Abstract
Using data from the National Pollution Discharge Elimination System compliance and the Clean Water State Revolving Funds (CWSRF) for wastewater treatment plants in nine states between 2010 and 2018, we examine (1) the effect of noncompliance on the distribution and size of awarded CWSRF loans, and (2) the effects of the CWSRF provision and award size on postfunding compliance. We observe that funded facilities have poorer compliance records than the unfunded ones and that funded facilities decrease violations within two years after receiving financial support. On average, a $50 million CWSRF loan decreases violations by one count within two postfunding years.
1. Introduction
Protecting water quality is one of the primary objectives of environmental policy and regulation in the United States. Passed by Congress in 1972, the Clean Water Act (CWA) has been one of the most prominent federal laws protecting water resources and environmental quality. Although significant progress has been made, major water quality challenges persist despite more than $1 trillion spent on abatement initiatives since the CWA (Keiser and Shapiro 2019; Keiser, Kling, and Shapiro 2019). Compliance of wastewater treatment plants with the CWA is a major component of efforts to improve surface water quality in the United States. Concerns about the adequacy of the wastewater treatment infrastructure, including investment needs, have been documented (EPA 2000, 2016a; Ramseur 2017).1
In 1987, the Clean Water State Revolving Funds (CWSRF) program was introduced to facilitate compliance with the CWA. The purpose of the program is to provide low-cost loans for a wide range of water quality infrastructure initiatives, including projects aimed at improving compliance with the National Pollutant Discharge Elimination System (NPDES) permits (Copeland 1999, 2012; Travis, Morris and Morris 2004). NPDES maintains point source discharge permits for a list of regulated water pollutants and corresponding records of reported violations. These permits restrict the discharge of regulated pollutants and specify monitoring and reporting requirements that form the basis for enforcement (Helland 1998a, 1998b; Gaba 2007; Chakraborti and McConnell 2012).
Despite the importance of financial support and the CWSRF program, the distribution and the impact of the CWSRF loans have not been studied. In this article, we examine the distribution of financial assistance across wastewater treatment plants and the effects on compliance with the NPDES permits. The objective of our study is twofold. First, we examine the distribution of CWSRF loans across wastewater treatment plants in terms of the likelihood of receiving the award and the size of the awarded loans. In particular, we focus on the role of prefunding compliance records. Is CWSRF support allocated to poorly performing facilities with more NDPES permit violations? Second, we evaluate the efficacy of CWSRF assistance in mitigating NPDES permit violations. Do CWSRF loans reduce NPDES permit violations in subsequent years? We examine the effects of CWSRF provision and award size on post-award noncompliance.
Adequate wastewater infrastructure financing is critical for compliance with the NPDES permit requirements and protecting water quality (Ramseur 2017). The CWSRF loans provide financial assistance to municipal, state, intermunicipal, or interstate agencies for construction or maintenance of publicly owned treatment works; for construction, repair, or replacement of decentralized wastewater treatment systems; for reduction, treatment, or recapture of stormwater or subsurface drainage water; for implementation of water conservation, efficiency, or reuse measures aimed at reducing the demand for publicly owned treatment works capacity; for implementation of measures to reduce energy consumption needs for publicly owned treatment works; and for measures supporting reuse and recycling of wastewater, stormwater, or subsurface drainage water (Copeland 1999; EPA 2016a, 2020). Although CWSRFs can be used for a variety of projects, the Government Accountability Office (GAO) (2006) reports that most of the funds (96%, approximately $50 billion) are used for wastewater treatment facilities and conveyance projects. States rely heavily on CWSRFs to finance water quality improvement projects (GAO 2006) and allocate funds according to various priorities, including wastewater infrastructure maintenance and replacement needs, population changes, the U.S. Environmental Protection Agency’s (EPA) enforcement requirements, and stricter water quality standards for temperature, nutrients, and sediments. States are required to match at least 20% of the federal contribution to the CWSRF loans.
The NPDES permits require facilities to sample their discharge and submit the Discharge Monitoring Reports, documenting permit violations, to the designated regulatory agency. The EPA and state regulatory agencies also rely on inspections to advance compliance with the NPDES guidelines and procedures. The EPA and the authorized state agencies have various enforcement tools at their disposal against violators. For example, the regulators may issue administrative orders that require facilities to correct the documented violations. The law also authorizes the regulators to take up civil and criminal proceedings, including mandatory injunctions, penalties, and jail sentences against individuals found to have violated the restrictions willfully.
We use the NPDES discharge violations data from wastewater treatment facilities in nine states between 2010 and 2018 to examine the distribution of CWSRF support and the efficacy of financial support for compliance with CWA. While most recent CWSRF annual reports are available for all 50 states, past reports that include facility-level data are available only for the 9 states in this study. The distribution of the CWSRF loans across facilities and over time presents a convenient setting that allows us to compare the performance of funded and unfunded facilities. We examine the difference between funded and unfunded facilities in terms of pre- and postfunding compliance with the NPDES permits. The prefunding analysis of violations sheds light on the role of noncompliance in the distribution of CWSRF loans, including award counts and magnitudes. The postfunding analysis examines the efficacy of CWSRF awards in improving compliance with the NPDES permits conditional on the award size. We also distinguish the effects of CWSRF awards provided once from the effects of awards provided twice or more on the postfunding compliance of funded facilities.
Water quality challenges vary across states and watersheds. Similarly, noncompliance of wastewater treatment plants varies across regulated pollutants and regions. Regional authorities may focus on different pollutants depending on regional priorities and watershed improvement goals. Therefore, we examine the aggregate number of violations instead of focusing on a subset of specific contaminants. Our objective is to examine noncompliance broadly rather than for a particular pollutant. We focus on aggregate violations because the use of particular pollutant violations would be too narrow to make conclusions about the relationship between CWSRF awards and compliance in general. Examining the aggregate number of violations is more likely to capture the effect of noncompliance on funding and the effect of CWSRF loans on compliance than would an examination of a subset of regulated pollutants.
We observe that the CWSRF loans are allocated to the facilities that appear to be in greater need of support according to the number of cumulative violations. Funded facilities have more violations than unfunded facilities before receiving CWSRF support. Conditional on controls, the results show that violations one year before funding differ significantly across funded and unfunded facilities. Although funded facilities continue to have more violations postfunding than the unfunded facilities, CWSRF loans reduce NPDES violations of funded facilities within two years of receiving support. The lagged effect of loan provision on improved compliance beyond two years is expected because updating wastewater infrastructure and treatment technologies takes some time after loans are provided (Keiser and Shapiro 2019). We also observe that loan size matters for the effects of one-time but not repeated awards on noncompliance within the first two years after funding. This result is likely due to a temporal lag in the impact of the awards on noncompliance, especially for larger awards that may require multiple years for project completion.
2. Review of Related Literature
Although water quality in the United States has improved significantly since the enactment of the CWA, the law has been one of the most controversial environmental regulations in the nation. Studies like Harrington and Malinovskaya (2015) and Keiser and Shapiro (2019) find that investments under the CWA have helped reduce water pollution. Other studies document a lack of sufficient improvement in water quality (Adler, Landman, and Cameron 1993; Knopman and Smith 1993; Harrington, Morgenstern, and Sterner 2009; Hayward 2011; Smith and Wolloh 2012). The EPA reports that over half of stream miles in the United States still violate water quality standards (Keiser and Shapiro 2019). In a national assessment of 1.2 million wetland acres, 658,000 acres were impaired for at least one designated use. Organic enrichment, oxygen depletion, mercury, arsenic and selenium are the top causes of impairment (EPA 2016b).
Limited existing literature examines the effect of financial support for water infrastructure. Keiser and Shapiro (2019) assess the effectiveness of the CWA grants in reducing ambient pollution and find that downstream water quality improves after facilities receive financial support. Harrington and Malinovskaya (2015) find that between 2008 and 2012, the CWSRFs were not necessarily allocated to water treatment plants with greater pollutant discharge. However, wastewater treatment plants that receive financial assistance experience a greater reduction in pollutant discharge than did unfunded plants. Flynn and Marcus (2021) find that the CWA grants awarded to wastewater treatment plants improve downstream infant health outcomes. Our work adds to the literature by evaluating the effects of the CWSRF on compliance with NPDES permits under CWA. Unlike prior studies, we examine the cumulative violations across all pollutants regulated by the NPDES permits rather than individual regulated pollutants.
Using the data from the EPA’s Grants Information and Control System, Keiser and Shapiro (2019) examine the effects of the CWA investments on ambient pollutant concentrations. Their results indicate that on average, each CWA grant to municipal wastewater treatment plants decreases downstream dissolved oxygen deficit, fecal coliforms, and the probabilities that downstream waters are not fishable and swimmable. Harrington and Malinovskaya (2015) evaluate the performance of the CWSRF in Iowa, Indiana, Maryland, and Texas. Using the Clean Watersheds Needs Survey, the Clean Water Benefits Reporting System and the Discharge Monitoring Report data, they examine the changes in discharge water quality across funded and unfunded wastewater treatment plants. They found that loans awarded between 2008 and 2012 improved plant performance in terms of biochemical oxygen demand (BOD) and organic nitrogen levels relative to the plants in the corresponding states that did not receive the loans. McConnell and Schwarz (1992) also argue that high BOD pollution by wastewater treatment plants is partly due to the lack of federal subsidies.
Previous studies examine the effects of inspection and enforcement on compliance with the CWA (Magat and Viscusi 1990; Earnhart 2004; Shimshack and Ward 2005; Gray and Shimshack 2011). Gray and Shimshack (2011) provide a helpful review of empirical studies on environmental monitoring and enforcement actions and find that monitoring and enforcement activities reduce emissions and violations. Magat and Viscusi (1990) examine compliance with water pollution regulations in the U.S. pulp and paper industry between 1982 and 1985 and find that inspection improves compliance. Sanctions, especially federal fines, have also reduced pollution by wastewater treatment plants and chemical facilities in the 1990s (Earnhart 2004). Shimshack and Ward (2005) empirically explore the impact of enforcement efforts on BOD and total suspended solids in pulp and paper production. They find a two-thirds reduction in statewide monthly violation rates in the years after a fine. On the other hand, nonmonetary sanctions, including formal administrative orders, formal notices of noncompliance, and administrative consent orders, have no effect on compliance. Malik (1993) shows that regimes with self-reporting can reduce regulatory costs if less frequent inspections are coupled with more consistent and frequent punishment. However, failures and inconsistencies of the CWA enforcement have been documented in the popular press, government reports, and the law and economics literature (Flatt 1997; Sigman 2005; Duhigg 2009).
Previous literature also documents the effects of special interests and corruption on compliance with NPDES permits (Flatt 1997; Grooms 2015). Using the Permit Compliance System and Integrated Compliance Information System (ICIS) data, Grooms (2015) finds that more corrupted states report larger decreases in documented violations than do less corrupted states after oversight transitions from federal to state control. Helland (1998a) examines the NPDES permit violations and special interest groups’ influence on state-level regulation using data from 232 pulp and paper plants from 1989 to 1993. He finds that state regulatory agencies are responsive to national and local interest groups. The extent of state residents’ involvement in environmental organizations affects violations and the probability of inspection. Grant and Grooms (2017) empirically examine nonprofit environmental groups’ influence on compliance with environmental regulations under the NPDES. Their results suggest that the presence of nonprofit environmental organizations reduces the number of violations. They also find that nonprofit groups reduce government inspections and severe effluent violations.
The EPA authorizes states to pursue enforcement actions under the CWA, which leads to inconsistencies in CWA enforcement across states (Grooms 2015). Unique environmental, financial, and cultural settings lead to enforcement heterogeneity (Travis, Morris, and Morris 2004). Flatt (1997) argues that without the threat of federal oversight, states are left to their own devices with no mechanism for uniform enforcement of the CWA laws. Lack of federal efforts to ensure the consistency of enforcement has been documented. We consider the data from nine states with different environmental, regulatory and fiscal contexts. We control for state heterogeneity using fixed-effects specifications and defer a detailed examination of particular state-level differences to future studies.
3. Methods
Our empirical approach takes advantage of the variation in NPDES noncompliance and CWSRF loan provision across wastewater treatment plants. We start with a t-ratio comparison of facility violations before and after receiving financial support, and across funded and unfunded facilities. For the prefunding violation analyses, we evaluate the difference between prefunding compliance of funded and unfunded facilities. The prefunding analysis examines the differences in violations of funded and unfunded facilities one, two, and three years before CWSRF provision. We use Heckman’s two-stage estimation to examine the effect of noncompliance on funding outcomes and award sizes. We also use a difference model with violations because the dependent variable and future funding as the independent variable for prefunding compliance analysis. Last, we examine postfunding compliance of funded and unfunded facilities using multiple-period difference-in-difference (DiD) techniques (Kirkpatrick and Bennear 2014; Yamazaki 2017). Postfunding analyses are repeated on a matched group of funded and unfunded facilities using propensity score matching (PSM). We include binary funding outcomes and award sizes in prefunding and postfunding analyses.
t-Ratio Analysis
The temporal variation in CWSRF loans across treatment plants enables a comparison of violations within and across funded and unfunded groups before and after funding. We start with a t-ratio comparison of (1) mean annual violations of funded and unfunded facilities after each funding year, (2) mean annual violations of funded and unfunded facilities prior to each funding year, (3) mean annual pre- and postfunding violations of funded facilities, and (4) mean annual violations of unfunded facilities before and after specific years.
Funded facility treatment groups include only the facilities funded in the respective year but not the previous two years. Each corresponding control group includes facilities that are not funded in the corresponding year, the preceding two, or the subsequent two years. For example, the 2014 treatment group, composed of the facilities financed in 2014 but not before, is compared with the control group with facilities not funded between 2012 and 2016. The comparisons are made using the average number of annual violations from three prefunding years and three postfunding years to assess differences between the funded and unfunded groups and changes in each group before and after funding. Each treatment year comparison excludes the violations from the corresponding year because the relative timing of the CWSRF award and violation occurrence is difficult to establish. Furthermore, violations documented in the same year as the CWSRF award are unlikely to be important determinants of awards or to reflect the effects of loan provisions on compliance.
Prefunding Compliance
We explore the effect of prefunding violations on CWSRF loan allocation by comparing the prefunding violations of funded facilities with corresponding violations of unfunded facilities. The prefunding analyses use violations as the independent variable to examine the effects on funding decisions and award sizes. Heckman’s two-stage model includes first-stage binary funding outcomes and second-stage award magnitudes conditional on receiving CWSRF support. For funded facilities, we include CWSRF provision observations from 2010 to 2018 and compliance data from the corresponding three prefunding years. For example, to examine the distribution of CWSRF support in 2010, we use compliance data from 2007 to 2009. Hence, our group of funded facilities includes those that received CWSRF support between 2010 and 2018. It would be preferable to ensure that those facilities did not receive support before 2010, but unfortunately, we do not have data on CWSRF before 2010. For treatment years 2012 through 2018, we ensure that neither treatment nor control facilities received CWSRF since 2010.
The Heckman model is used to examine the likelihood and magnitude of CWSRF in two stages. Selection bias occurs when certain facilities are systematically included in the second stage based on the first-stage selection process. This can lead to biased and inconsistent estimates in the second stage because the analysis is based on a censored sample. Heckman two-step selection model is a standard approach in these types of circumstances (Brounen and Kok 2011; Little and Rubin 2019). In the first stage, a probit model with state and year controls is used to examine whether facilities receive CWSRF support:
1
where, Φ(·) is the cumulative density function of the standard normal distribution. Fist takes the value of one when a facility i in state s receives a CWSRF loan in year t. Qs and λt are state and year fixed effects. yi,t−1, yi,t−22, and yi,t−3 are annual violations by facility i in state s in one, two, and three years before t. β1, β2, and β3 show the effect of lag violations on funding decisions. To account for the possibility of some facilities receiving CWSRF loans in more than one year, repeated funding binary control, RF, is included that takes value of one if the awarded loan is one of several loans and zero otherwise. This variable is excluded from the second-stage formulation because it is statistically insignificant and including at least one explanatory variable only in the selection equation is recommended (Sartori 2003). X represents a vector of facility-, county-, and state-level time-variant controls. εit is the standard idiosyncratic disturbance term.
In the second stage, linear specification is used to examine the magnitude of assistance as follows:
2
where Zist is the magnitude of the loan awarded to facility i in state s and period t. αs and λt are state and year fixed effects. The inverse of Mills ratio, ψ, is obtained from the first-stage probit model. All regressors from the first stage are used in the second-stage regression except for repeated funding dummy RF.
We also examine the prefunding compliance using panel count models including Poisson and logit with violations as the dependent variable and CWSRF support in the future three years as explanatory variables. Poisson regressions use facility annual number of violations, whereas logit models use transformed binary annual violation occurrence as dependent variables. Clearly, the purpose of these regressions is not to imply a causal inference in terms of the effect of future funding on violations in the preceding years. Instead, these regressions shed light on whether the future-funded facilities have more violations in the present. In other words, these models quantify differences in the present, prefunding violations between funded and unfunded facilities in the future. Although these “difference” models are not very intuitive in terms of causal inference, they offer insight for the difference in prefunding compliance of funded and unfunded facilities. The results from these models are provided in the Appendix.
We examine Heckman and count models because they use different empirical specifications and data. The methods use different data because the dependent variable in the count difference models is violations in year t (2009–2015) and control variables, including facility size and county income in year t, and future funding in t + 1, t + 2, and t + 3. On the other hand, the dependent variable in the Heckman models is funding in 2010–2018 with lag violations, county income, facility size, and repeated funding dummy as control variables. Hence, the dependent variable and the controls for flow and income in the Poisson models are for 2009–2015.2 On the other hand, the dependent variable and the controls for flow and income in the Heckman models are for 2010–2018. As a result, the data and the number of observations differ across the two approaches. Hence, some differences in findings are possible. Since the conclusions can differ, we feel that reporting the results from both approaches is helpful for the sake of robustness.
Postfunding Compliance
The DiD technique has been widely used to study similar intervention effects, where outcomes are compared over time and between treatment and control groups (Angrist and Pischke 2008; Brent, Cook, and Olsen 2015; DeAngelo and Hansen 2014; Finkelstein 2007; Lavaine and Neidell 2017). Post-funding analysis exploits the DiD estimation techniques to compare the NPDES permit violations across funded and unfunded facilities and across pre- and postfunding periods. Provision of the CWSRF loans cannot be assumed to be random, as would be required for a randomized experimental study with treatment and control groups. This is common in most social science research based on observational data. Nevertheless, the DiD framework allows us to analyze loan provision and noncompliance conditional on the parallel trends in violations of funded (treatment group) and unfunded (control group) facilities before providing CWSRF loans. We confirm that the pretreatment violation trends of funded and unfunded facilities do not differ significantly in most cases based on the parallel trends tests. We also present the results from the matched DiD analyses based on the nearest neighbor PSM.
We proceed with the multiple-period (Kirkpatrick and Bennear 2014; Yamazaki 2017) analysis of combined funding year treatments. We use DiD models to compare the violations within and across funded and unfunded groups before and after CWSRF awards. The control group includes facilities that are not funded in the corresponding pretreatment, treatment, and posttreatment periods, and the treatment groups include facilities that receive CWSRF in the corresponding year but not in the preceding years. The treatment group facilities are funded in one or more years in the relevant window.
The regressions use data from several prefunding and two postfunding years and exclude data from the corresponding funding year for funded facilities. Observations for treatment facilities in treatment years are not included in our analyses because relative timing of violations versus awards could not be established in the treatment years. For the control group, all observations are included. We ensure that none of the facilities in our treatment or control groups receive CWSRF support in prefunding years.
We perform the DiD analysis using two samples: one includes only one-time funded facilities between 2014 and 2016 as a treatment group, and the other includes facilities that received one or more loans between 2012 to 2016 as a treatment group. Between 2014 and 2016, only 10 facilities received multiple awards. Therefore, we focus on singular awards in these models and ensure that pretreatment (2010–2013) funding is not provided to any of the included facilities. This analysis disregards repeated funding and examines the effect of singular CWSRF awards. To account for multiple awards, we use the second sample (2012–2016), which includes 105 facilities funded more than once. However, the 2012–2016 analyses use observations from only two prefunding years (2010 and 2011) for parallel trend tests. Hence, the advantage of the 2014–2016 analysis focusing on singular awards is that these models include four rather than two prefunding years to confirm parallel trends of treatment and control groups pretreatment. On the other hand, 2012–2016 analysis includes repeatedly funded facilities but uses only two pretreatment years for parallel trend confirmation. We confirm the parallel trends of treatment and control facilities for both approaches. Conventional- and PSM-based DiD specifications are used for both (Chabé-Ferret and Subervie 2013).
The effects of CWSRF loans on posttreatment compliance are examined in terms of discrete funding and loan magnitude effects. First, we examine funding decisions disregarding the magnitudes of the loans. Subsequently, we include the effects of the loan magnitudes.
Dynamic estimation with lagged dependent variable as a regressor can result in a correlation between the unobserved effect and the dependent variable (Deininger, Ali, and Alemu 2011). We estimate our models conditional on both the initial value of the dependent variable and the facility-level average of the time-varying exogenous variables (Wooldridge 2005; Deininger, Ali, and Alemu 2011; Vesterberg 2018). All models are estimated using the standard facility-level random effects Poisson model.
Parallel Trend Assumption
The DiD analysis relies on the assumption that the control group serves as an adequate proxy for the counterfactual outcome that would have been observed in the treatment group if the treatment was absent (Ryan, Burgess, and Dimick 2015; Chan et al. 2017; Lai 2017; Oakes and Kaufman 2017). As is common in observational data, we cannot test this assumption directly because we do not observe the treated group in the absence of support from CWSRF. The identifying assumption in such a situation rests on the parallel trends of outcomes in the treatment and control groups in the pretreatment period (Wrenn, Klaiber, and Jaenicke 2016). This assumption holds if trends are similar across the treatment and control groups in the absence of the treatment (Beatty and Shimshack 2011).
The validity of the parallel trend assumption can be assessed by testing whether the differences in prefunding trends of violations between treatment and control facilities are statistically and economically indistinguishable from zero. We use equation [3] to test differences in trends during four pretreatment years in the models for the singular awards made between 2014 and 2016 as follows (Autor 2003; Agarwal and Qian 2014; Kearney and Levine 2015; Cerulli and Ventura 2019):
3
where yist is the violation of facility i in state s at time t, and λ and α are time and state fixed effects. We include an indicator for the three out of four pretreatment periods and leave out the indicator for the last pretreatment period (t + 1). Fi,t+2, Fi,t+3, Fi,t+4 are one if facility i is funded two, three, or four years after. Hence, the corresponding beta coefficients represent the prefunding difference in violations. Pretreatment indicators (β1, β2, and β3) are expressed relative to the omitted period, which serves as the base period for the parallel trends tests. If the trends in pretreatment violations between treatment and control facilities are similar during the pretreatment periods, then β1, β2, and β3 should be statistically insignificant (Kearney and Levine 2015). Fi,t−1 (Fi,t−2) is one if facility i receives CWSRF one (two) year(s) before t, and zero otherwise. β4 (β5) indicate the difference in violations between treatment and control facilities one (two) years after receiving the CWSRF support relative to the benchmark control group (Agarwal and Qian 2014).
A similar formulation is used to evaluate two pretreatment years’ violation differences in the models that include singular and repeated awards made between 2012 to 2016 (equation [4]). In this formulation β1, β4, and β5 are analogous to the corresponding coefficients in equation [3]. β2 and β3 are absent from equation [4] because this model includes only two prefunding years for parallel trend analysis. However, this model includes β6 for repeated funding effect pretreatment and β7 and β8 for repeated funding effects posttreatment:
4
Binary Funding Treatment DiD
The DiD strategy is applied in a setting with multiple periods and treatments (Imbens and Wooldridge 2009). First, we focus on the effects of binary funding decisions disregarding the magnitudes of the loans and account for repeated awards using equation [5]. Multiple indicator variables reflect the heterogeneity in the lagged effects of funding provision on noncompliance in subsequent years. The multiperiod regression model for the combined treatment analysis examines the difference in violations in the posttreatment years across treatment and control facilities:
5
where yist is the violation by facility i in state s in year t. F and RF are the dummies for one-time funding and repeated funding, respectively. Fi,t−1 (Fi,t−2) is one for facility-year observations with one and only loan provided one (two) year(s) ago. Similarly, RFi,t−1 (RFi,t−2) is one for facility-year observations with the last of the repeated awards provided one (two) year (s) ago. Hence, β11 and β12 show the effects of receiving a one-time award in treatment year t − 1 and t − 2 on violations in year t. Similarly, β21 and β22 respectively show the effect of receiving awards in multiple years in treatment years t − 1 and t − 2 on compliance in year t. αs is state fixed effect, and λt is the year fixed effect. X is a vector of facility and county-level time-variant variables. εit represents the standard idiosyncratic disturbance term.
Continuous Funding Treatment DiD
Next the effects of the award size are examined using the following model, which expands the previous specification by adding an interaction term for loan size:
6
where the notation is consistent with the notations in equation [5], except that this model includes interaction terms between binary funding variables and corresponding award magnitude, CWSRF. CWSRFi,t−1 (CWSRFi,t−2) indicates the magnitude of the loan provided to facility i one (two) year(s) prior to year t. For the facilities that receive multiple loans, the repeated funding amount is the total across multiple awards. β11 shows the magnitude effect of receiving support a year ago when funding was made only once. Similarly, β12 shows the magnitude effect of receiving support two years ago when funding was made only once. Interpretations for β21 and β22 are analogous except that they respectively show the cumulative effects of multiple loans provided to facility i up to one and two years before t.
PSM DiD
The DiD estimator is most suitable when treatment is random. However, treatment is often not randomized in observational data. In such cases, matched control groups can be used to reduce selection bias (Rubin 2008). We use the PSM technique (Rosenbaum and Rubin 1983) to control for unmeasured differences between the treatment and the control groups (Heckman, Ichimura, and Todd 1997). The DiD-PSM approach relies on matching treatment and control subjects in terms of observed characteristics to identify best match counterfactuals, making the common trend assumption more plausible (Gebel and Voßemer 2014).
We use a logit model to estimate the propensity score, which is the probability of receiving support from the CWSRF depending on several observable characteristics. The covariates include all available control variables that can influence the probability of obtaining the CWSRF loans. PSM entails forming matched sets of treatment and control subjects that share a similar value of the propensity score. We use the nearest neighbor matching technique to pair each treatment facility with the 20 closest control facilities in terms of the propensity score (Caliendo and Kopeinig 2008). After matching, the observable characteristics should be balanced between treatment facilities and matched control facilities. We formally test for the significance of differences in observable characteristics between the treatment and the matched control facilities samples using t-tests.
4. Data
We use data from 6,921 facilities in nine states (Arkansas, Arizona, California, Indiana, Nebraska, Oklahoma, Texas, Vermont, and West Virginia) from 2010 to 2018. The time frame of the analysis is based on the availability of the CWSRF loan data. Table 1 shows the annual summary statistics for the funded and unfunded facility-year observations. The funded group includes 3,362 facility-year observations, and the unfunded group includes 21,016 observations. The unfunded group includes observations for the facilities that were verifiably not funded between 2010 and 2018. Observations for the facilities that have no recorded funding but have missing data at least for one of the years during this time frame are excluded from Table 1. On the other hand, all available observations for the facilities that are funded at least once are included in Table 1 even if data for these facilities are missing in some years. The rationale for this construction is to make sure that the funded group includes facilities funded at least once, while the unfunded group includes only the facilities that were never funded in this time frame.
Summary Statistics for Wastewater Treatment Plants, 2010–2018
The states included in our analysis vary in size, violations, and funding. Appendix Table A1 provides summary statistics for violations, facility size, and funding broken down by state. For example, West Virginia has the largest average violation count of 17 per facility per year, and Nebraska has the lowest at 2.2. Such variation across states in our sample can be helpful for generalizing our findings. Still, generalization of the results requires caution. The mean violation in the United States (2) is comparable to the mean violations in some of the states in our sample. However, the range of annual facility violations in our sample is between 0 and 166, while for the U.S. sample it is between 0 and 246. Similarly, the facilities in our sample are smaller than some of the facilities in other states.
Wastewater treatment plants’ violations data are obtained from the EPA’s ICIS-NPDES database. The Effluent Limit Exceedances database provides discharge monitoring self-reported data, including violations of the NPDES permit effluent limits. The violations are reported per permit-holding facility. We use the aggregate number of annual violations across different pollutants rather than focus on individual effluents. The use of annual data mitigates the timing problem with violations and associated reporting (Grant and Grooms 2017). The quality of the data varies from state to state (GAO 2005).
The data on the CWSRF support received by the individual wastewater facilities are obtained from the state Departments of Environmental Quality. The state agencies publish annual reports that identify all disbursements of financial support under the CWSRF program in the respective year. We rely on these reports to identify the facilities that receive assistance.
We include two years’ lagged violations, annual wastewater flow, annual numbers of inspection activities, and county per capita income as facility-level covariates. Data on facility controls are obtained from the EPA’s ICIS-NPDES database, and the per capita income is obtained from the Bureau of Economic Analysis.
5. Results
The results are presented in the following order. We first compare the compliance records of funded and unfunded facilities using t-tests. Next, we report results from the prefunding analysis and from the postfunding DiD models. Postfunding DiD results include unmatched and matched sample models for robustness check. The unmatched models include all data, while the matched models include nearest neighbor-based propensity score-matched data to address concerns about the nonrandom funding treatment.
t-Ratio Analysis
Table 2 shows the mean annual violations of funded and unfunded facilities two years before and two years after each CWSRF assistance treatment year. The estimates are organized across columns A, B, and C and rows 1, 2, and 3. Block A1 shows the mean annual numbers of violations by the funded facilities during the two years before receiving support from the CWSRF in the respective years. Block B1 shows the corresponding noncompliance of the unfunded facilities. For example, in 2013, 92 funded facilities had 6.4 violations per facility per year during 2011–2012,3 while the 5,914 unfunded facilities had 3.5 annual violations per facility. Block C1 (C2) shows t-statistics for the hypothesis that there is no significant difference in the mean annual violations during the two years prior to (following) the loan provision year across funded and unfunded facilities. Blocks A3 and B3 show the t-test statistics for the hypothesis that no statistically significant difference exists between the mean annual violations before and after the treatment years for funded and unfunded facilities, respectively.
t-Tests for Mean Annual Violations
The results in block C1 show that the funded facilities have significantly higher numbers of prefunding violations than the unfunded facilities in two out of five loan provision (treatment) years. These results support the hypothesis that in 2013 and 2014, CWSRF was provided to the facilities that performed worse in terms of NPDES violations in the preceding years. However, block C2 shows that the mean number of annual postfunding violations is also higher for the funded facilities than for the unfunded facilities in two of the four treatment years. Based on these statistics, the efficacy of CWSRF in reducing noncompliance is ambiguous because the funded facilities have more violations than the unfunded facilities before and after funding. Block A3 shows a significant difference in mean annual prefunding and postfunding violations of the funded facilities in 2014. Similarly, block B3 suggests a significant difference in mean annual pre- and postfunding violations of the unfunded facilities from 2013 to 2016.
Conclusions about the efficacy of CWSRF loan provision are difficult to reach. First, it is difficult to assess whether the CWSRF reduced the violations of the funded facilities relative to the violations that would have been observed in the counterfactual case because noncompliance by the funded facilities is not observed in the absence of funding. Second, the t-tests do not take into account the confounding effects of other factors. Third, Table 2’s comparisons of funded and unfunded facilities before and after funding disregard the effects of repeated CWSRF provision. We turn to the multivariate regression analysis for a more detailed examination. Next we present the results from the prefunding analysis, followed by postfunding parallel trend analysis and multiperiod DiD results.
Prefunding Compliance
We examine the role of noncompliance for CWSRF distribution by comparing prefunding violations of funded and unfunded facilities. Table 3 presents the results from the Heckman two-stage selection models, including first step binary funding outcome (selection) and second step award magnitude (OLS) analysis. These models include all available observations regardless of whether a facility has missing data in some of the years. Unlike Table 1 and the forthcoming postfunding analysis results, no observations are excluded from these regressions.
Prefunding Violations
Results of selection models 1 and 2 show prefunding violation differences between funded and unfunded facilities. The estimates suggest that facilities with more violations are more likely to receive CWSRF assistance in a subsequent year. There is a statistical difference in violations of funded and unfunded facilities one year before receiving support. Facilities with more violations in t − 1 are more likely to be funded in year t than facilities with fewer violations in t − 1. These results suggest that perhaps prior noncompliance attracts the attention of regulators and signals the need for financial support.
While the probability of a facility being funded depends on NPDES violations in the previous year, stage 2 results show that the magnitude of the award is not associated with noncompliance. There is no statistical effect of noncompliance on the magnitude of the awards. Table 3 suggests that funding decisions depend on facilities’ compliance in the previous year, but the magnitude of the award may depend on other factors, like the size of the facility. Our results show a statistically significant association between facility size and the award’s probability and magnitude. Larger facilities are more likely to receive awards and tend to receive larger grants. Larger treatment plants may be prioritized because such facilities tend to be located in more populated areas (Keiser and Shapiro 2019).
The Poisson and logit regressions results with violations as the dependent variables and future funding as explanatory variables are provided in Appendix Table A2. We put these results in the Appendix because these models are not intuitive in terms of causal inference. Clearly we do not mean to imply that future funding outcomes influence noncompliance in the preceding years. Instead, these results represent another angle of view on whether compliance of facilities funded in future periods differs from compliance of future unfunded facilities. Statistical significance of funding in years t + 1, t + 2, and t + 3 support the hypothesis that facilities with poorer compliance records receive CWSRF assistance. The results consistently show a statistical difference in violations of unfunded and eventually funded facilities one, two, and three years before the funding was received. These results hold even when controlling for repeated provision of CWSRF awards in the preceding one or two years.
Postfunding Compliance
The results for the effect of CWSRF provision on facility violations are presented as follows. First, we report the results from parallel trend analysis (Table 4), followed by the results from binary funding treatment DiD (Table 5) and continuous funding treatment DiD (Table 6). All results are presented for matched and unmatched samples. The results are presented for one-time funding treatment in 2014–2016 and repeated funding treatment in 2012–2016. We include both of these results in all tables because the 2014–2016 treatment analysis has a longer pretreatment period (2010–2013 as opposed to 2010–2011) for parallel trend analysis, and the 2012–2014 treatment model accounts for repeatedly funded facilities. The 2014–2016 treatment regression models exclude repeated funding because only 10 facilities received CWSRF support more than once in 2014–2016. Hence, Table 4 shows four models, and Tables 5 and 6 show eight models, half of which (models 2, 4, 6, and 8) include control variables like facility’s size, previous compliance record, and county income. The rest (models 1, 3, 5, and 7) do not include additional controls. We include these models because, on one hand, regressions without control variables have more observations. On the other hand, the regressions with additional controls account for more factors besides the fixed effects and the treatment variables. Based on the DiD requirements, some of the available observations have to be excluded from the postfunding regressions. First, the facilities funded in the pretreatment periods have to be removed from the regressions because neither treatment and no control facilities can be funded pretreatment. Second, facilities with missing data in the pretreatment years have to be removed from the regressions to ensure that pretreatment period includes only the facilities that were not funded pretreatment.
Difference-in-Difference Parallel Trend Test
Postfunding Violations and Clean Water State Revolving Funds Loan Provision
Postfunding Violations and Magnitudes of Clean Water State Revolving Funds Awards
Parallel Trend Analyses for Postfunding DiD
Although the prefunding year compliance results indicate that funded facilities tend to have more violations in the prefunding years than the unfunded facilities, the parallel trend assumption requires a separate assessment. In particular, the results of prefunding violations show that funded facilities have more violations than unfunded facilities but do not necessarily reveal whether the trends in violations differ between treatment and control facilities.
We report the results of the formal parallel trend analyses in Table 4. The insignificance of lead indicators in most models (with and without matching) indicates that the difference in treatment and control group’s violations in the base year (one year before receiving support) is not significantly different from other prefunding years. This suggests that the difference in violations over time between funded and unfunded facilities does not change before funding. We have one exception in model 1 (2014–2016 treatment), where the prefunding violation differences four and three years before receiving support are significantly greater than the difference in the base year. However, the matched results (models 3 and 4) show no significant differences. Hence, the unfunded groups represent a reasonable benchmark counterfactual for the funded groups in terms of the trends in CWA violations pretreatment.
Binary Funding Treatment
We report the results from the multiperiod DiD models to examine the effects of binary funding (Table 5) and award sizes (Table 6) on noncompliance in the postfunding years. Funded groups in models 1, 2, 5, and 6 (Tables 5 and 6) include facilities funded only once between 2014 and 2016. In models 3, 4, 7, and 8, some facilities receive CWSRF support more than once between 2012 and 2016. These models separate treatment facilities into one-time and repeatedly funded groups.
Models 3, 4, 7, and 8 have more facilities than models 1, 2, 5, and 6 because the latter models exclude facilities that received support in 2012 and 2013, which are the pretreatment years in these models. Neither treatment nor the control facilities can receive support in the pretreatment years in the DiD approach. Observations for treatment facilities in treatment years are not included in any of the models because relative timing of violations versus awards could not be established in the treatment years. However, all observations from the control group are included in the regressions.
The 2014–2016 treatment year analyses (models 1 and 2 in Table 5) show negative effects of receiving a one-time CWSRF support on noncompliance of the funded facilities in the subsequent two years. Estimates in models 3 and 4 with the extended treatment period of 2012–2016 suggest that receiving a one-time CWSRF support helps the funded facilities reduce violations two years after receiving the awards, relative to the unfunded facilities. Model 4 also suggests that multifunded facilities decrease their violations two years after receiving their last award. However, model 3 indicates that violations increase one year after receiving the last CWSRF award. This may be due to the magnitude of projects associated with repeated funding, which may take longer to complete, producing more violations in the meantime. This effect is not statistically significant in our preferred model 4, which includes facility and county covariates.
Overall, Table 5 shows that the loans provided in 2012–2016 had the intended effect of reducing noncompliance. Postfunding violations decline two years after receiving support. Models 1 and 2 also suggest that facilities funded only once between 2014 to 2016 experienced a reduction in violations one year after receiving the loans. On the other hand, facilities that receive multiple loans may be pursuing larger projects, which take longer to complete and improve compliance. As a result, one-year lagged violations are statistically insignificant, whereas two-year lagged violations show a decrease relative to the unfunded facilities. These results are consistent with the matched sample regression results in models 5–8 in terms of signs and statistical significance of the estimated coefficients.
Table 5 shows that violations in the previous three years have a significant and positive coefficient. This suggests that NPDES violations are persistent as the facilities with more violations before receiving CWSRF continue to experience more violations in later years. The size of the facility (water flow) does not appear to correlate with noncompliance. Per capita income in the county where the facility is located is positively correlated with reported violations. One reason for this result may be that facilities in wealthier counties may follow more rigorous reporting practices than those in poorer counties. In such cases, facilities in wealthier counties may show more violations. Regulatory actions in terms of the number of lag inspection actions also help reduce noncompliance.4
Continuous Funding Treatment
Table 6, structured similarly to Table 5, reports the effect of award size on postfunding compliance. Interaction terms between binary funding variables and the corresponding CWSRF award size show the effects of award size on the number of violations. The results show that the size of one-time awards does not have a statistically significant effect on violations one year after funding. However, according to models 3, 4, 7, and 8, which account for repeat provision of CWSRF funds, the larger the one-time loans, the fewer the violations two years after funding. This result suggests that a one-time award’s magnitude has a lagged effect two years after funding on compliance, which is consistent with the results in Table 5. On the other hand, the results in Table 6 show no statistical effect of the size of the repeat awards on noncompliance within two years of the last award. Repeated loans are likely granted to facilities with larger projects that may take longer to complete. Therefore, reduction in noncompliance may take longer than two years. Replacing or updating treatment technologies and operations that improve compliance can take longer than two years after loans are provided (Keiser and Shapiro 2019).
We estimate marginal effects for funding provision and award size. Appendix Table A3 shows marginal effects at means corresponding to Poisson regressions results of models 4 from Tables 5 and 6, respectively. Marginal effect estimates suggest that each one-time award reduces violations by 0.139, and repeated awards reduce violations by 0.209. It takes, on average, seven one-time loans (1/0.139) and five (1/0.209) multi-awards to reduce violations of funded facilities by one count within two years of funding. The effects of CWSRF provision on noncompliance can be compared to alternative noncompliance interventions. For example, our results show that each inspection reduces violations by 0.005. Hence, it takes 200 inspections to reduce violations by one count.
Marginal effect estimates for the effects of award size indicate that a dollar of one-time CWSRF award decreases violations by 2.61 * 10−8 per facility two years after funding. This implies that on average, it takes a $50 million CWSRF loan to decrease violations by one within two years of award provision. In our sample, an average one-time grant is $7.6 million. Hence, it takes roughly seven awards to decrease violations by one. It is important to recognize that our postfunding compliance examination is limited to two postaward years. Improvements in compliance from CWSRF investments can take longer than two years (Keiser and Shapiro 2019). Therefore, limiting the analysis to only the first two years after funding provision is likely to undervalue the efficacy of CWSRF in improving long-term compliance.
Consistent with the results in Table 6, the marginal effect for the magnitude of the repeated CWSRF award is not statistically significant. Although the provision of multiple awards has a statistically significant and negative effect on violations (Table 5 and Appendix Table A3), the magnitude of the repeated awards has no statistically significant effect on violations within two years from the last loan provision. Large repeated CWSRF loans may require multiple years to complete. As a result, the impact of these awards may not be observed within the first two years of funding. Instead, one may need to look at longer time periods to detect a change in noncompliance. This is consistent with Keiser and Shapiro (2019), who document a positive effect of CWA grants on downstream water quality attributes several years after the grant is provided.
6. Conclusions
Compliance with the CWA is an important public health and environmental quality objective. The United States has spent over $1 trillion to control water pollution, and the economic efficacy of these investments remains debated (Keiser and Shapiro 2019; Keiser, Kling, and Shapiro 2019). Wastewater treatment facilities are responsible for treating wastewater before it is discharged into public water bodies. Pursuant to the CWA, all wastewater discharge facilities are issued permits that specify the type and the amount of the regulated pollutants they can discharge. Compliance with these permits depends on operational procedures and on technological infrastructure at the treatment facilities. Adequacy of operational procedures and especially technological infrastructure depends on the availability of finances. The CWSRF program was set up in 1987 in part to address this investment need. This study examines the allocation of CWSRF assistance across wastewater treatment facilities and the effects of these investments on compliance with the CWA in recent years.
While prior studies examine downstream water quality in terms of dissolved oxygen deficit, fecal coliform, and organic nitrogen (Earnhart 2004; Shimshack and Ward 2005; Earnhart and Harrington 2014; Harrington and Malinovskaya 2015; Chakraborti 2016; Keiser and Shapiro 2019), we focus on CWA compliance using cumulative discharge violations of NPDES requirements across all regulated pollutants. The analysis enables a broader evaluation of CWSRF loans, which can be used for various projects aimed at improving discharge water quality with respect to various pollutants. The limitation of using a cumulative number of violations is that the reporting quality can vary across different pollutants (Harrington and Malinovskaya 2015), which can mask the effects that may be present for particular pollutants. The disaggregated analysis of particular pollutants and financial support would be useful and should be pursued in future studies.
The NPDES violations by wastewater treatment plants in nine states between 2010 and 2018 reveal that funded facilities have poorer compliance records before receiving CWSRF support than do the unfunded facilities. This result is in contrast with Harrington and Malinovskaya (2015), who observe that financial support is generally not allocated to the facilities in greatest need in terms of compliance. Our results show a statistically significant difference between funded and unfunded facilities in prefunding compliance. We document that larger loans are awarded to larger facilities. Since larger facilities tend to be located in more populated areas, this result is consistent with Keiser and Shapiro (2019), who observe that larger loans are awarded to facilities in more populated areas.
We examine the effects of provision and magnitude of financial support on postaward compliance. We find that the CWSRF loans awarded between 2012 and 2016 reduced violations of funded facilities within two years of funding. This result is consistent with previous literature documenting a positive effect of financial support on water quality. Harrington and Malinovskaya (2015) find that CWSRF loans improve biological oxygen demand and organic nitrogen discharge of treatment plans in four states. We extend their results by considering the cumulative number of violations, rather than focusing on particular pollutants, and by explicitly examining the effects of single versus multiple awards and the effects of award magnitude. Keiser and Shapiro (2019) use a longer time horizon and a broader scope of analysis to show that CWA grants have a statistically significant and positive effect on downstream water quality several years after the grant is provided and continue to have a positive effect for many years after. The lagged effect of the grants is expected because some of the grants support projects that can take several years to complete. Our study differs by examining cumulative NPDES compliance rather than particular water quality attributes and focusing on the short-term effect on compliance within the first two years after funding. We find a qualitatively similar positive effect of CWSRF support even within the first two years after funding.
We observe some evidence of larger singular CWSRF awards having greater effects on reducing noncompliance than smaller awards two years after funding. We also estimate that on average, a $50 million one-time award decreases noncompliance by one violation two years postfunding. On the other hand, the cumulative size of repeated CWSRF awards is not statistically significant for reducing noncompliance within the first two years. Statistical insignificance of the value of multiple CWSRF awards is likely due to the data limitations. Our data include only two posttreatment years. Facilities that receive multiple CWSRF awards are likely to be engaged in larger infrastructure upgrades that take several years to complete after funding is provided. The average size of one-time awards in our sample is $7.6 million, and the average combined value of repeated awards is $28.7 million. Since projects funded by larger grants may take longer to complete, our results may not capture improvements in compliance that take more than two years to materialize. In this respect, our focus on the first two years of compliance postfunding undervalues the efficacy of CWSRF in improving compliance in the long run. Future examinations that use longer time horizons may detect longer-term impacts of CWSRF support, similar to Keiser and Shapiro (2019).
It is important to acknowledge that NPDES permits can be relaxed or tightened over time in response to changes in the water quality of the receiving watershed (Chakraborti and McConnell 2012). Hence, changes in compliance records can depend on the availability of adequate investments and management of the water treatment plants as well as on the changes in the requirements of the NPDES permits. Because the NPDES permit requirements can change over time, our results should not be interpreted in terms of the effect of the CWSRF support on water quality in general. Instead, the results in this study apply strictly to compliance with the CWA in terms of the NPDES violations, rather than to water quality.
Our results are limited by the temporal and spatial scope of the analysis. Although CWSRF has been in place since 1987, facility-level funding data are publicly available only from 2010 and only for a subsample of states. As a result, we cannot assess the effectiveness of the CWSRF prior to 2010 or generalize the result for the United States as a whole. Limited data availability prevents us from evaluating the long-run effectiveness of CWSRF support. We consider the effects of the CWSRF support on compliance with the CWA for only up to two years after the provision of the loan. However, it is possible that improvements in compliance may be observed in the longer run. The lagged effect is expected as infrastructure improvement projects can take time to complete. Nevertheless, this study provides important evidence supporting the efficacy of the CWSRF in improving compliance with CWA even within a couple of years after receiving financial assistance.
Our results are subject to the quality of the data used in the analysis. Although all point sources are required to obtain an NPDES permit under the CWA, not all facility, permit, or discharge monitoring data are provided in ICIS-NPDES, and reporting quality differs across states, facilities, and over time. In this regard, the inconsistency of the NPDES violation reports is a significant caveat. While this study offers a valuable initial assessment of the CWSRF program’s role in compliance with the CWA, possible strategic underreporting by wastewater facilities should be considered in the extrapolation and interpretation of our results.
Acknowledgments
Support for this study was provided by USDA-NIFA Hatch Grant WVA00691 and NSF Award 1903543. The authors are grateful for helpful suggestions by the conference participants at the annual meetings of the European Association of Environmental and Resource Economists (2020) and the Association of Environmental and Resource Economists (2021).
Footnotes
Appendix materials are freely available at http://le.uwpress.org and via the links in the electronic version of this article.
↵1 By some estimates, U.S. wastewater treatment plants will need $271 billion to meet CWA objectives, and infrastructure affordability will remain to be an issue of great concern (EPA 2016a; Ramseur 2017).
↵2 The earliest violation we include is from 2009 because our funding variable starts from 2010 and we include one, two, and three years of future funding as independent variables.
↵3 Some of the facilities have missing compliance information in some years. If there is also no other information including pollutant specific permit limits, enforcement, or inspection, then the observation is treated as missing. Otherwise, we treat the observation as a zero because compliance is mostly reported when violation exists. Missing data for some of the facilities in some of the years is the reason the numbers of observations for the same treatment and control groups differ across pretreatment (row 1) and post-post-treatment (row 2) in Table 2.
↵4 Funding may come with regulatory activities like inspection and enforcement, which in turn may influence compliance. To consider this line of reasoning following a reviewer’s suggestion, we examined whether funding triggers additional inspection or enforcement actions. Using the techniques similar to equations [5] and [6], we found no evidence of increased inspection or enforcement after funding. This result is available on request. We thank the reviewer for the suggestion.
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.