Estimating and Verifying United States Households’ Potential to Conserve Water

Urbanization and growing populations are placing increased demands on scarce, limited, municipal water supplies. Compared to expensive supply side options to expand municipal or regional water infrastructure, residential water conservation can cost effectively help demands match available supplies. Conservation can include technological changes, such as replacing old toilets, faucets, showerheads, dishwashers, and laundry machines with newer and more efficient appliances mandated by the Energy Policy Act of 1992. However, to include water conservation in water supply/demand planning, it is important to correctly know and forecast current and future water demands and the volume of water potentially saved by conservation actions.

Planners and water managers have forecasted water demand and estimated water savings for many conservation programs and measures (Berk et al. 1993; Buchberger and Wells 1996; Kenney et al. 2008; Michelsen et al. 1999; Renwick and Archibald 1998; Walski et al. 1985). For example, low flow showerheads and toilet dams were distributed among Hamilton Township residents in 1978; subsequently customers were surveyed to identify the number of devices actually installed, and coefficients obtained from this data were used in an algorithm to predict water savings (Walski et al. 1985). Data loggers have been installed on the supply line for single-family residences, recording the total instantaneous water demand of the household (Buchberger and Wells 1996). Models to estimate household-level water demands have been developed as a function of price, weather, house, and household characteristics, as well as other policy restrictions and interventions during the study periods (Kenney et al. 2008, Michelsen et al. 1999; Renwick and Archibald 1998). Water conservation program planners have also probabilistically described the volume of water saved from conservation actions by enumerating uncertainties associated with costumer demographic, behavioral, and technological parameters that influence water savings (Rosenberg 2007).

Tracking only the flow or usage rates of more efficient appliances doesn’t provide a direct way to estimate savings because human behaviors also play an important role—the duration and frequency of appliance use. Additionally, when people know they are using a water-conserving appliance, they may use the appliance longer or more frequently. This increased use may swamp expected water savings (Campbell et al. 2004).

Still, the demand forecasting and conservation estimation methods discussed above can be improved in several ways. First, estimates of water saved by structural conservation actions need to be empirically verified. More carefully gathering and storing observations of water use and pairing observations to estimates can help with empirical verification (Walski et al. 1985). Second, household heterogeneity must be more explicitly considered (Whitcomb 1990; Rosenberg 2007). Studies typically include a wide variety of explanatory variables (such as income, household size, lot size, age of house) culled from secondary data sources to characterize household heterogeneity, but use only one aggregate and primary-sourced dependent variable—monthly billed water use (Kenney et al. 2008). Using disaggregated end-use data for each water appliance can add more specificity. Third, technological and behavioral factors influencing water savings can be better described, disentangled, and considered as explanatory variables. For example, the duration and frequency each resident in a household uses an appliance may differ. At the same time, the flow, flush, or use rate of an existing appliance can depend on numerous factors including when the appliance was manufactured and whether it has been maintained. Similarly, the flow, flush, or use rate of a retrofitted appliance set by the manufacture may differ from the installed or actual rate. Thus, water use and savings depend on both technological and behavioral factors acting together.

Until recently, it has been difficult to separately measure and differentiate technological and behavioral factors. However, this separation is now possible using data logging and disaggregation methods deployed for a national study of nearly 1,200 households (Mayer et al. 1999) and a water appliance retrofit study for 96 households in Oakland, Seattle, and Tampa, between 2000 to 2003 (USEPA 2004). The latter retrofit study data include two weeks before and four weeks after each household was retrofitted with water efficient toilets, faucets, showers, dishwashers, and clothes washers. Prior work describes the disaggregation method (Mayer et al. 1999), organized the retrofit data, reported summary statistics such as average water saved by each appliance retrofit, and developed regression models to estimate demands using number of residents and house size as independent variables (USEPA 2004).

This case study presents analytical, hybrid, and regression models to estimate the water saved when retrofitting indoor water appliances and describe the prediction precision and efficacy among models. Models are built from and verified against the previously collected U.S. Environmental Protection Agency (USEPA) end-use retrofit study data (2005); model variables include preexisting and retrofitted flush and flow rates of water appliances, such as toilets, showerheads, faucets, and clothes washers. Also, the models use behavioral variables such as the duration and frequency of appliance use. In this way, the models identify separate and combined effects of technological and behavioral factors. We also show the distributions of water saved among households and regression models identify characteristics of households with the potential to save the most water from retrofits. Case study results highlight ways U.S. water utilities can target retrofits to households with potential to save the most water.

This work uses end-use data previously collected from 96 single-family houses in Seattle, East Bay Municipal Utility District (EBMUD), and Tampa between 2000 and 2003 before and after each household was retrofitted with water efficient appliances (USEPA 2005). Water use was recorded by placing data loggers on each participating household’s water meter and recording water flow through the meter at 10 s intervals. Flow signals were then postprocessed to determine the duration, water volume, and frequency of household leaks, outdoor and indoor water uses including toilets, showers, clothes washers, faucets (USEPA 2005). For details on the methods used to monitor and disaggregate end uses, see Mayer et al. (1999).

The houses selected for the study used more than 227 L (60 gal.) per capita per day and were representative of households in the three cities. Participating homes averaged 46 years in age. Old homes are less likely to have water-conserving appliances; however, participating homes initially showed a wide range of appliance flow and flush rates indicating varied old to newer, water-conserving appliances. Also, additional sociodemographic data was collected on each participating household, including adults per household, children per household, number of bedrooms, number of bathrooms, floor area, and price paid for water.

Water use data collected before the retrofit constituted the baseline water use for each household. Next, water appliances were retrofitted with more efficient ones, i.e., existing toilets were replaced with low flush volume toilets. One month after the retrofit, water use was again recorded for two weeks. Finally, six months after the retrofit, water use was logged for two more weeks to identify behavioral changes and the persistence of water savings from the retrofits.

In general, households reduced the water use after they were retrofitted with the new appliances (Fig. 1). The initial analysis found annual water savings averaged 79, 21, 6, and $5{\mathrm{m}}^3$ per year, respectively, for toilet, clothes washer, showerhead, and faucet retrofits (USEPA 2004). Yet, six homes didn’t save any water after retrofits, while other households saved more than $750{\mathrm{m}}^3$ per year. Overall, 93% of the households saved water, showing that retrofits can be effective. Fig. 1 also suggests that utility companies can improve conservation program effectiveness with reduced effort if they can target programs to households on the right tail of the distribution with the most potential to conserve water.

Analytical and regression methods are used to estimate water saved and estimates are compared to the actual water saved by households participating in the USEPA study. The methods used in this research have as an objective to predict the savings observed. In the following, methods to calculate actual savings and develop analytical, hybrid, and regression models are presented. Hybrid models embed an analytical equation in the regression model.

Model results are presented by appliance. For each hybrid and regression model developed, regression coefficient values and the $R^2$ showing the fraction of variation in the dependent variable (water saved) that is explained by the model variables are reported. A Kolmogorov-Smirnov test (K-S test) was also used to compare resulting distributions of water saved among households by the analytical, hybrid, and regression models to the actual water saved. The K-S test indicates if the distributions of two datasets differ significantly, and makes no assumption about the distribution shapes or the sample size (Chakravarti et al. 1967). The K-S test gives a $D$ value, $D$ being the maximum difference between the two cumulative density functions tested. The null hypothesis that the two distributions are the same when the associated probability $P$ is less than 0.05 was rejected.

One of the purposes of this study is to identify households with the potential to save more water. These houses are located in the right tails of the water savings distributions shown in Figs. 1, 2, and 3. Surveys were used and and end-use data were collected about and at the houses to identify characteristics of households that will likely save the most water from retrofits.

Households were ranked by water volume saved by retrofitting each appliance, then the largest savers (top 20 households) for each appliance were separated from the rest. We chose this breakpoint to allow sufficient degrees of freedom to run regressions for each group. This segregation also means a certain household could be in the high savings group for one appliance, but not for other appliances. For each appliance, separate linear regressions were made for the groups of largest and smallest savers. For each group, water savings was regressed against the independent variables (water price, number of residents, number of full and three-quarter bathrooms, flow rate, and frequency of use).

For each appliance, households that saved the most water by retrofitting the appliance had more residents than households that saved less water (Table 4). On average, large savers had approximately one more person per household than small savers. Similarly, households that saved the most water from retrofits had, prior to retrofits, appliances that delivered larger volumes per use and were less water efficient. This difference was significant for the toilet, shower, and faucet models ($P<0.05$). Another factor that differentiated high savers from low savers across most appliances was behavior. High savers tended to use appliances more frequently than low savers with this difference significant for the toilet and faucet models.

Interestingly, results also show that both high and lower savers used laundry machines with about the same frequency. Both groups also typically had the same number of bathrooms; as such, these two factors are not particularly helpful to differentiate users in the two groups.

Although not statistically significant, the largest savers generally faced lower water prices than smaller savers. This finding runs counter to microeconomic theory that suggests higher water prices should encourage larger reductions in water use. However, in this study, the participating cities paid for and installed all water efficient appliances for households. Thus, the counterintuitive result may occur because larger savers had a lower financial incentive prior to the study to invest in water conservation measures and possibly used more water then (price was also constant through the study period). Together, the tails analysis suggests that a large family size combined with a water efficient appliance and high frequency of use result in large water savings from retrofits.

Analytical, regression, and hybrid models were developed to estimate water savings by retrofitting toilets, showerheads, clothes washers, and faucets. K-S test results show the three types of models all reasonably predict distributions of water saved except for the hybrid model for shower retrofits and the analytical model for faucet retrofits. Regression results indicate regression and hybrid models for toilet and clothes washer retrofits explain more of the variations in water saved than models for shower and faucet retrofits. This result is expected because toilets and clothes washers have a fixed water volume per use, whereas users can regulate shower and faucet flow rates during each use. Additionally, regression models for toilets and clothes washers explained more of the variations in water savings than the hybrid models. In these instances, behavior varied before and after retrofits and it was not possible to separate the behavior component in the hybrid models. For shower and faucet models where the behavioral variable was separated, hybrid models explained as much or more of the variation in water savings as regression models (Table 1). These comparisons emphasize the importance to include behavioral factors in models to estimate water savings by retrofits. They also suggest that a simple analytical approach to estimate savings may suffice when data are available to describe the technological and behavioral factors that influence savings.

More elaborate regression and hybrid models are warranted when it is important to show the separate and combined effects of technological and behavioral factors on water use and water savings (Table 2). Elasticity values shown in Table 2 provide a way to normalize and compare the relative effects of technological and behavioral factors on water savings. In all models, technological and behavioral factors are important and significant (Table 2). In several models such as for showers and faucets, the technology component contributes more to overall water savings than the behavioral component as represented by larger elasticity values in the regression results. This effect is magnified for appliances where behavior did not change preretrofits and postretrofits. In a few cases, households increased their water use after the retrofits, as a consequence of behavioral changes that offset technological improvements. For appliances such as toilets and clothes washers where household use behavior changed preretrofit and postretrofit, it was not possible to differentiate in the hybrid models effects of technology and behavioral factors. Instead, an average change in water use variable was introduced that aggregated behavioral and technological factors prior to regression. Otherwise, both the regression and hybrid models use the same natural log of the independent variables and the main difference is that the hybrid model regresses against the natural log of the dependent variable (water saved), whereas the regression model regresses against the actual value.

When comparing characteristics of households that saved the most water by retrofits to households that saved less water, in general, high saving households had more residents than households that saved less water. Also, prior to retrofitting, these households had less efficient appliances and used appliances more frequently than households that saved less water by retrofits. Utility companies can use these findings to identify and target households with high potential to save water. To target these households, utility companies should seek houses with larger number of residents and with least efficient appliances. Utility companies can determine high occupancy from household surveys or census records. Utility companies could also use property parcel or permitting records as an indicator of the age (and likely flow rates) of appliances inside a house. Study results also show that houses with a higher frequency of use save more water, but this household behavior may be difficult for a utility to identify and target in campaigns to motivate behavioral change. Utilities cannot install data loggers in the houses of all their customers. To identify households with potential to save the most water, water utilities can and sometimes do rely on large billed water use as a proxy for behavior; however, demographic and technologic factors such as household size and appliance flow rates can confound this practice. The finding and current practices identify the need for further research to identify household behavioral indicators that utility companies can readily measure and reliably act upon to target households with large potential to save water from retrofits.

It is important to point out that the models use as a variable the number of permanent residents and do not account for visitors during the study period. Unobserved visitors could alter the frequencies with which appliances are used. Also, the data collected by the loggers were aggregated by appliance type, which means technological and behavioral variable values for each appliance type were results of averaging across all appliances of the same type in a house. Further, the analysis excluded the leakage volume saved by retrofitting leaky appliances because intermittent leaks from faucet drips, unsealed toilet flappers, and other sources prior to retrofit were disaggregated separately apart from appliance water use. In the disaggregation, these leaks are easily identified by their very low flow rates (too low to be faucets), association with other events that might initiate a leak, or duration different than faucet use (Mayer et al. 1999). This exclusion means actual water savings from retrofits are larger than reported savings; it also identifies a further need for improved leak prediction and modeling capabilities. Finally, for hybrid models, some households were excluded from the analysis because it was not possible to calculate the log of water savings for households that increased use after retrofits and had negative savings. Thus, sample sizes were different among the analytical, hybrid, and regression models.

Analytical, hybrid, and regression models were developed to estimate water savings using detailed, disaggregated water end-use data collected before and after retrofitting toilets, showers, faucets, and clothes washers in 100 households in Oakland, Seattle, and Tampa. Water savings result from a combination of the technology installed in the households and the use of appliances. Our models use technological and behavioral variables and represent an improvement on prior models that rely on only demographic and household characteristics.

Results show analytical, hybrid, and regression models can all reasonably characterize distributions among households of water saved by retrofits. Use of analytical models to estimate water savings in a simple, dimensionally consistent way when data exist to describe the technological and behavioral factors that influence savings. Alternatively, use regression and hybrid models when there is a need to (1) estimate indoor residential water savings based on technological and behavioral components; and (2) separate out the effects of each component. For all the appliances, technological and behavioral variables such as flow rates, durations, and use frequency are significant with technological variables having a larger effect on water saved for appliances such as showers and faucets where the user regulates the appliance flow rate during each use event.

Houses that saved more water from retrofits tended to have more residents than those who saved little or no water. They also had less efficient appliances prior to retrofits, used appliances more frequently, and experienced lower water prices than lower savers. U.S. water utility companies can use these case study findings in conjunction with census data, customer surveys, and property parcel and permitting records to identify customers with high potential to conserve water indoors.

U.S. water utility companies can also use these case study results and findings to target customers with more residents, less efficient appliances, and high use frequency to replace old appliances for newer and more efficient ones to conserve water. With targeting, utility water conservation programs can save more water with less effort. The case study results also help improve engineering methods to estimate water savings from retrofits.