Environmental justice and drinking water quality: are there socioeconomic disparities in nitrate levels in U.S. drinking water?

Drinking water quality is regulated in the United States under the Safe Drinking Water Act (SDWA), which establishes national monitoring and reporting requirements and maximum contaminant levels (MCLs) for 88 contaminants. In 2013, 9800 public water systems, serving 26.5 million Americans, had violations of health-based standards [1]. In 2015, lead contamination in the drinking water supply of Flint, Michigan, caused elevated blood lead levels in children following the use of a new drinking water source, the Flint River, as a cost-saving measure [2]. Because 60% of Flint’s residents are African American and 40% live below the poverty line, this crisis sparked a nationwide debate about environmental justice—equal treatment and protection from environmental harm regardless of race, ethnicity, or income—and drinking water quality.

Low-income and minority communities often face disproportionate burdens of exposure to contamination sources and environmental pollution, and associations with race and ethnicity persist even after accounting for differences in income [3]. While few studies have looked for links between drinking water and environmental justice indicators (e.g., poverty, race/ethnicity) [4], existing studies have found associations between poorer drinking water quality and these indicators [4–8]. Community water systems (CWSs) that serve communities with lower median incomes, lower rates of home ownership, and higher proportions of Hispanic or non-white residents have been associated with higher levels of nitrate and arsenic [5–7]. Among small rural water systems in Quebec, those serving areas with more material deprivation (based on income, education, and employment) were more likely to have lead levels of health concern and less likely to have advanced water treatment [4]. Health-based violations of the SDWA were more common in poorer communities with higher proportions of Hispanic or African-American residents; the effects of race and ethnicity were not apparent in more affluent communities [8]. Environmental justice associations with drinking water have not been consistently observed, and may depend on the spatial scope and individual contaminants studied. For instance, Cory and Rahman [9] concluded there was limited evidence for environmental justice disparities in exposures to arsenic in water systems in Arizona. The likelihood of studies finding environmental justice associations with the siting of hazardous waste facilities strongly depends on the unit of analysis (e.g., county, census block) and overall scope (e.g., state, national); studies with a small unit of analysis and large scope were most likely to find significant associations [10]. To date, studies of environmental justice and drinking water contaminants have considered individual states (e.g., Arizona, Oregon) or sections of states (e.g., California’s Central Valley); however a nationwide assessment is lacking.

According to Balazs and Ray’s Drinking Water Disparities Framework [11], there are a wide range of natural, built, and sociopolitical factors that can cause and perpetuate disparities in water quality, reliability, and infrastructure. Small water supplies, particularly those that serve low-income and minority communities, may have poorer source water quality due to closer proximity to pollution sources. In addition, such supplies may have diminished technical, managerial, and financial (TMF) capacity to properly manage their drinking water, so these systems may lack the resources necessary to comply with testing requirements. Indeed, a nationwide analysis indicated that small CWSs were more likely to have management-related SDWA violations [12]. When problems are identified, small systems with limited TMF may struggle to address these problems, such as through the installation of new treatment systems or development of better protected sources [1]. In communities of color, institutional barriers in local planning and zoning practices may lead to lower rates of drinking water and wastewater infrastructure improvement [13]. These can relate to both internal factors (diminished ability to raise rates for customers) and external factors (ability to apply for loans). These factors are especially apparent in unincorporated areas, which have no tax base and lie outside of the municipal boundaries overseen by county or state entities.

Nitrate is one of the contaminants most frequently found in violation of health-based standards in U.S. drinking water [14]. Nitrate naturally occurs in aquatic systems at low concentrations (< 1 mg/L NO₃-N), while concentrations greater than 1 mg/L NO₃-N are considered elevated above background and indicative of human activity [15]. Common anthropogenic sources of nitrate include fertilizers used for agricultural production and landscaping, animal manure, wastewater discharges from sewage treatment plants and septic systems, and fossil fuel combustion. Elevated levels of nitrate can signal the presence of other contaminants of concern; a study of over 2000 private wells found that wells with > 1 mg/L NO₃-N were more likely to have levels of pesticides and volatile organic compounds (VOCs) above one-tenth of an MCL or health-based screening level [15]. Among public and private wells in sand and gravel aquifers, nitrate concentrations were correlated with pharmaceuticals and other unregulated drinking water contaminants [16, 17].

As part of the original implementation of the SDWA in 1974, the U.S. EPA established a nitrate MCL of 10 mg/L NO₃-N (45 mg/L NO₃⁻) based on case studies of methemoglobinemia in infants who consumed formula mixed with water containing nitrate [18]. More recent epidemiological studies have found associations between nitrate concentrations in drinking water and bladder cancer [19, 20], thyroid cancer [21, 22], colon cancer [23, 24], kidney cancer [25], birth defects [26, 27], low birth weight [28], and preterm birth [29, 30]. Some of these effects were significant for exposures at or above 5 mg/L, particularly over longer exposure periods [21, 23, 26, 27]. The International Agency for Research on Cancer classified “ingested nitrate or nitrite under conditions that result in endogenous nitrosation” as a probable human carcinogen (Group 2A) [31]. Exposure to nitrate in drinking water has also been linked to thyroid dysfunction [32], although the Agency for Toxic Substances and Disease Registry concluded that there is “limited evidence” for nitrate-induced thyroid dysfunction [33]. The U.S. EPA’s Integrated Risk Information System (IRIS) is undertaking a broad re-evaluation of the health effects of nitrate and nitrite [34].

In light of growing epidemiological evidence for nitrate health effects below the MCL and evidence on a local level for socioeconomic disparities in nitrate exposure, our study was designed to evaluate whether nitrate concentrations are elevated in public water supplies that serve communities with higher proportions of low-income and/or minority residents. We hypothesized that CWSs serving communities with higher proportions of Hispanic residents would have higher nitrate levels because 80% of U.S. farmworkers are Hispanic [35] and because synthetic fertilizers used in agriculture are the largest source of nitrogen inputs in the U.S. [36]. We also anticipated that the high cost of removing nitrate from contaminated drinking water would lead to socioeconomic disparities in nitrate exposures. This study represents the first investigation of socioeconomic disparities in drinking water contaminants at the national scale and provides new insights into the interplay of system characteristics and demographic parameters.

The majority of estimated annual mean nitrate concentrations were below 1 mg/L for all system sizes, regions, and source water types. However, 1647 CWSs, serving over 5.6 million Americans, had average nitrate concentrations at or above 5 mg/L from 2010 to 2014. The West and Midwest had the highest proportions of high nitrate systems (≥5 mg/L) and the highest 95th percentile nitrate concentrations, while the South had the lowest (Table 1). The proportion of high nitrate CWSs was nearly four times higher for systems that rely on groundwater sources compared to CWSs that rely on surface water sources, and the 95th percentile concentration for groundwater systems was more than 2 mg/L higher. Relative to other system sizes, very small systems (≤500 people) had the highest 95th percentile concentration and the highest proportion of high nitrate systems.

Among environmental justice indicators, race and ethnicity differed among regions and across system sizes, while wealth-related factors were less variable. Black and Hispanic residents made up a larger proportion of residents served by very large systems than smaller systems. Systems in the South served the highest proportion of Black residents, and CWSs in the West served the highest proportion of Hispanic residents. The proportion of residents with annual incomes below the U.S. Census poverty threshold did not vary considerably as a function of system size, while rates of home ownership were lowest in counties served by very large systems. Poverty rates were highest in the South and West, and rates of home ownership were lowest in the West.

We observed complex relationships between nitrate levels and system size. In the binary model, very small systems had a greater likelihood of high nitrate compared to small and medium systems. By contrast, very small systems were predicted to have lower concentrations than other system sizes when nitrate was evaluated as a continuous variable (compared to very small systems, nitrate concentrations were 9.0% higher in small systems and 51% higher in very large systems; Table 3). In unadjusted models, the same observations were noted; very small systems were predicted to have greater likelihood of high nitrate but lower continuous nitrate concentrations than all other system sizes (Additional file 1: Table S7). For other variables related to water system characteristics and land use, systems relying on groundwater sources had 34% higher nitrate concentrations compared to systems relying on surface water (95% CI: 26, 42%) and were more than four times as likely to have high levels of nitrate (odds ratio = 4.1; 95% CI: 3.0, 5.6). The extent of cropland coverage and livestock production had significant but small associations with nitrate concentrations, while a one-unit increase in the percent of land area used as cropland had a more substantial effect (3.3% increase, 95% CI: 3.0, 3.7%) on the likelihood that a CWS had high nitrate.

To evaluate whether differences between the city- and county-level analyses were related to the refined spatial scale of cities or trends specific to the portion of the country that provided city information, we conducted an additional analysis using county-level demographics for only those CWSs that provided city served information (“county-level subset”). Overall, the results of this county-level subset model were similar to the results of the nationwide analysis using county-level data. Although the magnitude of some coefficients in the county-level subset analysis changed relative to the national analysis (for instance, race/ethnicity had approximately half of the effect on nitrate concentrations and poverty had about twice the effect), the statistical significance and directionality of the predictors were comparable (Table 4). The results of this comparison suggest that the differences between the county- and city-level analyses are primarily due to different relationships between nitrate levels and demographic predictors at various spatial scales, rather than being an artifact of the part of the country evaluated in the subset analyses.

We included region as a covariate in our nationwide models because of regional differences in nitrate concentrations and demographic characteristics (Table 1) and significant differences among some regions in unadjusted models (Additional file 1: Table S7). In the unadjusted models, the Midwest and West had the highest proportions of high nitrate systems, while the West had higher nitrate concentrations than the Midwest and South. In the adjusted models, the Midwest had significantly lower nitrate concentrations than other regions, while only the West had a higher proportion of high nitrate systems than the Midwest. To investigate potential regional differences in associations among demographics, water system characteristics, land use, and nitrate concentrations, we stratified our nationwide model by region.

This study represents the first nationwide analysis of socioeconomic disparities in exposures to contaminants in public drinking water. We found that 5.6 million Americans relied on a public water supply with an average nitrate concentration ≥ 5 mg/L, one-half of U.S. EPA’s drinking water standard, over the five-year period spanning 2010–2014. Epidemiological studies have suggested that long-term exposure to water with nitrate concentrations above 5 mg/L may be associated with some types of cancer, birth defects, and preterm birth [19, 23, 27, 29]. We found that the proportion of Hispanic residents was significantly associated with nitrate levels, while the proportion of Black residents was inversely associated with nitrate levels. The associations with poverty and home ownership were mixed; when we used the demographics of the counties served by each water supply, we found that poverty was negatively associated with nitrate, while home ownership, an indicator of wealth and political empowerment, was inversely associated with nitrate when we used city-level demographics. Very small water systems (serving ≤ 500 people) had the highest nitrate levels overall, but after adjusting for demographics and local land use (cropland, livestock production, and urbanicity), very small systems were predicted to have lower nitrate levels than larger systems.

We found that the percent of Hispanic residents was associated with higher nitrate levels in our nationwide analysis and in all U.S. regions except the Midwest. These associations were modest; nationally, a 10% increase in the proportion of Hispanic residents (i.e., increasing from 10 to 20%) served by a CWS was associated with a 19.6% increase in nitrate concentration. Balazs et al. [6] also saw an association between percent Hispanic residents and nitrate levels in small public water supplies in California’s Central Valley, an agriculturally intensive area. We had hypothesized that proportion of Hispanic residents would be associated with nitrate because many agricultural communities have a high proportion of Hispanic residents. However, our observed association persisted even after we adjusted for agricultural activity by including cropland and livestock production as covariates in our models (Table 3) and our nationwide correlation analysis found a negative correlation between the proportion of Hispanic residents and percent cropland (Additional file 1: Table S2). These results suggest that the association between Hispanic residents and nitrate is not solely explained by proximity to agricultural sources, although adjusting for county-level cropland and livestock production will not eliminate residual confounding of the association between proportion of Hispanic residents and nitrate concentration by agricultural contamination of source waters since our covariates do not account for other factors such as soil type, rates of fertilizer use, and adoption of best management practices to control fertilizer runoff. Nevertheless, while agriculture is the largest source of land-based nitrogen inputs, major sources are also present in urban areas, including wastewater treatment plants, leaking sewer lines, and urban runoff [14]. Our correlation analysis found a positive correlation between percent Hispanic residents and percent urban households (Additional file 1: Table S2), suggesting that some of the association between Hispanic residents and nitrate levels may be related to nitrate sources in urban areas.

In addition to proximity to nitrate pollution sources, the observed relationship between proportion of Hispanic residents and nitrate may be indicative of disparities in TMF resources related to source water protection and water treatment. Communities with higher proportions of minority residents, particularly those who are non-native English speakers, may have less political influence and may be disenfranchised from political and budgetary decision-making processes [11], and therefore may have fewer resources to install new treatment technology or develop new source waters in response to contamination. In our analysis, we are not able to identify the relative importance of proximity to nitrate sources and management-related factors. Nevertheless, our findings are consistent with prior studies in which Hispanic communities were found to have higher drinking water exposures to arsenic, another contaminant regulated under the SDWA. In Oregon, communities served by CWSs in violation of the arsenic MCL had a much higher proportion of Hispanic residents [5], and in Arizona, the proportion of Hispanic residents served by a public water system was positively associated with the likelihood that that system violated the arsenic MCL [9]. Since arsenic in groundwater often comes from geogenic rather than anthropogenic sources [5, 7], these associations may be indicative of disparities in the ability of communities to afford enhanced drinking water treatment technologies, and taken together, suggest that Hispanic communities may experience elevated exposures to multiple drinking water contaminants.

In contrast to our results for Hispanic residents, we observed that the proportion of Black residents was inversely associated with nitrate on a national level, although this association was only observed in the South in our regional analysis. The inverse association observed nationally may be heavily influenced by the South, which has the lowest proportion of high nitrate systems, the most water systems, and the highest proportion of Black residents of all regions. In the South, the negative relationship between the proportion of Black residents and nitrate levels may be explained in part by biogeochemical factors. Pennino et al. [14] suggested that the lack of nitrate MCL violations in Louisiana, Mississippi, and Alabama—states that all have > 25% Black residents—may be associated with biological uptake and transformation processes and regional geological factors. The finding of no significant association between the proportion of Black residents and nitrate levels in the West is consistent with findings of Balazs et al. [6], who observed no significant relationship between non-Hispanic people of color and nitrate levels in California’s Central Valley.

The associations between wealth-related parameters (poverty and home ownership) and nitrate levels differed among U.S. regions and various spatial resolutions of the demographic data. In our national adjusted models using county-level demographics, poverty was inversely associated with nitrate levels while home ownership was not associated with nitrate. By contrast, when we used demographic data corresponding to cities and towns rather than counties, we observed that poverty was not associated with nitrate and that home ownership was inversely associated with nitrate, implying that cities and towns with higher proportions of renters tend to have higher nitrate levels. The effect estimates for poverty and home ownership varied between unadjusted and adjusted models, although our model building approach with a priori variable selection does not indicate which covariates were associated with the most substantial changes in these estimates. As with race/ethnicity, wealth can be expected to relate to levels of contamination in two ways: proximity to pollution sources and ability to treat contaminated source water. While we adjusted our models for agriculture and urbanicity, these variables may not have captured proximity to other important sources of nitrate inputs, such as landfills, industrial facilities, fossil fuel combustion, and home building [36, 45]. Such factors may be related to wealth to the extent that they are driven by economic activity. Discrepancies between our city- and county-level analyses could be explained if county-level wealth operated in our model as the best proxy for nitrogen sources not accounted for by other variables, while city-level wealth better represented civic engagement, capacity to raise customer rates, and, by consequence, ability to treat contaminated water.

We hypothesized that smaller water systems would have higher nitrate concentrations. Smaller water systems may have fewer financial and technical resources to address contamination issues when they arise [1], and the cost of water treatment per household is considerably higher for smaller systems because of a lack of economies of scale [9]. Indeed, very small systems had higher 95th percentile nitrate concentrations than larger systems and were more likely to exceed 5 mg/L nitrate in unadjusted models. Additionally, after adjusting for demographic and land use parameters, very small systems were more likely to have high nitrate compared to small and medium sized systems. However, in our regression analyses with nitrate as a continuous variable, very small systems had lower nitrate concentrations compared to larger system size categories nationwide (Table 3) and in the Midwest, Northeast, and West (Table 5). Taken together, these results suggest that larger systems have higher nitrate on average, but that very small systems are more likely than other system sizes to have nitrate concentrations at the high end of the distribution. Previous studies have found inconsistent relationships between system size and contaminant violations. Switzer and Teodoro [8] identified a negative relationship between the population served by a system and the system’s number of health-based SDWA violations (MCL and treatment technique violations) in a subset of CWSs across the U.S., while Rahman et al. [46] reported a positive association between MCL violations and the number of people served by water systems in Arizona. In a purely statistical sense, larger water systems may be more likely to detect elevated nitrate levels because they are required to test more frequently and, in the case of groundwater systems, might draw from a greater number of source water wells.

Beyond health-based violations, very small CWSs were reported to have more frequent violations of monitoring and reporting requirements than larger systems [47]. This observation is consistent with our data: of the 41,781 CWSs we paired with county demographics, very small systems were significantly more likely than larger systems to lack nitrate sample results over the five-year study period (5.2% of very small systems missing nitrate data compared to 1.8% of larger systems; Pearson χ² = 301). This difference is unlikely to be due to differences in testing requirements, since CWSs are required to test for nitrate annually or more frequently [48]. Such difficulties in adequately monitoring drinking water contaminants likely stem from limited financial resources and/or managerial expertise, and may signal concurrent challenges in conforming to SDWA health-based guidelines. In this regard, Balazs and Ray [11] reported that very small water systems in Fresno County, California, that had failed to monitor for drinking water contaminants under county governance were found to have MCL violations when state officials investigated.

Strengths of our study include the extensive scale and completeness of our dataset for both demographics and water quality data, and our use of information about purchasing water systems to link water quality data with entire areas served by CWSs. A major limitation to our analysis is the potential for exposure misclassification. Because we lacked information about flow volumes from multiple sources within CWSs, we weighted all samples collected for each CWS equally. In some cases, this may have led to overestimates of nitrate concentrations in systems where more contaminated sources are pumped less frequently or only maintained for backup; this overestimation may affect groundwater systems more since we anticipate that they may have more intake points than CWSs with surface water sources. Furthermore, nitrate concentrations may show substantial intra-annual variability, so depending on when nitrate samples were collected within each year, our aggregated metrics may not capture the true average nitrate concentration for each CWS. As with any study of population-level data, we are limited in our ability to draw conclusions about individual exposures and thus limited in our ability to infer causal relationships between the EJ variables of interest and exposure to nitrate in drinking water. Nitrate levels in CWSs may not accurately reflect the exposures among residents in those areas because some residents may rely on bottled water. Use of bottled water may exacerbate disparities in pollutant exposures because lower-income residents have lower ability to pay for bottled water. Another limitation was that we lacked information about cities and towns served for about half of the CWSs in our analysis and so we had to rely on county-level demographics in most analyses, which is a limitation because demographic data for a county may not accurately reflect the demographics of all cities and towns within each county. Our analysis using city-level demographics is limited in its geographic scope, although this did not seem to account for the differences in model results using city- and county-level demographic data. Ideally, we would use census block level information to provide the best resolution of demographic data. However, because geocoded information that specifies the CWS serving each census block is not available, we could not analyze data at the census block level. For small CWSs that serve part of a large city, we used the demographics of the whole city, but the demographics of the population served by the CWS may vary from those of the city overall. Some CWSs were not included in our analysis because we could not match the names of the cities and towns served with locations in the Census Bureau data, although this accounted for a very small proportion of systems. We also could not include unincorporated areas in our city-level analysis because they are not included in Census Bureau data. Finally, we were unable to compile nitrate data for all CWSs. In some cases, this was related to data handling problems; for instance, some records were only available in paper reports, and in other cases, this may reflect a lack of compliance with monitoring requirements, which is more likely for small rural communities who may not be able to afford testing or where there is less enforcement of testing requirements.

Our study did not include the 44 million Americans who rely on a private well for their drinking water, for whom water quality testing is not required under the SDWA. Private wells are shallower than public wells, and shallow wells are more vulnerable to nitrate contamination [49]. They are also more likely to be located in rural areas and may be in closer proximity to agriculture and livestock production sources. Private well owners are usually not required to test for nitrate or other drinking water contaminants, so their presence may go undetected. Among nearly 4000 private wells tested in rural Wisconsin by a state water quality laboratory, nearly 10% exceeded the nitrate MCL [50]. There is evidence of environmental justice disparities in communities using private wells or lacking piped-water entirely. A review by VanDerslice [41] summarized case studies of minority communities reliant on contaminated private wells. For instance, in a low-income Hispanic community of 25,000 in the Yakima Valley in Washington State, more than 10% of private wells exceeded the nitrate MCL [51]. These case studies are further evidence of impaired water quality in communities reliant on private wells and indicate potential socioeconomic disparities in these communities as well.

This study represents the first nationwide analysis of socioeconomic disparities in exposures to drinking water contaminants, and the framework that we developed in this study can be extended to investigate disparities in exposures to other drinking water contaminants. We found that communities with higher proportions of Hispanic residents tend to be served by community water systems with higher nitrate and greater likelihood of being over 5 mg/L. Our regression analyses indicate that this association is not completely explained by proximity to cropland and livestock production. While > 99% of CWSs do not exceed the nitrate MCL of 10 mg/L, 5.6 million Americans are served by CWSs with nitrate concentrations above 5 mg/L. Nitrate data for private wells, which are even more vulnerable to nitrate contamination, are lacking. Understanding the extent of current exposures, particularly among vulnerable subpopulations, is critical for developing effective strategies to reduce exposures in these communities. Our findings suggest that programs intended to help low-income and small CWSs may not be adequately assisting communities with high proportions of Hispanic residents. Epidemiological evidence for adverse health effects associated with consumption of drinking water above 5 mg/L nitrate raise concerns about increased risk in people exposed at this level and support a re-evaluation of the federal nitrate standard. Even well below the standard, nitrate levels of 1 mg/L or higher are associated with anthropogenic impact; thus nitrate may be an inexpensive indicator to identify drinking water systems that may also contain other contaminants of concern.