Tobacco Documents Online

414 Harvard Environmental Law Review Table 3: Benefits for Alternative Strategies ['Vol. 8:395 Benzene Coke Ovens Acrylonitrile Percentage of BAT Results Relaxed Uniform Standard° Rwnr~tc iw w 62 dosts 57 61 29 Differential Standardc Benefits . 96 81 60 Costs , 37 33 18 Cost per Life Saved (in $/ million) Relaxed Uniforrnb 3.9 1.8 64.2 incremental BAT 41.6 4.7 274. Differential` 2.5 0.93 42.1 incremental BAT 80.4 8.3 286. Net Benefits ($ million/year) BAT -2.2 -13.9 -28.8 Relaxed uniform -1.1 -6.4 -8.0 Differential -0.6 0.5 -4.9 a. Estimates are based on data available to EPA when the standard was proposed. b. Defined as: maleic anhydride: 90 percent coke ovens: doors only acrylonitrile: AN monomer and nitrile clastomer plants c. Defined as: maleic anhydride: 97 percent control for plants with exposure factors greater than 0.6 coke ovens: doors and topside for plants with factors greater than 2.0 acrylonittile: BAT controls for AN monomer and nitrile clastomer plants with exposure factors Qreaterthan 0.2 broad source categories. Section 112 is typical; the BAT standards apply to all plants within the source category. This approach ignores the fact that plants located in high density areas affect many more people and produce much greater exposure reduction for the same amount of emis- sion control.16 Table 3 summarizes the application of these alternate regulatory strategies to the three pollutants. Alternatives that target controls on the high-exposure plants are referred to as "differential standards." Both the relaxed standards and the differential standards reduce costs much more than they reduce benefits. The cost-per-life-saved estimates, however, are still quite high. In fact, the only alternative that yields positive net benefits at a value per life saved of $1 million is differential standards for coke oven emissions. The other alternatives result in net losses ranging 116. The maleic anhydride plant located in St. Louis, for example, accounts for approximately 80% of the overall benefits. See Haigh, Harrison & Nichols, supra note 12, at 27. 19341 {Huzardou.r Air Pollutants 415 from $0.6 million for differential standards for maleic anhydride plants to $28.8 million for the BAT standards for acrylonitrile plants. The wide range in net benefits demonstrates the need for more detailed analysis of alternative regulatory strategies for the specific -1- h-etantc rn °'~dit:0^ the d°~.:,:ls '~f a°ti ~~:"° the L_--C•_ . » , v c.o ua.~ i..e: veuenw illid cosis of alternatives differ considerably among specific pollutants. Since the anal- ysis of the effect of uncertainty presumes a familiarity with the derivation of the estimates, a more comprehensive description of the case study results is presented below. /. Benzene Of the five maleic anhydride plants that would need new control equipment to meet a ninety-seven percent control standard, two already achieve ninety percent control."' Therefore, the marginal cost of increas- ing control efficiency in these plants by seven percent is quite high. EPA would save a substantial amount of money with little change in benefits by relaxing the standard to a ninety percent control level. The estimated exposure reduction is only six percent lower than at ninety-seven percent, but costs fall forty-three percent."8 The cost per statistical life saved drops to $3.9 million, a substantial improvement over the BAT propdsal. The cost per statistical life saved of BAT standards rises to $41.6 million when ninety-seven percent controls are compared to ninety percent con- trols. Therefore, unless the value of a statistical life saved is taken as greater than $41.6 million, the stricter standard is unjustified."y A uniform standard of ninety percent control improves cost-effec- tiveness by screening out plants for which the proposed standard has little impact on emissions or exposure. Differential standards, which set tighter requirements for plants with high exposure factors, offer a more ambitious and controversial way of increasing efficiency.'10 In extreme form, differential standards based on exposure factors lead to plant- specific standards. Limited categorization is a more practical approach. The eight plants emitting benzene, for example, could be split into four "high-exposure" plants and four "low-exposure" plants.1z' A regulation requiring ninety-seven percent controls on only the high-exposure plants, and no additional controls on the other plants, yields ninety-six percent of the benefits of the proposed uniform standard at thirty-seven percent of its cost.122 The differential standard also surpasses the uniform ninety 117. See Benzene Emissions Background Information, supra note 58, at Table I-5. 118. See Haigh, Harrison & Nichols, supra note 12, at 28-29. Unfortunately, the EPA has not developed cost estimates for 90% controls. A conservative estimate of the net benefits of relaxing the standard results from assuming that 90°!o controls would cost just as much as those achieving 97% for the three plants that currently have no controls. 119. Id. 120. See generally Harrison & Nichols, supra note 115 (discussing the advantages of varying standards in response to inter-plant differences in the marginal benefits of emission control). 121. See Haigh, Harrison & Nichols, supra note 12, at 29-30. 122. Id. U4a9VSSZoz

416 Harvard Environmental Law Review [Vol. 8:395 percent alternative, achieving slightly greater benefits at seventy-one percent of the cost.12' Thus, even a crude, two-level differential standard sigrri ficaniey improves the cost-effectiveness of benzene standardc124 2. Coke Oven Emissions The EPA could improve the cost-effectiveness of BAT controls on coke oven emissions by eliminating controls on some sources of emis- sions.123 Controls on charging are substantially less cost-effective than those for doors or topside leaks./26 Eliminating the charging standard reduces costs by twenty-nine percent, but cuts benefits by only nine percent. Controls on door leaks are the most cost-effective component of the BAT standard, with a cost-effectiveness ratio of less than $1.8 million per statistical life saved. By imposing BAT standards solely on door leaks, the EPA would cut costs thirty-nine percent while retaining eighty percent of the benefits of the complete BAT standard.1zJ A total of fifty-four plants would be subject to BAT control require- ments, but seventeen plants currently meet the requirements.'Ia The exposure to coke oven emissions varies widely across the remaining thirty-seven plants, with the exposure factor ranging from a low of 0.58 to a high of 5.93.'r' The wide range in exposure factors offers an oppor- tunity to increase efficiency by restricting the standard - or portions of it - to plants with relatively high exposure factors. Of the thirty-seven plants, twenty-one have exposure factors greater than 2.0 µg/m'-person- years/kg.10 A regulation imposing the door and topside standards only on those plants yields eighty-one percent of the benefits at only thirty- three percent of the cost of the uniform BAT standard."' 3. Acrylonitrile The thirty plants currently emitting acrylonitrile can be divided into four source categories: AN Monomer, acrylic fiber plants, nitrile elasto- 123. Id. 124. Of course, the cost-efl'ectiveness of the differential standards will vary with the categorization of the "high exposure" plants. Id. at 31. 125. Although the data available to the EPA pennit consideration of the individual cumponents of the BAT standard for coke oven emissions, it is insufficient to analyze alternative levels for the different sources within plants. 126. See Haigh, Harrison & Nichols, supra note 12, at 34-35. 127. !d. The door standard still does not yield positive net benefits, however, unless the value ascribed to saving a life is at least $1.8 million (based again on the CAG risk estimate). Id. 128. See id. at 16 (citing U.S. EPA. Draft Tables on Maximum and Minimum Emission Estimates from By-Product Coke Oven Charging, Door Leaks, and Topside Leaks on Wet- C'oaI Charged Baneries (Apr. 1983)). 129. See id. at 33 (citing 1981 Background Information, supra note 90, at app. E). 130. See 1983 Research Triangle Cost Estimate, supra nole 90. 131. See Haigh, Harrison & Nichols, aupra note 12, at 35-36. 1984) Hazardous Air Pollutants 417 mer, and ABS/SAN resin plants.12 The cost-effectiveness estimates vary widely among these source categories. A regulation rPst"etb::o ihe B AT standards tn the twm m~et ~nat_~fF rtoo0 :ci iiytcgiyilCS, the nltrlle elastomer and AN monomer plants, would yield sixty-two percent of the benefits of the complete set of standards at twenty-nine percent of the cost."' The average cost per life saved, however, would still be qver $64 million.t34 Controls on even the most cost-effective category, nitrile elastomer plants, yield a cost per life saved of almost $48 million. Thus, none of the BAT standards for controlling acrylonitrile emissions can be justified on benefit-cost grounds. EPA model plant data indicate that a flare to control column-vent emissions from AN monomer plants would reduce emissions about sev- enty-six percent below uncontrolled levels at a cost of less than !i0.032 per kilogram of acrylonitrile."s Using the average exposure factDr for those plants of 0.248 µg/m'-person-years/kg, the implicit cost per life saved would be under $290,000, a relatively modest sum."6 All of the AN monomer plants, however, already have such flares.13" This fact affords at least one indication that manufacturers have already installed those control devices that are least expensive. As in the other two case studies, widely varying exposure factors offer opportunities to improve cost-effectiveness by limiting standards to high-exposure plants."" Regulations restricting BAT standards to AN monomer and nitrile elastomer plants with exposure factors greater than 0.2 µg/m'-person-years/kg, for example, yield sixty percent of the ben- efits of the complete set of BAT standards at only eighteen perc.-nt of the cost.19 The most cost-effective plant, however, has a cost-effective- ness ratio of approximately $18 million per life saved.10 Thus, although 132. Plants in Ihe last three catesories all use AN monomer as a feedstock. The largest fcedstock use is acrylic fibers. cmptoycd primarily to manufacture rugs and ctothing. ABS and SAN are l.ah resins used to produce hard plastics for such items as pipes, apptiances. disposable utensils, and packaging. Nitrile ctastomcr is a type of rubber used extensively in the automobile industry for hoses, gaskets, and seals. fd. at 17-18. See also Energy and Envtl. Analysis. Inc., supra note 100, at I-1 to 1-9. 133. See Haigh. Harrison & Nichols, supra note 12, at 39-40. 134. Id. 135. See Key & Hobbs, supra note 100, Table VI-2. 136. See Haigh. Harrison & Nichols, supra note 12, at 39-40. 137. See Key & Hobbs, supra note 100, at app. F, at Table F-1. 138. Another possibility is to consider less stringent regulations for the individual source categories. The EPA, however, has not analyzed such alternatives. 139. See Haigh, Harrison & Nichols, supra note 12, at 40-41. The estimated ratuction in emissions from controlling those plants is 312,000 µg/m'-person-ytars, while the csti- mated control cost is $8.4 million. 140. Id. The estimated reduction in exposure from controlling that plant is 98.000 µg/ m'-person-years, while the estimated cost is 5800.000. V4m9VSSzoz

, 418 Han•ard Enpironmentcal ,Lae•vl3rdieev [Vol. 8:395 differential standards substantially improve the cost-effectiveness ratios of acrylonitrile controls, they do not yield benefits commensurate with the costs of control. E. Summarv The results of the three case studies indicate that uniform technol- ogy-based controls have vastly different net benefits depending upon the pollutant and the sourcd category. The implicit cost per life saved by BAT standards varies by a factor of almost 100 among the three pollut- ants. Moreover, in each of the three cases, alternate standards yield higher net benefits than BAT for any plausible value of risk reduction. For two of the three cases, however, even the most cost-effective stan- dards considered fail any reasonable benefit-cost test. In the third case, coke oven emissions, regulation produces positive net benefits for a value per life saved of $1 million only by relaxing the control standard and restricting it to high-exposure plants. These conclusions must be viewed as tentative, for they do not take into account the substantial uncertainties associated with estimating the benefits of controlling airborne carcinogens. lll. UNCERTAINTIES IN ESTIMATING BENEFITS The benefit estimates discussed in the case studies employ point estimates of parameter values based on EPA data. Most of the estimates, however, are highly uncertain; the plausible range for the unit risk esti- mate in each case covers several orders of magnitude. Critics argue that such uncertainties render quantitative analysis too unreliable to guide policy. The key issue, however, is not whether the estimates are precise - clearly they are not - but how robust the conclusions are in the face of substantial uncertainties and potential errors. This Part evaluates each of the four steps in benefit estimation, beginning with the estimation of emission reduction. It addresses both the generic problems and specific examples from the case studies for each step. Additionally, it considers the potential importance of non-cancer control benefits that have not been quantified. A. Uncertainties in Estimating Emissions In theory, estimating emission reductions involves nothing more than monitoring the pollutant source before and after control, and subtracting the results. Despite this apparent simplicity, estimates of the reduction in emissions are far from precise. Several sources of uncertainty, common to the vast majority of regulations likely to be considered under section 112, arise in measuring emissions. In the case of coke oven controls, emissions estimation may be the largest source of uncertainty in esti- mating the benefits of regulation. S~Q~~~~zoz 19841 Hazardous Air Pollutants 419 The uncertainties in estimating emissions and emission reductions are particularly great at the level of individual plants. The EPA bases its emission estimates on a model plant and projects them to actual iodivida:~t sourees using a limited number of nlaut_-s-cific e*fi, cacii of the three cases, for example, EPA assumed that all plants within a given category had the same uncontrolled emission rate. In reality, however, plants are likely to vary widely. An EPA contractor estimated that maletc anhydride plants vary by a factor of three in the amount of benzene that is not converted in the manufacturing process, and that would thus be emitted in the absence of controls.12 Nitrile elastomer plants emitting acrylonitrile show a similar range."' Another factor creating uncertainty in model plant projections is the lack of adequate knowledge about the effectiveness of existing controls. Although many plants already have emission controls of some kind, due to state regulations, Occupational Safety and Health Administration (OSHA) standards, or economic self-interest in recovering valuable feed- stock or by-products, the EPA has made only rough estimates of the effectiveness of such controls. I" Finally, model plant estimates do not consider the effects of varying production levels on eventual emissions. Emissions depend on both the emission rate and the percentage of plant capacity used.'as Few plants operate at full capacity; thus, benefit estimates must be adjusted down- ward to compensate for actual production levels. This problem is most severe when control techniques are capital-intensive because control costs are then fixed across all production levels while benefits vary directly with production levels.16 Therefore, the EPA model plant pro- jections may be highly inaccurate predicters of emission reductions at actual plants. - Even if emission estimates are accurate at the time they are made, they may not provide reliable projections of the impact of a proposed regulation. The effects are most dramatic in the case of maleic anhydride plants, where all of the uncontrolled plants identified by EPA when the regulation was proposed have since closed, switched feedstocks, or in- stalled controls.1O In the case of coke ovens, the depressed state of the 141. See Nichols, supra note 42, at 184-86. 142. Benzene Emissions Background Information, supra note 58, at 1-7. See al.so, Nichols, supra note 42, at 181. 143. See Radian Corporation supra note 100, at 43. 144. See, e.g.. Benzene Emissions Background Information, supra note 58, at Table 1-5 (presenting estimates of current benzene emissions from maleic anhydride plants). 145. Obviously, as capacity utilization declines the production process uses less of the substance and therefore emits less of it. 146. Benefits are proportional to the amount of emissions reduced and the emission reduction is related to the production level. Hence, if production levels drop, so do total benefits. Because capital costs are fixed, the benefit-cost ratio also drops. 147. See iqfra notes 216-218 and accompanying text.

420 Harvard Environmental Law Review [Vol. 8:395 1984J Hazardous Air Pollutants 421 steel industry suggests that additional plants may close over the next few j%'car8. t41 Emission estimates are likely to be most uncertain when each plant has multiple "fugitive" sources (such as leaking doors), as the coke oven case jllustrates. An EPA contractor presented minimum and maximum estimates, which vary by a factor of II for doar leaks, 6.4 for'topside leaks and over 300 for charging leaks."9 The results for coke ovens presented in Part 11 usa a simple average of the minimum and maximum estimates.13O Substitution of the maximum estimates reduces the cost per life saved by less than a factor of two. Use of the minimum estimates, however, increases the cost per life saved by more than a factor of six for the BAT standard.'s' Uncertainties about emissions appear to be most important for coke ovens because: (1) the uncertainties are much greater for coke ovens than for either of the other cases; and (2) the coke oven decision is the "closest" one, with cost-effectiveness ratios in the plausible range, Even with the maximum emission estimates, however, it is not clear that the uniform BAT standard yields positive net benefits. These results suggest that it would be useful to narrow the range of estimates of emissions, from coke ovens, particularly if the tentative decision was to proceed with regulation. A plausible benefit-cost case for the BAT standard is possible only if actual emissions are in the upper end of the estimated range. 2. Uncertainties in Estimating Exposure The dispersion models used by the EPA to predict pollutant exposure contain pervasive uncertainties. In particular, critics question the relia- bility of these models at substantial distances from sources and their ability to predict concentrations indoors, where individuals spend most of their time. Dispersion models for toxic air pollutants combine source charac- teristics, like the height and velocity of releases, with meteorological inputs, including wind speed, direction, and turbulence.12 Although the methodology is straightforward, the accuracy of these dispersion models is uncertain. Model accuracy is difficult to evaluate empirically because, in many cases, measured ambient concentrations at a particular location 148. See U.S. Envtl. Protection Agency. Draft Tables on Maximum and Minimum Emission Estimates from By-Product Coke Oven Charging, Door Leaks, and Topsidc Leaks on wet-Coal Charged Batteries (Apr. 1983) (provided by S. Grove, Office of Air Quality Planning & Standards). 149. See, e.g., 1981 EPA Draft Coke Oven Regulation, supra note 90, at I. The charging standard under consideration for coke ovens, for example, sets an upper bound on the number of seconds of visible emissions during the charging cycle. Jd. 150. See supra notes 90-99. 125-131 and accompanying text. 151. See Haigh, Harrison & Nichols, supra note 12, at 61. 1' e Benzene Emissions Background Information. supra note 58, at 4-1 1 to 4-17. are hard to relate to emissions front the individ;:ul so::rces ,,;odeied.°?- ThP 2cc:..:'w^,. ^ f .1;° a-'- ~ - . ., ~.~.o,:,,,~,~ vctertoraies as the distance from the source increases.'1' As a result, dispersion modeling usually is not carri:d out more than thirty kilometers from the source plant."' In theory this truncation introduces a bias, understating total exposure levels. Concen- trations at greater distances, however, are typically very low, making tht resulting bias very minor as well."~ t Dispersion models are designed to predict outdoor concentrations, but most people spend the vast majority of their time indoors. Recent studies of "indoor air pollution" suggest that concentrations of pollutants indoors may be very different from those outdoors.'S' Many of these studies, however, have involved pollutants that have indoor as well as outdoor sources.'s° Pollutants emitted solely by outdoor sources will have equal or lower average concentrations indoors than those outdoors. "v Therefore, the use of outdoor concentrations to estimate exposure levels may overstate the benefits of the regulations. Another source of uncertainty arises from the failure to use plant- specific data in estimating exposure from individual plants. Exposure levels around a particular plant critically depend on whether prevailing winds blow toward or away from densely populated areas. Variables like stack height, exit velocity, gas temperature and local meteorological data also affect actual exposure." None of the case studies, however, used such plant-specific data to calculate exposure factors.161 153. C. Miller, Exposure Assessment Modeling: A State-of•the-Art Review (19't8) (report prepared for U.S. EPA) (EPA-60(k/3-78-065). 154: See Haigh, Harrison & Nichols, supra note 12, at 62-63. 155. See, e.g., 1979 Assessment of Exposure to Acrylonitrile, supra note 100, at Table VI-5; Benzene Emissions Background Information, supra note 43, at app. E-8. The modeling for maleic anhydride plants was carried out only to 20 kilometers, which my distort comparisons with the other cases. Id. To check for possible bias, exposures tor coke ovens and acrylonitrilc were estimated using data carried out to only 20 kilometcrs and the results were compared with the original estimates. The comparisons were reassur- ing: the differences were only 9~o for coke ovens and 11% for the acrylonitrile plants. See tlaigh, Harrison & Nichols, supra note 12, at 63. 156. See, e.g., 1979 Assessment of Exposure to Acrylonitrile, supra note 100, at Table VI-5. 157. See, e.g.. Spengler & Sexton,lndoorAir Pollution: A Public Jlead'h PnspecYit•e, 221 ScvencE 9( July 1983) (compiling the various primary studies on indoor air pollutants). 158. Id. at I1. 159, ld. 160. Greater accuracy could be achieved by using more plant-specific parameters, some of which could be measured with very low decision costs. It would seem particulatly easy and cost-effective, for example, to use local meteorological data. 161. See 1981 Background Information supra note 90, at app. E(extrapolating frem Pittsburgh mcterological data to all coke oven plants); Benzene Emissions Background Information, supra note 58, at app. E, at E-8 (extrapolating from Pittsburgh mcterologic ad data to all malcic anhydride plants); 1979 Assessment of Exposure to Acrylonitrile. supra note 100, at 26 (basing acrylonitrile results on generalized conditions rather than actual data from any particular area). 9Z.®9VSSzoz

[Vol. 8:395 19841 Hazardous Air Pollutants 423 422 Harvard Environmental Law Review Finally, EPA estimates implicitly assume that individuals spend all of their time close to their homes; the population data are based on place of residence.162 This assumption is accurate for children who attend nPnrtiv crhnnlcy . .vnr rn.r nnn_u.nrL:r.n o~ amo.t.Jtac av6~ ~ vpe...im m st ~.C t6o:.~ N ...e ..-.....) ..~..vv... ..v.. ..v.r.o..~ u .o ev .o sav •a at or near home. It may, however, create larger inaccuracies for adults who work far from their homes. To the extent that concentrations where people work are different from those at home, the exposure factors will be inaccurate. Plants located in areas where more people work than live create higher than estimated exposure levels, but the opposite occurs if plants are located in areas where more people live than work. Uncertainties about the exposure factors used in these case studies have not been quantified. The uncertainties are greatest, however, at the level of individual plants, because of the failure to use plant-specific values for any parameters other than population.16' No systematic sources of upward or downward bias are apparent in the case study exposure estimates. . C. Uncertainties in Estimating Risk Estimating the unit risk factor is the most uncertain step in analyzing carcinogens. Evidence of carcinogenicity typically comes from either high-dose animal studies or from epidemiological studies of workers ex- posed to relatively high concentrations of the substance. All three of the case studies described above relied on epidemiological evidence of car- cinogenicity as the primary basis for risk assessment.'6' Thus, none involves the difficult and controversial task of extrapolating carcinoge- nicity from animals to humans.'65 Risk estimates in the case studies did, however, require substantial extrapolation from high-dose to low-dose exposure.'66 The problem of extrapolating from high-dose data to low-dose ex- posures arises because neither epidemiological studies nor laboratory experiments with animals are capable of detecting low-level risks.'11' Several mathematical models have been developed to perform the nec- essary extrapolations.'68 Unfortunately, neither current theory nor em- pirical evidence provides unambiguous support for any one model.169 162. See, e.g., Benzene Emissions Background Information, supra note 58, at app. E, at E-6. 163. See supra notes 152-62 and accompanying text. See also Harrison, Distribu- tional Objectives in Health and Safety Regulation, in THE BENEFITS OF HEALTH AND SAFETY REGULATION 177-201 (A. Ferguson & E. LeVeen eds. 1981) (estimating exposure to automotive air pollution at work as weu as at home). 164. See supra note 81 and accompanying text (benzene). See supra note 92 and accompanying text (coke ovens). See supra note 103 and accompanying text (acrylonitrile). 165. See E. CROUCH & R. WILSON, R1S[/BENEFIT ANALYSIS 64--a (1982). 166. These studies often measured risk, however, at doses 1000 or more times higher than the exposure levels affected by the regulation. Id. at 114-16. 167. Id. 168. See Nichols, supra note 42, at 164-70 (discussing the various models). 169. Id. Most regulatory agencies, including the EPA, use the "one-hit" model or a variant of it.10 That model assumes that cancer can be induced by a single "hit" of a susceptible cell by a carcinogen. Thus, the model uoes rtui yieid a threshoid beiow which there is a zero risk of cancer_ At low exposure levels, the predicted risk is proportional to the dose; if the relevant dose is 1000 times lower than that at which the risk was mea- sured, for example, the estimated risk is also 1000 times lower. Because1 of this property, the "one-hit" model is often called the "linear" model., It is difficult to tell how much of the linear model's popularity is due to scientific belief in its accuracy as opposed to a value judgment that decisionmakers should be conservative in the face of great uncertainty. In any event, most scientists accept the linear model as providing an upper-bound estimate of cancer risk."' The other models commonly used in estimating cancer risk are con- vex at low doses; as the dose is reduced, risk falls more than propor- tionately.1z Given the same data, these models all predict smaller low- dose risks than the linear model. "' In fact, when the extrapolation from measured risk covers two or more orders of magnitude, as typically happens in EPA regulation, the other models' estimates may be treated as zero because they are so much lower than the linear model's projec- tions. "• Thus, regulations to reduce low dose exposure to environmental carcinogens must rest on a belief that the linear model has a significant probability of being correct. Ideally, experts could assess the probability that each of the possible models is correct, and then use those probabilities to compute an ex- pected dose-response function. Unfortunately, such assessments are not available. If they were, it is likely that the expected dose-response func- tion would be approximately linear at low doses, because the nonlinear models,predict such small risks that the linear model component would dominate so long as the probability assigned to the linear model's cor- rectness was nontrivial. Note, however, that the unit risk factor for the expected dose-response function would not be as large as that estimated by the linear model alone; the estimated risk would equal approximately the pure linear estimate times the probability that the linear model is correct. Thus, while it may be reasonable to assume that the expected benefits of control are proportional to the reduction in exposure, esti- mates of reduced mortality in this article are probably too high, perhaps 170. See, e.g., E. CROUCH AND R. WILSON, supra note 165, at 115. 171. In its preliminary report on benzene, for example, the CAG said that the linear model "is expected to give an upper limit to the estimated risk." See Carcinogen Assessment Group, Office of Health & Envtl. Assessment, U.S. Envtl. Protection Agency, Carcinogen Assessment Group's Preliminary Report on Population Risk to Ambient Benzene Exposures 1 (1977) (unpublished paper). 172. See Nichols, supra note 42, at 164-70. 173. See id. (providing equations for the various models and an example of their widely different predictions at low doses when estimated from the same high-dose data). 174. See fd. fig. 7.2, at 168. ~t~.®~~~~zoz

424 Harvard Environmental Law Review [Vol. 8:395 by a substantial margin, because they rely exclusively on the linear model. Even if one accepts the linear model, controversies about the inter- j.r2tation of 2NiuciTl/oioglcai data make the unit risk estimates uncertain E;xposure levels in epidemiological studies often cause the greatest dif- ficulties because the exposures typically occurred many years earlier when few measurements were made.13 The controversy surrounding the CAG's risk estimate for benzene illustrates this problem and others that can arise in interpretin4 epidemiological studies.16 The CAG based its unit risk estimate for benzene on data from three epidemiological studies. "' In each case, it had to make assumptions about exposure and other factors. Many of these assumptions have been criti- , cized for overstating the risk."' Two EPA analysts, for example, con- cluded that the CAG risk estimate was too high by a factor of four."y An occupational physician testified that the CAG estimate should have been lower by more than a factor of ten.'aa The differences between these estimates and the CAG's are particularly startling because they were based on the same studies and model. Disputes about the appropriate dose-response model and the inter- pretation of highly imperfect epidemiological studies make it impossible to develop unit risk estifnates for any of the three substances that can be defended rigorously. The unit risk estimates used in Part 11, however, probably reflect an upward bias, primarily because they were derived solely from the linear model.'a' 175. See Address by S. Lamm to the EPA in Washington. D.C. (Aug. 21. 1980) (testimony for the American Petroleum Institute at hearings on the proposed standard for maleic anhydride plants) [hereinafter cited as Address by Lamm); see also R. Luken & C. Miller, Regulating Benzene: A Case Study (Sept. 1979) (U.S. EPA unpublished paper). 176. See supra notes 80-89 and accompanying text (discussing the cost-effectiveness of benzene). 177. See Final EPA Benzene Assessment, supra note 65. Studies included: one of workers in two plants using benzene as a solvent to make a transparent film, see Infante, supra note 81, at 76-78; another of Turkish shoe workers using bcnzene-bascd adhesives, see Aksoy. Leukemia in Shoe Workers Exposed Chronically to Benzene, 44 BLOOD 837 (1974); Aksoy, Types oJ'Leukemia in Chronic Benzene Poisoning: A Study in Thirty-Four Patients. 55 AcrA HAEMATOLOOICA 65 (1976); Aksoy, testimony before Occupational Safety and Health Administration, Washington, D.C. (July 13, 1977); and the third of workers in chemical plants using benzene. see Ott, Townsend. Fishbeck & Langner, Mortality Among Individuals Occupationally Exposed to Bcnzene (Exhibit 154) (OSHA Benzene Hearings July 19-Aug. 10, 1977). 178. Critics have raised issues including the CAG's exposure estimates for all three studies, its inclusion of the deaths of two workers not in the original cohort of the Infante study, its failure to exclude workers exposed to other hazardous chemicals in the Ott study, and its estimate of the baseline risk in the Aksoy study. See Nichols, supra note 42, at 170 (summarizing the criticisms of the CAG study); Address by Lamm, supra note 175. 179. See R. Luken & C. Idiller, supra note 175. 180. Address by l-Amm, supra note 175, at 4. 181. See supra notes 170-75 and accompanying text. For benzene, several studies suggest further that the CAG has overestimated the linear model's coefficient. See supra notes 176-80 and accompanying text. 1984] Hazardous Air Pollutants 325 To the extent that the unit risk factors are too high, the expected benefits of controls are overestimated. Revising those estimates down- psrylnnilrile ward reinforces the earlier conclusions that benzene and controls are nnt r ncr_Pffa_fiyn ~az Ir ~1~~ oe.+F-o -s c-e ~--- . to.: c.;o~twaloii inai uniform BAT standards on all three sources of emissions from coke oven plants would not be cost-effective relative to less stringent regulatiorr.103 D. Uncertainties in Valuing Risk Reduction Critics of the use of benefit-cost ana9ysis to evaluate environmental policy often focus on the difficulty of assigning a "value to life."'" TThe empirical studies of wage premiums for occupational risk cited in Part II cover a wide range, from several hundred thousand to several million dollars per life saved. Even that wide range, however, is sufficient to reject BAT standards for maleic anhydride plants and for all four types of plants emitting acrylonitrile. It is also sufficient to indicate cost-hen- eficial modifications of the coke oven regulations, though not sufficier-Itly precise to determine if more limited regulation of coke ovens is justified. Several objections can be raised to the use of wage premium studies to value risks reduced through environmental regulation. If workers are not fully aware of the risks they run, wage premiums will not reflect the workers true willingness to accept risk in exchange for higher pay.1N3 In addition, dangerous jobs tend to be filled by individuals willing to accept risks for lower compensation."" Thus, even if the wage premium studies accurately measure trade-offs acceptable to workers studied, they may underestimate the general population's willingness to pay for reduced risk. Despite these criticisms, some simple examples suggest that the higher end of the range of values estimated by the wage premium studies is more likely an overestimate than an underestimate. If the value per life saved is $5 million, for example, the government should impose auto safety regulations that cut the risk of traffic fatalities in half as long as the control cost per new car is less than $12,500."" With that same value 182. See supra notes 117-24 and accompanying text, See also supra notes 132-140 and accompanying text. 183. See supra notes 125-31 and accompanying text. 184. See, e.g., I)oniger. supra note 23, at 518-19; Rodgers, Benefits, Costs, and Risksr Oversight ojHealth and Environmental Decisionmaking. 4 HAttv. ENVrL. L. ttev. 191, 196-98 (1980). 185. See Raiffa, Schwartz & Weinstein, Evaluating Health F,J)'ects of Socieral Daci- sions and Programs, in DECISION MAKING IN THE ENVIRONMENTAL PROTECTION AGENCY (1977). 186. !d. at 37. 187. As there are roughly 50.000 automobile-related fatalities each year, such a:cch- nology would save 25,000 lives annually. If the value per life saved is $5 million, thai the value of the technology would be $125 billion. If we assume further that there are 10 m;llion new cars sold each year, then the technology would be worth up to $12,500 per car. See Haigh, Harrison & Nichols, supra note 12, at 68. SlG..iJ.J k/ -s1!b'tr0Z

426 Harvard Environmental Law Review [Vol. 8:395 per life saved, a family of four with the median yearly income should be willing to give up about one half of that income in order to face the average overall death rates that prevailed in 1975 rather than those from 1970.'" A more FvnwlnGmental pi,i osopiicai ujeciiun is based on the distine- iion between voiuntary and involuntary risks.'" Individuals are free to choose their jobs (and their cars). In contrast, people, as individuals, have little choice about the quality of the air that they breathe. Society should be willing to pay much more to avoid such involuntary risks, the argument continues, than individuals would spend to reduce hazards over which they have personal control. Supporters of benefit-cost analysis reply that it makes little sense for the government to make fundamentally different trade-offs than individuals would when confronted with similar private choices.190 Decisionmakers, however, may be especially con- cerned about distributional implications if the risks are unusually large and concentrated among a small group of individuals.j9' Two factors suggest that, in general, a lower value should be ascribed to lives saved through the regulation of environmental carcinogens than to many other public choices involving risk. First, cancer is dispropor- tionately a disease of the elderly, so each life "saved" represents rela- tively few additional years of life.'92 Regulatory programs should be evaluated in terms of 'years of life saved, not total lives saved.'y' This suggests that the value per life saved should be lower for evaluating regulations to control carcinogens than for analyzing other programs, such as highway safety, that prevent the deaths of younger people. The second factor is the substantial delay between control expen- ditures and reductions in risk due to time lags between exposure to carcinogens and the onset of disease. Conventional benefit-cost analyses discount streams of benefits and costs to reflect the time value of money. Economists differ as to whether discounting should be applied to health 188. See Bailey, supra note 70, at 45-46. 189. See E. CROUCH AND R. WILSON, supra note 165, at 85. 190. See, e.g., Zeckhauser, supra note 67, at 419. 191. For a general discussion on distributional effects of environmental regulations, see Harrison, supra note 163; Harrison and Portney, Who Loses from Reform of Environ- mentat Regulation in REFORM OF ENVIRONMENTAL REGULATION (W. Magat ed. 1982). See also D. HARRISON, WHO PAYS FOR CLEAN AIR? (1975) (discussing the cost and benefit distribution of federal automobile emission standards). 192. The death rate for the type of leukemia associated with benzene, for example, is more than 26 times higher among people aged 70 to 74 than among children aged I to 5. See Final EPA Benzene Assessment, supra note 65, at Table 1. 193. Zeckhauser and Shepard argue that mortality benefits should be summarized in terms of the discounted number of "Quality Adjusled Life Years" (QALYs) saved. Their QALY measure adjusts benefits to include reductions in the quality of life due to disability, for example. See Zeckhauser & Shepard, Where Now for Saving Lives?, 40 LAw AND CON rEMPORARY PROUs. 5 (Autumn 1976). 19841 Hazardous Air Pollutants 427 benefits.'y' Most theoretical discussions support discounting,''" but in common practice the timing issue is ignored.'% Discounting reduces the relative value of saving lives through control of a.nvi:vinu2iual cauciTiogctis, fircause the benefits of reductng exposure are realized many years after the costs are incurred. At a discount rate of five percent, for example, a twenty-year time lag reduces the value ,of risk reduction by sixty-two percent compared to an immediate risk rt- duction, say through improved fire protection.197 The valuation of risk reduction remains uncertain and highly conten- tious, with little prospect for agreement on any particular dollar. value for saving a life. The problem is at least as much one of ethics and politics as it is one of science and the interpretation of empirical evidence. EPA, however, cannot avoid making trade-ofTs between protection and control costs, whether it does so explicitly or implicitly. Fortunately, precision may not be very important because many decisions are correct over wide ranges of values, Moreover, it is possible to narrow the range ptesented earlier by reducing the high end. Values much in excess of $1 million per life saved appear difficult to justify, particularly for airborne carcinogens for which the benefits are delayed and the lives saved are relatively short. E. flnquantifted Benefrts18 EPA's procedures almost certainly overstate the cancer-reduction benefits of controlling hazardous air pollutants. By focusing solely on cancer in its quantitative estimates for section 112 pollutants, however, the EPA may miss other important health and environmental benefits. Many carcinogens, including the three considered here, have also been associated with non-cancer health effects at relatively high doses.'" For most of these non-cancer effects, however, scientists generally accept 194. See, e.g., Raiffa. Schwartz & Weinstein, supra note 185, at 42-49. 195. See, e.g., id. at 49. 196. See, e.g., Page, Harris & Bruser, Removal of Carcinogens from Drinking Water: A Cost-Bencfit Analysis (Jan. 1979) (Social Science Working Paper #230, California Insti- tute of Technology, Pasadena, Cal.). 197. The equation for discounting is B/(I+r)` = PV, where B is the benefit in current dollars, r is the discount rate, x is the number of years from today in which the benefit accrues, and PV is the present value of the benefit. In the example given, r equals .05 and x equals 20; the present value of the benefit today (PV) is 37% of B. 198. This article, and therefore this section, considers only human health benefits; ; no consideration is given to benefits related to reduced wildlife and plant damage from these toxic substances. See, e.g., Acrylonitrile Assessment Document. supra note 77, at 88-100 (describing the effects of acrylonitrile on plants, domestic wildlife and aquatic organisms). 199. Oltice of Research & Dev., U.S. Enval. Protection Agency, Assessment of I Health Effects of Benzene Germane to Low-level Exposure 48-65 (1978) (EPAfi00/I-78- I 061) (noting benzene's association with aplastic anemia and other serious blood disorders) GL09VS~~~z

428 Han'ard Environmental Law Review (Vol. 8:395 the concept of zero-risk thresholds, and current environmental exposures ..~.,n°3r to !;e £r ~e:..Y: tti..o r.. ~l..v...u..a Y ,. I tlJ.mo ~.r . ~M Chromosomai damage - mutagenic effects - may be an exception, as scientists are less willing to assume that such effects have thresh- olds.2°' All three of the case study pollutants appear to cause chromo- somal damage."2 None, however, has been associated with birth 'defects, and analyses by EPA's health experts emphasize mutagenic evidence as corroborating the carcinogenicity of the substance, rather than as a sep- arate concern.203 The "conventional pollutant" benefits associated with controlling some hazardous air pollutants may be more significant. States have reg- ulated benzene and acrylonitrile to help meet the ambient standard for ozone.20' Coke ovens have been regulated to meet the particulate ambient standard.203 In addition, controls on section 112 pollutants may also control other pollutants. If maleic anhydride plants use incineration to control benzene emissions,-for example, they would also reduce carbon monoxide, a "conventional" pollutant covered by an ambient standard.z06 Occupational exposure represents still another potentially important omitted benefit category. Some controls designed to reduce emissions to the ambient environmetqt also reduce the exposure of workers. This effect is most likely to be significant when the emissions are from low-level, fugitive sources, as is true of coke ovens. If the sources are elevated stacks, as with maleic anhydride plants emitting benzene and the acry- lonitrile plants, environmental controls are unlikely to have much impact on workers. The importance of these omitted benefit categories'varies widely across specific regulations. In the three cases discussed, they do not affect the basic conclusions for maleic anhydride plants and acrylonitrile, primarily because the cancer benefits are so small in those cases and the Ihereinal'ter cited as Health Effects of Benzene); EPA Coke Oven Assessment, supra note 90, at 54-63 (noting the acute and chronic toxicity of coke oven emissions); Acrylonitrile Assessment Document, supra note 77, at 116-48 (noting the acute, subacute and chronic toxicity of ucrylonitrile). 200. See, e.g., Nichols, supra note 42, at 152 (benzene). 201. Id. at 162. 202. See Acrylonitrilc Assessment Document, supra note 77, at 156-66; Final EPA Benzene Assessment, supra note 65, at app. I-S; EPA Coke Oven Assessment, supra note 90, at 27-52. 203. See, e.g., Final EPA Benzene Assessment, supra note 65, at app. 1-S. 204. See, e.g., [3 State Air laws) ENVT REt'. (BNA) 521:0621, 521:0631 :0664 (1983) (Texas' regulation o(volatile organic compound emissions); II State Air laws/ ENVT REP. (DNA) 346:0501, 346:0521 (1983) (Florida's regulation of volatile organic compound emissions). 205. See, e.g.. II State Air t.awsl ENVT REt•. (BNA) 301:0501, 301:0513-:0515 (1982) (Alabama's restrictions on coke oven emissions): fd. at 336:0501, 336:0512 (1984) (Dela- ware's restrictions on coke oven emissions); 12 State Air Laws) ENVT REP. (BNA) 411:0501, 411:0516 (1982) (Michigan's restrictions on coke oven emissions). 7 '5 Fed. Reg. 26,660, 26,661 (1980). 1984J Hazardous Air Pollutants 429 only potentially important omitted benefits appear to be those assoriatPa with conventional nnttutants. Tn thp rYtPnt tt;-r ----rti br„u.. ~.w aic i-pur- tant, benzene and acrylonitrile are probably best addressed by the fratne- work established for other conventional pollutants - state implementa- tion plans for existing sources and new source performance standards for new ones. ' I The omitted benefit categories are more troubling for the coke ovet{ case, primarily because it is a closer decision on the basis of cancer reduction benefits alone. The quantitative significance of the additional benefits from reduced worker exposure and reduced particulate emissi•ans cannot be evaluated, but it seems unlikely that they would be sufficient to justify the uniform BAT standard over the alternatives of a less strin- gent uniform standard or a differential stategy targeted at high exposure plants. F. Summary Huge uncertainties pervade estimates of the benefits of regulating airborne carcinogens. As a result, the figures presented in Part II tr us,t he viewed with a strong dose of skepticism; they may well be in error by orders of magnitude. These uncertainties, however, do not alter the major conclusions of the case studies. The clearest conclusions emerge for the four source categories etnit- ting acrylonitrile. The cost-effectiveness ratios for emission controls were ten or more times higher than the plausible range of values for risk reduction.'O7 Nothing in this section has suggested that benefit estimates err by that margin.='8 The calculations for benzene emitted from maleic anhydride plant~ gave u substantially narrower result, although the estimated cost per life saved was still in excess of $6 million.219 Several factors suggest that an accurate estimate of the expected cost-effectiveness ratio would be sub- stantially higher. They include: (I)-the general issue of the approprtate dose-response model;=1° (2) evidence that the CAG overestimated the linear model's risk factor;=" and (3) a significant rise in the cost per life saved when the estimates are adjusted for less than full capa.:ity operation.'''- The most ambiguous results arise in the case of coke ovens, although a BAT standard for charging emissions almost certainly would fail a 207. See supra notes 132-140 and accompanying text. 208. Unless, of course, one favors a nonlinear dose-response model, but that would cut in the other direction. 209. See supra notes 87-89 and accompanying text. 210. See supra notes 167-175 and accompanying text. 211. See supra notes 176-180 and accompanying text. 212. See supra notes 145-146 and accompanying text. osro_q-,V_C~szoz

430 Harvard RnvironmentaP Law Review [Vol. 8:395 198A] .f:azardous Air Polfcr:ants 431 benefit-cost test.21 Whether the uniform door and topside standards generate positive expected net benefits remains in doubt. Two issues raised in Part 111, however, weigh against those standards: (1) the likeli- .. fo.1 .•.oL.214 ....,d tinM that the nurP linear mw[pl nv^rwctimnte~ tFt. o..~r....... .. r_... (2) the GvidGfi~;G Su~Csllug that a valne on rlsk reduction much in'excess of $1 million per life saved cannot be justified.21S In fact, it is unclear whether even differential standards limited to high-exposure coke plants would yield positive net benefits. Such standards, however, unquestion- ably represent an alternative superior to uniform BAT standards. G. Postscript Recent developments reinforce our conclusions regarding benzene emitted from maleic anhydride plants and cast further doubt on the wisdom of imposing standards on coke ovens. After the maleic anhydride standard was proposed in 1980, five important changes took place: (1) four plants shut down; (2) two plants converted to n-butane, appar- ently in response to higher benzene prices; (3) the largest plant installed controls and began to convert all of its capacity to n-butane; (4) an additional plant was "discovered;" and (5) EPA reduced the BAT stan- dard to the equivalent'of ninety percent control.216 As a result, had the standard been imposed, it would have applied only to the newly discov- ered plant, a small one located in a lightly populated area, and the estimated health gain would have been to prevent approximately one case of cancer every 3..00 years.21 Citing those minimal potential health impacts, the EPA withdrew the proposed standard for maleic anhydride plants in early 1984.218 More recent estimates from the EPA indicate that coke oven plants also pose a smaller threat than estimated earlier. Data in a recent EPA report suggest that the BAT standards would save less than five lives per year, in contrast to over ten lives per year estimated on the basis of the earlier data. The newer EPA estimates rely on higher emissions but much lower exposures, based on newer modeling using meteorological data for each plant.219 Even more recently, the CAG lowered its estimate of unit 213. See supra notes 125-127 and accompanying text. 214. See supra notes 167-175 and accompanying text. 215. See supra notes 67-72 and accompanying text. 216. See A. NICHOLs, TARGETINO ECONOMIC INCENTIVES FOR ENVInONMENTAL PROTECTION 157 (1984). 217. Id. 218. 49 Fed. Reg. 8386 (1984). 219. See Office of Air Quality Planning & Standards, U.S. Envtl. Protection Agency, Coke Oven Emissions from Wet-Coal Charged By-Product Coke Oven Batteries - Back- ground Information for Proposed Standards (Sept. 1983) (draft EIS) (Research Triangle Park, N.C.). This document does not calculate reductions in fatalities or exposure. It does, however, include estimates of unit risk and baseline emissions and cancer cases, from which it is possible to measure average exposure per unit of emissions. The document also gives estimates of emission reductions, from which reductions in cancer cases can be estimated. risk and the EPA learned that additional plants have shut down,.so the estimated annual reduction in cancer cases has fallen to about two.220 It thus appears that coke ovens are no longer a "close" case; although no cost estimates are available for the closed Dlants, the estimated cost 'per cace avnidrrt fnr ths. nAT ,t~~.~-QaS ~., .ot t:~ c Pc .. oae.o.a o.v a., a,A..VJJ Vl yJ 1111111V11. IV. FINDINGS , 1 The three case studies illustrate many of the problems and uncer- tainties involved in estimating the benefits of environmental regulation. Although benefit-cost analyses of such regulations can never be very precise, these studies suggest that quantitative assessments of benefits can provide valuable information to regulators interested in improving the efficient use of society's resources. In- this Part, some of the lessons from the case studies are summarized, first with respect to section 112 of the Clean Air Act and then with respect to the more general use of benefit-cost analysis to evaluate strategies for regulating health-threat- ening pollutants. A. Section 112 In dealing with "hazardous air pollutants" covered by section 112 of the Clean Air Act, the EPA has consistently followed a technology-based approach to regulation. The "generic" policy proposed in 1979 would have formalized this approach in an attempt to speed up and routinize the process of listing and regulating such substances.221 More recently, some members of Congress have suggested forcing EPA regulation of sectiort 112 pollutants by giving the agency a deadline for making deci- sions on a list of thirty-seven substances.222 The BAT approach to regu- lation is flawed because it implicitly treats airborne carcinogens as a homogeneous class. The case studies indicate that airborne carcinogens are a very heterogeneous class, with wide variations in benefits (and costs) across substances and source categories. 1. Heterogeneity Even within a small sample of three pollutants studied, the risk reduction benefits from controlling emissions• vary enormously because of differences in carcinogenic potencies and in exposure patterns. Each kilogram of coke oven emissions, for example, causes about 500 times as much cancer risk as a kilogram of acrylonitrile or a kilogram of benzene emitted from a maleic anhydride plant.27' Regulatory analyses that focus 220. Personal communication from Teresa Gorman, Office of Policy, Planning & ~ Evaluation, U.S. Envtl. Protection Agency, Washington, D.C. (Apr. 12, 1984). 221. See supra notes 43-50 and accompanying text. 222. See supra text accompanying notes 55-56. 223. See Haigh, Harrison & Nichols, supra note 12, at Table 2.1. rN®9G/~~C.®z

432 Harvard Environmental Law Review [Vol. 8:395 on the feasibility and affordability of controls ignore these critical differences. The cost per unit of risk reduction also varies greatly across the three cases, differing by a factor of more than Inn hPtwPP., pntcv r,lan r ... ..tc _ ., and the least cost-effective acrylonitrile category. These wide variations suggest that a policy of applying BAT standards to all sources emitting airborne carcinogens imposes higher than necessary costs to achieve any given level of overall risk reduction. Individual substances and source categories must be conjidered on their own merits, taking account of potencies and exposure levels as well as technology and affordability. 2. Modest Benefits From Control The desirability of strict regulations on airborne carcinogens is easily overstated. in both the benzene and the acrylonitrile cases, for example, a small number of sources emit millions of kilograms of proven human carcinogens each year. Moreover, the controls being considered are em- inently affordable; their costs are estimated at less than two percent of total sales.22' The case studies show, however, that only modest health benefits are likely to result from the regulations. BAT standards for both acrylo- nitrile and maleic anhydride plants would have a combined effect of avoiding less than one cancer death per year. The coke oven standards would provide substantially larger benefits, but even in that case the gain in public health seems rather modest for standards that apply to a major industry on a nation-wide basis. Of course, it is not certain that all section 112 regulations would yield similarly small benefits. The case studies, however, cast doubt on the proposition that control of airborne carcinogens will lead to major reductions in the nation's cancer burden. The fact that the pollutants considered here have been assigned relatively high priority by EPA re- inforces this skepticism. B. 2'Dre Role ofBenefit Cosi Analysis l. Evaluating Proposed Regulations Existing methods of quantitative assessment may not yield clear answers as to the cost-effectiveness of regulations in all, or even most, cases. Many of the components in benefit estimation are highly uncertain. Because the final estimate typically is a multiplicative function of these individual components, the overall level of uncertainty is extremely high. Nonetheless, robust conclusions often can be drawn to help regulators avoid imposing some regulations for which the benefits are far smaller than the costs. Benefit-cost analyses may also identify regulations that clearly provide positive net benefits, although none of the instant case studies identified such a regulation. 224. See, e.g., supra text accompanying notes 87-88. 19841 Hazardous Air Pollutants 433 2. lmproving Regulations Most discussion of benefit-cost analysis focuses on its role as a"test" for ,.r+...r,., ri.r.^s. °,d '." ...b:,•lot'vi.••o.Iu _..,• • .a:i~cnr•t-CC.St ae~a~y~iS~~y~iS iS CVen more USC(U1, GOW- ever, as a tooi for designing regulations. In all three of the case studies, less stringent controls yielded most of the benefits of the BAT standards at far lower cost. Although none of these modified uniform stand3rds resulted in clearly positive net benefits, all were more efficient than the original BAT standards. If benefit-cost principles were applied early•in the regulatory process and used to guide the selection of control options for detailed analysis, even larger gains could be realized. The case studies indicate that regulatory efficiency is maximized by exploiting marginal differences in the benefits of control among sources. These differences arise primarily because of differences in population densities around plants; the public health benefits of controlling emissions are far larger in cities than in lightly populated rural areas. In all three cases, restricting standards to areas where the marginal benefits of control are relatively high led to impressive efficiency gains over unifilrm standards. 3. Information Requirements and Delays If they are to be useful to decisionmakers. analytic techniques can not rely on data that are unduly expensive or time consuming to obrain. Analysis is not free; it consumes scarce resources that could be put to other uses and may cause delays in an already lengthy regulatory process. Fortunately, a great deal can be done with information that is alrc:ady collected by EPA. Also, a sharper set of decision criteria should speed tip rather than delay the regulatory process. Note that the technical data for ap three case studies were based on information developed as pat1. of EPA's BAT strategy for controlling hazardous air pollutants. Thus, per- forming the kinds of analyses presented in this article should not signif- icantly increase either the costs or the delays of the regulatory process. For relatively close decisions, such as the coke oven case, additional information could prove useful, particularly in four areas: (1) cost and emissions estimates for a wider range of control options; (2) more plant- specific data for exposure estimates (as were recently developed by EPA for coke ovens); (3) estimates of non-cancer benefits, particularly those associated with conventional pollutants (such as ozone and particulate matter); and (4) development of techniques for estimating the expected level of cancer as well as the "plausible upper bound" now used by E:PA. Adoption of benefit-cost principles could reduce the amount of in- formation required to regulate in many cases. Current efforts, fat ex- ample, typically include studies of the "economic impact" of regulations, attempting to predict their effects on plant closings, product prices, and the like. Such impacts are of second-order importance relative to the direct benefits and costs of control. Application of benefit-cost principles in allocating agency resources may also reduce the costs of analysis by leading to the curtailment of the regulatory process before large expenses ZS®9W s,SZ®Z