National trends in drinking water quality violations

Maura Allairea,1, Haowei Wub, and Upmanu Lallb,c

a
Department of Urban Planning and Public Policy, University of California, Irvine, CA 92697; bColumbia Water Center, Columbia University, New York,
NY 10027; and cDepartment of Earth and Environmental Engineering, Columbia University, New York, NY 10027

Edited by Susan Hanson, Clark University, Worcester, MA, and approved January 9, 2018 (received for review November 16, 2017)

Ensuring safe water supply for communities across the United a strategic objective to achieve consistent compliance for 91% of the
States is a growing challenge in the face of aging infrastructure, population served by CWSs (4). However, from 1993 to 2009, com-
impaired source water, and strained community finances. In the pliance with Safe Drinking Water Act (SDWA) health-based regu-
aftermath of the Flint lead crisis, there is an urgent need to assess lations ranged from 79 to 94% of the population served by CWSs (4).
the current state of US drinking water. However, no nationwide Targeting utilities that are underperforming is one approach
assessment has yet been conducted on trends in drinking water to improve compliance and consistently provide safe drinking
quality violations across several decades. Efforts to reduce viola- water. Currently, state enforcement agencies lack a systematic
tions are of national concern given that, in 2015, nearly 21 million procedure to select systems for additional monitoring and in-
people relied on community water systems that violated health- spection. Routine quality monitoring is specified at the federal
based quality standards. In this paper, we evaluate spatial and level and allows for more frequent sampling at systems with re-
temporal patterns in health-related violations of the Safe Drinking cent violations. However, no guidelines exist for identifying sys-
Water Act using a panel dataset of 17,900 community water tems without recent reported violations that could benefit from
systems over the period 1982–2015. We also identify vulnerability additional oversight. Identifying hot spots and vulnerability fac-
factors of communities and water systems through probit regres- tors associated with violations could better direct enforcement
sion. Increasing time trends and violation hot spots are detected in
activity to struggling utilities and allow for increased compliance
several states, particularly in the Southwest region. Repeat viola-
across the country.
tions are prevalent in locations of violation hot spots, indicating
This paper presents a national assessment of trends in drinking
that water systems in these regions struggle with recurring issues.
water quality violations across several decades. Currently, there
In terms of vulnerability factors, we find that violation incidence in
is not a good understanding of quality violations and few peer-
rural areas is substantially higher than in urbanized areas. Mean-
while, private ownership and purchased water source are associ-
reviewed studies have been done on SDWA compliance. Past
ated with compliance. These findings indicate the types of
analyses of the association between water system characteristics
underperforming systems that might benefit from assistance in and violations have generally been limited in terms of geographic
area and/or study period. Several studies focus on a single state
Downloaded from https://www.pnas.org by 219.75.104.102 on February 4, 2024 from IP address 219.75.104.102.

achieving consistent compliance. We discuss why certain violations

might be clustered in some regions and strategies for improving (e.g., refs. 3 and 5–7). The two studies in the peer review liter-
national drinking water quality. ature that address SDWA violations nationwide have study pe-
riods limited to a single year (8) or 7 y (9). Rubin (8) describes
|
drinking water water quality | violations | Safe Drinking Water Act | summary statistics of violations for four SDWA rules for the year
regulatory compliance 2011. A key finding is that similar proportions of small and large
systems incur health-based violations. However, given that only

E nsuring access to safe drinking water poses a challenge for

US water systems in the face of aging infrastructure, im-
paired source water, and strained community finances. Com-
Significance

munities across the country have been impacted by recent cases Drinking water contaminants pose a harm to public health.
of impaired water quality. The Flint crisis exposed as many as Some can cause immediate illness, such as the 16 million cases
98,000 residents to elevated levels of lead, disinfection by- of acute gastroenteritis that occur each year at US community
products (DBPs), and Escherichia coli and Legionella bacteria. water systems. Here, we show that health-based drinking
In addition, substantial populations have coped with interrup- water quality violations are widespread, with 9–45 million
tions to potable water supply due to the Elk River chemical spill people possibly affected during each of the past 34 years.
in West Virginia and unregulated toxins formed during algal While relatively few community water systems (3–10%) incur
blooms near Toledo, Ohio. These well-publicized events motivate health-based violations in a given year, improved compliance is
needed to ensure safe drinking water nationwide. Currently,
the need to assess the current state of US drinking water quality.
state enforcement agencies lack a systematic procedure to se-
How widespread are violations? Are violations more prevalent in
lect systems for additional inspection and monitoring. We
vulnerable communities, such as low-income and rural areas? Do
identify hot spots and vulnerability factors associated with
utility characteristics, such as private ownership and origin of violations, which can allow public policies to target under-
source water, influence the likelihood that violations occur? performing water systems.
Identifying threats and potential improvements to water ser-
vices are needed, given that an estimated 16.4 million cases of Author contributions: M.A. and U.L. designed research; M.A. and H.W. performed re-
acute gastroenteritis each year in the United States are attrib- search; M.A. and H.W. analyzed data; and M.A. wrote the paper.
uted to community water systems (CWSs) (1). Equity concerns The authors declare no conflict of interest.
are also gaining recognition as evidence builds regarding lower- This article is a PNAS Direct Submission.
income and minority communities receiving poor quality water Published under the PNAS license.
(2, 3). Generally, water systems in the United States provide Data deposition: Data associated with this research is available at Dataverse (https://
reliable and high-quality drinking water. Violations tend to be dataverse.harvard.edu/dataverse/allaire).
infrequent. However, in a given year, about 7–8% of CWSs re- 1
To whom correspondence should be addressed. Email: [email protected].
port at least one health-based violation. While this rate is rela- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
tively low, improved compliance is needed to ensure safe drinking 1073/pnas.1719805115/-/DCSupplemental.
water nationwide. The Environmental Protection Agency (EPA) has Published online February 12, 2018.

2078–2083 | PNAS | February 27, 2018 | vol. 115 | no. 9 www.pnas.org/cgi/doi/10.1073/pnas.1719805115

 summary statistics are presented, results do not isolate the as- housing density, and percent nonwhite population. Since CWSs are not re-
sociation between violations and specific utility characteristics. quired to collect demographic information about customers, census in-
Meanwhile, Wallsten and Kosec (9) develop count regressions formation can provide some insight into possible association between water
quality violations and community characteristics. Data availability was suf-
that relate community and water system characteristics to
ficient to interpolate values for intercensus years. A more detailed de-
SDWA violations from 1997 to 2003. The study found that pri-
scription of the data is provided in SI Data.
vate ownership does not generally affect compliance.
In this study, we assess spatial and temporal trends in health- Hot-Spot Analysis of Violations. We use spatial autocorrelation to describe and
based violations as well as vulnerability factors that may influ- map spatial clusters of SDWA violations. The local Gi*(d) statistic (local Getis–
ence violation occurrence. We assemble a panel dataset that Ord statistic) is used to test the statistical significance of local clusters (13).
includes 17,900 CWSs in the United States from 1982 to 2015. For each county, we estimate a local Getis–Ord statistic as a function of
We combine demographic information from the US Census with distance between neighboring counties and violation occurrence. Further
data on SDWA violations and CWS characteristics from the description is provided in SI Materials and Methods.
EPA Safe Drinking Water Information System (SDWIS). Spatial
trends and hot spots are assessed via local spatial autocorrela- Vulnerability Factors Associated with Violations. Probit regression is used to
tion. Vulnerability factors, which include CWS and demographic assess the relationship between drinking water quality violations and char-
characteristics, are assessed using probit regressions. Our main acteristics of water systems and communities. The likelihood of a water
supplier incurring a health-based quality violation is modeled as follows:
research objectives are to (i) characterize the extent and severity
of drinking water quality violations across the United States, and  
Prðyit = 1jXÞ = Φ β0 + βx xi + γ jt Cjt + αt Tt + ϕk Sk , [1]
(ii) identify characteristics of water systems and communities
associated with violations that can be used to better target where yit is a binary indicator of violation occurrence for water supplier i in
enforcement activity. year t. The probability of a CWS having a violation is estimated as a function
of time-invariant water supplier characteristics (xi) and county-level com-
Background munity characteristics (Cjt) for county j in year t. Each water supplier i is as-
The SDWA, passed in 1974, authorizes the EPA to regulate sociated with a county j and state k. The model includes dummy variables for
drinking water quality. Standards are set at the federal level and each year (Tt) and state (Sk). Year dummy variables control for changes in
states have primary responsibility for enforcement. When a water federal regulations or compliance over time. Meanwhile, controlling for
system fails to meet an EPA-mandated drinking water standard, state-level effects is important because the SDWA is enforced primarily at
a drinking water violation can be issued. Violation occurrence the state level. State-level enforcement can differ due to variation in sam-
pling protocols, technical capacity, and financial resources. The coefficient
and reporting may differ across states due to differences in en-
for the state dummy will then indicate the average propensity for water
forcement, financial resources, source water quality, and treat-
quality violations being reported in a state due to either water quality
ment costs (10). In addition, states are able to set stricter violations or enforcement effort. Covariates included in the probit model
standards than federal SDWA rules. were selected based on a careful review of the literature, lasso regression,
Downloaded from https://www.pnas.org by 219.75.104.102 on February 4, 2024 from IP address 219.75.104.102.

Not all violations pose immediate health concerns. This study and specification tests.
focuses on health-based violations, which include maximum Models are specified for total violations and for total coliform violations.
contaminant levels (MCLs), maximum residual disinfectant lev- The total coliform models may provide more accurate estimates of associa-
els, and treatment techniques. Standards for contaminants that tions between violations and covariates since (i) total coliform is regulated in
can trigger health-based violations are specified by the National a consistent manner throughout the study period and (ii) Total Coliform
Primary Drinking Water Regulations. We exclude monitoring Rule (TCR) violations are reported more accurately than other types of MCL
and reporting violations, which have poor data quality since only violations (12). We estimate coefficient values and average marginal effects.
a small portion of these violations (9–23%) are reported to Average marginal effects are useful for interpretation; they provide a single
the national SDWIS database (11). Both naturally occurring estimate of the effect of each covariate on Pr(Y = 1). For more detail on our
regression analysis and calculation of average marginal effects, see SI Ma-
and manmade contaminants can trigger health-based violations.
terials and Methods.
These contaminants or contaminant indicators include total co-
liform, turbidity, DBPs, radionuclides, and organic and inorganic Results
chemicals.
Summary Statistics. We find that water quality concerns extend
Materials and Methods well beyond Flint. In 2015, 9% of CWSs in our study sample

SUSTAINABILITY
Data. We construct a balanced panel dataset that includes CWSs with con- violated health-based water quality standards, affecting nearly

SCIENCE
sistent reporting to the EPA SDWIS over the study period, 1982–2015. To be 21 million people. During each of the past 34 y, 9–45 million
included in the study, a public water system must be classified as a CWS, people were affected, representing 4–28% of US population.
serve over 500 people, begin reporting violations in 1982 or prior, and be Our balanced panel dataset contains 34 y and 17,900 CWSs,
located within the continental United States. Violation records and CWS which serve 87% of the population supplied by CWSs in the
characteristics were obtained from the SDWIS. Community characteristics at continental United States. Full summary statistics and variable
the county level were obtained from the US Census. The unit of analysis in
definitions are provided in Table S1. About 8.0% of the
this study is the utility year.
608,600 utility-year observations had some type of health-based
Violation records from the SDWIS indicate whether or not a MCL was
exceeded or a treatment technique was not met. Contaminant concentra-
violation, while 4.6% had a total coliform violation. In total,
tions or sampling results for regular monitoring are not available. Another there were 95,754 health-based violations and total coliform was
limitation of the SWDIS is underreporting of some violations. To address the most prevalent type of violation, representing about 37% of
underreporting, we emphasize the regression results for total coliform vio- all violations (Fig. 1). Fecal coliform violations are relatively
lations, which are more accurately reported than other types of violations rare; only 2,138 occurred during our study period. Violations
(12). Also, we exclude very small systems serving fewer than 500 people since categorized as “other” contaminants were also prevalent, rep-
these are more likely to have inadequate reporting practices (8). Overall, resenting about 36% of violations. DBPs comprised the majority
these violations data represent the best national information that is cur-
of other violations. Violations of treatment rules and nitrate are
rently available for drinking water quality.
much less common (21% of total). Descriptions of violation
Water system characteristics include type of source water, service pop-
ulation, and ownership type. County-level locations of each CWS are defined categories are provided in Table 1 and SI Data.
based on the CWS address provided in the SDWIS. Community characteristics Differences in the rate of violation occurrence exist across
are county-level variables from the US Census and represent annual values for characteristics of CWSs and communities (Table S2), as dis-
the 1982–2015 study period. Characteristics include household income, cussed in SI Results.

Allaire et al. PNAS | February 27, 2018 | vol. 115 | no. 9 | 2079
 rose after 2002, when the Stage 1 DBPs Rule became enforce-
able. The time dummy variables in our regression analysis cap-
ture federal rule changes, as evidenced by positive and significant
coefficient values after major regulation changes (Fig. S3).
Substantial differences across time also exist between rural
and urban areas, in terms of number of violations per CWS (Fig.
3). Furthermore, low-income rural areas have a larger compli-
ance gap than higher-income rural areas. DBP violations account
for much of this gap. Differences between rural and suburban
areas became pronounced and statistically significant after new
DBP rules in the early 2000s.

Spatial Trends. Violations also vary considerably across geographic

locations. Fig. 4A shows a map of the total number of violations
Fig. 1. Number of health-based violations, 1982–2015, by contaminant (1982–2015) per CWS in a given county. Some of the counties with
type. the highest prevalence of violations are rural, located in Texas,
Oklahoma, and Idaho. Total coliform violations (Fig. 4B) are es-
Time Trends. We find that significant time trends exist for some pecially prevalent in the West and Midwest. A variety of factors
states. State trends are estimated as deviations from the national could contribute to differences across counties, including the
time trend (Table S5). Twenty states deviate above the national quality of source water and state-level enforcement.
trend, including Nebraska, Arkansas, North Carolina, and Fig. 4 depicts high violation prevalence throughout Oklahoma
Southwest states—New Mexico, Texas, and Oklahoma (Fig. S4). and parts of Idaho. In some areas, total coliform comprises
Only three states are below the national time trend—Wash- a large portion of total violations. This is especially true in
ington, Iowa, and Illinois (Fig. S4A). Similar results are found for Nebraska (56% of all violations during the study period are total
time trends of total coliform violations, except more states have a coliform), Missouri (42%), New Hampshire (35%), and Mas-
declining trend (Fig. S4B). At the national level, the linear time sachusetts (81%). There are several locations that have high
trend of violations is insignificant once we control for the year prevalence of total violations, but not total coliform, such as
dummy variables (Table S5). Texas and Kansas. The proportion of total coliform violations
Violation occurrence has fluctuated over the history of the has implications for interpreting regression results from models
SDWA. Fluctuations in number of violations can be attributed to specified for total coliform and total violations.
a variety of factors, including regulatory changes, enforcement Spatial differences in violation incidence vary considerably
Downloaded from https://www.pnas.org by 219.75.104.102 on February 4, 2024 from IP address 219.75.104.102.

capacity, raw water quality, and treatment capabilities. Fig. 2 across time. This is particularly true for the Southwest region
depicts the number of health-based violations across the CWSs (includes Arizona, Oklahoma, New Mexico, and Texas), which
included in the analysis. Total violations more than double over had similar violation incidence as other regions until the early
the 34-y study period. This increase is partially driven by new 2000s. After the Stage 1 DBP rule in the early 2000s, violation
regulations. The number of regulated contaminants has dra- incidence dramatically increased and is more than triple that of
matically increased from 22 (in 1974) to 91 (at present). Table 1 other regions (Fig. S1). DBPs account for the vast majority of
provides a list of EPA rules and enforcement dates. violations in the Southwest. This region might be particularly
Spikes in violations appear to occur immediately after new susceptible to DBP violations due to high summer temperatures
federal regulations. Utilities might undergo a learning and ad- (14) and high levels of total organic carbon in source water.
justment process when coping with new regulations. For exam- Furthermore, Oklahoma requires a high level of minimum dis-
ple, violation counts of DBPs and other chemicals dramatically infectant residual (1.0 mg/L total chlorine). This could cause

Table 1. Description of categories of drinking water quality violations

Study category Contaminants SDWA rules

Total coliform Fecal coliform (E. coli) NIPDWR, TCR, RTCR

Total coliform
Treatment rules plus nitrate Nitrate NIPDWR
Nitrite Phase II
Turbidity NIPDWR, SWTR, IESW, LT1ESW
Cryptosporidium IESW, LT1ESW, LT2ESW
Giardia lamblia SWTR, GWR
Viruses SWTR, GWR
Arsenic Arsenic NIPDWR, Arsenic
Lead plus copper Lead, copper NIPDWR, LCR, RLCR
Other DBPs Trihalomethanes, Stage 1 DBPs, Stage 2 DBPs
Radionuclides NIPDWR, Radionuclides
Inorganic and organic chemicals NIPDWR, Phase I, Phase II, Phase V

Notes: SDWA rule names and enforcement dates are as follows: Arsenic, Arsenic Rule (2006); GWR, Ground Water Rule (2009); IESW,
Interim Enhanced Surface Water Treatment Rule (2002); LCR, Lead and Copper Rule (1992); LT1ESW, Long Term 1 Enhanced Surface
Water Treatment Rule (2005); NIPDWR, National Interim Primary Drinking Water Regulations (1977); Phase I Rule (1989); Phase II Rule
(1992); Phase V Rule (1994); Radionuclides, Radionuclides Rule (2003); RLCR, Revised Lead and Copper Rule (2007); RTCR, Revised Total
Coliform Rule (2016); Stage 1 DBPs, Stage 1 Disinfectants and Disinfection By-products Rule (2002–2004); Stage 2 DBPs, Stage 2
Disinfectants and Disinfection By-products Rule (2012–2013); SWTR, Surface Water Treatment Rule (1993); TCR, Total Coliform Rule
(1990); and Trihalomethanes, Total Trihalomethanes (1981–1983).

2080 | www.pnas.org/cgi/doi/10.1073/pnas.1719805115 Allaire et al.

 Fig. 2. Number of health-based violations, 1982–2015. Gray bars represent dates that major components of rules became enforceable.

DBP issues during high temperatures, when effective chlorine privately owned utility is 0.5% lower than a government-owned
levels decline and additional disinfectant might be needed. utility. A larger association was found for total violations (1.0%).
This result might be attributable to private firms facing the
Violation Hot Spots. Intense hot spots of SDWA violations are possibility of takeover by a municipal government, which could
identified in Oklahoma and parts of Texas, based on local spatial lead to greater compliance (15). Large private utilities in particular
autocorrelation (Fig. 5). While these findings are somewhat have considerable resources at stake should they deliver poor-
similar to Fig. 4A, the hot-spot analysis offers the advantage of quality water and face lawsuits or takeover. Our coefficient esti-
determining whether clusters of violations are significant. Similar mates suggest that the interaction term between private and large
spatial clusters are found for violation counts per CWS and for firms is significantly associated with fewer violations (Table S3).
binary indicators of violation occurrence. Hot-spot locations Less urbanized areas are associated with greater likelihood of
tend to include water systems with repeat violations. In our violation occurrence (Table S6). Meanwhile, our indicator of mi-
Downloaded from https://www.pnas.org by 219.75.104.102 on February 4, 2024 from IP address 219.75.104.102.

overall sample, only 11% of the CWSs have repeat violations, nority, low-income populations is associated with higher likelihood
which involve two or more subsequent years of a violation. The of total coliform violations. These findings might indicate envi-
states with the greatest proportion of CWSs with repeat viola- ronmental justice concerns for rural areas and minority commu-
tions are Oklahoma (43% of CWSs in the state), Nebraska nities. Further research is warranted to understand environmental
(35%), and Idaho (33%). justice issues and which communities are particularly at risk.
When assessing hot spots over time, we find that spatial clusters Utilities in more rural, less urbanized areas tend to have less
of violations have shifted during the 34-y study period (Fig. S2). In capacity to comply with quality regulations and face financial
the earliest decade of the study period (1982–1992), significant hot strain due to declining populations and lower incomes (16).
spots of violation counts are present in the Northwest, southern Furthermore, small, rural CWSs might especially struggle, given
California, and Pennsylvania. Meanwhile, in the following decade that we estimate the highest predicted probability of violation
(1993–2003), the magnitude and significance of the estimated occurs at small, rural CWSs relying on surface water sources
z-scores increases in Oklahoma, Tennessee, and Idaho. Finally, (Fig. S5). Last, we find larger negative associations for total
in more recent years (2004–2015), significant hot spots appear
in Texas and increase in parts of Oklahoma.

SUSTAINABILITY
Regression Results. Regression results are presented both as co-

SCIENCE
efficient estimates (Table S3) and average marginal effects
(Table S4). Our major findings are that violation occurrence is
significantly associated with less urbanized areas and lag viola-
tions, while compliance is associated with purchased water
source and private ownership.
Violations in the prior year are significantly associated with
current year violations. This provides further evidence that re-
curring violations at a given water system are a concern. Repeat
violations are prevalent in locations of spatial clustering of viola-
tions, as previously discussed in the section on violation hot spots.
Water systems that purchase treated water from other utilities
show a lower propensity for violations. Wholesale water providers
may have greater capacity to achieve regulatory compliance (9).
The probability of a total coliform violation for a utility purchasing
water is 0.9% lower than a utility with a groundwater source (Table
S4). Similar results are found for total violations, although the Fig. 3. Total violations per water system, by housing density category and
estimated marginal effects are smaller in magnitude. income group. Low-income counties have median household income below
Privately owned utilities appear to be less vulnerable to vio- 75% of national median household income. In year 2015, national median
lations than public ownership, which agrees with previous find- household income was $55,775 and 45% of rural CWSs are located in
ings (5, 9, 15). The probability of a total coliform violation for a counties defined as low-income.

Allaire et al. PNAS | February 27, 2018 | vol. 115 | no. 9 | 2081
 Water systems in these locations are prone to recurring issues, as
evidenced by our regression analysis that found violations in the
prior year are significantly associated with current year viola-
tions. Repeat violations have been a focus of EPA regulation in
recent years.
Beyond temporal and spatial trends, we also find major vul-
nerability factors. Violation occurrence is found to be associated
with rural areas, while compliance is associated with purchased
water source and private ownership. These findings indicate the
types of communities that might benefit from greater regulatory
oversight and assistance in achieving consistent compliance with
water quality standards.
Regulatory compliance can be a challenge for rural systems
due to limited financial resources and technical expertise. A
smaller customer base can mean less revenue for infrastructure
improvements, repayment of debt, and salaries to attract tech-
nically skilled operators. Small systems also face restricted access
to loans and outside financing, compared with larger systems
with higher credit ratings (18). This presents a challenge for
complying with stricter standards, especially those that require
Fig. 4. Number of violations per CWS, 1982–2015, by county. (A) Total vi- capital modifications or new operational procedures. A new rule
olations. (B) Total coliform violations. Intervals in legend are selected based for DBPs in the early 2000s led to a dramatic increase in viola-
on the Jenks natural breaks classification method. tions in rural areas. The compliance gap was especially large
between low-income rural areas and more urban counties. We
violations, compared with total coliform. This might be partly also find that low-income, minority communities may face higher
attributable to rural systems facing challenges in complying with likelihood of certain violations, such as total coliform.
DBP rules. Lower housing density and water flows through dis- Meanwhile, water systems that rely on purchased water show a
tribution networks can lead to older “water age,” which enables lower propensity for violations. Purchased water is produced by
DBP formation (17). Rural systems may also have less technical wholesale water providers, which may have greater capacity to
capacity and financial resources to implement infrastructure achieve regulatory compliance (9). In addition, privately owned
upgrades or frequent flushing of the system. utilities are found to be less vulnerable to violations than gov-
Downloaded from https://www.pnas.org by 219.75.104.102 on February 4, 2024 from IP address 219.75.104.102.

In our analysis, urbanization is the combination of housing ernment ownership. In particular, large private firms are associated
density, median household income, and percent nonwhite pop- with lower likelihood of violation. This might be attributable to
ulation. It is not informative to interpret the individual co- large private firms having considerable resources at stake should
efficient estimates of these three variables since they are they deliver poor-quality water and face lawsuits or takeover by
moderately correlated. The correlation matrix of coefficient es- a municipal government.
timates for model 2 indicates correlations between housing Several policy implications emerge from our findings. First,
density and median income (r = −0.54) as well as housing density prioritization of technical guidance and financial assistance
and nonwhite population (r = −0.37). Therefore, we conduct could benefit underperforming systems. Expansion of training
principal-component regression to address these correlations, as and assistance could address a variety of operational issues, in-
described in SI Results. In this way, the association between vi- cluding source water protection and development of procedures
olations and these three correlated variables can be better un- for monitoring and maintenance, which are the most common
derstood since the variables are transformed intro uncorrelated system deficiencies (19). Second, merging and consolidation of
components. We interpret the first principal component (PC) as systems, where feasible, could provide a way to achieve econo-
an indicator of urbanization; it represents the average effect of mies of scale for adequate treatment technologies. Feasibility of
all three variables. Meanwhile, the second PC represents the consolidation will be influenced by existing infrastructure, dis-
contrasting effect, and we interpret it to be an indicator of mi- tance, and liability concerns. The electricity sector has undergone
nority, low-income populations. substantial restructuring since the mid-1990s, through consolida-
In all models, the P value of the likelihood ratio χ2 is less than tion and separation of generation and delivery. Third, purchased
0.01, which indicates that each model as a whole is statistically water contracts might provide a cost-effective way for utilities to
significant. For all model specifications, bootstrapped estimates comply with regulations, especially small water systems.
produce similar results. Our preferred specifications are the total
coliform models with interaction terms (models 4 and 6), since total
coliform violations have more accurate reporting and are regulated
more consistently than other violations during the study period.

Discussion
Targeting utilities that are underperforming is one approach to
improve compliance and consistently provide safe drinking water.
Identifying hot spots and vulnerability factors associated with vio-
lations could better direct enforcement activity and inform the
allocation of federal grant funds that assist state-level enforcement.
Intense hot spots of SDWA violations and increasing time
trends are detected in some states, especially Southwest states Fig. 5. Spatial clusters (hot spots) of health-based violations, 1982–2015.
such as Oklahoma and Texas. Notably, repeat violations are Hot spots for number of total violations per CWS, by county. Intervals in
prevalent in locations of intense spatial clustering of violations. legend are selected based on the Jenks natural breaks classification method.

2082 | www.pnas.org/cgi/doi/10.1073/pnas.1719805115 Allaire et al.

 These findings are relevant for larger questions regarding reported to the national SDWIS database. Capturing a more
national-level water quality regulations: (i) How can regulatory complete record of violations is crucial for addressing potential
rules consider linkages between the SDWA and the Clean Water public health concerns. A major barrier to improved reporting is
Act?, (ii) Should drinking water standards be set at a subnational substantial reductions in state and federal funding for enforcement
level?, and (iii) What strategies can improve data collection and activities. State funding declined a third from 2001 to 2011, while
monitoring of violations? workloads have increased due to newly promulgated rules (21).
Regulation and enforcement activities that better consider the Beyond restored funding, technology developments could en-
linkages between the SDWA and the Clean Water Act could able improved monitoring and data management. In-pipe sensors
address concerns over impaired source water. Vulnerable com- may offer a game-changing way to continuously monitor water
munities may face the additional challenge of rising compliance
quality within a distribution system. However, sensors are in the
cost that is driven by source water impairment. Further research
is needed regarding the extent to which treatment costs at public early stages of development and their use is rare in practice due
water systems might be affected by emerging contaminants and to low accuracy and high costs of capital and operation. A more
potential changes to Clean Water Act standards or enforcement. immediate development is improved data management. A new
Considering the interactions of state and federal regulations is cloud-based reporting system, SDWIS Prime, might allow greater
also crucial. For example, the combined effects of state-level consistency in reporting across states. Currently, each state main-
residual disinfectant requirements, poor quality source water, tains its own database, which can lead to inaccuracies during data
and high summer temperatures might cause the Southwest re- transfer from utilities to the state to the federal SDWIS. This new
gion to be particularly susceptible to DBP violations. system also automates compliance determination, which can re-
National SDWA rules ensure that all water systems share the duce labor expense. In combination with cloud-based reporting,
same minimum water quality standards. However, variation in algorithms should be developed that can automatically check for
standards across states or municipalities might increase effi- outliers. This will facilitate analytics on large volumes of data and
ciency since this would reflect local preferences for trade-offs allow emerging problems to be identified at an individual utility or
between risk reduction and intervention cost (20). Subnational across a region.
standards might allow net benefits to be realized in specific lo- Overall, this study informs a more directed approach to increase
cations, since net benefits will vary based on compliance cost, con- compliance with drinking water quality regulations. While quality
taminant prevalence, vulnerable populations, and other factors. In violations are generally infrequent, some do reflect a risk to human
addition, subnational rules might allow a broader range of contam-
health. Identifying hot spots and vulnerability factors associated
inants to be regulated, especially those that only affect specific lo-
with violations can allow public policies to target underperforming
calities. At present, the SDWA only regulates contaminants that
frequently occur at harmful levels across the country. An important water systems. Reducing water quality violations can lead to im-
area for future research is assessing the welfare implicates of uni- proved health outcomes and less disparity in water service.
Downloaded from https://www.pnas.org by 219.75.104.102 on February 4, 2024 from IP address 219.75.104.102.

form, national-level standards compared with subnational standards.

ACKNOWLEDGMENTS. The manuscript has benefited from helpful discus-
Last, there is a nationwide need for improved data collection sion and comments from Katherine Alfredo, William Becker, Alan Roberson,
and monitoring of violations. Underreporting of SDWA viola- two anonymous referees, and the associate editor. Financial support is
tions deserves attention, given that an estimated 26–38% of gratefully acknowledged from a NatureNet Science Fellowship, National
health-based violations are either not reported or inaccurately Science Foundation Award 1360446, and an Earth Institute Summer Grant.

1. Messner M, et al. (2006) An approach for developing a national estimate of water- 12. US Environmental Protection Agency (2002) Data reliability analysis of the EPA safe
borne disease due to drinking water and a national estimate model application. drinking water information system/federal version (SDWIS/FED) (US Environmental
J Water Health 4:201–240. Protection Agency, Washington, DC), EPA 816-R-00-020.
2. Stillo F, MacDonald Gibson J (2017) Exposure to contaminated drinking water and 13. Getis A, Ord J (1992) The analysis of spatial association by use of distance statistics.
health disparities in North Carolina. Am J Public Health 107:180–185. Geogr Anal 24:189–206.
3. Balazs C, Morello-Frosch R, Hubbard A, Ray I (2011) Social disparities in nitrate-contaminated 14. Rodriguez MJ, Sérodes JB (2001) Spatial and temporal evolution of trihalomethanes
drinking water in California’s San Joaquin Valley. Environ Health Perspect 119:1272–1278,
in three water distribution systems. Water Res 35:1572–1586.
and erratum (2011) 119:A509.
15. Konisky D, Teodoro M (2016) When governments regulate governments. Am J Pol Sci
4. US Environmental Protection Agency (2009) Factoids: Drinking water and ground
60:559–574.
water statistics for 2009 (US Environmental Protection Agency, Washington, DC), EPA-
16. National Research Council (1997) Ensuring small water supply system sustainability.

SUSTAINABILITY
816-K-09-004.
5. Rahman T, Kohli M, Megdal S, Aradhyula S, Moxley J (2010) Determinants of environ- Safe Water from Every Tap: Improving Water Service to Small Communities (National

SCIENCE
mental noncompliance by public water systems. Contemp Econ Policy 28:264–274. Academy Press, Washington, DC), p 153.
6. Guerrero-Preston R, Norat J, Rodríguez M, Santiago L, Suárez E (2008) Determinants 17. Al-Jasser AO (2007) Chlorine decay in drinking-water transmission and distribution
of compliance with drinking water standards in rural Puerto Rico between 1996 and systems: Pipe service age effect. Water Res 41:387–396.
2000: A multilevel approach. P R Health Sci J 27:229–235. 18. US Environmental Protection Agency (2001) 1999 Drinking Water Infrastructure Needs
7. Pike W (2004) Modeling drinking water quality violations with Bayesian networks. Survey Modeling the Cost of Infrastructure (US Environmental Protection Agency,
J Am Water Res Assoc 40:1563–1578. Washington, DC).
8. Rubin S (2013) Evaluating violations of drinking water regulations. J Am Water Resour 19. Oxenford J, Barrett J (2016) Understanding small water system violations and defi-
Assoc 105:E137–E147. ciencies. J Am Water Works Assoc 108:31–37.
9. Wallsten S, Kosec K (2008) The effects of ownership and benchmark competition: An 20. Dinan T, Cropper M, Portney P (1999) Environmental federalism: Welfare losses from
empirical analysis of U.S. water systems. Int J Ind Organ 26:186–205.
uniform national drinking water standards. Environmental and Public Economics, eds
10. Noll R (2002) The economics of urban water systems. Thirsting for Efficiency: The Economics
Panagariya A, Portney P, Schwab R (Edward Elgar, Cheltenham, UK), pp 13–31.
and Politics of Urban Water System Reform, ed Shirley M (Pergamon, Amsterdam), pp 43–63.
11. US Environmental Protection Agency (2004) EPA claims to meet drinking water goals 21. Association of State Drinking Water Administrators (2013) An analysis of state drinking
despite persistent data quality shortcomings (US Environmental Protection Agency, water programs’ resources and needs (Association of State Drinking Water Adminis-
Washington, DC), Report No. 2004-P-0008. trators, Arlington, VA).

Allaire et al. PNAS | February 27, 2018 | vol. 115 | no. 9 | 2083