Professor Sunstein’s fuzzy math

Professor Sunstein’s fuzzy math

McGarity, Thomas O

INTRODUCTION

For many years, Professor Sunstein has been one of academia’s most persistent and persuasive advocates of federal agency use of cost-benefit analysis in health, safety, and environmental decisionmaking. A cost-benefit balancing approach to governmental decisionmaking squares nicely with civic republican values that acknowledge the important role that government must play in achieving a fair distribution of resources. At the same time, it urges informed and fair-minded professionals to decide, after due deliberation, what is best for the rest of us.1 Professor Sunstein has always preached a “soft” version of cost-benefit analysis that takes an honest stab at quantitative assessment of the costs and benefits of major health, safety, and environmental regulations, but does not necessarily allow the result to dictate the ultimate outcome of any given rulemaking effort. Agencies should, in Professor Sunstein’s view, use the knowledge gained from cost-benefit analysis for guidance in setting regulatory priorities and defining the outer bounds of rational decisionmaking.

In The Arithmetic of Arsenic,2 Professor Sunstein goes beyond the theory of cost-benefit analysis to examine in some detail an important application of that approach to regulatory decisionmaking in the real world. This altogether commendable exercise leads Professor Sunstein to what, apparently for him, is the surprising conclusion that quantitative risk assessment and monetization techniques yield a very broad range of plausible benefits for nearly all of the available regulatory options. Rather than shake his faith in the value of quantitative cost-benefit analysis as a decisionmaking tool in risk regulation, this revelation leads Professor Sunstein to lessons that are fully consistent with the soft cost-benefit approach that he has always advocated, with the additional caveat that courts should give agencies a great deal of leeway in reviewing health, safety, and environmental regulations. Ultimately, the source of Professor Sunstein’s unwillingness to abandon the paradigm altogether is his profound and abiding lack of confidence in the capacity of an uninformed and simpleminded public to make wise decisions about the magnitude of health, safety, and environmental risks, and the steps that should be taken to reduce those risks.

Parts I and II of this Response will examine the EPA’s arsenic risk assessment and benefits analysis in light of Professor Sunstein’s “Questions and Doubts” about the proper shape of the dose-response curve at low levels and his observations on the proper monetary sum that the EPA should assign to statistical life and limb. Part III consists of an in-depth critique of what Professor Sunstein characterizes as a “peer review” of the EPA’s benefits assessment by an economist and a research associate at the American Enterprise Institute. The point of this exercise is to demonstrate that leaving the benefits assessment function to the professionals, as Professor Sunstein is inclined to do, will, in practice, bestow large amounts of regulatory discretion on self-proclaimed experts who lack expertise in matters about which they freely opine and who have sharp ideological axes to grind.

In Part IV, I highlight some of the curiosities and contradictions that adhere to Professor Sunstein’s continued commitment to the synoptic paradigm despite his clear understanding of the huge uncertainties that necessarily attend attempts to quantify and compare the costs and benefits of health, safety, and environmental regulation. For example, after fully acknowledging that the benefits of the EPA’s arsenic standard could vary over an extraordinarily wide range, Professor Sunstein insists on carrying some of the numbers to the point of meaningless precision. Although he declines to provide his own best estimate of the benefits of the regulation on the ground that it would be “too speculative,”3 he makes it clear that he believes that the EPA’s estimate was too high. Professor Sunstein believes that the numbers that lack the precision to inform decisions about which communities should receive government subsidies for treating drinking water are precise enough to support a “sliding scale” of regulations capable of protecting some communities more than others.

Part V addresses Professor Sunstein’s claim that quantitative cost-benefit analysis is useful, despite its severe limitations, as a cure for the “intuitive toxicology”4 that guides an uninformed and simple-minded public to demand bad government policies that lead to irrational health, safety, and environmental standards. Put simply, I believe that Professor Sunstein places too much faith in professionals like the individuals critiqued in Part III and he does not give enough credit to the common sense of ordinary citizens.

Finally, Part VI examines some solutions to the dilemma posed by large uncertainties in the decisionmaking process. Agreeing with Professor Sunstein that “soft glance” judicial review is appropriate in this area, I suggest that he think twice before abandoning informational solutions, technology-based approaches, and the margin-of-safety concept in favor of a cost-benefit balancing approach that is dominated by experts and inaccessible to the lay public.

I. PROFESSOR SUNSTEIN’s ARSENIC CASE STUDY

Act (“SDWA”), which he correctly observes reflect “a strong commitment to cost-benefit balancing.”5 However, the enactment of the 1996 amendments by no means signaled a “dramatic shift”6-much needed in Professor Sunstein’s view-from earlier approaches to risk regulation. Cost-benefit analysis has played a role in environmental regulation from the outset of what Professor Sunstein calls “1970s environmentalism.”7 Cost-benefit balancing formed the core regulatory concept for the Federal Insecticide, Fungicide and Rodenticide Act Amendments of 1972 (FIFRA),8 and the 1976 Toxic Substances Control Act (TSCA)9-two of the least successful statutes of the environmental decade. The congressional decision to employ a cost-benefit decision criterion in both of those early statutes is arguably the reason that the EPA has canceled so few of the old “bad actor” pesticides under FIFRA and has not taken any significant action to limit exposure to toxic chemicals under TSCA. The process of gathering health, environmental, and cost data, dealing with large uncertainties in the data and associated models, quantifying and monetizing benefits, comparing costs and benefits of realistic alternatives, and providing support for the agency’s conclusions in an administrative record has thoroughly stymied government action under both of these statutes.

Cost-benefit analysis, however, does not play as prominent a role in standardsetting under the SDWA as it does under FIFRA and TSCA. The starting point for establishing a “primary drinking water regulation” is a “maximum contaminant level goal” (MCLG), which is defined as “the level at which no known or anticipated adverse effects on the health of persons occur and which allows an adequate margin of safety.”10 This absolutist health-based goal is aspirational only, and it is frequently below the level at which the EPA establishes the legally binding “maximum contaminant level” (MCL). The next step is to specify an MCL as close to the MCLG as is “feasible.”11 Thus the MCL is a technology-based limitation on the health-based aspirational goal of the MCLG. In 1996, Congress added to the mix two balancing-based limitations by providing for two exceptions to the feasibility-limited, heath-based standard. First, the EPA Administrator may establish an MCL at a level other than the feasible level if the treatment needed to meet a feasible MCL would increase the risk from other contaminants or the technology would interfere with the treatment of other contaminants.12 This sensible limitation allows the Administrator to employ risk-risk balancing to ensure that the MCL does not cause more harm to public health than good. Second, if benefits at the feasible level would not justify the costs, the EPA may propose and promulgate an MCL “that maximizes health risk reduction benefits at a cost that is justified by the benefits.”13 This final option represents a cost-benefit balancing adjustment to the otherwise applicable feasibility-based standard. Thus, the extent to which cost-benefit considerations will override both health limitations and feasibility considerations for any given drinking water contaminant will depend upon the extent to which the costs of healthier but feasible alternatives outweigh the health benefits derived from those alternatives. As Professor Sunstein’s analysis suggests, this in turn, will depend almost entirely on the assumptions that the agency builds into the cost-assessment, risk-assessment, and monetization models.14 The arsenic rulemaking thus presents an opportunity to debate whether the move toward a greater role for cost-benefit-based decisionmaking in the SDWA is a good idea that Congress should emulate when it revisits other environmental statutes.

II. SCIENTIFIC UNCERTAINTIES AND WIDE BENEFITS RANGES

A cost-benefit analysis of the sort that Professor Sunstein advocates consists of an estimate of the costs that the standard will impose on both regulatees and society in general and a quantitative assessment of the value of the benefits to society of the protections that the standard will provide. A health benefits assessment, in turn, consists of two basic operations: (1) quantitative risk assessment and (2) monetization. A risk assessment takes into account the toxicity of the regulated substance and the extent of human exposure to the chemical through various means (for example, dietary, inhalation, or dermal). The toxicity assessment identifies the diseases that the substance is capable of inducing in human beings and attempts to calculate the dose-response relationship between exposure to the chemical and the onset of each of those diseases. The risk assessor then uses the predicted dose-response curves to calculate the incidence of diseases caused by exposure to the substance at the levels identified in the exposure assessment.15 As Professor Sunstein discovered, daunting scientific uncertainties greatly hinder virtually all of these exercises.

ments into benefits assessments by assigning dollar values to reduced risks to health and the environment.16 When addressed to credible risks to human health, however, this monetization function encounters not only large measurement uncertainties, but also serious theoretical impediments, including the daunting question of whether the proper measure is “willingness-to-pay” or “willingness-to-sell.”17

A. THE ARSENIC RISK ASSESSMENT

Different disease endpoints may have different dose-response relationships both in terms of the potency of the response and in terms of the shape of the dose-response curve. For arsenic, the EPA identified several disease endpoints, including skin cancer, various internal cancers (such as bladder, kidney, skin, and lung cancers), alterations in gastroenterological function, hardening of the arteries to the arms and legs, diabetes, and adverse effects on reproduction.18 Different disease endpoints can result from different routes of exposure, including the following: dietary exposures, such as drinking water; dermal exposures, such as contact during bathing; and inhalation exposures. If a single endpoint may be reached through different exposure routes, then the contribution of each exposure route to the overall increase in disease attributable to the chemical may be relevant to risk management decisions. For example, if arsenic can cause skin cancer through both dietary and dermal exposures and if dermal exposures are attributable largely to sources of arsenic in drinking water, then the contributions of both routes are relevant to an assessment of the health risks posed by arsenic in drinking water.

The EPA’s arsenic risk assessment also required an assessment of the extent to which individuals are exposed to arsenic through drinking water. This often extremely complex task was rendered somewhat simpler because the drinking water of regulatory interest came from a single source-public water treatment systems. Matters were slightly complicated, however, because some public treatment systems rely upon groundwater as a source of drinking water, some rely upon surface water, and some rely upon both. The risk reduction attributable to any particular standard will vary depending upon the amount of arsenic that is present at “background” levels in the source water. If background levels vary from region to region, then the risk reduction benefits of a uniform standard will vary even though the standard does not.

The EPA discovered that the background levels of arsenic in ground and surface water vary from place to place.19 Sometimes the presence of arsenic in source water is attributable to “natural” causes like high levels of arsenic in mineral deposits, and sometimes it is attributable to human activities like mining, wood treatment with pesticides, or hazardous waste disposal.20 On the basis of information collected from a twenty-five state database, the EPA estimated that 5.4%Io of the existing groundwater treatment systems and 0.7%70 of the surface water treatment systems treated source water containing average arsenic levels above the 10 parts per billion (ppb) level at which the agency proposed to set the MCL.21 The EPA further estimated that seven systems nationwide had arsenic levels above the existing standard of 50 ppb.22 This analysis matched other estimates by the U.S. Geological Survey and the National Arsenic Occurrence Survey reasonably well.23 The potential consequences of human exposure to arsenic can be explored through laboratory animal studies and epidemiological investigations.

When animal studies form the primary basis for risk assessment, the dose levels eliciting the greatest responses are typically much higher than the exposures that humans encounter in the real world. The risk assessment must, therefore, extrapolate from high exposures to low exposures. Similarly, when risk assessments are based upon epidemiological studies of highly exposed human beings, the risk assessor must extrapolate the predicted disease rates at low exposure levels from the disease rates at high exposure levels. When animal testing is employed, the chemical is usually tested at several dose levels, and the study results can provide risk assessors with some indication of the shape of the dose-response curve. Despite repeated attempts, however, arsenic has not caused cancer in laboratory animals.24 This has served as a great hindrance to the EPA’s risk assessment efforts because it has deprived the agency of a valuable source of information derived from controlled exposures.

factors complicate attempts to arrive at a theoretical dose-response curve. First, the actual exposure levels of the humans in the exposed cohorts under examination are not always known with much precision.26 Second, because scientists conducting epidemiological studies are not able to control all aspects of the activities and exposures of the humans under observation, unintended biases can undermine the conclusions of epidemiological studies.27 One especially prevalent form of bias in epidemiology is the “confounding factor,” which is a risk factor for the disease at issue that is associated with the exposure under study in the population of interest, but is not otherwise affected by the exposure or the disease.28 If confounding factors unrelated to the exposure of the substance at issue can account for the observed differences in disease rates, then it may be inappropriate to conclude that the exposure to the substance is responsible for those differences.29 Third, cost and feasibility limitations generally yield epidemiological studies that are not especially powerful. A negative epidemiological study, therefore, rarely warrants a strong conclusion that there is no causal relationship between disease and human exposure.30

noted, in particular, that the studies might not have accounted adequately for all “confounding factors” that potentially contributed to the increased incidence.33 In addition, the exposure of the individuals in the Taiwan study was uncertain because of difficulties in measuring the varying arsenic levels in village wells.34 The EPA, nevertheless, agreed with a special committee appointed by the National Research Council of the National Academy of Sciences (the “NAS committee”) that the Taiwan data provided an adequate basis for assessing potential risks from lower concentrations of arsenic in drinking water.35 More limited studies conducted in Chile and Argentina corroborated the Taiwan studies.36 An epidemiological study conducted in Utah found no association between arsenic in drinking water and bladder or lung cancer, but the EPA concluded that the Utah study was not powerful enough “to estimate excess risks” with sufficient precision or reliability.37 The bottom line was that the EPA faced huge uncertainties in defining a dose-response relationship for arsenic,38 and Professor Sunstein correctly cautions that the EPA’s estimates based upon these studies “should be taken with some grains of salt.”39

NAS committee and it frankly acknowledged that “[t]he use of a linear procedure to extrapolate from a higher, observed data range to a lower range beyond observation is a science policy approach that has been in use by Federal agencies for four decades.”43 The linear approach “is intended to identify a level of risk that is an upper limit on what the risk might be.”44 The agency noted that modes of action that did not involve direct reactions with DNA could also produce linear responses at low doses.45 The EPA, as a result, believed that the linear model had a sound basis in both science and policy.46

Professor Sunstein generally accepts the EPA’s assessment to this point. He concludes:

On the basis of what is known about carcinogens generally, the best scientific judgment seems to be that the dose-response curve for arsenic is sublinear. But this is a speculative judgment, not based on direct evidence. In addition, we certainly do not know how sublinear the dose-response curve is, if indeed it is sublinear.47

To provide a “prudent” estimate of the cancer risks attributable to arsenic in drinking water, the EPA applied the linear model to the epidemiological data from the Taiwan studies and concluded that setting the drinking water standard at 10 ppb would eliminate approximately twenty-eight lung and bladder cancers per year.50 The EPA first used the Department of Agriculture’s Continuing Survey of Food Intakes by Individuals (CSFII) to estimate that the daily mean per capita consumption of community tap water by individuals was 1.2 liters per person per day.51 The agency further estimated that people in the top ten percent of water consumption consume 2.3 liters per person per day.52 The agency then applied Monte Carlo statistical techniques53 to posit a distribution of consumption levels, and this, in turn, yielded lifetime “relative exposure factors” for each of the age categories used in the water consumption analysis.54 The EPA applied the relative exposure factors to the risk distribution derived from the Taiwanese data by the NAS committee for bladder cancer and by the more recent study by Professor Morales and colleagues, for bladder and lung cancer.55 The EPA did not, however, employ one of the Morales models because it yielded supralinear dose-response curves and the agency knew of no biologically plausible reason to expect that the dose-response curve was supralinear.56 The final result of these calculations was the table upon which Professor Sunstein relies in portraying the potential benefits of various MCLs.57

that choice on sound regulatory policy.60 Professor Sunstein acknowledges this well-known characteristic of decisionmaking under conditions of scientific uncertainty61 and he persuasively argues that it would be more revealing for the agency to characterize the risks posed by arsenic in drinking water with a range of plausible estimates.62 Professor Sunstein then attempts his own characterization of the range of risks using the EPA’s estimate (based upon a linear extrapolation) as the high extreme and an estimate prepared by Burnett and Hahn63 as the low extreme. This yields estimates of lung and bladder cancers avoided ranging from six to twenty-eight.64

The foregoing analysis is limited to lung and bladder cancers because those were the only cancer risks associated with arsenic that the EPA believed could be quantified.65 The EPA was confident that reducing the concentrations of arsenic in drinking water would reduce other kinds of cancer as well, but it declined to estimate the number of other cancers that would be prevented because that would involve an entirely speculative exercise. The statute required the EPA to consider both quantifiable and nonquantifiable benefits, and the EPA was willing to do so.66 Professor Sunstein, however, is not satisfied with this posture. Although the EPA was not “irresponsible” in declining to quantify these additional benefits, “it would also be responsible to attempt to specify an upper and lower bound.”67 Professor Sunstein notes that Burnett and Hahn estimated, again for no particular reason, that twice as many unquantifiable cancers would be avoided as quantifiable cancers. This seems especially speculative, in Professor Sunstein’s view, in light of the NAS committee’s estimate that the risk of all types of cancer could be eight times greater than the quantifiable bladder cancer risk.68 Professor Sunstein, therefore, posits a high end estimate of four times the EPA estimate for lung and bladder cancers. Adding these nonquantified cancers to the mix, Professor Sunstein’s calculations yield a range of 24 to 112 total cancers avoided. However, he later concludes that “the high estimate of 112 lives saved is unrealistically high and a bit of a scare tactic in light of the problems in the Taiwan data and the probability that the dose-response curve is sublinear.”69 Thus, nonquantifiable cancers fade in and out of Professor Sunstein’s analysis like an aircraft warning beacon.

In addition to carcinogenic effects, the EPA concluded that arsenic in drinking water was also associated with noncancer effects like hypertension and diabetes, as well as adverse reproductive effects.70 Both the EPA and the NAS committee also understood that sensitive populations like infants and pregnant females were likely to be much more susceptible to the carcinogenic effects of arsenic in drinking water, but there was insufficient scientific information to permit separate cancer risk estimates for these sensitive subpopulations.71 The EPA did not attempt to quantify the impact of its proposed drinking water standard on the incidence of those adverse effects, nor did it attempt to quantify nonhealth adverse effects of arsenic exposures.72 The agency simply noted that “if the Agency were able to quantify additional arsenic-related health effects and non-health effects, the quantified benefits estimates may be significantly higher than the estimates presented in this analysis.”73 Apparently, Professor Sunstein is willing to accept the EPA’s decision to leave these potentially quite extensive benefits unquantified, and he does not suggest a range of estimates for each of the noncancer effects associated with arsenic in drinking water. This omission from his quantitative analysis means that his low-end benefits estimates are too low, but his high-end estimates may only be slightly too high or perhaps not high enough.

B. MONETIZATION

The EPA concluded that “[a]rsenic ingestion has been linked to a multitude of health effects, both cancerous and non-cancerous.”74 Because different disease endpoints may have different monetary values (for example, the monetary benefit of reducing the incidence of one disease may be higher or lower than the monetary benefit of reducing the incidence of another disease), a thoroughgoing benefits assessment would consist of the sum of several risk assessment and monetization exercises.75 Just as he discovers a great deal of uncertainty in the quantitative risk assessments, Professor Sunstein finds huge uncertainties inherent in this monetization exercise.76

cal human lives and human illness. The EPA initially chose $6.1 million as the “value of statistical life” on the basis of a number of studies of the risk premiums that workers are allegedly able to extract from employers for undertaking hazardous work.78 As Professor Sunstein notes in Table 5, the EPA selected its number based on those studies from a range of $0.7 million to $16.3 million in 1997 dollars.79 The EPA adjusted its choice to $6.77 million to reflect growth in income and it applied that number to twenty-six percent of the bladder cancers avoided and eighty-eight percent of the lung cancers avoided to reflect the mortality rates of those two cancers.80

these adjustments, nor [were] there adequate empirical data to support definitive quantitative estimates for all potentially significant adjustment factors.”83 Consequently, the agency’s estimate for the value of a statistical life remained unadjusted for these admittedly important considerations.84

Professor Sunstein addresses many of these factors in more detail than the EPA does in its preambles and contractor-prepared background document. In most cases, Professor Sunstein concludes that fairly significant adjustments are probably warranted.85 For example, he cites a study by Professor Revesz suggesting that “the value of avoiding a death from an involuntary, carcinogenic risk should be estimated as four times as large as the value of avoiding an instantaneous workplace fatality.”86 With this in mind, Professor Sunstein multiplies the EPA’s estimate of the value of a statistical life by four to reach a value of $26.8 million. This exercise, however, takes into account at most only three of the six factors that the EPA identified as being beyond the wage-premium– based calculation. Noting that workers exposed to hazardous workplaces are still paid much less that most of the population even with the wage premiums, and that “the median salary of all wage earners is twenty-three percent higher than the median salary of most workers involved in the willingness-to-pay studies,”87 Professor Sunstein concludes that the value of statistical life must be least $33 million.gg This being said, Professor Sunstein also believes that an estimate of $1.1 million is within the realm of credibility.89

appropriate.92 This reduces the EPA’s $6.77 million figure to only $4.5 million.93

Professor Sunstein is certainly correct to emphasize that existing wage premium studies produce a very wide distribution of estimates and that they surely do not encompass every consideration that should go into monetizing the value of a statistical life. Whether those problems are cured by picking a number in the middle of the range of peer reviewed studies, multiplying that number by four because another law professor thought that was a sensible way to account for a few of the neglected considerations, and boosting that number by an additional twenty-three percent because rich people assign a higher monetary value to their lives than modest wage earners do is certainly an open question. Some experts may agree with Professor Sunstein’s gedanken exercise; many probably would disagree in at least one regard. I would venture that most sensible lay people would find the whole business more than a bit bizarre.

When it comes to discounting future benefits to present value, the argument among the experts immediately turns to the discount rate-perhaps because it really matters to them as a theoretical question or perhaps because they know that the selection of the discount rate determines any project’s fate.94 As Professor Heinzerling has very persuasively demonstrated, the EPA was by no means irrational in concluding that it is simply inappropriate to discount life-saving interventions with impacts in the future.95 Once again, sensible lay people would undoubtedly wonder at the care with which the risk assessors and monetizers carry the grossest of quantitative estimates out to six figures and then discount future benefits back to present value to facilitate comparisons with discounted costs. They may even wonder whether the experts carried out the exercise with such apparent precision to draw attention away from the nakedness of the assumptions that fueled the models that begat the numbers in the first place.

somewhat arbitrary.”96 He goes on to criticize the EPA’s $6.1 million figure as “too high,” even though he argues that other considerations would justify multiplying the number derived from wage premium studies by a factor of four.97 Thus, although Professor Sunstein is more willing than most risk monetizers to pay lip service to considerations like the difference between voluntary and involuntary risks, when push comes to shove, he is prepared to abandon such “soft” considerations and accept a very low “hard” estimate of $4.5 million per statistical life saved.

Because the suffering of persons who survive cancer was not incorporated into the value of statistical lives calculated from wage premiums, the EPA decided to base the value of reducing survivable cancer on the willingness of individuals to pay to avoid nonfatal cancer.98 Sadly, the agency lacked any economic studies of the willingness of people to pay to avoid cancer. It instead used a number that some economists had derived for the willingness to pay to avoid chronic bronchitis, which is apparently close enough to nonfatal cancer for government work.99 That number turned out to be $607,162.(100) Professor Sunstein notes that the EPA calculated that a 10 ppb MCL would prevent sixteen to twenty-six cases of curable cancer, but he otherwise ignores both of these calculations (which would yield another $15 million in benefits) in his assessment of the overall monetary benefits of the standard. Professor Sunstein also declines to incorporate into his quantitative analysis other nonquantifiable benefits that the EPA considered, such as the noncancer health effects that probably fit a threshold dose-response curve at low levels, but may not yield a zero response at the current standard of 50 ppb.101

III. THE BURNETT & HAHN “PEER REVIEW”

he credulously takes its conclusions more or less at face value as representing a legitimate expert assessment of the scientific and economic bases for the EPA rule.

The Burnett and Hahn pamphlet can hardly be characterized as a peer review of the EPA’s multidisciplinary work. Mr. Burnett is identified as a “researcher” at the Joint Center, and Mr. Hahn is an economist well known for his strong anti-interventionist views with respect to the health, safety, and environmental regulation of business activity. Though both may claim expertise in monetization-an important step in the EPA’s analysis-neither of the authors has expertise in toxicology or dose-response modeling. Although Professor Sunstein is too polite to say so, the Burnett and Hahn tract is actually a hatchet job and, for the very reasons that he perceptively articulates, is unworthy of the attention that he devotes to it.103 It is, however, a fine example of why ordinary people should be skeptical of Professor Sunstein’s conviction that self-proclaimed “risk experts” like Burnett and Hahn ought to play a greater role in health, safety, and environmental regulatory decisionmaking.

Burnett and Hahn predict that the costs of the EPA’s 10 ppb MCL will exceed the benefits by about $190 million each year and, for that reason, will, in John Graham’s memorable phrase, “statistical[ly] murder” ten people per year.104 The Burnett and Hahn analysis consists of several steps, all of which are highly questionable and some of which only could have been motivated by a strong desire to achieve a preordained result.

the difference between one and four. Their arbitrary choice of a factor of two cannot be explained as an exercise of scientific judgment because the two economists had no scientific judgment to exercise.

Second, Burnett and Hahn reject the EPA’s conservative assumption that the dose-response curve for arsenic carcinogenicity is linear at low doses as “not realistic because the human body can metabolize arsenic at low levels, rendering it nontoxic.”107 The authors do not elaborate upon their qualifications to make this toxicological assessment. Although they cite the NAS committee’s report for the proposition that the body can metabolize arsenic,108 they do not explain why or how this would affect the dose-response curve. The NAS committee concluded that it was possible that the dose-response curve for arsenic was sublinear at low doses, but it was also possible that the doseresponse curve was linear.109 Indeed, the 2001 update to the NAS report noted that “the fact remains that there is some empirical evidence suggesting that a supralinear model might indeed hold,” and it posited that a supralinear model was “plausible” because “it could result from a subpopulation that is susceptible to low doses of arsenic.”110 That update reaffirmed the committee’s conclusion that “[t]he data on the mode of action of arsenic do not indicate what form of extrapolation should be used below the exposure range of human data.”111 Burnett and Hahn apparently conclude that the data on mode of action indicate that dose-response curve is sublinear at low doses.

From this conclusion-which is not actually a finding of the NAS committee– Burnett and Hahn speculate that “[a] possible sublinear model is to assume that only 20 percent of the ingested arsenic is toxic at low levels while almost all of it is toxic at much higher levels.”114 Burnett and Hahn’s model thus yields a risk that “is only about a fifth of that suggested by a linear response model.”115 Multiplying the fifty-six fatal cancers avoided (from the earlier calculation) by twenty percent yields a total of eleven cancers prevented by the standard116-a number that, given the large population of exposed people, “is so small as not to be worth addressing, given the uncertainties in the data and the EPA’s limited resources to develop regulations.”117

In deriving a highly simplistic sublinear model for the arsenic dose-response curve from statements about arsenic’s mode of action in the NAS committee’s report, Burnett and Hahn ignore the committee’s caution at the outset that “the mechanisms or modes of action by which inorganic arsenic causes toxicity, including cancer, is [sic] not well established.””18 In fact, the committee later addressed this question specifically and concluded that “[t]he data on the mode of action of arsenic do not indicate what form of extrapolation should be used below the exposure range of human data.”119

entirely possible that the metabolic byproducts of the liver’s attempt to render arsenic less acutely toxic were themselves implicated in the carcinogenic response.

The NAS committee’s report does not support Burnett and Hahn’s suggestion that the metabolic process responsible for converting inorganic arsenic in drinking water into organic metabolites does not function (or, in Burnett and Hahn’s terms, becomes “saturated”124) at the 100-500 ppb exposure levels present in some of the “positive” epidemiological studies.125 The exposed cohorts in the epidemiological studies were not being forced to consume concentrations of arsenic hundreds or even thousands of times greater than the concentrations that humans ordinarily consume, as is often the case in animal studies in which primary metabolic pathways become saturated. The epidemiological studies demonstrate an association between arsenic and many forms of cancer in humans at exposure levels in the range of 100 ppb.126 In the committee’s view, this suggested that “the distance of extrapolation is very small-less than 1 order of magnitude.”127

break point somewhere between 50 ppb and 100 ppb. Nowhere in either of the two NAS reports is there even the slightest suggestion that the epidemiological data fit this form of carcinogenesis dose-response curve.

The reason that the committee believed that the “most plausible” shape of the dose-response curve was sublinear at low doses had nothing to do with the percentage of ingested arsenic that the body is able to metabolize. Rather, it had to do with the fact that arsenic and its metabolites do not appear to interact directly with DNA to mutate genes.128 Although it is unclear exactly how arsenic or its metabolites induce a carcinogenic response, the committee believed that the mechanism is probably indirect, and “[a]n indirect mechanism of mutagenicity suggests that the most plausible shape of the carcinogenic doseresponse curve is sublinear `at some point below the level at which a significant increase in tumors is observed.'”129 However, the committee reasoned that it was “prudent not to rule out the possibility of a linear response because there was “insufficient scientific evidence to identify the dose at which sublinearity might occur.”130 The committee, in its 2001 update, ultimately concluded that although “[t]he results of the mode-of-action studies do not provide a clear picture of the shape of the dose-response curve at low doses,” the evidence “is accumulating that relatively low concentrations of arsenic, potentially achievable through consumption of drinking water containing 10-50 ppb of arsenic, can alter biochemical pathways relevant to carcinogenesis.”131

Nowhere in the subcommittee’s original report or in its 2001 update is there even the remotest suggestion that the proper way to estimate the carcinogenic response to arsenic exposure at low levels is to extrapolate linearly from the response at higher levels and divide by five. The Burnett and Hahn exercise is the sort of back-of-the-envelope calculation that, even if appropriate for some kinds of economic analysis, has no place in scientific decisionmaking. To the extent that science is relevant to science policy decisions about where to establish an MCL, the NAS committee’s 2001 update has the following bottom line advice:

No credible scientist with expertise in chemical carcinogenesis has suggested a model as simplistic as that advocated by Burnett and Hahn.

The fourth step in the Burnett and Hahn analysis is to discount the benefits of any cancers prevented by a lower MCL on the ground that typically there is a long latency period between the exposure to a carcinogenic substance and the onset of bladder cancer.133 Whatever amount a person would be willing to pay to prevent an immediate death is, in their view, more than that person would be willing to pay to eliminate the certainty of death at some point in the remote future. Burnett and Hahn, therefore, assume a latency period for carcinogenesis of thirty years and employ a seven percent discount rate to reduce the value of the statistical lives saved by the EPA’s arsenic standard from the $6.1 million figure that the EPA used to $1.1 million.134 Professor Sunstein does not disagree with the proposition that latency periods should be factored into the monetization analysis via some discount rate, but he finds “some evidence” to support a discount rate of two to three percent and therefore discounts the EPA’s $6.1 million to $4.5 million.135

Professor Sunstein quite properly criticizes Burnett and Hahn for assuming “that future health benefits should be discounted at the same rate as future monetary costs.”136 Others have questioned whether future health benefits should be discounted at all.137 I will leave that debate to others and note only that to the extent that discounting for latency periods is a proper exercise in an economic analysis of health protections, Burnett and Hahn have erred on the high side. This exercise results in an estimate of monetary benefits that errs on the low side.

benefits. According to Professor Sunstein’s “very high” estimate, which, in my mind, is at least as plausible as Burnett and Hahn’s estimate, the net benefits are $3.584 billion.141

Burnett and Hahn, however, do not stop here. They postulate that a $190 million net economic loss142 will indirectly result in the loss of around ten statistical lives per year.143 This remarkable conclusion stems from the assertion that increased regulatory costs result in increased unemployment and unemployment is associated with higher mortality rates. The notion that the costs imposed by life-saving regulation kill people has been bandied about in economic literature for many years, but has not been taken very seriously by the government or many others.144 Professor Sunstein, however, does take this idea seriously,145 perhaps because it solves a serious incommensurability problem for a person of his civic republican sensibilities. If dollars spent on regulation are easily convertible to statistical deaths, then cost-benefit analysis converts just as easily to risk-risk analysis with lives on both sides of the ledger. I have elsewhere critiqued the “regulation kills” argument and do not propose to repeat that critique here.146

spell the difference between life and death.148 In short, the economists who claim to find a correlation between health and wealth have not-like the public health scientists who attempt to interpret similar epidemiological data– suggested any mode of action that would tend to explain the correlation. Finally, at the very least, the argument seems aimed at the wrong agencies. If there is any truth at all to the “regulation kills” thesis, then its advocates should take aim at the Federal Reserve Board, whose conscious efforts to maintain high unemployment rates have been far more effective in keeping people out of work than all the regulations of the EPA and OSHA combined.

Professor Sunstein concludes that the Burnett-Hahn study “is worth close attention” because “the AEI-Brookings Joint Center is highly respected for its careful work on CBA, and Hahn is an especially able and influential observer of the regulatory process.”149 Although Mr. Hahn may be especially influential, his work is notorious for hostility to federal regulatory programs and solicitude for the position of regulated entities. I have elsewhere criticized his highly questionable effort to belittle the value of federal health and safety regulation on the basis of what he misleadingly portrays as the government’s own cost and benefit estimates.150 Given the terrorist acts of September 11, 2001, perhaps the most compelling reason for taking what Mr. Hahn has to say about federal health and safety regulation with a large grain of salt is his 1997 critique of the recommendations of a blue ribbon commission to enhance security at airports, in which he argued against adopting the Israeli model of airport security based on his conclusion that “[iln the case of preventing airline terrorism, a very sophisticated approach may not be desirable given its high cost.”151

attack agency decisions, and Professor Sunstein provides a blueprint for the way in which lawyers for the mining industry should go about undermining the arsenic standard that relies to some extent upon the Burnett and Hahn critique.153

IV. PROFESSOR SUNSTEIN’S CONTINUED COMMITMENT TO THE SYNOPTIC PARADIGM

Professor Sunstein’s deep immersion in the arsenic documents leads him to the conclusion that a very wide range of potential benefits can be derived credibly from the existing data. As a result, “[s]ometimes the best that can be done is to specify an exceedingly wide `benefits range,’ one that does not do a great deal to discipline judgment.”154 Professor Sunstein’s obvious enthusiasm for this “insufficiently appreciated” discovery should prove gratifying to aging observers of the regulatory process who have appreciated this point for decades. The reality of huge uncertainty in quantitative risk assessment existed before Professor Sunstein discovered it, but his excitement about that discovery is contagious.

threshold for arsenic’s carcinogenic response above 50 ppb, to $3.794 billion, for those who are persuaded by Professor Sunstein’s efforts to quantify the benefits of a 10 ppb MCL and accept a $33 million value for a statistical life.157

The speculative nature of the enterprise does not prevent Professor Sunstein from reaching the bottom-line conclusion that “the quantified benefits are most unlikely to be enormously higher than the $210 million price tag.”158 On the other hand, the quantified benefits “might well be higher whether or not they are much higher.”159 One gets the impression that Professor Sunstein has concluded that the 10 ppb MCL is probably not justified on cost-benefit grounds, but it is also apparent that quantitative techniques did not pave the way to that conclusion.

Professor Sunstein’s excursion into the arsenic rulemaking record leads him to suggest seven general lessons about cost-benefit analysis, all of which are fully consistent with what he has been saying about cost-benefit analysis for the last fifteen years. This journey also leads him to suggest that cost-benefit analysis could facilitate more efficient regulatory approaches to removing arsenic from drinking water. For example, the EPA could “[i]mpose a targeted rule, with a sliding scale of regulations, ensuring that the cost-benefit ratio supports the outcome in each area.”160 One marvels at the speed with which an analytical technique that produces a range of overall benefits extending from $0 to $3.8 billion is transformed into a powerful analytical engine capable of justifying nonuniform standards that provide more protection to some citizens than others. At the same time, Professor Sunstein recognizes that one alternative to his suggestion for nonuniform standards is a uniform standard with government subsidies for systems with especially high removal costs. This solution will not work, however, because the numbers cannot tell us “whether this is the best way to spend limited taxpayer dollars.”161 Somehow these mysterious numbers can tell government who or who not to protect, but they cannot tell it who or who not to subsidize.

the way that ordinary people-unaided by experts wielding quantitative tools– make decisions about risk.

V. COST-BENEFIT ANALYSIS AS A CURE FOR “INTUITIVE TOXICOLOGY”

In his brief history of the arsenic rule, Professor Sunstein makes it clear that he believes that public reaction to President Bush’s decision to suspend and reexamine the Clinton Administration’s arsenic rule was both uninformed and simple-minded. Because the chemical was arsenic and because the debate was about something as familiar as drinking water, Professor Sunstein concludes that the controversy appeared “highly accessible” to the public162 and it was therefore “easy to be outraged.”163 Rather than trust the new Administration’s assessment of the costs and benefits of the former Administration’s standard, the public relied upon “intuitive toxicology” to conclude that the new Administration was doing to little to protect it from the risks posed by arsenic in its drinking water.164

Professor Sunstein believes that ordinary people employ a set of “simple rules” that do “not easily make room for issues of degree.”165 Lay risk assessors rely upon the “affect heuristic,” through which judgments about risks are “greatly affected” by apparently irrelevant factors like whether or not the sun is shining.166 A heuristic “operates as a kind of mental shortcut, substituting itself for a more careful inquiry into consequences.”167 Such simple-minded rules of thumb can lead the public to make “systematic errors.”168 Among the simple rules employed by these simple people is the absolutist principle that all carcinogens should be banned. The sadly uninformed public “does not show an understanding of the different imaginable dose-response curves and the possibility of safe thresholds or even benefits from low exposure levels.”169 Consequently, common folk “often make mistakes in thinking about the seriousness of certain risks,”170 and an aroused public then demands action from their government. This, in turn, results in “inadequate or even counterproductive” regulatory decisions.171

scientific connection between arsenic contamination in drinking water and public outrage was tenuous, there were many very good reasons for the public to distrust the Bush administration’s decision beyond the unthinking application of a few simple heuristics.

Professor Sunstein concludes that cost-benefit analysis is the cure for the ill effects of the irrationality that infuses the “intuitive toxicology” through which the American public addresses health risks. Unlike the uninformed and simpleminded public, the practitioners of cost-benefit analysis can bring to the table “a form of toxicology that is actually supported by data.”177 This, in turn, “should help government resist demands for regulation that are rooted in misperceptions of facts.”178 The solid scientific basis underlying cost-benefit analysis should prevent interest groups from strategically using the public’s cognitive problems to “[fend] off regulation that is desirable, or [press] for regulation when the argument on its behalf is fragile.”179 Professor Sunstein welcomes the emergence of what he calls “the cost-benefit state” because it “can lead to more stringent regulation of serious problems, less costly ways of achieving regulatory goals, and a reduction in expenditures for problems that are, by any account, relatively minor.”180 In his great enthusiasm for cost-benefit analysis, Professor Sunstein misses the possibility that the synoptic paradigm is itself a heuristic, a simplifying framing device “substituting itself for a more careful inquiry into consequences.”181

Consider, for example, the bias inherent in the application of the expertise of some economists to the monetization exercise. The experts always employ “willingness-to-pay” for risk reduction as the measure of the value of credible risks to health and safety, even though most of them surely know (because of the efforts of observers like Professor Sunstein) that an equally valid measure is “willingness-to-accept” the risks imposed by others. This goes beyond the minor disputes over whether the reduction of involuntary risks is more valuable than the reduction of voluntary risks. It goes to the heart of the enterprise, and the adoption of the willingness-to-pay test can (and usually does) bias the analysis against regulatory intervention. When converted to dollars, the willingness-to-accept measure invariably yields much larger values that the willingnessto-pay measure. People have limited resources to draw upon in deciding how much they can pay to reduce risks, but there is no limit to the amount that they can demand to accept those risks.182

At first glance, drinking water standards appear to be one of the few subjects of environmental regulation in which willingness-to-pay is the most appropriate measure. In areas of the country in which arsenic is a “natural” constituent of the source of drinking water in such high amounts that a 10 ppb MCL would require its removal, the primary beneficiaries of any drinking water standard will be the community-water-provider customers who will pay for the cleaner water.183 Nature put the arsenic in the water and if a community wants to get the arsenic out, it will have to pay for the technologies that are capable of making the world safer than nature designed it to be.

In some (perhaps many) areas of the country, however, the arsenic is in the source water because human activities caused the arsenic to be there. To some extent, arsenic can be found in sources for drinking water because of human carelessness in conducting hard-rock mining, applying pesticides, and disposing of hazardous wastes.184 In these situations, it is much less clear why some people-the residents in communities with arsenic contaminated source water– should have to pay other people-the owners of the activities that caused the arsenic to contaminate the source water-to make the water safe. The question quickly devolves to an essentially legal one regarding the initial allocation of “rights”: Does the mining company have a “right” to pollute the water up to the point at which the public is willing to pay to induce it to stop? Or do the members of the affected community have a right to source water at least as clean as the water that nature provided up to the point at which the mining companies can persuade them to accept the additional cancer risks?

When a regulatory program addresses credible morality risks for which there are no markets and for which the difference between the willingness of the affected people to pay to reduce such risks and their willingness to accept the imposition of such risks is quite large, then the initial allocation of entitlements is critical to the value that the agency assigns to those risks in a cost-benefit analysis. The value of risk-reducing governmental interventions will be much greater in a world in which people can demand to be paid to accept risks than in a world in which they must pay to have them reduced. By relying exclusively upon willingness-to-pay as the measure of the value of a statistical life, both the EPA and Burnett and Hahn biased their analyses against regulation in those parts of the country in which human beings played a role in contaminating source water with arsenic.

Professor Sunstein leaves unanswered the question of why the public should trust experts not to bias their analyses on questions like the shape of the dose-response curve when those same experts are inclined to bias their analyses against governmental protections on the monetization question. As Professor Sunstein observes, Burnett and Hahn erred on the side of underestimating the cancer risks posed by arsenic in drinking water when they assumed that the dose-response curve was sublinear at low doses. I have demonstrated elsewhere that Mr. Hahn has biased his analyses to de-emphasize the benefits of environmental regulation for specious reasons, like the baldly asserted high probability that scientists will develop a cure for cancer in the next fifty years.185 Why should ordinary people be inclined to place their lives in the hands of people who clearly favor industrial activity over protective governmental intervention just because they claim to be “experts” in risk assessment and monetization?

VI. SUGGESTIONS FOR IMPROVING HEALTH, SAFETY, AND ENVIRONMENTAL REGULATION

Professor Sunstein’s in-depth inquiry into arsenic rulemaking has inspired him to probe in a brief and tentative way several suggestions for improvement. I will address two of those suggestions in light of the foregoing, rather dismal assessment of the potential for cost-benefit analysis and make two suggestions of my own for Professor Sunstein’s consideration.

A. “SOFT GLANCE” JUDICIAL REVIEW

“courts should play an exceedingly deferential role in overseeing CBA at the agency level.”186 In evaluating agency cost-benefit balancing, the courts “should give agencies the benefit of every reasonable doubt.”187 Professor Sunstein notes that courts are not specialists in toxicology or economics and wisely cautions that intensive judicial review can only further impair an already 11 ossified” rulemaking process. Because “[a]ny choice has a large policymaking dimension…. courts should be reluctant to displace the judgments of administrators, who have advantages both as technocrats and public representatives.”188

Although Professor Sunstein’s observations on judicial review are right on the mark, I am not confident that judges will take that message away from his article. Throughout the article, he criticizes the EPA for failing to quantify more aspects of the decisionmaking process, but carefully massages the assumptions underlying the EPA’s models to create slightly different models that produce somewhat smaller numbers than the EPA and somewhat higher numbers than Burnett and Hahn. One gets the strong sense that Professor Sunstein rather enjoys this intellectual repartee. And why not? Arguing over such hard-headed topics as how best to derive the value that workers assign to their lives and the extent to which children are more valuable to society than old folks can enhance self-esteem in a morbid sort of way. Participants in such arguments can demonstrate (at least to themselves) that they are capable of rising above the emotionalism of the masses to make the cold calculations necessary for rational decisionmaking.

I suspect that more than a few current members of the federal bench also take some pride in their ability to rise above the politics and emotions of the moment to base judicial decisions on hard analysis. Professor Sunstein strongly warns judges to resist this tendency when reviewing agency decisionmaking in the context of health, safety, and environmental regulation. This is an area, he warns, in which huge scientific uncertainties preclude hard analysis and in which policy must necessarily play a very large role. At the same time, however, Professor Sunstein cannot resist the temptation to correct what he perceives to be errors in the analysis. With that in mind, why should judges with equally powerful intellects (and even stronger egos) find it any easier to resist the temptation to wade in and set the errant agency straight? In any event, Professor Sunstein should finish the thought that he quite properly begins: When judges play an active role in this policy laden area, they are properly subject to criticism for advancing their own hidden policy agendas.

B. INFORMATIONAL SOLUTIONS

tion.”189 Informed consumers of public drinking water may demand sensible action upon learning that their water exceeds the federal standards for arsenic, especially given the clear knowledge that they will be footing the bill for any improvements.190 Professor Sunstein suggests the possibility of a sliding-scale approach under which public water suppliers would have to ensure that arsenic levels never exceeded a mandatory cap (he suggests 30 ppb) and would have to disclose the presence of arsenic in drinking water to their customers at levels lower than the cap but higher than some agency-determined de minimis level.191 Public pressure would then force water systems to remove more arsenic if the customers felt strongly enough about those risks. This proposal would have the advantages of “ensuring that people will learn that some carcinogenic substances are not especially dangerous at low levels and also alerting people to the need for tradeoffs (in the form of a higher water bill).”192

C. TECHNOLOGICAL FEASIBILITY AS A STARTING POINT

Professor Sunstein continues to be skeptical about the technology-based approach to health, safety, and environmental regulation even though his recent education on the limitations of cost-benefit balancing as a decisionmaking criterion would seem to lead him to a technology-based solution. Under that approach, the regulator carefully segments and studies the relevant industry and requires regulatees to meet standards based upon an administrative determination of the degree to which the relevant pollutants can be removed through the application of the best available (or some intermediate level of) pollution reduction technology.195 As noted above, the second step in promulgating an MCL is to determine an MCL that is as close to the MCLG as is “feasible.”196 In 1996, however, Congress added a cost-benefit balancing adjustment to this otherwise applicable technology-based standard.197 Professor Sunstein’s analysis has aptly demonstrated that this adjustment will not be easy to apply in a rational and consistent fashion.

In the case of arsenic, feasible technologies exist to reduce arsenic concentrations far below the 10 ppb concentration at which the EPA set the MCL.198 As Professor Sunstein notes, many countries insist that their water treatment systems provide water with lower levels than the 50 ppb standard that has been in place in the United States for twenty-five years.199 That other countries employ more stringent standards is not, as Professor Sunstein suggests, a convenient “mental shortcut, showing what it is right to do-notwithstanding the reasonable questions that might be asked about the scientific bases for those practices.”200 It is a clear demonstration of the technological feasibility of more stringent standards. Nevertheless, Professor Sunstein is still not satisfied that such technologies will not cost too much.

dards are over the toxicity of the contaminant, the shape of the dose-response curve at low levels, and the value of human life. A decisionmaker can be more confident at the end of the day that he or she has made a rational choice when the choice is about what technology works in the real world rather than when the choice is about matters that are clouded by huge economic uncertainties and moral ambiguity.

D. MARGIN-OF-SAFETY ASA BACKUP

The NAS committee in 2001 concluded that evidence “is accumulating that relatively low concentrations of arsenic, potentially achievable through consumption of drinking water containing 10-50 ppb of arsenic, can alter biochemical pathways relevant to carcinogenesis.”202 If this statement is accurate, then an MCL of 10 ppb does not leave a very large margin of safety, and it should not, in the view of adherents to the cost-benefit paradigm. Margins of safety throw the cost-benefit equation out of balance and force companies to spend too much for pollution controls. For cost-benefit proponents, the standard should be set as close as possible to the point at which marginal costs equal marginal benefits.

The margin-of-safety concept reflects a cautionary approach of erring on the side of safety when large uncertainties render precise risk projections impossible. The D.C. Circuit long ago observed that:

If administrative responsibility to protect against unknown dangers presents a difficult task, indeed, a veritable paradox calling as it does for knowledge of that which is unknown then, the term `margin of safety’ is Congress’s directive that means be found to carry out the task and to reconcile the paradox.203

How many of us want to drive over a bridge or ride in an airplane for which the last dollar spent on safety just equaled the projected monetized lives saved discounted to present value? A margin of safety provides a backup level of safety as a hedge against catastrophe when the experts turn out to be wrong.

EPA is at liberty to adopt a linear nonthreshold model for carcinogenesis.204 Erring on the side of safety even to that very limited extent may be inconsistent with the cost-benefit constraint that now binds the EPA.

If Professor Sunstein is correct, then a 10 ppb arsenic standard may be unjustified under the statute even though the science does not clearly support a deviation from the linear model. Even a court that is willing to heed Professor Sunstein’s injunction against overly intrusive judicial review may find it hard to resist his criticism of the EPA’s interpretation of its newly amended statute. If, as Professor Sunstein suggests, the cost-benefit limit that Congress has placed on standard setting under the SDWA prevents the agency from erring on the side of safety in the assumptions that it adopts, then it will join section 6 of TSCA and section 6 of the pre-1996 FIFRA in the garbage heap of environmental statutes that are so difficult to implement and accomplish so little when implemented that they have been effectively discarded. I am persuaded that the 1996 amendments do not go that far, but if they do, then the statute may again require amendment in the near future.

AN INVITATION AND A CONCLUSION

In his arsenic piece, Professor Sunstein has made a commendable effort to wade into the muck and mire of health and safety regulation. He has discovered that regulatory decisionmaking is a very messy business with few clear “scientific facts” to guide decisionmakers and many legitimate disputes about the nature of the default assumptions and inference guidelines that must necessarily serve as substitutes for facts if decisions are to be made in a timely fashion. Professor Sunstein has undertaken a detailed examination of cost-benefit analysis in the context of the arsenic rulemaking, and he has discovered that the synoptic paradigm leaves much to be desired when it encounters the real world.

A close encounter with the real world of the sort that Professor Sunstein has now experienced is an altogether useful exercise for those who purport to be scholars of administrative law. I would invite other scholars-Professors Revesz, Adler, and the younger Posner (the elder is probably lost beyond redemption) come to mind-to join Professor Sunstein in this useful enterprise. These scholars have demonstrated a willingness to work very hard and they bring a keen intelligence to the task.205 They should not confine themselves to the academic literature and the tomes of the think tanks. Is it too much to ask that a scholar of administrative law, at least once in her career, dive into the swamp and try to make sense out of real-world administrative decisionmaking? Professor Sunstein certainly has done so and now others should join him.

Having taken this commendable step, Professor Sunstein emerges with his pre-existing views largely intact. He still has very little respect for the way that ordinary people go about making decisions regarding risk, and he retains a great deal of confidence in the ability of risk assessment and monetization experts to guide public decisionmakers to the “best” results. He offers no guidance whatsoever, though, on what counts as expertise. Presumably, a degree in law or economics is not sufficient to convey the status of expert in health and environmental risk assessment. Yet, one gets the impression that, having gone to the considerable effort of studying the relevant documents in detail, Professor Sunstein may feel qualified to sit at the table at which the experts gather. Finally, convinced that a “soft” form of cost-benefit analysis is the most rational approach to health, safety, and environmental standard-setting, he concludes that Congress was wise in 1996 to place a cost-benefit limit on the technology– based approach that had previously dominated EPA rulemaking under the SDWA.

THOMAS O. MCGARITY*

* W. James Kronzer Chair in Trial and Appellate Advocacy, University of Texas School of Law.

Copyright Georgetown University Law Center Jul 2002

Provided by ProQuest Information and Learning Company. All rights Reserved