Contribution to scholarship

Our study contributes to the literature by examining how bills for drinking water vary within and across states and by sub-state region, defined in terms of combined metropolitan statistical areas (MSAs). Comparisons using these geographic units allow us to account for potential similarities in regulatory environment, climate, water source availability, and other factors that influence bill amounts. To our knowledge, only one previous study, Thorsten, Eskaf and Hughes (2009), takes a similar data collection approach with a near census of systems with available billing data across one entire state in the U.S [1]. Notably, they find that water systems’ bills are significantly and positively correlated with bill amounts charged by other nearby systems. Chica-Olmo, González-Gómez and Guardiola (2013) take a similar study approach in Southern Spain and find similar results (2009) [1, 2]. This finding is somewhat intuitive; one might expect relative parity in price levels for an essential service such as drinking water, as is supported for other household staple goods and services.

Our use of a cross-state sample of systems with available billing data complements work by Teodoro and Saywitz, who report trends in drinking water and sewer affordability based on a nationally-representative sample of household water bills, and additional work in certain states by Teodoro [3–5]. Other recent studies analyzing water bills also identify multi-state or national trends in drinking water affordability over time [6–8]. Each of these studies has a slightly different focus and contribution to the literature; we discuss our findings in the context of these studies. One drawback of this literature is that studies either examine affordability in very large systems or use a convenience sample of systems.

Our study differs from other recent analyses in at least three key respects. First, our dataset of systems, as described below, represents nearly all systems that report billing data in four regionally diverse states across the U.S. where the most extensive data are available. This is valuable given that most studies which examine variation in water bills focus predominantly on large systems or small geographies, or use very limited data [9]. For instance, previous comparative bill work in North America is largely limited to individual metro areas—Los Angeles, Chicago, and British Columbia [10–12].

Second, we focus on explaining disparities in bill amounts rather than exceedances of any specifically defined affordability thresholds. We focus on bills to relate water affordability to the broader concept of water poverty, which provides a more holistic assessment of water access [13]. Previous literature acknowledges that water affordability is shaped by local community contexts and social inequalities [13, 14]. In particular, bill amounts interact with city, community, and household factors to influence water affordability and access [14]. We further differentiate our analysis from past studies by identifying bill amounts that are outliers at the low end and the high end of the state and MSA distributions, the former of which could suggest low system financial capacity and the latter of which could suggest customer affordability concerns. We use 4,000 gallons per month as our primary billing comparison point to capture a relatively modest level of household consumption. Prior research on affordability and conservation has also used this threshold for household consumption [7, 15].

Third, in addition to using available system characteristics, we collect data on water quality violations from the Safe Drinking Water Information System (SDWIS), available from the US Environmental Protection Agency, and customer base characteristics from the U.S. Census to jointly explore the relationship between these factors and billing levels for water to an extent beyond that in existing studies. Several studies have shown a relationship between race-ethnicity and water quality outcomes [16, 17]. The relationship between a system’s technical, managerial, and financial (TMF) capacity, including its ability to respond to regulatory water quality violations, however, is much discussed and assumed in affordability conversations, but rarely documented empirically [18].

Preview of findings and policy implications

In this study, we find notable local-level variation in bills that differs based on system characteristics. We find that municipally-owned systems are more likely to have lower water bills relative to for-profit systems. Our results also suggest that systems that serve populations with higher levels of income inequality and higher proportions of Non-White population typically have lower water bills. Additionally, factors such as purchasing water and having neighboring systems with higher water bills significantly correlate with higher water bills.

Although our findings have implications for state and national drinking water system assessment efforts, the extent to which these results are generalizable across the U.S. is unclear. Collecting and analyzing bill data from additional states will be essential to assess national trends. Notwithstanding, this study fills a gap in the literature by comparing bill amounts across and within states, and examining the factors that drive variation in water bills across state boundaries [9]. It also calls attention to some system-level characteristics that drive some of the observed variation in bills. As such, the findings of this study provide insight for water system TMF capacity, local economic capacity, and affordability policies. This study also identifies future research and data collection needs to make those efforts more robust.

Water rates data

The UNC EFC dashboards are a unique and valuable set of water bill data. First, they are carefully compiled using a refined and well-tested data collection and standardization method. UNC EFC compiles these data primarily for benchmarking and comparison purposes between local systems in a given state. UNC EFC explicitly advises users to “compare [bills] with caution. High rates may be justified and necessary to protect public health.” Though we acknowledge the validity of this statement, we also recognize that high rates may create affordability challenges for low-income households and in turn create different public health risks [19]. Second, even compared to other credible state-level and national-bill-level surveys (such as surveys conducted by AWWA-Raftelis and Circle of Blue) and dashboards (Duke’s Nicholas Institute), the UNC EFC dashboards provide a near census of all systems with available billing data in a given state. The data contained in UNC EFC’s dashboards effectively represent 80–90% coverage of all systems serving 500 or more people and historically have achieved the highest response rate of any such effort. The actual number of community water systems in each state is larger than the number reported in each UNC EFC dashboard (AZ = 742; WI = 1,034; GA = 1,725; NH = 710). This equates to 40% of all systems, but 86% of non “very small” systems serve a population of 500+ (GA = 1,100; WI = 541; AZ = 434; NH = 581). The 500-customer population threshold is the cutoff for EPA’s “very small” system designation. Third, the dashboards also provide bill amounts for multiple levels of consumption; this provides a more accurate and consistent estimate of bills than estimates from water systems or household self-reporting. Lastly, they merge in additional valuable contextual information from several sources. This information includes system size, billing cycles, rate structures, and CDP.

As of late 2021, the UNC EFC currently had rates dashboards available for 18 states. All the dashboards employ very consistent, albeit not uniform, methodologies and some contain multiple years of data [20]. For the purposes of this study, we focus on data for Arizona, Georgia, New Hampshire, and Wisconsin (see Fig 1). The UNC EFC made available a near-standardized Excel spreadsheet version of data for each state analyzed, with the exception of Wisconsin. In each spreadsheet, data on most variables available in the online dashboards are available for each system, with one key exception. For Wisconsin, collecting data from the dashboard required manual scraping from PDF forms into Excel.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Four states included in this study and data available.

https://doi.org/10.1371/journal.pwat.0000190.g001

We focus our analyses on documenting the variation in bill totals at 4,000 gallons of consumption. We examine this variation within and across states and adjacent communities, defined in terms of metropolitan statistical areas (MSAs).

With respect to rate structures, a flat fee or charge refers to systems charging customers a single monthly amount regardless of water use. Uniform rates charge a single volumetric rate per unit of consumption (e.g., the rate is multiplied by monthly consumption). Uniform rates differ from block or tiered rates, in which the volumetric rate changes based on which ‘block’ of consumption a customer falls in. Increasing-block rates occur when the volumetric rate increases for customers in higher consumption levels, while decreasing block rates charge lower rates for customers with higher water usage. Many systems have bills that include a combination of flat fixed charges and volumetric charges. The flat fixed charge is often called a base charge in this situation, as it is assessed even for customers with no water usage, with volumetric charges (whether uniform or block rates) then included on top of this base charge.

There are clear differences in water rate structures for systems in each state. In Arizona, 73% of systems use increasing-block rate structures, whereas 78% of Wisconsin systems use decreasing-block rate structures, and 75% of New Hampshire systems use uniform rate structures. In Georgia, there is a more even split between uniform rate structures (47%) and increasing-block rates (48%). Flat and other rate structures are uncommon in all four states.

The bill amounts in the UNC EFC dataset reflect monthly-equivalent water bills for the various consumption points. Given that systems use billing structures with differing frequencies, the actual bills customers face in each system may be different. Notwithstanding, the monthly equivalents provide accurate estimates of the amount of monthly expenditure necessary for a household for comparative purposes.

There are significant differences in the method that water systems use for determining fixed charges across states. The majority of systems in Arizona calculate base pricing by meter size (67%), although some use constant pricing (32%). New Hampshire systems show a more even distribution between constant base pricing (48%), pricing by meter size (33%) and no base pricing (19%). Most systems in both Georgia and Wisconsin use constant base prices (84% and 99.7% respectively).

The median water bills at typical consumption points of 4,000 gallons per month are strikingly similar across states. Arizona, New Hampshire, and Wisconsin have nearly identical median water bills ranging $31-$33 whereas Georgia has consistently lower median bills at around $23.

The UNC EFC data also includes key data on water system characteristics, including how each system sources its water. Water systems are classified into three categories: groundwater, surface water, or purchased water. EFC obtains this information from the U.S. EPA’s publicly-available SDWIS database. The U.S. EPA categorizes systems as surface water if any of their sources are surface water. Groundwater is the most common water source for water systems across all four states (85% in Arizona, 63% in Georgia, 67% in New Hampshire, 89% in Wisconsin). Purchased water and other sources are uncommon in Arizona (9%) and Wisconsin (4%) but more prevalent in Georgia (19%) and New Hampshire (26%).

The UNC EFC data also include information on ownership structure. System ownership includes four categories: for-profit, municipal, other government, and other. For-profit refers to private systems including investor-owned systems which, depending on the state and size of the system, may or may not be regulated by a state or state-level public systems commission. Municipal refers to city-owned and operated water or sewer systems; while “other government” refers to special districts, county authorities, or joint powers authorities, depending on the state. “Other” encompasses all other system types, particularly mutual water companies that do not fit in the for-profit, municipal, or other government categories. Not-for-profit (e.g., mutual and cooperative water companies), as classified in the EFC database, only accounted for 24 systems total. Due to the very small size of this category compared to the overall sample, these were combined with the 3 systems classified as ‘other’ into a single other category. The states exhibit similar makeups of ownership type diversity, except for Arizona. Over 50% of all systems in Georgia, New Hampshire, and Wisconsin are municipal; in contrast, 69% of systems in Arizona are for-profit and only 3% are municipal (see Fig A in S1 Text).

Water quality data

Given the lack of readily available water quality data from state-level databases, we queried the U.S. EPA’s publicly-available SDWIS search function by state [21]. We then use the system names provided in the EFC database to identify and scrape data system by system [1, 8]. We only consider and scrape violation records data between the years of 2009 and 2020. To determine year of violation, we use the compliance period start date in the SDWIS data. As with rate data, SDWIS data on water violations and system characteristics must also be compared with caution, based on potential missing or incorrectly classified data [22].

For each system (referred to as community water systems in SDWIS) in each of the four states (1,558 systems- see Table 1), we collect and record each violation as an individual row in a spreadsheet to allow for maximum flexibility in analysis by violation type/time period per system. For analytical purposes, we define health-based compliance shortcomings (our primary water quality category of interest) in terms of Maximum Contaminant Level (MCL) violations, Lead Copper Rule exceedances, and treatment technique violations. We also include an additional measure of water quality compliance shortcomings that includes all “monitoring and reporting,” “notification”, and other miscellaneous violation types. The difference between compliance start and end dates captures length of time out of compliance during the time period of interest. But this was not used as a variable of interest in analysis given variability and degree of missingness in the quality of compliance date entry. While the original intention of the effort was to code the full detail of each violation, it quickly became impossible given some systems (especially in Arizona) have dozens of monitoring and reporting violations; we only coded full detail of each violation for systems with less than 5 violations. Combining these data allows us to study water quality in relation to other system characteristics which may be explored in future studies. In Supplementary Text 1, we provide bivariate correlations between water quality data and system-level characteristics. In this study, however, the primary outcome of interest is the residential monthly bill for water services.

In terms of health violations, most systems (1,158 of 1,558 systems) incurred no primary health-based violations from 2009 to 2020. Resulting health-based violation averages vary from 1 to 2 violations per system for all states, with the lowest average of 1.05 violations per system in Georgia and the highest average of 2.17 violations in Arizona. However, the median number of violations for systems in each state is 0, with several outliers (a maximum of 290 for one system in Wisconsin). Further analysis would be required to ascertain the extent of data limitations from SDWIS violation data. Observed differences in water quality compliance data across states may in part reflect actual compliance variation but also may reflect potential inconsistencies in state programs monitoring compliance or in the way in which violation data were coded and entered between states. Efforts were made to clean data to ensure unique violation entries, but SDWIS data should also be used with caution [22].

Demographic data

We obtain demographic data for each water system using CDP boundaries as the relevant geographic unit. Though geographic information system (GIS) shapefiles for water systems currently represent the best currently available approach (albeit still imperfect) to approximating water system boundaries, shapefiles are only available for the state of Arizona among the four states analyzed [23]. We explore our choice of using CDP boundaries relative to other potential approaches using shapefiles for Arizona (see Supplementary Text 2 for boundary comparisons). This approach follows Berazher et al. in their use of CDP boundaries for water systems in North Carolina [24]. We explored the possibility of using zip codes and geocoding address information but found these approaches to be inferior to our selected method. Using a single valid address for each system (i.e., geocoding) is not a reliable strategy given that the public-facing SDWIS is missing any address for some systems, and some of the addresses provided are out of county or even out of state P.O. boxes [22].

In addition to better approximating geographic boundaries, our use of CDP boundaries confers two additional benefits. First, the choice of CDP boundaries is consistent with collection efforts by the UNC EFC for some of its state dashboards, as well as other recent studies which evaluated similar alternatives [24]. Second, using CDP boundaries allows us to compile and approximately match income and race-ethnicity data from the U.S. Census to characterize the socioeconomic status of customer bases, as well approximate each system’s total population, the proportion of population which is Non-White, its median household income, the proportion of its population under 100%, 150%, and 200% of poverty level, and the Gini Inequality Index, which is a measure of income inequality.

We collect socioeconomic data from the U.S. Census (American Community Survey 2015–2019, 5-year estimates) at both the CDP and county-levels: race, ethnicity, median household income, and poverty status. The ACS data for these variables were downloaded using the National Historical Geographic Information System (NHGIS) database [25]. Using these data, we create the following five variables: Non-White population proportion, Hispanic/Latino population proportion, median household income, and proportion of the population under 100% and 200% of the federal poverty level (FPL). We match these census variables to systems using the system’s primary CDP, as listed on the UNC EFC dashboard or SDWIS. If the EFC dashboard did not provide a primary CDP/county or one that did not match census data (144 systems), the CDP was obtained from SDWIS. In the event the SDWIS CDP did not match available census data (35 cities), the primary county was used.

Census demographic data was matched to systems based on the primary location served. Where CDPs could not be obtained (no primary CDP served was noted in the UNC EFC data), we supplemented these data with the primary county served from SDWIS codes. We also ran a sensitivity test which excludes systems with county-matched data. We recognize substantial shortcomings with this approach; there is no perfect way to match system customer base demographic data to system boundaries. Any method to attribute population characteristics from the census to small water systems is likely to have a high degree of inaccuracy, given that the smallest census geography at which population characteristic data are available (the block group, serving between 600 and 3,000 people) is larger than any very small systems. For very small and some small systems, only manually-collected socioeconomic characteristic survey data will be sufficient.

Regression model specifications

Our analysis provides basic descriptive statistics across states to look for general state-level and regional trends for water bills at the 4,000-gallon consumption level. We then estimate a linear model to examine what system-level characteristics most significantly influence water bills. We estimate our model using ordinary least squares linear regression to shed light on the contribution of different system characteristics, including ownership type, rate structure, and water source type to a system’s monthly water bill.

To study the factors that influence bill amounts, we estimate the following linear model: where the outcome of interest Y_i is the total monthly bill amount corresponding to 4,000 gallons of water for residential customers at the system-level. We choose 4,000 gallons as the relevant amount because we argue that it represents a level of consumption that meets the standard for modest indoor household use. In other words, 4,000 gallons allows for essential needs but does not extend to substantial discretionary use (i.e., outdoor irrigation). Monthly consumption of 4,000 gallons reflects roughly 45 gallons (0.17 m3) per capita per day for a three-person household, which represents modest use. The decision to use 4,000 gallons as our main outcome of interest is also motivated by the fact that all systems in the dataset had bill data at that level (some states lacked data at other consumption levels).

We acknowledge that essential indoor needs vary by household size; and some studies find indoor use to be greater in some U.S. metro areas than others [26]. Notwithstanding, our choice of 4,000 gallons is consistent with the range presented in affordability literature which examines bills for “reasonable” levels of consumption. Such previous studies examine consumption levels ranging from 3,000 gallons in North Carolina (Thorsten, et. al, 2009), to 3,740 gallons nationally (5 CCF, commonly used in AWWA-Raftelis surveys), 4,488 gallons in California (Pierce, Chow and DeShazo, 2020), to up to 6,200 gallons for municipalities across the U.S. (Teodoro and Saywitz, 2020) [1, 4, 19].

The vector X_i represents the following system-level characteristics to explain variation in bills:

• water quality compliance:
  1. ○ total number of health-based violations (2009–2020)
  2. ○ total number of monitoring and reporting violations (2009–2020)
• socio-demographics of the customer base:
  1. ○ proportion of major racial ethnic groups
  2. ○ median household income,
  3. ○ proportion of population below federal poverty level
• nearest neighbors’ cost (i.e., average bill for all other systems in the county of the system and systems in the counties contiguous to the system’s county)
• system ownership type: municipal, other government type, private, or other
• system customer base size (i.e., approximated system service population)
• rate structure type: flat, increasing block, other block or uniform volumetric
• presence of a free monthly water allowance (i.e., binary variable where a 1 indicates system provides a baseline water allocation with the fixed charge)
• type of water source: groundwater, surface water, or purchased

Hypotheses

Each of the model specifications includes data on system-level characteristics hypothesized to affect water bills, based in part on similar studies cited in a recent U.S. Government Accountability Office (GAO) report [9]. Despite limited previous research on drivers of variation in bills, we draw on prior studies to generate hypotheses on the expected influence of the system-level variables on system bills.

Hypothesis 1: A review of multiple rate studies by the GAO suggests that public or quasi-public system ownership types (e.g., municipal, other government) will correlate with lower bills than for-profit systems [8, 9]. We anticipate that larger system customer base size will correlate with lower bill amounts based on potential economies of scale [10].

Hypothesis 2: We anticipate that higher nearest neighboring system bills will lead to higher bills, as local similarities to some extent reflect cost and political similarities [1, 2].

Hypothesis 3: We expect that purchased water sources will also lead to higher bills due to the costs of purchasing raw water which may be reflected in the bill price for customers, compared to systems with their own ground or surface water source rights. However, we note that we do not have data on source water quality, which may also impact bills. In particular, levels of salinity or contaminants may increase treatment costs [27]. Geographic or topographical factors can also influence costs of purchased water delivery [9].

Hypothesis 4: While limited, prior research in Michigan and North Carolina found higher bills faced by minority populations; if these trends extend beyond these states we would expect higher proportions of Non-White customer populations to correlate with higher water bills [28, 29].

Hypothesis 5: Finally, we expect household bills to be lower for systems that provide a free baseline allocation of water each month, as customers are not charged for this initial allocation that is a portion of the 4,000 gallons.

The vector γ_i represents geographic controls to assess intra-state variation. Including these fixed effects allows us to examine the influence of system-level characteristics on water costs while taking into account variation due to state location (state-level cost variation). We also estimate an additional specification which includes an additional set of fixed effects that represent local areas. As previously noted, we define local areas in terms of MSAs, using combined statistical areas where applicable. The last term in our model, ϵ_i, represents an error term that captures any remaining unexplained variation.

Influence of system characteristics

In support of Hypotheses 2 and 5, water bills are significantly higher among systems where neighboring systems charge more for water and significantly lower for systems that provide a baseline monthly allocation of water [1, 2]. Table 4 demonstrates that, when controlling for state location (fixed effects) and holding all else constant, a $1 increase in the average monthly bill for 4,000 gallons by neighboring systems will predict a $0.80 increase in average monthly bill for a system. Meanwhile, the presence of a baseline monthly allocation of water (some water provided without charge) predicts a decrease of $4.73 for a system’s monthly bill level for 4,000 gallons, holding all else constant. These effects are slightly stronger when the local MSA of a system is controlled for instead of the state in Table 5 ($0.96 increase and $4.94 decrease for an increase in neighboring water bills or a monthly allowance respectively). Additionally, municipal systems tend to levy significantly lower water bills than for-profit systems (the baseline of for-profit systems is not included as a variable in the model to enable comparison) [6]. The “other” category for government ownership type was not a significant correlate in the model compared to the for-profit system type baseline. In support of Hypothesis 1, a larger service population is associated with significantly lower bills. We also find that water systems that purchase water sources have significantly higher bills than systems with groundwater sources, which supports Hypothesis 3. This is demonstrated by the values in Table 4 and 5; all else constant, when a system purchases water as its main source, the model predicts an increase in the monthly water bill for 4,000 gallons of $1.78 or $2.14 depending on fixed effects (controlling for a system’s state or MSA respectively).

The presence of a tiered water rate structure is also significantly negatively correlated with bill amounts in the model. While further research is required to assess the reasons behind this trend, this does follow prior research that finds increasing block rates for water may improve water affordability [31]. Often increasing block rates include a ‘lifeline’ or lower rate for lower levels of consumption, thus modest levels of usage (e.g. our selected threshold of 4,000 gallons) may result in lower bills than the same usage under certain flat or decreasing block rate structures [32]. In the regression models, the use of an increasing block rate structure predicts a system will have a $2.44 or $2.85 decrease in monthly bill when controlling for state or MSA system location respectively (all else constant).

Influence of demographic variables

However, higher income inequality, which is highly correlated with a greater proportion of Non-White population (see Supplementary Text 1), is significantly associated with lower bill amounts. Higher Non-White population in our model is significantly associated with lower bills, contrary to our expectation (Hypothesis 4) after controlling for other factors. Table 4 demonstrates that, holding all else constant, a 1% increase in the proportion of Non-White population served by a water system predicts a decrease of $6.38 or $6.27 in the system’s monthly water bill for 4,000 gallons, when controlling for the system’s state and MSA respectively.

Influence of water quality and monitoring variables

We did not find evidence of water quality violations, either health or procedural, influencing bill totals. We do note, however, that the proportion of Hispanic/Latino residents served by a system is significantly positively correlated with more health violations (see results of bivariate correlations in Table 1 in S1 Text). This finding echoes existing research regarding potential inequities in water quality by race and ethnicity [23]. Similarly, we do not find significant associations between health-violations and system size (i.e., service population). The relationship is positive and non-significant, despite evidence that smaller system size predicts more health violations than much larger systems [10, 33]. Further research is required to better assess the relationship between water quality and bills; as noted above, there may be differences in how systems and regulators monitor, report, and enforce water quality violations among states which may impact the data and ultimate correlations.

Review of results and contributions

In this study, we compile data on water system bills from four states to analyze variation in local bills at modest levels of consumption (4,000 gallons). We then examine characteristics of systems and water bills across the four states selected for analysis: Arizona, Georgia, New Hampshire, and Wisconsin. Though median bill amounts for drinking water are similar across all four states, we find major differences in bills within regional and local areas, influenced by system-specific characteristics.

We find large variation in bill amounts at each geographic scale that we analyzed, including within MSAs. Variation in water bills across systems can in part be explained by system ownership type, size, water source, rate structure, and the bills of neighbors, as well as race-ethnicity and income inequality of customers. Municipal ownership type, larger service population, and a baseline monthly allowance of water all predict systems with lower bills, supporting hypotheses from existing studies. Additionally, we find that, as hypothesized, higher bills of nearby systems and using purchased water as the main water source predicts higher bills for a system. Our results suggest that water systems with high levels of poverty have higher water bills. Counter to our expectations, including evidence from previous studies, our results also suggest that local systems that serve populations with higher levels of income inequality and higher proportions of non-White population typically have lower water bills. These relationships, however, appear to be relatively weak statistically, and thus deserve further scrutiny. Overall, our finding that some system characteristics are strongly correlated with water bills has important implications for policy interventions in the context of evaluating systems’ TMF capacity, as well as customer affordability, and community environmental justice outcomes.

Limitations and future directions

Much of the variation in bills, however, remains unexplained in our modeling. One potential reason for the unexplained variation is limited water system-specific data. Comparisons with existing available data sources require ‘caution’ [20, 22]. We could not capture all drivers of water costs nor could we consider historic drivers of water rates since our bill data is derived from a single point in time. Other potentially interesting variables, suggested by Teodoro and Saywitz and the UNC EFC, for example, are not readily available for collection and matching at the local system level but may be collectible for pilot analyses for individual or small groups of states [4, 34]. These include system capital replacement rates; receipt of general fund, state, or federal assistance; physical and uncollectible non-revenue water levels; customer class composition ratios; staffing levels and compensation; and quality treatment requirements. Recent efforts by researchers have resulted in the Municipal Drinking Water Database (MDWD) which may prove useful for capturing such variables at the water system level [35]. The MDWD contains variables on water system revenues, expenditures, and capital outlay, although initial review finds numerous systems in the UNC EFC database missing from the MDWD. A null finding regarding water quality compliance and bills, despite multiple specifications, also deserves further exploration, particularly to understand links between water affordability and access to safe, clean water. With respect to water quality, another avenue for future research would include a comparison of pre and post pandemic situations as several recent studies have observed that water quality improved during periods when lockdown measures were in place [36–40].

Our study expands upon previous work examining variation in water bills at the state and metro-specific levels and provides additional insight which can inform future policy efforts. This study does not, however, provide definitive answers on the best metrics to compare intra-MSA bill variation at the low or high end, or factors contributing to this variance. Local water bill amounts have historically been, and largely remain, a local matter with some state oversight. While we find consistent averages across states, we find very high levels of bill variation at the local and regional scales. Significant variation in bill amounts, as opposed to water quality—where the Safe Drinking Water Act imposes common minimum standards—is inevitable and may be justifiable. Most of the correlates of high bills which we identify are not conducive to direct policy reforms. Notwithstanding, extreme disparities may justify policy efforts from states to promote consolidation, evaluate TMF capacity, provide customer assistance, or implement rate revision guardrails. Regardless of the extent to which policymakers seek to intervene to address local bill disparities, understanding this variation is an important step in the context of affordability and larger environmental justice efforts.