Abstract
This paper provides evidence that the Clean Water Act implemented through effluent limits responded to local water quality. We choose biological oxygen demand as the pollutant and dissolved oxygen as a water quality indicator. We use a panel of permits for 100 plants in Maryland, Virginia, and Pennsylvania for 1990 to 2004. We estimate that decline in water quality by 1 mg/L lowers permits by 5 mg/L. This finding demonstrates greater flexibility than might be expected in an effluent standards–based approach. It suggests efficient resource use, with permits relaxed with water quality improvements and tightened with water quality declines. (JEL Q52)
I. Introduction
The Federal Water Pollution Control Act Amendments of 1972 (FWPCA-72) marked a significant departure from the Water Quality Act of 1965. The focus shifted from attainment of ambient water quality standards, determined by the states, to establishing a national goal of fishable and swimmable waters, by July 1983. This amended law, followed by the two revisions in 1977 and 1987, is known as the Clean Water Act (CWA).
The 1972 legislation shifted responsibility in favor of federal authority. The amendments vested qualified states with the responsibility of issuing permits and enforcing them, under the supervision of the U.S. Environmental Protection Agency (EPA). The regulation of point sources came from technology-based effluent standards rather than local cost-benefit analysis (Freeman 2002). Effluent limitation guidelines were national standards based on the type of production facility, that is, the degree of pollutant reduction attainable by an industrial category. For sewage treatment plants, the monthly average limit of 30 mg/L or at least 85% removal of biological oxygen demand (BOD) was implemented. This uniform requirement is known as the secondary treatment standard.
The key regulatory tool of the CWA for controlling point source pollution is the National Pollution Discharge Elimination System (NPDES), which was designed to issue pollutant discharge limits for point sources. The permits, issued by authorized state governments or the EPA, specified limits on pollutants discharged into surface waters by industrial facilities as well as wastewater treatment plants (also called publicly owned treatment works [POTWs]). They give end- of-pipe concentration and quantity limits for pollutants discharged by major plants.1 States were also required to designate uses to each water body and maintain water quality levels adequate to support these uses. Failure to achieve ambient water quality supportive of the designated uses could replace technology-based effluent standards with water quality– based standards.
From the beginning, economists have been critical of the Clean Water Act. It appears rigid and inconsistent with regulations based on benefits and costs. Because technology-based permits were end-of-pipe requirements—secondary standards for POTWs and best professional judgment for industrial plants—they were viewed as unresponsive to local water quality and thus inconsistent with benefit-cost analysis. Discussing the role of technology-based standards, Freeman (2002, 136–37) noted that “the standards were to be based strictly on technological factors, such as what kind of pollution abatement equipment was available, rather than water quality objectives. Under the Act, regulators did not need to estimate the capacity of bodies of water to assimilate pollution nor to consider the relationship between individual dischargers and water quality.” This view of the CWA portrayed the regulations as rigid; despite the initial success with technology-based permits, implementation of additional controls required for water quality–based permitting system was largely absent.2
The empirical evidence on the functioning of the CWA has been limited and mixed. McConnell and Schwarz (1992) model POTW permit levels as a function of local variables that might increase the demand for environmental quality, such as state income, population density, and stream flow or velocity. These factors would support higher dissolved oxygen (DO), other things equal. However, the McConnell-Schwarz work employed cross-section data, so it deals with only one permit cycle and could not be expected to uncover the relationship between water quality in the previous permit cycle and current effluent limits. Gray and Shadbegian (2004) find that estimates of the dollar benefits of improved water quality did not exert a significant influence on water inspections and enforcements of U.S. pulp and paper mills. The results of their regulatory activity models were less consistent than the pollution equations, since variables related to local benefits were noticeably less significant. They conclude that “perhaps regulators use other, unmeasured, mechanisms to control emissions levels, such as the details of the air and water permit requirements for each plant” (532).
Earnhart (2004a) describes BOD concentration limits for a sample of municipal plants in Kansas as varying across facilities and years, and within years. He notes that, according to state officials, limits are sometimes lowered to address ambient water quality concerns associated with DO. In his 2007 paper, Earnhart documents that these facilities faced limits that differed from federal standards in 24% of the sample of monthly data, over 1990–1998. However, using the same panel of plants, Earnhart (2004b) did not find any significant impact of a time-invariant, watershed-level index of ambient water quality on inspections and enforcement decisions of the state and federal regulators.
The prevailing view of the CWA and the empirical evidence lead to the core question about setting permits under the CWA. Did local water quality influence permit levels during the 1990s and early 2000s? In this paper we provide evidence that ambient water quality directly influenced the effluent limit chosen by the regulator, during a period when technology-based permits were apparently the common practice.
Our empirical approach defines a time period by dividing up the 14-year period of data into distinct plant-specific permit cycles. The permit cycle of each plant is the unique period of time for which a permit is valid for a particular plant. The time-series set-up—with information from past permit cycles—yields a richer analysis because in practice the permit writer must use factors observed in the preceding permit cycle, rather than the current cycle, to determine new permit levels. Using a panel of permits for industrial plants and POTWs, located in Maryland, Pennsylvania, and Virginia, we estimate the impact of DO on permits for BOD.
To our knowledge, this is the first empirical evidence that permit levels were responsive to downstream ambient water quality. We find evidence that permit writers reduced permits when water quality declined and increased permits when water quality improved; at the same time, setting effluent limits such that ambient standards were maintained. This result matters for two reasons. First it demonstrates a greater flexibility than might be expected in a regulatory system that introduced a command and control approach in the 1972 amendments. Second, the consistent result that permits increase when water quality improves and decline when water quality deteriorates may not always indicate efficient regulations, but at least it is not inconsistent with efficient management.
Within the literature on regulatory behavior, there is little direct evidence that ambient environmental quality influences agency actions. This literature typically provides a complex picture of regulation, which admits the influence of political factors but is not completely without regard to the goal of maximizing social welfare. Cropper et al. (1992) studied the reregistration of pesticides. They find that the EPA appears to account for both lives saved from reducing pesticide exposure and benefits from pesticide use in reregistration decisions, with wide variations in the implied values of lives saved. Further they find ample room for political influence in the reregistration process. Viscusi and Hamilton (1999) investigated target risk reductions for cancer and expenditures on risk reduction at Superfund sites. Their findings are less supportive of efficient resource allocation. Their estimates imply that target risks increase with an increase in toxicity of chemicals, or when the pathway of the chemical is residential. Viscusi and Hamilton find an even wider variation in the cost of a cancer case averted. Moore, Maclin, and Kershner (2001) studied relicensing of power plants. They find no evidence for efficient resource allocation in their empirical results. Our concern also lies with agency behavior. We searched for evidence that regulators heed downstream ambient water quality in making their decisions about effluent permit levels. This response of regulators is consistent with efficient allocation of resources.
II A Model of Permit Setting
The objective of modeling the permitting decision is to test whether regulators responded to downstream ambient water quality at a time when technology-based permits were the predominant tools for pollution control. The CWA stipulates that technology-based standards will be set for each major polluter. POTWs must meet the secondary treatment requirement of no higher than 30 mg/L, or 85% removal of BOD, or whichever is more stringent. For industrial plants, best professional judgment (BPJ)3 is exercised to determine the relevant technology-based limits. For streams that do not meet water quality standards necessary to maintain designated uses, water quality–based effluent limits (WQBELs) would need to be set for all dischargers in the stream segment. However, according to the NPDES Permit Writers’ Manual (USEPA 1996) the CWA did not achieve significant progress in accomplishing the goals of setting and meeting water quality standards. The 1987 amendments to the CWA, also known as the Water Quality Act, required all states to identify waters that were not expected to meet water quality standards after technology-based controls on point sources had been imposed. “These plans were expected to address control of pollutants beyond technology-based levels” (USEPA 1996, 6). The first 303(d) list of impaired waters was started in 1996, across all states in the United States. This list has an update schedule of two years to reflect changes in ambient water quality conditions and reclassify waters, as impaired or otherwise.
The Permit Writers’ Manual served as a training manual for all new and experienced state regulators, since there was a renewed focus on developing water quality standards required for meeting designated uses and implementing (technology- or water quality– based) permits to meet these standards. For waters not meeting standards, more stringent water quality–based permits would need to be enforced on all polluters. For high-quality wa- ters,4 that is, water bodies where existing conditions are better than necessary to support the fishable/swimmable use, permits may be relaxed if certain conditions are met and as long as existing uses are not threatened. For high- quality waters, permits might also be made more stringent in response to a decline in water quality under the condition that uses are projected as threatened (even though minimum standards are currently met).
Receiving-stream water quality would have to be evaluated while determining discharge limits for a plant. Computer models simulate the impact of the wastewater discharge from a plant on the receiving stream’s DO under the assumption of technology-based standards. If the model predicts that the receiving stream’s DO will remain above the stream standard, then technology-based limits are selected. If the model predicts that the receiving stream’s DO standard will be violated, then stricter BOD limits would need to be chosen. The permit writer would have access to these engineering reports when determining permit levels.
In high-quality waters, water quality can be lowered based on considerations of economic benefits or municipal and electrical services provided to the community. In no case may water quality be lowered to a level that would interfere with existing or designated uses. Thus, permit levels can be made less stringent (given that DO was higher than the minimum standard) based on local water quality conditions.5 However, actual evidence or records of such socioeconomic impact analyses conducted by permit writers is rarely publicly available.
A review of the Permit Writers’ Manual also reveals that there are plant-specific factors such as age of the plant or vintage of its technology that cannot be observed by outside researchers. Information on plant discharges, permit levels, effluent design flow, type of plant, and monitoring and enforcement actions comprise most of the publicly available data maintained by the EPA and the states. To address this potential bias due to omitted variables, we estimate a model with past water quality as the only explanatory variable and plant fixed effects to control for all other plant- or location-specific, time invariant, factors that influence abatement or wastewater treatment processes. This simple model specification can be justified based on the focus of this paper, which is to find evidence on whether permit writers chose effluent limits that were flexible to ambient water quality.
In equation [1], we specify a simple model where the permit-level Pics for plant i, in cycle c and season s, is a function of past permitting cycle water quality DOi(c − 1)s:
[1]We choose permits for BOD5,6 since data records of monthly average pollutant discharges, tracked by the EPA and states, are most abundant for this pollutant. Further, experience suggests similar discharge behavior for total suspended solids and BOD. Technology for control of BOD is also linked with reductions in nitrogen pollution.
Under the NPDES program, industrial plants discharging BOD5 in their effluents face limits on only quantity (loading), or both concentration and quantity. Municipal discharge permits limit both the concentration and the quantity of BOD5. We model permit levels for concentration and quantity of BOD5 separately.7
A permit cycle and season determine a unit of time that is specific to each plant in our model. A permit cycle is defined by the phasing out of an old permit, which is replaced by a new permit level that is assigned to a polluter. NPDES permits are typically issued with a five-year cycle. Within a cycle, the monthly average permit level assigned to each polluter is constant, except for seasonal variation allowed for some plants. These are incorporated in the permits if a water quality evaluation shows that secondary limits are not adequate during the high-temperature (low- flow) months.8 The permittee is required to meet the more restrictive limit because the assimilative capacity of the receiving stream is lower during summertime. By contrast, water bodies can accommodate the less stringent, wintertime limits normally set at a maximum allowable BOD5 of 30 mg/L for POTWs. Hence we divide each plant’s permit cycle into two seasons—high temperature (low flow) and low temperature (high flow)—in order to accommodate the plants that exhibited seasonal variability.
In equation [1], the core explanatory variable is the measure of downstream ambient water quality at the monitoring location/site nearest to plant i, and prevailing during the previous permit cycle (c-1) and season s. All other plant-specific characteristics that are time invariant are captured by plant fixed effects β i.
Based on our discussions of permit setting, we expect water quality prevailing in the previous permit cycle to have a positive impact on new effluent limits. If past permit cycle DO declines, and water quality evaluation reveals that under critical stream conditions ambient standards will not be met, then regulators reduce effluent limits in the current permit cycle. For example, higher non–point source loads affecting the monitoring station downstream to a plant in Cycle 1 might lead to lower DO in Cycle 1 and possibly a lower effluent limit chosen in Cycle 2. The same positive relationship would hold for increases in DO. If DO rises, and water quality analysis reveals that ambient standards for the stream segment will be met, regulators increase effluent limits because abatement is costly to plants.9
The error term for plant i in cycle c and season s, ε ics in equation [1], is expected to exhibit temporal dependence. Serial correlation arises because unobserved time-varying aspects associated with pollution generation are captured by the error term. Information on time-varying factors such as abatement/treatment technology (e.g., the vintage of the plant) used by the regulator is not accessible to outside researchers. Other potentially timevarying factors for sewage treatment plants are technological/treatment practices such as collection system types, location-specific factors such as areas served, total population served, number of industrial plants discharging into its systems, and volume and type of influents.
Consider the situation in which a plant reports to the regulator regarding a scheduled improvement in treatment technology that is not captured by the design-flow characteristic. The regulator chooses a lower effluent limit for the plant in Cycle 1, after taking into account the status of water quality in the past permit cycle. The error term in Cycle 1 captures this negative effect. In Cycle 2, the regulator will again choose a lower effluent limit, after accounting for ambient water quality in Cycle 1, given that the plant has already installed an improved technology (in Cycle 1). Error term in Cycle 2 will again be small and thus positively correlated with the error term in Cycle 1. Hence, changes in unobserved factors that influence the choice of more (or less) stringent permit levels in Cycle 1 explain why more (or less) stringent permit levels are also chosen in the subsequent Cycle 2.
III. The Plant and Monitoring Station Data
The data consist of a sample of 100 major NPDES plants tracked by the EPA’s Permit Compliance System (PCS) database (USEPA 2001a). There are 26 plants from Maryland, 22 from Pennsylvania, and 52 from Virginia. The maps in Figures 1, 2, and 3 show the locations of these plants in the three states. These maps are adapted from the EPA’s BASINS software (USEPA 2001b). It integrates the spatial information on major point source polluters and water quality monitoring stations.
Location of Sampled National Pollution Discharge Elimination System Plants in Maryland
Location of Sampled National Pollution Discharge Elimination System Plants in Pennsylvania
Location of Sampled National Pollution Discharge Elimination System Plants in Virginia
Figure 1 shows that all but one plant in Maryland discharge into Chesapeake Bay. Some of the sampled plants in Pennsylvania and Virginia, particularly the ones in the west, discharge their effluents into streams and rivers that ultimately flow into the Allegheny, Tennessee, and the Ohio Rivers. Other plants discharge in waters that flow into the Delaware River and then the estuary, while others such as the Chowan and Roanoke Rivers flow into the Albemarle Sound. The remaining plants discharge in waters that ultimately drain into Chesapeake Bay.
Monthly permit levels for these 100 plants are obtained for a time period of about 14 years, from January 1990 to February 2004.10 The universe of this sample is all NPDES “majors” releasing pollutants into public water bodies. The sample includes only those facing BOD5 limits that discharge into a free- flowing stream in Maryland, Pennsylvania, and Virginia. From the EPA’s PCS database, we obtain monthly data on effluent limits, permits start date, design flow, and SIC code (i.e., the type of plant: 55 sewage treatment, 6 electrical services, 3 public sector, and 36 manufacturing plants11).
In order to determine the relevant ambient water quality for each plant we begin by locating the appropriate stream/river segments.
We identify the nearest downstream (to the plant’s discharge point) water quality monitoring station with DO data, using the EPA’s BASINS maps.
The EPA’s central database STORET is the primary source for ambient water quality data. This database incorporates data from a network of monitoring stations on almost every stream and river segment mapped. We obtain monthly DO data for the stations in Maryland and Pennsylvania from the Legacy Data Center, which contains historical (1990 to 1998) water quality data under the STORET system. For the subsequent time period, multiple sources are used. For the monitoring stations in Maryland, we use monthly DO data available under the Chesapeake Bay Monitoring Program. For stations in Pennsylvania, the STORET Data Warehouse provided monthly records of DO data for the time period beyond 1998. For the stations in Virginia, an official of the Virginia Department of Environmental Quality (VADEQ) provided monthly data for the entire time period (1990 to early 2004).
In the current sample, there are 79 downstream monitoring stations with data on monthly DO, over the approximately 14-year period. Three plants were not mapped in the BASINS maps.12 Of the 97 manufacturing and sewage treatment facilities, 59 had a unique ambient water quality monitoring station downstream of its discharge point. For the remaining 38, 26 of them had 2 plants with the same downstream station. The other 12 had 3 plants discharging upstream to the same downstream station.
For each plant, we categorize the time period of about 14 years from January 1990 to February 2004 into plant-specific permit cycles. In the initial years of this dataset, most plants were facing permit levels that were assigned to them in the late 1980s. Revised permits were assigned around the mid-1990s. For most of the 1990s, the plants faced about a five-year permit cycle, as specified by the statutes of the existing regulation. Starting in the late 1990s a number of the plants faced permit cycles that were shorter than the usual five- year time span, ranging from one year to three years. In our sample, the number of cycles per plant varies from one to six. Two plants faced 6 permit cycles, 5 plants had 5 permit cycles, 22 plants had 4 cycles, 51 plants had 3 permit cycles, and 15 plants had 2 cycles. Five other plants had data on only one cycle, most likely on account of phasing out of permits.
In the entire sample of 100 plants, 86 plants have BOD5 monthly concentration limits and 94 plants have monthly quantity limits data. Figure 4 shows that at the state level, permit writers appear to behave differently, despite common pressures from the CWA. As seen in Figure 4, the annual average of concentration permits (mg/L) were declining (over 1990– 2003) in Maryland and Virginia, while in Pennsylvania the annual average of concentration permits went down first and then became less stringent. The annual means of load limits (lbs/day) were more or less constant for the plants in Maryland and Virginia relative to Pennsylvania, where limits were relaxed earlier and made more stringent recently.
Annual Average Concentration and Quantity Permits by State
The average of monthly concentration permits (by season) was 27.8 mg/L, while the minimum and maximum ranged from 5 mg/L to 111 mg/L (Table 1). The average of monthly quantity permits (by season) was 1,871.9 lbs/day, while the minimum and maximum ranged from 9.7 lbs/day to 42,400 lbs/ day for BOD5 loadings (Table 1). The average of monthly quantity permits (by season) seems to exhibit much more variability across plants.
Summary Statistics of Permits and Dissolved Oxygen (DO) for Past Cycle, Three, Two, and One Year(s)
In the sample of 86 and 94 plants that faced concentration and quantity limits, there were 21 and 68 plants that witnessed at least one change in their concentration and quantity permits, respectively. Table 2 shows that the average of the change in concentration permits was - 0.92 mg/L with a range of -70 to 20 mg/L; while the average of the change in quantity permits was 83.55 lbs/day with a minimum and maximum of - 2,186.99 and 5,010 lbs/day.
Table 2 also shows that the average of the upward revisions in concentration permits was 5.38 mg/L, while the average of the downward revisions was - 13.71 mg/L. For quantity limits, the average of the upward revisions was 459.45 lbs/day, while the average of the downward revisions was - 348.37 lbs/ day. For concentration limits, the number of upward revisions in permits (24) was close to the number of downward revisions in permits (30). For quantity limits, permits were made less stringent many more times than permits were revised to more stringent levels (131 vs. 79).
Summary Statistics of Changes in Permits: Upward and Downward Revisions
Our data show that permits were relaxed almost as frequently as they were made more stringent, if not more, especially in the case of quantity limits. This is contradictory to the notion that during the late 1980s and 1990s permit writers were predominantly revising permits to more stringent levels to tackle poor ambient water quality. For high-quality waters, the Permit Writers’ Manual accommodates less stringent revisions of permits based on benefits accrued to a plant’s operation. However, actual evidence of such analyses could not be found in the regulatory literature. In our paper, we find that permit writers relaxed limits in high-quality waters, leading to a decline in water quality due to increase in discharges, as long as uses were not threatened.
The key regressor is the seasonal average of the monthly DO data observed during the past permit cycle of each plant. Since the time period defining a permit cycle is specific to each plant, the monthly DO data is averaged over distinct time periods, across plants in our sample.
Permit writers most likely incorporate the entire past cycle water quality in their permitting decisions, since the minimum ambient water quality standard is necessary at any point in time and hence every month and cycle must be considered. Alternate time spans of three or two years of past permit cycle were considered because states conduct water quality assessments over a period of two to three years. For example, the historical 303(d) list of impaired waters that assesses water quality in all streams and rivers within a state has an update schedule of two years. The Chesapeake Bay Program, which oversees water quality of the Chesapeake drainage basin, tracks water quality status over a period of three years. Water quality during the last year was also considered because conversations with permit writers revealed that effluent limits are occasionally revised in case new data on critical flow becomes available and even after the draft permit has been approved in the public notice period.13
We take the mean14 of the monthly DO data over four different time periods of the preceding permit cycle. We calculate this average by season. The minimum and the maximum of the seasonal average DO recorded over the entire preceding cycle ranged from 4.32 to 13.8 mg/L (Table 1). On average, the DO in the previous cycle was recorded at 9.62. This value was well above what is required to maintain “aquatic life support” use (which is 4–6 mg/L, depending on conditions such as temperature, flow, and salinity).15,16 It is the use directly related to ambient concentrations of DO. It is also part of the “baseline” use (fishable/swimmable) that every stream or river system across the United States needs to meet. The seasonal averages for the last three, two, and one year(s) of the previous permit cycle have mean and standard deviation very close to those for the entire cycle, as seen in Table 1.
The annual average DO values fluctuate on an annual basis, across all three states. However, no clear pattern emerges at the state level (Figure 5). For stations in Maryland, the annual average DO levels were recorded between 9 and 9.8 mg/L over the 14-year time phase. Similarly, for Pennsylvania and Virginia the annual average DO levels fluctuated between 9.3 and 10.5, and 9.2 and 9.9 mg/L, respectively. The mean quality varied little at the state level. However, as Table 1 shows, the state-level averages mask variation over time and across plants.
Annual Average Dissolved Oxygen by State
There were three plants with concentration limits and one plant with quantity limits that had information on permits and past water quality for non-overlapping cycles. These dropped out of our regression sample leaving us with 83 concentration and 93 quantity plants.
IV. Estimation Results
We estimate a fixed-effects model with the log17 of concentration or quantity permits regressed on the past cycle of seasonal average water quality.18 We assume a correlated error structure within plants, and from one permit cycle to the next (by season), primarily in order to account for time-varying unobserved influences that arise once the time-series element of multiple plant-specific permit cycles is introduced.
The estimation procedure begins by transforming the fixed-effects model to get rid of the plant-specific parameters. The within- transformed equation can then be used to obtain a consistent estimate of the serial correlation coefficient by regressing the residuals, obtained from running ordinary least squares (OLS), on residuals for the preceding cycle and same season. Given an estimate of the autocorrelation coefficient, a Cochrane-Orcutt (1949) transformation is done to address autocorrelation of the error term. Finally, OLS on the transformed data will produce an estimate of the coefficient on water quality, conditional on the estimated serial correlation coefficient.
Table 3 reports the fixed-effects estimation results for concentration and quantity limits with AR(1) disturbance in the error terms. The coefficients show that the seasonal mean DO levels prevailing in the preceding permit cy- cle19 had a significantly positive effect on BOD permits chosen by the regulators.
Fixed-Effects Estimation of Concentration and Quantity Permits Model with AR(1) Disturbances
The coefficient on DO of 0.165 in Column (1) of Table 3 means that if mean water quality during the past cycle falls by 1 mg/L, then concentration permits in the subsequent cycle would be reduced by 16.5%, in other words, made more stringent. For a plant with average BOD permits of 27.8 mg/L this translates into a reduction of about 4.6 mg/L. This coefficient can also be interpreted such that a one standard deviation decline in past cycle DO would lead to a 9.0 mg/L reduction in the concentration permits for a typical plant.
For quantity limits, the estimated coefficient on DO of 0.331 in Column (2) of Table 3 implies that if mean DO during the past cycle increased by 1 mg/L, quantity permit chosen in the next cycle and for the same season would be raised by 33.1%. For an average plant with quantity permits of 1,871.92 lbs/day, this translates into an increase of about 619.6 lbs/day. The coefficient can also be interpreted such that a one standard deviation increase in past cycle DO would increase quantity permits by 1,214.4 lbs/day in the current cycle.
Reduction in concentration and quantity permits by 16.5% and 33.1%, respectively, is not that severe in terms of additional costs incurred due to inflexibility in abatement technology or in terms of jeopardizing compliance, given that plants are polluting at only about 30% of their permits.20 ,Earnhart (2007) found evidence on inflexibility in treatment technology by showing that plants cannot adjust their discharges drastically in a short time period, for example, from one month to the next. He found that more stringent limits, at the time of a new permit, increase discharges relative to limit levels, since facilities cannot adjust their treatment fully to the new lower limits. Similarly, new less stringent limits improve performance levels, in other words, plants choose not to or cannot immediately adjust their treatment fully to match the higher limits. In addition, plants are found to significantly overcomply despite increasing costs of abatement. For example, McClelland and Horowitz (1999) have shown that marginal cost of abatement is positive (i.e., plants incurred substantial costs to overcomply). Studies such as those by Bra¨nnlund and Lo¨ fgren (1996) and Houtsma (2003) investigate stochastic water pollution discharges as a possible reason why plants overcomply despite costly abatement. Firms not wanting to violate their permit levels ever will have mean emissions lower than the permit when emissions are random. Bandyopadhyay and Horowitz (2006) also find evidence that plant operators did not exert enough control over the wastewater treatment process to allow for higher effluent levels—driving manufacturing and sewage treatment plants to overcomply.
Hence, these apparently large coefficients can be justified based on the impact of the change in the permit levels on plant discharges, which in turn has implications for abatement costs and degree of compliance of the plant. In response to a stringent revision in permits, following a decline in water quality, the plant is unable to adjust its discharges to the same extent as the change in its permit levels. However, previous studies have shown that these additional costs incurred are not negligible and are increasing with increases in limits stringency. At the same time, the typical plant’s degree of compliance would not be jeopardized with the stringent permit revisions, because of its significant degree of overcompliance with the initial permit levels.
In terms of understanding the impact of this change in discharges on water quality itself, studies like that of Gray and Shadbegian (2004) have shown that, for most plants in their sample, only large increases in pollution discharges generated measurable impacts on downstream water quality. Pollutant discharge data were combined with stream flow data to calculate the transport of pollutants downstream and the resulting water quality on a mile-by-mile basis for every stream. The water quality model used has its origins in simplified models such as the Streeter-Phelps (1925) equation. From this we can infer that large reductions in discharges are necessary to produce a measurable positive impact on downstream water quality. However, the inability of plants to reduce discharges proportionately in response to more stringent permits immediately is common knowledge to regulators. This in turn means that permit writers would need to choose more stringent permit levels in order to influence discharges such that water quality improves at a future time period.
In the case of an improvement in water quality, the large coefficients are feasible, first, because the permit writer is obligated to ensure that the simulated water quality as a result of this new less stringent permit level does not threaten ambient standards as predicted by a water quality model. The permit writer would have access to these engineering reports at the time when the new permit level is chosen. Second, in response to the relaxed permits, plants most likely cannot increase their discharges proportionately due to inflexibility in technology, as shown by Earnhart (2007). However, as previous studies have shown, the cost savings from lower levels of abatement are not negligible since marginal costs of abatement are positive.
In terms of the impact of this increase in discharges on downstream water quality itself, we know that assigning higher limits by 16.5% and 33.1% most likely did not affect the designated use of the water body, since permit writers are obligated to ensure that the simulated water quality under the new less stringent permits does not threaten ambient standards. In response to such lenient revisions in permits, plants most likely cannot increase their discharges proportionately due to inflexibility in abatement technology. In addition, previous studies have found evidence that only large increases in pollution discharges produce a measurable negative impact on water quality. Thus it seems unlikely that increasing concentration or quantity discharges by a magnitude that is strictly smaller than the permit revisions would threaten ambient standards.
During the period of study, the CWA as implemented through the NPDES permits came under considerable scrutiny. We find evidence of the regulation’s flexibility by showing that permit writers revised permits in a predictable way in response to changes in ambient water quality. If past water quality improved, then concentration/quantity permits were made less stringent, that is, revised to higher permissible effluent levels. This would lead to a decline in water quality due to increase in effluent discharges. However, permit writers assigned less stringent effluent limits while attempting to ensure that the new permits did not threaten ambient standards required to maintain designated use.21 If past water quality declined, then concentration/ quantity permits were made more stringent, that is, revised to lower permissible levels, so that water quality would be improved through reduction in discharges.
Robustness Tests
In this section, we report robustness tests conducted on different subsamples of the data. First, we investigate if there is a distinct pattern between permit writers assigning effluent limits for streams that had a history of poor water quality, in contrast to those where technology-based standards seemed to meet designated use. To accomplish this, we focus on the easily identifiable subsample of POTWs that most likely faced water quality–based permits. If a POTW faces a BOD5 limit that is less than 30 mg/L it means that the receiving water is a water quality–limited segment. These reaches do not have sufficient waste load assimilative capacity to allow the discharge of secondary-level treated wastewater. However, for industrial plants, an effluent limit lower than 30 mg/L is not indicative of a water quality–based permit, since one of the crucial factors that determine the permit level chosen by the regulator is the type of production facility.22
During the time period of this study, concentration permits were not revised as often as quantity permits. Perhaps these were revised to water quality–based limits prior to the time period of this study, in the late 1980s, when the regulators were obligated to begin addressing poor water quality. Following concerns of few permit revisions, mostly for concentration limits, the second subsample of plants includes only those plants that received permit revisions. We estimate our models on this subsample including both POTWs and industrial plants.
Lastly, we estimate the impact of water quality on concentration and quantity limits by separating out the subsample of plants that are sewage treatment plants from the industrial plants. POTWs have less control over their pollution generation process since plant operators cannot control fluctuations in the influents they receive from municipalities and industrial dischargers. Managers of industrial plants, on the other hand, can exercise much more control over the pollution generation process. Industrial managers determine how much to pollute based on their plants’ effluent limits and the potential risk of facing more stringent permits in case water quality is declining. We investigate if permit writers accommodate such broad differences in pollution generation process between treatment plants and industrial plants.
Table 4 presents the estimation results focusing on sewage treatment plants that were most likely facing water quality–based limits, during the time period of our study. Of the 55 sewage treatment plants, there were 32 plants that either faced concentration permits more stringent than the 30-mg/L standard, or they faced seasonal limits23 (i.e., 30 mg/L during winter and lower than 30 mg/L during summer). For both concentration and quantity permits, the results are not significantly higher than the coefficients for the entire sample. These are stream segments that had a prior history of poor water quality, which led to their initial assignment of more stringent water quality–based limits from technology-based limits. This means that a rise in DO levels might not lead to as lenient a choice of permit levels as might be observed in a stream segment with no history of poor water quality. On the other hand, decline in DO levels might not be met with extremely stringent choice of permit levels given that current status of water quality meets the minimum standards and that abatement is costly to plants.
Fixed-Effects Estimation of Concentration and Quantity Permits Model with AR(1) Disturbances: POTWs That Most Likely Faced Water Quality–Based Limits
Table 5 estimates the permit models on the subsample of plants (manufacturing as well as POTWs) that faced at least one change in their permit level, from one cycle to the next. Results show that the impact of past water quality is robust to focusing on the subset of plants that witnessed permit revisions during the time period of this analysis. The empirical results are not significantly higher for concentration permits when compared to the coefficient for the entire sample. For quantity permits, the coefficient is somewhat smaller than the entire sample. Perhaps some of these quantity permit revisions are capturing changes in the levels of technology-based limits assigned to the plants.
Fixed-Effects Estimation of Concentration and Quantity Permits Model with AR(1) Disturbances: Plants That Witnessed Permit Revisions
Finally, Table 6 shows that the impact of water quality on concentration permits for POTWs is somewhat higher, compared with industrial plants. For quantity limits, permit writers seem to be revising permit levels for treatment plants by a much higher magnitude, factor of 7, than they are revising permit levels for industrial plants.
Fixed-Effects Estimation of Concentration and Quantity Permits Model with AR(1) Disturbances: Sewage Treatment (POTWs) versus Industrial Plants
Consider the situation where non–point source pollution in the stream segment increases, leading to a decline in DO in Cycle 1. Permit writers know that sewage treatment plant operators find it more difficult to adjust concentration of pollutants in their effluent discharges compared with loads of pollutants. When faced with a decline in DO, a manager of a treatment plant can reduce its BOD loads, keeping concentration constant (presuming that it can exercise less control over its pollutant concentration) by reducing effluent flows. So, for concentration discharges, there is no significant difference in the magnitudes of the coefficient for POTWs compared with that for industrial plants.
For quantity discharges, POTWs do not face the same constraint of less control over the pollution generation process compared with the industrial plants. And, given that a single POTW processes wastewater from municipalities and multiple industrial plants, reducing its pollutant loads might have a greater impact on ambient water quality. A single industrial plant, on the other hand, operates under similar constraints when making its own concentration and load discharge decisions and hence might be facing less stringent load permit revisions in response to a decline in DO.
V. Conclusions
This paper contributes to the literature on regulatory behavior by showing that regulators respond to local water quality when setting effluent-based limits. We present the first direct evidence that both concentration and quantity permits were utilized as policy instruments, by demonstrating the flexibility of these permits to ambient water quality. We accomplish this by estimating the impact of downstream DO, prevailing during the past permit cycle, on the effluent limits chosen in the new cycle. In our sample, water quality in general was good in terms of maintaining the fishable/swimmable standard. We find that an improvement in water quality led to less stringent permits even if this meant a decline in water quality, given that standards were met. On the other hand, a decline in water quality-led to more stringent permits even if current uses were met, for example, if water quality was projected as threatened under conditions of growing population.
Results show that for a 1 mg/L decline in the seasonal mean water quality, prevailing during the past permit cycle, concentration permits in the new cycle would be made more stringent by 16.5%. For quantity permits, a decline in water quality measured in the preceding cycle (and season) of 1 mg/L results in permits chosen in the current cycle to be made more stringent by 33.1%. These results are justifiable based on the inability of plants to adjust discharges drastically in response to more stringent limits and in terms of no significant changes in their degree of compliance, since they were significantly overcomplying with their initial permits. However, previous evidence of increase in marginal costs of abatement with limits on stringency means that the additional abatement costs incurred by the plants would not be negligible. At the same time, the typical plant’s degree of compliance would not be jeopardized with the stringent permit revisions.
The impact of this lower reduction of pollutant discharges on downstream water quality, relative to the reduction in permits, seems unlikely to threaten ambient standards. This is because inflexibility of abatement technology is common knowledge to regulators. And, the previous evidence that only large reductions in discharges produce a measurable positive impact on downstream water quality means that permit writers choose stringent permits such that water quality can be improved at a future time period.
We perform three robustness tests on different subsamples of the data. Results yield similar coefficients on responsiveness of permits to past water quality when focusing on sewage treatment plants most likely facing water quality–based limits, that is, on stream segments with a past history of poor water quality, and for sewage treatment and industrial plants that faced revisions in their permit levels. Estimating the permits model on sewage treatment plants and industrial plants separately yielded a significantly higher coefficient on past DO for quantity limits of treatment plants as opposed to industrial plants. Treatment plants are unable to control their concentration discharges, in contrast to loads of pollutants, while industrial plants operate under similar constraints for both concentration and quantity discharges. Consequently, permit writers revise load limits more drastically for treatment plants, since they process wastewater from municipalities and multiple industrial dischargers and hence might have a bigger impact on ambient water quality.
From our analysis, we provide evidence that water pollution regulation in the United States was responsive to local water quality conditions at a time when it was cast as adopting a command and control approach. Failure in implementing local water quality–based (cost-benefit) decisions led to this change in regulation to end-of-pipe effluent standards since the early 1970s. We find evidence that during the 1990s and early 2000s, state permit writers succeeded in incorporating downstream water quality into their permitting decisions. This evidence is in contrast to the general view of the CWA, since water pollution regulators seem to have responded to ambient stream conditions even prior to the era of implementation of total maximum daily loads for impaired stream segments.
Lessons could be drawn for large developing countries, where the command and control approach has failed because of excessive costs of monitoring and implementation of standards. A framework of designated uses and anti-degradation policy, driven by local water quality considerations, might be more cost-effective to implement. Problems with local water quality threatening designated uses might serve as the trigger for state permit writers about abatement behavior of industrial and municipal polluters. Alternately, in countries with rampant noncompliance with effluent standards, regulators can prioritize enforcement based on local water quality conditions. Of course, it entails the upfront cost of an elaborate monitoring network for water quality that might need to be established, to implement anti-degradation of local waters.
Acknowledgments
We thank John Horowitz for guidance at the inception of this research and Maureen Cropper and Erik Lichtenberg for comments on earlier drafts. Two reviewers have given generous and detailed comments, helping us to clarify our findings.
Footnotes
The authors are, respectively, postdoctoral fellow, Department of Economics and the Murphy Institute, Tulane University, New Orleans; and professor, Department of Agricultural and Resource Economics, University of Maryland, College Park.
↵1 Major municipal dischargers include all facilities with design flows greater than 1 million gallons per day, or facilities serving populations greater than 10,000, or facilities with EPA/state-approved industrial pretreatment programs (i.e., they receive industrial process wastewater). Major industrial facilities are determined based on specific ratings criteria developed by the EPA or the authorized state (USEPA 1996). Some of these factors are nature and quantity of pollutants discharged, character and assimilative capacity of the receiving waters, presence of toxic pollutants in the discharge, and compliance history of the discharger.
↵2 See Andreen (2004) for a discussion on difficulties of implementing the second stage of the NPDES permits: water quality-based effluent limits. Boyd (2000) mentions that during the period July 1998–August 1999, a number of reports and revisions to the NPDES program and anti-degradation policy were introduced in support of water quality planning and management.
↵3 The BPJ guidelines for industrial facilities specify that the performance of each plant will be compared with the best existing performance within that industrial subcategory, and the cost of reducing pollution by a certain amount for the industrial plant is compared with the cost of a sewage treatment plant (with similar design flow).
↵4 For example, the state of Pennsylvania classifies its waters as high quality if, based on at least one year of data, water quality “exceeds levels necessary to support the propagation of fish, shellfish and wildlife and recreation in and on the water by being better than the water quality criteria” for at least 99% of the time for parameters including dissolved oxygen, pH and metals (www.pacode.com/secure/data/025/chapter93/s93.4b.html). The Chesapeake Bay Program established a target concentration for DO at 5 mg/L at all times. Specifically, if the median for the most recent three-year period summer bottom DO exceeds 5 mg/L, then water quality is rated as “good.”
↵5 More details on this can be obtained from the following web sites: www.epa.gov/waterscience/standards/about/adeg.htm, www.mde.state.md.us/programs/water/tmdl/Water%20Quality%20Standards/Pages/Programs/WaterPrograms/tmdl/wqstandards/index.aspx, www.mde.state.md.us/programs/water/tmdl/Integrated303dReports/Pages/Anti-Degradation.aspx, www.deq.state.va.us/wqs/, and www.pacode.com/secure/data/025/chapter93/chap93toc.html.
↵6 In this paper, BOD5 and BOD are used interchangeably. BOD5 is a 5-day test that measures the amount of dissolved oxygen consumed by the decomposition of carbonaceous and nitrogenous matter in a sample of the wastewater (under laboratory conditions, e.g., 20°C) over a five- day period. It has a detection limit of 1 mg/L.
↵7 Quantity limit = Design flow × Concentration limit × Conversion factor. Effluent design flow captures the expected scale of the production or treatment process of each plant and hence is an upper limit to the volume of effluents that the plant is designed to process or generate.
↵8 The well-documented seasonal trend of DO is that during the high-temperature (impact on solubility of oxygen) and low-flow (impact on flushing and re-aeration) summer and fall months (as observed in Maryland, Pennsylvania, and Virginia), ambient concentration of DO is lower.
↵9 Studies like that of McClelland and Horowitz (1999) show that, for their sample of paper and pulp plants, marginal cost of abatement was positive, that is, additional abatement cost incurred is higher for each incremental unit of abatement.
↵10 The EPA’s PCS database provides effluent discharge data from the monthly discharge monitoring reports that are required to be filed by major sewage treatment and manufacturing plants that face a BOD5 limit in terms of concentration (in mg/L) and/or in quantity (lbs/day).
↵11 There are 4 food and kindred products, 4 textile mill products, 7 paper and allied products, 10 chemicals and allied products, 6 petroleum refining and related industries, 1 rubber and miscellaneous plastics products, 2 leather and leather products, 1 fabricated metal products, except machinery and transportation equipment, and 1 transportation equipment.
↵12 For these three plants, information on receiving waters from the PCS database was matched with the location of monitoring stations in the station descriptions to identify the relevant downstream monitors.
↵13 Personal communication, Kyle I. Winters, Virginia Department of Environmental Quality.
↵14 Using the median rather than the mean measures of water quality for the past permit cycle gives similar results.
↵15 In Maryland, this minimum standard is 5mg/L, as specified in note 6.
↵16 The underlying monthly average downstream water quality also had a mean (median) value of 9.6 (9.4) mg/L, much above the minimum standard of 4–6 mg/L (with a range of 1.41–19 mg/L and only 1% of the data under 5 mg/ L).
↵17 BOD5 has been well documented in the literature (Bandyopadhyay and Horowitz 2006) to follow a log normal distribution.
↵18 An OLS model with plant-level dummy variables and standard errors clustered within plants was also estimated. In the model, log of permit levels measured by concentration or quantity limits was regressed on past cycle seasonal average water quality, design flow, type of production facility, sociodemographic aspects, state-level controls, and control for plant-specific effects. The results were much more conservative than the fixed-effects model with serial correlation. OLS with plant dummy variables does not account for changes in unobserved factors in Cycle 1 that lead to change in permit levels (in Cycle 1), both of which lead to a greater change in DO (in Cycle 1) and hence greater change in permits in the subsequent cycle (Cycle 2).
↵19 Using last 3, 2, or 1 year(s) of water quality from the past permit cycle, did not change the coefficient on water quality significantly.
↵20 In our sample, plants discharge at 33% of their concentration limits and at 27% of their quantity limits on a monthly average basis.
↵21 An example of observed permit revisions in the sample of plants illustrates this hypothesis (which is based on Tier 2 of the anti-degradation policy). For one of the plants in our sample DO went up from 8.2 to 9.1 from Cycle 1 (Season 2) to Cycle 2 (same season), quantity permits were relaxed in Cycle 3 (Season 2), which resulted in a fall in DO to 7.6 mg/L in the same cycle and season.
↵22 For example, plants that belong to the subcategories of rayon fibers, thermoplastic resins, and other fibers have technology-based BPT (Best Practicable Control Technology Currently Available) limitations of 24, 24, and 18 mg/ L, respectively, which are all less than 30 mg/L.
↵23 In our sample there are 15 such plants with seasonal variation in their permits for BOD5 concentration and 17 with seasonal variation in their permits for BOD5 quantity.
