Environ. Sci. Technol. 2001, 35, 4414-4420
Addressing Uncertainty and Conflicting Cost Estimates in Revising the Arsenic MCL
excess cancer risks of between 1 in 100 and 1 in 1000, which is well above the EPA’s target risk range for drinking water standards of between 1 in 10 000 and 1 in 1 000 000. Arsenic exposure is widespread as roughly half of U.S. public water supplies have detectable arsenic concentrations (>0.5 µg/L) in their finished water.
PATRICK L. GURIAN* AND MITCHELL J. SMALL Department of Engineering and Public Policy and Department of Civil and Environmental Engineering, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213
The costs of implementing a lower MCL are (as illustrated herein) subject to debate but will clearly be significant. Figure 1 summarizes three alternate cost estimates: one by the EPA (1), one by an independent group of researchers sponsored by the American Water Works Association Research Foundation (4), and one from our previous research (5) sponsored by EPA grants. Mean predictions are shown for our model with brackets indicating the 95% credible region. The AWWARF study estimates shown are the “residuals influenced” estimates with the brackets indicating the “cost envelope” from this study. (The cost envelope reflects the range of national costs due to uncertainties in the selection of compliance options. The upper bound is based on the assumption that water systems will in aggregate adopt the most-expensive mix of compliance options considered credible by the study’s authors, while the lower bound technology assumes the least-expensive, credible mix of compliance options.) The EPA estimates shown are based on an interest rate of 7% as this is comparable to the 7.8% interest rate used in the other studies. For most of the MCLs, the AWWARF study estimates are several times higher than the EPA estimates, with our model estimates between them. The exception is at 20 µg/L where the AWWARF predictions are lower than the other two estimates.
JOHN R. LOCKWOOD AND MARK J. SCHERVISH Department of Statistics, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213
The current effort to revise the arsenic drinking water standard is one of the first times that the promulgation of a Maximum Contaminant Level (MCL) for drinking water has been influenced explicitly by benefit-cost considerations. Different stakeholders have developed different estimates of the costs, benefits, and appropriate decision-making criteria for a lower standard. In this study, alternative analyses prepared by the U.S. EPA and by independent researchers are compared. The large discrepancies in the aggregate national cost estimates are shown to result largely from differences in the engineering cost estimates for arsenic treatment processes. Further research is needed to resolve these discrepancies. Alternative regulatory approaches, such as providing point-of-use treatment or exempting water systems with high household compliance costs, yield only modest improvement in the overall cost-effectiveness of lower standards but are effective at addressing serious affordability problems for the small percentage of (primarily small) water systems where these problems are predicted to occur. The U.S. EPA may wish to provide more explicit guidance to state regulators and to water utilities as to the conditions under which these options will be acceptable.
Introduction In the final days of the Clinton Administration, the U.S. EPA finalized a rule lowering the drinking water standard, or Maximum Contaminant Level (MCL), for arsenic from 50 to 10 µg/L (1). The Bush Administration has since delayed the effective date of this rule until February 22, 2002, citing the need for further study, including an independent review of the EPA’s Regulatory Impact Assessment (RIA) (2). Delaying the rule has been controversial, provoking much public comment and discussion. The arsenic MCL is worthy of this attention as both the expected benefits and the costs of the proposed regulation are high. The National Research Council Subcommittee on Arsenic in Drinking Water (3) estimated that the current standard of 50 µg/L is associated with lifetime * Corresponding author present address: Dept. of Civil Engineering, University of Texas at El Paso, Engineering Bldg., Room E-201, El Paso, TX 79968-0516; phone: (915)747-6922; fax: (915)747-8037; e-mail:
[email protected]. 4414
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 35, NO. 22, 2001
Such discrepancies in cost estimates are particularly important because the decision-making process for revising the arsenic MCL has been influenced explicitly by benefitcost analysis. Prior to the 1996 Amendments to the Safe Drinking Water Act, drinking water standards were required to be as protective of human health as “feasible”, with feasible defined to include both technological and cost considerations (6). Under the 1996 Amendments, the EPA is still required to identify this feasible level, but the the EPA Administrator is given discretionary authority to establish a less stringent standard if a benefit-cost analysis indicates that this is appropriate. In the case of arsenic, the EPA has used this discretionary authority. The Agency determined that a standard of 3 µg/L would be feasible but not justified on benefit-cost grounds. This is a fundamental change to the decision-making criteria, and the effects of this change merit careful study. The only other MCL that has been influenced explicitly by benefit-cost considerations is the uranium MCL. The EPA determined that a uranium MCL of 20 µg/L would be feasible, but in December 2000, the EPA issued a final rule setting the MCL at 30 µg/L (7). The greater role for benefitcost analysis has not won the support of all stakeholders. The Natural Resources Defense Council (8) has advocated a standard of 3 µg/L, accepting the EPA’s analysis as to the feasible level, but being less receptive to the proposed adjustment to a less protective level determined from benefit-cost arguments. This paper addresses two topics relevant to the current regulatory debate. First, we consider the sources of the discrepancies in estimates of the national costs produced by the EPA and the AWWARF-sponsored study. Second, we consider alternative regulatory approaches as a means of managing uncertainty and avoiding unintended consequences. 10.1021/es001899n CCC: $20.00
2001 American Chemical Society Published on Web 10/18/2001
FIGURE 1. Comparison of cost estimates. For the authors’ model, the brackets indicate the 95% credible intervals. For the AWWARF results, the brackets indicate the cost envelope.
Statistical Simulation Methodology As the details of our benefit-cost model are given elsewhere (5), we provide only a brief summary here. The general approach is that of a system-by-system simulation of the costs and benefits of a proposed arsenic regulation on the nation’s community water systems. Bayesian statistical models are used in a “two-dimensional,” or nested, variability and uncertainty analysis (9). An inner loop, describing variability across water systems, is nested in an outer loop that describes uncertainty in the overall population characteristics as reflected by variation in the model parameter values. Each iteration of the inner loop involves simulating each of the approximately 55 000 community water systems in the nation. Raw water arsenic concentrations, existing treatment type, and current arsenic removal efficiencies are sampled from statistical distributions fitted using available databases (10) and subsequently used to calculate a current finished-water arsenic level for the community water system. For those community water systems with finished water arsenic levels above the proposed MCL, the costs and removals of various compliance options are simulated. Removals are based on information compiled by a previous study, the National Arsenic Occurrence Survey (NAOS) (1113). Nominal removal fractions from the NAOS study are modified by randomly generated factors to reflect variability in removals from system to system and uncertainty in the central tendency of the removals (5). The costs are based on treatment technology cost curves developed by the NAOS study (11, 13) and are modified by a separate set of variability and uncertainty factors. The water supplier’s choice of compliance option is simulated by assuming that the least costly option that achieves the proposed MCL is chosen. After a compliance option is selected,, the arsenic removal achieved by the technology is used to calculate the change in arsenic concentration in the finished drinking water and the resulting exposure reduction. Fatality risk reductions for lung cancer and bladder cancer are estimated using risk factors from a recent study (14) and a linear dose-response model. (These risk factors have been updated since our previous research in ref 5. A summary is provided in Appendix A in the Supporting Information.) Costs and risk reductions achieved at individual community water systems are aggregated to produce national estimates or estimates for subpopulations of interest. The analysis does not consider
other health effects associated with arsenic (e.g., other types of cancer) and a wide variety of potential secondary effects of the regulation such as increases or decreases in exposure to other contaminants that occur as a result of the treatment modifications made to address arsenic; the use of private wells as an alternative to public supplies in locations where the public supply becomes more expensive (or is forced to close due to affordability problems); the environmental and health implications of the transportation of treatment chemicals and disposal of residuals (15); and impacts on site remediation requirements at locations where cleanups are targeted to the drinking water standard.
Sources of Discrepancies in Alternative Cost Estimates The three models considered here (1, 4, 5) differ both in methods and in input parameters. We examined different possible sources of discrepancies to determine which contributed most substantially to the overall differences in national cost estimates. Important possible sources of disagreement are the treatment technology cost curves. These curves indicate the typical cost of a treatment process (e.g., activated alumina, microfiltration) as a function of the size of the drinking water system. The EPA national compliance cost estimate uses technology cost curves developed by an earlier EPA study (16). The AWWARF study uses a set of cost curves, termed the “case-study” cost curves (17), because they were developed by generalizing the costs estimated in case studies of compliance with lower arsenic MCLs for specific utilities. Our model uses cost curves from the NAOS study (11). Previous research (18) has shown that the treatment technology cost curves used by these studies vary greatly. The AWWARF cost curves are generally several times higher than the EPA cost curves, with the NAOS cost curves having intermediate values. This relationship holds across a wide range of treatment types and water supplier sizes. To assess the impact of different cost curves on the model results, we modified our model to use the cost curves employed by the other two studies. These modifications are detailed in Appendix B in the Supporting Information. As shown in Figure 2, when our model was run using the EPA cost curves, the model results closely resembled the EPA’s results. When the AWWARF study’s cost curves were used, our model gave results similar to the AWWARF study. This indicates that the differing cost curves are in large part VOL. 35, NO. 22, 2001 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4415
FIGURE 2. Impact of alternate cost curves on model results for an MCL of 10 µg/L. For the authors’ model, the brackets indicate the 95% credible interval. For the AWWARF results, the brackets indicate the cost envelope. responsible for the large differences in the national compliance cost estimates produced by these different studies. In contrast, we found that the three studies, for the most part, derived similar estimates of the number of systems exceeding various potential MCLs. (Differences in the number of systems affected can explain much of why the AWWARF estimate for 20 µg/L is lower than the others and why there is still some difference between our model results and EPA’s estimate even when our model is run with the EPA cost curves. See Appendix B in Supporting Information for details.) This raises the question of why the treatment technology cost estimates differ so greatly. While it is difficult to identify a single factor, the feasibility of different disposal methods for treatment residuals appears to be an especially important factor. (All of the arsenic treatment technologies transfer arsenic from the source water to another medium which must then either be treated or disposed. These waste products of the treatment process are termed residuals.) The EPA assumed that in many cases treatment residuals, such as ion-exchange regeneration brines, could be discharged to sanitary sewers, an inexpensive option. The AWWARF cost estimates assume that a more costly treatment and disposal process will be required (e.g., precipitation, evaporation, and landfilling). This issue is discussed in more detail in Appendix B in the Supporting Information.
TABLE 1. Effect of System Size (Population Served) and Location on the Number of Systems with Predicted Arsenic Exceeding an MCL of 10 µg/L and the Number of These Predicted To Have Compliance Costs of Greater Than $500/Householda category
all systems
systems exceeding 10 µg/L
systems with costs >$500/ household
Size (Population Served) small (10 000)
38 600 (71%) 12 500 (23%)
1 490 (76%) 389 (20%)
813 (98%) 14 (1.7%)
3 570 (6.5%)
87 (4.4%)
2 (0.22%)
10 100 (19%)
1 210 (62%)
541 (65%)
44 600 (81%)
752 (38%)
288 (35%)
Location top 10 states (AZ, NV, OK, CA, MT, NE, AK, NM, ND, IN) other 40 states
Location and Size small systems in top 10 states other systems
8 500 (16%)
947 (48%)
532 (64%)
46 200 (84%)
1 010 (52%)
297 (36%)
total
54 700 (100%)
1 960 (100%)
829 (100%)
a
Alternative Regulatory Approaches
Mean simulation results are shown. The percentages of systems in each category (shown in parentheses) are relative to the overall total, shown at the bottom of each column, and do not add up to 100% in all cases due to rounding.
Much of the public debate over arsenic in drinking water has focused on the aggregate national costs and benefits of alternative MCLs. In reality, the impacts of the rule vary greatly from water system to water system. Figure 3a shows one example of this variability as predicted by our model, the distribution of annual household compliance costs for an MCL of 10 µg/L. (Household costs are computed assuming 2.7 people per household based on the 1990 U.S. census. The NAOS cost curves are used.) The cumulative distribution of household costs is shown for both population (brown) and water systems (black). The EPA’s estimated affordability limit of $500 in additional costs per household (19) is also shown. While over 95% of community water systems will incur no compliance costs (their finished water is already below 10 µg/L), 1.5% of community water systems (95% credible interval of 1.1-2.1% or between 580 and 1200 systems) are predicted to have household compliance costs exceeding the EPA’s affordability limit. The fraction of the national
population served by these community water systems is only 0.03% (95% credible interval of 0.01-0.1%) as they are predominantly smaller systems. The unequal distribution of community water systems predicted to be out of compliance with a 10 µg/L MCL and predicted to have affordability problems is further emphasized in Table 1. Noncompliant systems tend to be smaller than average and are concentrated in a few states. The concentration of systems with predicted affordability problems is especially high among small systems (population served
