- ° 20 r nology, etc.). 15 In addition to the factors defined earlier (and to shown in Table 1), our demand model includes two additional terms. First, standard microeconomic the- s ory predicts that households will be less responsive to 0 gay ~~,e ~~y q~9G s Ooro tioL~ oacQ changes in price when constrained by restrictions tea,` ab 441, (Howe and Goemans, 2002). To account for this we a2 Ze2 a s~ e~be ~6, included a price-restriction interaction term (ln(ave- High Pre-Drought -Mod Pre-Drought Low Pre-Drought pricej,t_1)*restrictt), which explicitly accounts for any - W High Drought - - - Med Drought - Low Drought differences in responsiveness to price when restric- tions are in place. When restrictions are not in place, FIGURE 2. Average Consumption per Billing Period by User Type and Drought Condition. the term ln(avepricet,t_1)'*restrictt equals zero and the price elasticity of demand is equal to f31. Alterna- tively, when restrictions are in place, ln(avepricei, t_1) management programs, and to take full advantage of `restrict, does not equal zero. During these periods, the richness of the dataset, the study team chose to the price elasticity of demand is equal to f'1 + f32. fl2 expand our analysis of water demand in two additional from Equation (1) is a measure of the change in price ways. The first compares households with respect to elasticity when restrictions are in place. This their relative levels of water consumption; the second approach was originally suggested, albeit for different compares consumption during the predrought and reasons, by Moncur (1987) and later by Michelsen drought periods. Our rationale for doing so is largely et al. (1999); however, both of these studies omitted evident in Figure 2, which plots system-wide residen- this variable from their final analysis (due largely to tial water demand over the study period. a lack of variation in the dataset resulting in high The data presented in Figure 2 are organized in collinearity between the interaction term and other two ways which we find highly illuminating. First, it variables). To our knowledge, our study is the first to is disaggregated into predrought (2000-01-01 to 2002- test for the difference in price elasticity when restric- 04-30) and drought (2002-05-01 to 2005-04-30) peri- tions are in place. Second, we include a block rate ods to test the changing influence of price in these dummy variable (blockrate) to allow for the possibil- two periods. (Other variables, such as restrictions ity that, for reasons other than the direct price effect, and rebates, are impossible to analyze in this way as household consumption patterns differ under 11311 they did not exist in both time periods.) Second, we structures (Olmstead et al., 2003). divided our population into three groups based on Many studies of water demand utilize the regres- each household's average summer consumption sion technique known as ordinary least squares between 1997 and 1999, a period that experienced (OLS) to estimate demand for water. However, use of relatively normal summer weather conditions. House- OLS to estimate Equation (1) is likely to produce holds whose average summer use was in the bottom biased results due to both the likely endogeneity of 25% of all households are classified as "Low" volume ln(avepricei,t_1) and ln(avepricet,t_1) restriett, and users, while those in the highest 25% comprise the omission of unobserved individual effects (Arbues "High" volume users; the rest of the households are and Barberan, 2004). To account for this, we utilize designated as "Med" (medium). This was done so that fixed effects, instrumental variables (FE-IV) we could investigate how the influence of price, approach. This approach addresses both the potential restrictions, and price-restriction interactions varied endogeneity associated with price and the omission of among each of the three groups. In both cases, these unobserved individual effects, guaranteeing unbiased, subsets of data are analyzed using the same model of efficient parameter estimates. Appendix 1 contains a demand and statistical methodology that were more detailed discussion of this approach. applied to the full population, and are supplemental to that analysis. Comparison of Water Use Between Groups Select descriptive statistics associated with these and Across Time Periods. As part of our efforts to three subgroups are presented in Table 2. While it is generate findings that can be useful to managers worth noting that income is highly correlated with in the design and implementation of demand household type, our delineation of household types is JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 199 JAWRA KENNEY, GOEMANS, KLEIN, LOWREY, AND REIDY TABLE 2. Summary Statistics by Type of Household (averages). households into the three groups based on their pred- rought consumption levels represents our attempt to Household Type All identify those households that are more (or less) - Variable Households Low Middle High likely to be constrained by the outdoor water use restrictions. Factors Under Utility Control consum 10.25 4.90 9.34 14.80 cpilagap 2.20 2.19 2.20 2.22 Factors Outside of Utility Control Economic-Demographic (block-level) RESULTS AND DISCUSSION hhinc 54,874 50,680 53,967 58,928 medage 34.77 33.66 34.33 36.35 pph 2.85 2.81 2.87 2.82 The demand model used in this analysis performs houseowned 0.79 0.75 0.78 0.81 newhome 0.02 0.01 0.02 0.03 well, as evidenced by the fact that all but one coeffi- oldhome 0.32 0.36 0.34 0.22 cient exhibits the expected sign and is significant at numbedrooms 1.44 1.40 1.44 1.46 1%. Moreover, the adjusted r2-value of 0.40 is on the # of households 10143 1015 6594 2534 high end of the range presented in past studies that have utilized household level data (e.g., Hewitt and not done to identify how income differences influence Hanemann, 1995; Renwick and Archibald, 1998; Pint, responsiveness to utility policies, nor are we attempt- 1999). The results of the data analysis are presented ing to estimate different portions of a representative below in Tables 3, 4, and 5, which disaggregate consumer's demand curve. Rather, we wish to distin- between those factors that are (Tables 3 and 4) and guish among customers based on the quantity of are not (Table 5) under the control of the utility and water they use for outdoor purposes; high-volume thus subject to management intervention. water users are large outdoor water users. The mag- nitude of the effect that restrictions (and price changes made when restrictions are in place) have on Items Under Utility Control any given household is dependent on what its water use would have been absent the restrictions (i.e., its Table 3 provides results (including coefficient esti- unconstrained demand). Stated differently, when mates, z-test statistics, and significance levels for restrictions are in place, a representative consumer Equation (1) utilizing the FE-IV technique) for those does not exist. Rather, for every level of restriction items under utility control, namely price, restrictions, there are two different types of households: those rate structures, and rebates. impacted by the restrictions and those that are not Table 4 provides a summary of these findings as (i.e., households that either do not water outside or they relate to the influence of price and restrictions whose outdoor watering is already less than that on water demand. allowed by the outdoor watering restrictions). Unfortunately we cannot directly observe who is, and Influence of Price. Under the assumed log-log who is not, bound by the restrictions. Splitting relationship between consumption and price TABLE 3. Results for Utility Controlled Variables. By Type of Household Before vs. During Drought Dependent Variable: All In(consum) Households Low Middle High Before During Factors Under Utility Control In(epilagap) -0.60 (156.57)*** -0.34 (28.68)*** -0.57 (126.99)*** -0.75 (98.84)*** -0.56 (12.22)*** -1.11 (96.11)*** ln(epilagap) *restrict 0.23 (34.54)*** -0.11 (6.03)*** 0.19 (24.56)*** 0.51 (36.64)*** NA 0.85 (62.31)*** restrict -0.31 (57.9)*** 0.03 (1.84)* -0.28 (44.45)*** -0.57 (49.7)** NA -0.85 (-67.25)*** blockrate -0.05 (31.22)*** -0.01 (2.22)="'" -0.04 (23.55)"* -0.08 (25.64)**" NA -0.09 (49.38)*** ln(blprddays) 0.61 (114.8)*** 0.58 (34.69)*** 0.62 (97.38)*** 0.61 (57.02)*** 0.57 (95.09)*** 0.76 (73.26Y`** outdoorrebate 0.01 (0.69) -0.05 (1.27) 0.02(l.43) 0.03 (2.08)*** NA -0.11 (7.26Y"** indoorrebate -0.10 (15.54)*** -0.16 (5.7)*** -0.10 (12.54)*** -0.07 (6.65)*** NA -0.14 (15.93)*** wsr 0.16 (9.38)*** 0.17 (2.35)*** 0.15 (6.82)*** 0.13 (4.7)*** NA -0.25 (4.61)*** Number of observations 679,134 68,059 441,833 169,242 274,671 364,237 Number of households 10,143 1,015 6,594 2,534 10,143 10,143 Overall R2 0.4 0.18 0.45 0.59 0.42 0.36 Note: Absolute value of z statistics in parentheses. *Significant at 10%; **significant at 5%; significant at 1%. JAWRA 200 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION RESIDENTIAL WATER DEMAND MANAGEMENT: LESSONS FROM AURORA, COLORADO TABLE 4. Effectiveness of Price and Restrictions by Type of User. alter opportunities for price-based demand manage- % (see Martinez-Espineira, 2002; Goemans, 2006). Change Price in Demand Also having significant management implications Elasticity Due to is the comparison of predrought price elasticities Price During Restrictions (-0.56) to those during drought (-1.11). We are Elasticity Restrictions Only* unable to conclusively determine why customers were All -0.60 -0.37 -12.12 more than twice as sensitive to price during drought Low users -0.34 -0.46 -6.49 than before, but two possible explanations are worth Middle users -0.57 -0.39 -12.11 considering. First, these differences in elasticity may High users -0.75 -0.24 -13.82 derive, at least in part, from the wealth of media cov- erage and public education programs that accompany *Assuming average prices during periods with restrictions. drought (Moncur, 1987; Nieswiadomy, 1992). Second, these differences might indicate that the price elastic- presented in Equation (1), the coefficient on price, /i1, ity of demand is highly nonlinear outside of the range provides a direct estimate of price elasticity of of prices experienced prior to drought (Pint, 1999). As demand when restrictions are not in place. Consistent noted earlier, the magnitude of price increases with prior research we find price elasticity of demand observed over the study period was significant, with to be significant and inelastic (-0.60) throughout the the price per thousand gallons for water purchased in year. That is, given a 10% increase in price, demand the highest block increasing by more than $7 during can be expected to decrease by 6%. This result is well the drought. These are very different explanations within the range of past estimates: e.g., in 15 studies suggesting very different demand management reviewed by Brookshire et al. (2002), price elasticity approaches; thus, this result is a subject worthy of ranged from -0.11 to -1.59 (average of -0.49), while further research. One key element of that research Espey et al.'s (1997) review of 24 studies found 75% agenda is presented in the following section, and con- of price elasticity estimates fell between -0.02 and cerns the interaction of pricing policies with drought- -0.75. Note that while our estimates reflect the high inspired water restrictions. demands associated with Colorado's summer irriga- tion season, they represent a "year-round" estimate Restrictions and Price-Restrictions Interac- of price elasticity. This estimate likely would have tions. The coefficient on restrict, N, provides an esti- been higher had we confined our focus to the irriga- mate of the percentage change in demand, absent the tion season. influence of price, associated with imposing restric- The analysis by type of user confirms the hypothe- tions. In other words, it identifies how effective sis that price elasticities vary considerably among restrictions would be if the price of water were zero, user groups (perhaps explaining some of the range in which is shown in Table 3 as -0.31 (31% reduction). price elasticity estimates in previous studies), with While this conceptualization is certainly unrealistic, high water users generally more responsive to price it is theoretically useful when you consider that as (elasticity of -0.75) than low water users (-0.34). the price of water increases from zero, the effective- This observation can be important for planning pur- ness of restrictions will be reduced as more and more poses in many ways, such as in estimating how exist- users will find price, rather than restrictions, to be ing user populations are likely to respond to price the more significant controlling factor on their interventions, and also in assessing how long-term water-using behavior. It is impractical, therefore, to changes in demographics and housing/land-use may think about the effectiveness of restrictions without TABLE 5. Results for Variables Outside Utility Control. By Type of Household Before vs. During Drought Dependent Variable: All In(consum) Households Low Middle High Before During Factors Outside of Utility Control Irrigation 0.30 (133.19)*** 0.15 (21.19)*** 0.29 (108.02)*** 0.38 (86.33)*** 0.33 (88.03)*** 0.30 (102.22)*** Holiday 0.07 (39.66)*** 0.08 (13.96)*** 0.08 (34.81)*** 0.06 (17.33)*** 0.08 (30.02)*** 0.06 (24.22)*** avemaxt 0.02 (341.39)="** 0.01 (58.77)*** 0.02 (278.47)*** 0.03 (216.09)*** 0.02 (223.35)*** 0.02 (222.2)*** totprecip -0.04 (67.07)*1°r -0.03 (15.34pk:r* -0.04 (57.66)*** -0.04 (35.41)*1"' -0.03 (28.1)*** -0.05 (72.65)="'** constant -1.18 (63.31)*** -1.18 (20.09)*** -1.20 (54.33)*** -1.11 (29.69)*" -1.02 (30.01)*** -1.11 (30.04)*** Note: Absolute value of z statistics in parentheses. *Significant at 10%; **significant at 5%; significant at 1%. JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 201 JAWRA KENNEY, GOEMANS, KLEIN, LOWREY, AND REIDY explicitly considering their relationship to price, differently, if the goal of demand management is to which we have primarily done herein with the control the high users, pricing policies may provide price-restrictions interaction term (discussed below). the best long-term option whereas restrictions may Nonetheless, it is worth noting that if we use the provide the most logical drought-coping strategy. average price conditions observed when restrictions were in place in our model of water demand, the Rate Structures. Also of note is that the coeffi- water savings that can be attributed solely to restric- cient on blockrate is significant and negative (-0.05), tions would be estimated at roughly 12% (which is indicating that when faced with an IBR pricing struc- generally consistent with other studies considering ture, households consumed 5 percent less than they the relatively moderate restrictions utilized in Aur- would have under a uniform rate pricing structure. ora). We caution against applying this result in other This is consistent with previous research (Olmstead settings, as the effectiveness of restrictions is closely et al., 2003), and supports the common argument linked to case-specific factors including price, the dis- that, in addition to price levels, rate structures them- tribution of customer types (also discussed below), selves can be valuable in promoting conservation weather conditions, and customer familiarity with (e.g., see Western Resource Advocates, 2003). One (and support for) the restrictions. argument for why this might be the case is that Rather than considering how prices influence the although households do not often have detailed effectiveness of restrictions, it is perhaps more useful knowledge of the rate structure, they are generally to consider the problem in reverse: how does the aware that excessive consumption will result in adoption of restrictions modify the influence of price excess costs. This awareness causes them to consume on demand (as measured by changes in price elastic- less in an attempt to limit this possibility. ity)? Consistent with economic theory, the interac- tions term in our model of water demand is positive Rebates and Water Smart Readers. Indoor and and significant (+0.23), meaning that as restrictions outdoor rebate programs and the use of WSR are are implemented, consumers are less responsive to admittedly a diverse category, but are grouped price. Again, the explanation is clear: for any given together for discussion as their datasets share two customer, either price or restrictions (but not both) similar limitations. First, participating individuals will be controlling, depending on which provides the self-selected themselves for the particular programs; lowest (i.e., first-encountered) threshold. Summing thus, while we can track how participation influenced the price elasticity (-0.60) with the interactions term water demand among these individuals, it is problem- (+0.23) yields an effective price elasticity of demand atic to assume that a similar response would occur during restrictions periods of -0.37. among all members of the population. Second, while The policy ramifications of this observation are the indoor rebate programs (e.g., toilet rebates) are particularly evident by looking at the results for each designed to cover retrofitting activities, the outdoor user group, which show the adjusted price elasticity programs covering the installation of more efficient during restrictions to range from -0.24 for high users sprinkler technologies likely cover a mix of both ret- to -0.46 for low users. Managers wishing to reign in rofits and new construction, perhaps including signifi- the high users during drought, therefore, may be cant system expansions. As we have no data on these wise to focus on restrictions; whereas low water users other activities, assessing the effectiveness of the out- are perhaps better targeted (if at all) with price modi- door rebate programs is difficult. In fact, the coeffi- fications - although these users, by definition, have cient calculated for the outdoor programs, #6, is less opportunity to reduce consumption than others, statistically insignificant (and slightly positive), and and these price increases may therefore be more thus is not discussed further in our analysis. punitive than pragmatic. In any case, it is important The coefficient calculated for the indoor rebate pro- to appreciate that the theoretical savings from pric- grams, k is significant, large, and shows the ing policies and drought restrictions are not additive, expected negative sign (-0.10), suggesting that, all the impact of each policy can vary significantly else constant, participation in the indoor rebate pro- among user groups, and the choice of policy has rami- gram reduces household demand by approximately fications that go beyond water savings to include 10%. This finding is nearly identical in magnitude to issues of equity and revenue generation. Similarly, it those reported in other investigations, particularly is important to note that price elasticities among the Renwick and Green (2000) and Renwick and Archi- three groups go in opposite directions depending on bald (1998), and provides further empirical justifica- whether drought-inspired restrictions are in place, tion for using indoor rebate programs as a demand suggesting that the appropriate tool for drought man- management tool. agement is not necessarily the appropriate tool for The calculated coefficient for the Water Smart long-term (baseline) water conservation. Stated Reader, wsr, is also highly significant (+0.16), JAWRA 202 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION RESIDENTIAL WATER DEMAND MANAGEMENT: LESSONS FROM AURORA, COLORADO although the positive sign of the result was initially Seasonality of Water Demand. As is intuitively confusing. As noted above, WSR households self obvious, demand for water is shown to be highly sea- selected into the rebate program. One can control for sonal and dependent on climate and weather condi- the potential endogeneity of the WSR variable using tions. Water use in the irrigation season is methods similar to those utilized for average price. fundamentally and significantly higher than the rest However, the discrete nature of the WSR variable of the year (as shown earlier in Figure 2), a fact that complicates the use of these techniques, both compu- makes demand management in summer a particular tationally and technically. Thus, while we have omit- point of management emphasis. The coefficient on ted this type of analysis from this paper, we have irrigation is significant and positive (+0.30), indicat- applied these techniques to a random subset of Aur- ing that, irrespective of the influence of temperature ora households in a subsequent paper (in production) and precipitation, household water use increases by in which the effect of WSR ownership is the focus. 30% just by virtue of being in the irrigation season. There we estimate a FE-IV model where, in the first As expected, this effect is most pronounced among stage, we instrument for both price and WSR using a high-volume users (+0.38). Also as expected, the coef- fixed effects logit model to estimate the probability of ficient on holiday is significant and positive (+0.07). owning a WSR. The fact that a random subset of Although this effect is clearly outside the scope of households was sent an additional advertisement for management, including this factor in models of water the WSR is used as an instrument for WSR owner- demand is worthwhile in improving the accuracy of ship. Preliminary results from that work support the all estimated variables. finding presented here: namely that WSR ownership had a positive effect on demand for water. Weather. Also intuitive is the observation that, Conventional wisdom is that providing customers all else being equal, demand for water increases as with real-time information about water use increases temperatures rise, and decreases as precipitation their ability to track consumption and charges, and increases. Specifically, the model predicts that for thus should help convey the deterrent effect on exces- every one degree Fahrenheit increase in average sive use provided by the IBR structure. Why, then, daily maximum temperature over the course of the did the water consumption of our population of WSR billing period, water use increases about 2%. Simi- customers increase by 16%? The answer, we believe, larly, for every inch of precipitation, water use lies in the observation that although total use went decreases by roughly 4%. Understanding this rela- up among this group, the frequency with which these tionship awaits additional research on household- users entered into the most punitive pricing tier (the level decision-making (particularly associated with third block) diminished. It appears that prior to lawn watering) and the types of irrigation technolo- obtaining a WSR, users fearful of entering the third gies employed. (These questions are central to the block would err on the side of caution by consuming emerging Phase 2 of research). less than they would have otherwise preferred, but Findings that relate climate and weather condi- when armed with the ability to track consumption, tions to residential demand can be useful in several these same users skillfully budgeted consumption to facets of planning and management, especially in take full advantage of the lower priced blocks. The light of research suggesting that climate change will result is more extensive use of water in Blocks 1 and likely mean fundamental changes in average temper- 2 (and thus higher net consumption), and less con- atures (clearly increasing), precipitation, and the fre- sumption in Block 3. This observation should be quency of extreme events such as droughts and floods heartening to water managers, as it suggests that (Wagner, 2003). Considering climate change issues is informed consumers will adjust their behavior in particularly challenging for water managers along accordance with the water budget provided by the Colorado's Front Range, where water source and utility, adjusting use to fully utilize their apportion- demand areas are often separated by great distances ment in the low priced blocks (or tiers) that presum- and elevations. But regardless of what climatic ably reflect some notion of reasonableness while changes are in store for Aurora and other Front avoiding those blocks associated with excessive use. Range cities, a growing reliance on demand manage- ment to cope with extreme conditions and stresses (including those associated with population growth) Items Outside of Utility Control only underscores the need to understand all facets of residential water demand. Table 5 provides results for those influences on water demand that are beyond the control of water Demographic Considerations. Table 5 does managers, namely the seasonality of water demand not provide any statistics regarding the influence and weather. of household and house characteristics (i.e., JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 203 JAWRA KENNEY, GOEMANS, KLEIN, LOWREY, AND REIDY demographic considerations) on residential water that total water savings are not additive of each demand, a consequence (as noted earlier) of our program operating independently; second, by showing method of data analysis that relies on fixed effects. that the effectiveness of pricing and restrictions poli- Fixed effects models estimate household demand for cies varies among different classes of customers (i.e., water in each period as deviations from the house- low, middle, and high volume water users) and hold's average use over the period of record. This between predrought and drought periods; and third, approach effectively "averages-out" time-invariant in demonstrating that real-time information about unobserved effects such as i1i, allowing the researcher consumptive use (via the WSR) helps customers to obtain unbiased parameter estimates for the reach water-use targets. At each point in the analy- remaining variables (i.e., the One obvious down- sis, we have identified relevant management implica- side to this approach is that we are unable to recover tions of these findings. parameter estimates for any time-invariant variables To the extent that future water demand research (i.e., the (ps). is pursued with the aim of further informing and As discussed in Appendix 1, this is a small price to empowering water managers to better predict and pay for insuring that we produce unbiased parameter manipulate residential water demand, investigators estimates for the remaining variables. The central will need to make additional progress illuminating objective of this paper is to develop a better under- the interplay among the many factors now known to standing of how utility policies introduced by Aurora influence demand. This suggests a need to better influenced demand for water. Changes in income, age understand water-use decision-making processes at of homeowner, etc. are outside of the control of utili- the household level, which in turn will necessitate ties. Although interesting, it is of greater importance the assembly of improved datasets. This seems partic- to us to be sure that we have controlled for these ularly important as water utilities (like Aurora variables, something the fixed effects approach offers. Water) adopt dynamic, customer-specific water bud- Note that we are not omitting the effect of income gets, with the competing aims of managing water from the analysis; rather, we are controlling for it demand (and water revenues) in both normal and along with numerous unobserved individual charac- emergency settings, all within a framework that cus- teristics. tomers can readily understand and endorse as equita- Nonetheless, we would be remiss if we did not reit- ble. To simultaneously achieve these goals is a erate that some literature already exists to document formidable challenge, and is deserving of the same demographic effects, and similarly, if we failed to level of intellectual effort as has traditionally acknowledge that our division of customers into three been devoted to understanding and managing water user groups suggests that high-volume water users supplies. tend to be wealthier, older, and live in newer and lar- ger homes than other customers (see Table 2). We believe that a better understanding of demographic ACKNOWLEDGMENTS factors may be useful in designing and targeting demand management programs and in projecting This work was supported by the Western Water Assessment, a future demand patterns as cities age and evolve. NOAA sponsored program located at the University of Colorado dedicated to helping water managers better anticipate and respond to challenges associated with climate variability and change. Also essential to this research was the active and full participation of Aurora Water, a leader in municipal water demand management. CONCLUDING THOUGHTS Overall, our findings are consistent with the litera- LITERATURE CITED ture in demonstrating that residential water demand Arbues, F. and R. Barberan, 2004. Price Impact of Urban Residen- is largely a function of price, the impact of nonprice tial Water Demand: A Dynamic Panel Data Approach. Water demand management programs, and weather and cli- Resources Research 40, Article No. W11402. mate. Our unique contributions derive from the depth Arbues, F., M.A. Garcia-Valinas, and R. Martinez-Espineira, 2003. of the household-level dataset, the presence of the Estimation of Residential Water Demand: A State-of-the-art Review. Journal of Socio-Economics 32:81-102. extreme drought event in the study period, and the Aurora Water Department, 2005. Aurora Water 2005 Management diversity of associated management interventions. Plan. http://www.aurorawater.org, accessed June 1, 2005. Substantively, this study increases the knowledge of Baltagi, B.H., 2005. Econometric Analysis of Panel Data (Third residential demand in at least three salient ways: edition). J. Wiley & Sons, West Sussex, England. Billings, R.B. and D.E. Agthe, 1980. Price Elasticities for Water: first, by documenting that pricing and outdoor water A Case of Increasing Block Rates. Land Economics 56(1):73-84. restriction policies interact with each other ensuring JAWRA 204 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION RESIDENTIAL WATER DEMAND MANAGEMENT: LESSONS FROM AURORA, COLORADO Billings, R.B. and W.M. Day, 1989. Demand Management Factors Maidment, D.R. and S.P. Miaou, 1986. Daily Water use in Nine in Residential Water Use: The Southern Arizona Experience. Cities. Water Resources Research 22(6):845-885. Journal of the American Water Works Association 81(3): Mayer, P.W., W.B. D'Oreo, E.M. Opitz, J.C. Kiefer, W.Y. Davis, 58-64. B. Dziegielewski, and J.O. Nelson, 1999. Residential End Uses Brookshire, D.S., H.S. Burness, J.M. Chermak, and K. Krause, of Water. AWWA Research Foundation, Denver, Colorado. 2002. Western Urban Water Demand. Natural Resources Jour- Michelsen, A.M., J.T. McGuckin, and D. Stumpf, 1999. Nonprice nal 2(4):873-898. Water Conservation Programs as a Demand Management Tool. Carter, D.W. and J.W. Milon, 2005. Price Knowledge in Household Journal of the American Water Resources Association 35(3):593- Demand for Utility Services. Land Economics 81(2):265-283. 602. Cavanagh, S.M., W.M. Hanemann, and R.N. Stavins, 2002. Muffled Moncur, J.E.T., 1987. Urban Water Pricing and Drought Manage- Price Signals: Household Water Demand Under Increasing- ment. Water Resources Research 23(3):393-398. Block Prices. FEEM Working Paper No. 40.2002, Milan, Italy. Nichols, P.D. and D.S. Kenney, 2003. Watering Growth in Colo- Espey, M., J. Espey, and W.D. Shaw, 1997. Price Elasticity of Resi- rado: Swept Along by the Current or Choosing a Better Line? dential Demand for Water: A Meta-Analysis. Water Resources University of Denver Law Review 6(2):411-452. Research 33(6):1369-1374. Nieswiadomy, M.L., 1992. Estimating Urban Residential Water Foster, H.S. and B.R. Beattie, 1979. Urban Residential Demand for Demand: Effects of Price Structure, Conservation, and Educa- Water in the United States. Land Economics 55(1):43-58. tion. Water Resources Research 28(3):609-615. Gaudin, S., 2006. Effect of Price Information on Residential Water Nieswiadomy, M.L. and D.J. Molina, 1988. Urban Water Demand Demand. Applied Economics 38:383-393. Under Increasing Block Rates. Growth and Change 19(1):1-12. Gaudin, S., R.C. Griffin, and R.C. Sickles, 2001. Demand Specifica- Olmstead, S.W., M. Hanemann, and R.N. Stavins, 2003. Does tion for Municipal Water Management: Evaluation of the Stone- Price Structure Matter? Household Water Demand Under Geary Form. Land Economics 77(3):399-422. Increasing-Block and Uniform Prices. NBER Research Paper, Gegax, D., T. McGuckin, and A. Michelsen, 1998. Effectiveness of March, Cambridge, Massachusetts. Conservation Policies on New Mexico Residential Water Pielke, Sr., R.A., N. Doesken, O. Bliss, T. Green, C. Chaffin, J.D. Demand. New Mexico Journal of Science 38:104-126. Salas, C.A. Woodhouse, J.J. Lukas, and K. Wolter, 2005. Goemans, C., 2006. Optimal Policy Instruments for Utilities Using Drought 2002 in Colorado: An Unprecedented Drought or a Increasing Block Rate Policy Pricing Structures, WWA Work- Routine Drought? Pure and Applied Geophysics 162:1455-1479. ing Paper 102006, http://wwa.colorado.edu/about/homepages/ Pint, E.M., 1999. Household Responses to Increased Water goemans/IBRO71006.pdf, accessed August 1, 2006. Rates During the California Drought. Land Economics 75(2): Gutzler, D.S. and J.S. Nims, 2005. Interannual Variability of Water 246-266. Demand and Summer Climate in Albuquerque, New Mexico. Renwick, M.E. and S.O. Archibald, 1998. Demand Side Manage- Journal of Applied Meteorology 44:1777-1787. ment Policies for Residential Water Users: Who Bears the Con- Hanke, S.H. and L. de Mare, 1982. Residential Water Demand: A servation Burden? Land Economics 74(3):343-359. Pooled Time-Series Cross-Section Study of Malmo, Sweden. Renwick, M.E. and R.D. Green, 2000. Do Residential Water Water Resources Bulletin 18(4):621-625. Demand Side Management Policies Measure Up? An Analysis of Hewitt, J.A. and W.M. Hanemann, 1995. A Discrete/Continuous Eight California Water Agencies Journal of Environmental Eco- Choice Approach to Residential Water Demand Under Block nomics and Management 40:37-55. Rate Pricing. Land Economics 71(2):173-192. Rhoades, S.D. and T.M. Walski, 1991. Using Regression Analysis to Howe, C.W. and C. Goemans, 2002. Effectiveness of Water Rate Project Pumpage. Journal of the American Water Works Associ- Increases Following Watering Restrictions. Journal of the Amer- ation 83(12):45-50. ican Water Works Association 94(10):28-32. Shaw, D.T. and D.R. Maidment, 1987. Intervention Analysis of Howe, C.W. and F.P. Linaweaver, 1967. The Impact of Price on Water Use Restrictions, Austin, Texas. Water Resources Bulle- Residential Water Demand and its Relationship to System tin 23(6):1037-1046. Design and Price Structure. Water Resources Research Shaw, D.T. and D.R. Maidment, 1988. Effects of Conservation on 3(1):13-32. Daily Water Use. Journal of the American Water Works Associ- Hsiao, C., 2003. Analysis of Panel Data (Second edition). Cam- ation 80(9):71-77. bridge University Press, Cambridge, England. Shin, J., 1985. Perception of Price When Price Information is Johnston, J., and J. DiNardo, 1997. Econometric Methods (Fourth Costly: Evidence From Residential Electricity Demand. The edition). McGraw-Hill, New York, New York. Review of Economics and Statistics 67(4):591-598. Jones, C.V. and J.R. Morris, 1984. Instrumental Price Estimates Syme, G.J., B.E. Nancarrow, and C. Seligman, 2000. The Evalua- and Residential Water Demand. Water Resources Research tion of Information Campaigns to Promote Voluntary Household 20(2):197-202. Water Conservation. Evaluation Review 24(6):539-578. Jordan, J.L., 1999. Pricing to Encourage Conservation: Which Wagner, F.H., Ed, 2003. Preparing for a Changing Climate: The Price? Which Rate Structure?, Water Resources Update, Illinois Potential Consequences of Climate Variability and Change. A District Newsletter 114:34-37. Report of the Rocky Mountain/Great Basin Regional Assess- Kenney, D.S., R.A. Mein, and M.P. Clark, 2004. Use and Effective- ment Team for the U.S. Global Change Research Program, ness of Municipal Water Restrictions During Drought in Colo- Washington, D.C. rado. Journal of the American Water Resources Association Western Resource Advocates, 2003. SMART Water: A Comparative 40(1):77-87. Study of Urban Water Use Efficiency Across the Southwest. Lee, M.Y., 1981. Mandatory or Voluntary Water Conservation: A December. http://www.westernresourceadvocates.org/media/pdF Case Study of Iowa Communities During Drought. Journal of SWForewordAcknowledgements.pdf, accessed January 15, 2004 Soil and Water Conservation 36(4):231-234. Woodard, G.C. and C. Horn, 1988. Effects of Weather and Climate Lee, M.Y. and R.D. Warren, 1981. Use of a Predictive Model in on Municipal Water Demand in Arizona. Report prepared for Evaluating Water Consumption Conservation. Water Resources Arizona Department of Water Resources and Tucson Water, Bulletin 17(6):948-955. Tucson, Arizona. Lyman, R.A., 1992. Peak and Off-Peak Residential Water Demand. Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Water Resources Research 28(9):2159-2167. Panel Data. MIT Press, Cambridge, Massachusetts. JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 205 JAWRA KENNEY, GOEMANS, KLEIN, LOWREY, AND REIDY APPENDIX 1: METHODOLOGICAL meaningless parameter estimates that are inconsis- CONSIDERATIONS tent and/or inefficient (Hsiao, 2003). Luckily, tech- niques to account for i1i in the panel setting are well developed. Under the assumption that i1i is uncorre- No model includes every relevant factor. Hence our lated with the right hand side regressors, random inclusion of an error term (eit), which is intended to effects models may be used to generate parameter capture the net influence of everything omitted from estimates that are consistent and efficient (note that the model. Under the assumption that Fit is indepen- this is not the same as saying that i1i is random). dently and identically distributed (across both time Alternatively, if t1i is correlated with any of the right and individuals) with mean zero and constant vari- hand side regressors FE models must be used ante, Ordinary Least Squares (OLS) is the best linear (Johnston and DiNardo, 1997; Wooldridge, 2002; unbiased estimator (Johnston and DiNardo, 1997). Hsiao, 2003; Baltagi, 2005). In short, which model is However, as defined in Equation (1), Fit is not likely appropriate depends on the assumptions one is to satisfy this assumption. Potential problems associ- willing to make about iyi. ated with the inclusion of lagged average price and We choose to address i1i within the FE framework. the omission of unobserved individual effects will FE models estimate household demand for water in cause OLS to produce parameters estimates that are each period as deviations from the household's aver- biased and/or inefficient. Therefore, we utilize a age use over the period of record. This approach effec- mixed fixed effects (FE), instrumental variables (IV) tively "averages-out" time-invariant unobserved approach to account for these likely problems. In effects such as rli, allowing the researcher to obtain doing this we address both the potential endogeneity unbiased parameter estimates for the remaining vari- associated with price and the omission of unobserved ables. Thus, we can recover parameter estimates for individual effects, resulting in unbiased, efficient those variables which change over time (i.e., the f') parameter estimates (Wooldridge, 2002). by comparing individual households with themselves We address the need for and appropriateness of over time. The downside to this approach is that we this approach in subsequent subsections within this are unable to obtain parameter estimates for any appendix. We first discuss problems associated with variables that remain constant across time, even if omitting unobserved individual effects and the moti- they vary across households. As a result, our demo- vation behind modeling these unobserved effects as graphic terms drop out of the analysis and we are FE. This is followed by a discussion of the problems unable to estimate the parameters for these variables associated with demand estimation under block rate (i.e., the (p's). However, as noted earlier, this is a pricing and the motivation behind instrumenting for small price to pay and in some ways advantageous price. given our reliance on Census data for these variables. It is unclear the extent to which census blocks are homogenous; as such, use of Census data might intro- Unobserved Individual Effects duce additional error or bias into the model depend- ing on the distribution of characteristics across any Our dataset undoubtedly omits numerous factors given block. unique to each household that are relevant to deter- It is tempting to utilize random effects so that one mining water demand. Likely omissions include data can recover parameter estimates for time invariant about the presence/absence of evaporative coolers variables (e.g., Gaudin et al., 2001). However, use of (Hewitt and Hanemann, 1995), and data about irriga- random effects models would only be appropriate if i1i ble acreage and the type of sprinkler systems was uncorrelated with all of the regressors in Equa- employed in their maintenance. In comparison to tion (1). We believe that this is highly unlikely, in issues dealing with price, accounting for the potential which case modeling ni as random effects would presence of unobserved individual effects has received result in inconsistent parameter estimates. Omitted less attention within the water demand literature. variables such as lot size, type of irrigation system, Consistent with standard panel data models devel- and type of cooling system are likely correlated with oped in Johnston and DiNardo (1997), Baltagi (2005), income, age of home, and the other right hand side Hsiao (2003) and Wooldridge (2002), these omissions variables include in Equation (1). A Hausman-Wu can be represented in Equation (1) by decomposing Fit test, comparing random and FE estimates (after in- into two parts so that eit = ni + pit, where )1i corre- strumenting for price) on all households and for each sponds to the unobserved individual effects men- of the subgroups supported this belief. This is not tioned above and Ftit is the standard error term surprising. Referring to the preference of many econ- satisfying the typical assumptions corresponding to omists for FE estimates, Johnston and DiNardo the classic linear model. Ignoring qi can lead to (1997: 403) note: "This preference seems to be a JAWRA 206 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION RESIDENTIAL WATER DEMAND MANAGEMENT: LESSONS FROM AURORA, COLORADO consequence of the reasonable belief that, apart from purely experimental or quasi-experimental situations, it is unlikely that the FE are uncorrelated with the regressors of interest." Our choice of FE is also motivated by the potential consequence of being incorrect. When the random effects model is valid, the FE models will still produce consistent parameter estimates. Alternatively, when a FE approach is appropriate, use of random effects techniques will NOT produce consistent parameter estimates. Average Price Under block rate pricing, the relationship between price and consumption is unusually complicated, as price (either average or marginal) not only influences consumption, but the level of consumption influences price. After more than 50 years of research, this issue is well documented; however, little consensus exists as to how to estimate con- sumer demand under such rate structures (e.g., see Pint, 1999; Arbues et al., 2003). Given the relationship between price and quantity, price is likely to be endogenous and OLS will produce biased estimates (Nieswiadomy and Molina, 1988). In our case, concern arises over the fact that both ln(avepricei,1_1) and ln(avepricei,t_1)*restrict, are likely to be correlated with el,t. This correlation may exist both through Ili when not using the FE model and prt_1 when using the FE model framework (Arbues and Barberan, 2004). To account for this, we follow a common practice of using IV techniques. (For a more detailed discussion of this approach in the panel set- ting, see Wooldridge 2002.) Consistent with Nieswiadomy and Molina (1988) and others, we use the parameters of the rate struc- ture as instruments for ln(auepriceht_1) and ln(aue- pricei,t_1)'restrictt. Specifically, in the first stage the price of each block, by itself and interacted with restrictt, are used to generate estimates of ln(aue- pricei,t_1) and ln(avepricei,t_1) `restrictt that are uncor- related with Fi,t. These estimates are then used in the estimation of Equation (1). Valid instruments for ln(avepricejt_1) and ln(aue- pricei,t_1)`Erestrictt must possess two qualities: they must be correlated with average price and uncorrelat- ed with the error term. The instruments described above satisfy both of these requirements. Our deci- sion not to include the variables corresponding to the width of each block is worth noting. In this case the use of block width as an instrument for average price would be inappropriate, as the width of each block is likely to be correlated with Et,t during periods when water budgets are used. JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 207 JAWRA