INTRODUCTION
I. RISK ASSESSMENT: A PRIMER
  A. History and Structure
  B. What Is Risk? The Frequentist Answer
II. RISK REGULATION AND "INDIVIDUAL RISK": A SURVEY OF
      GOVERNMENTAL PRACTICE
  A. "Individual Risk" and Agency Practice: The Environmental
      Protection Agency
    1. Cancer Risk




				

				
				 
				


 Assessment and "Individual Risk"
      a. Air pollution (Clean Air Act section 112)
      b. Water pollution (Clean Water Act and Safe Drinking
         Water Act)
      c. Solid waste (RCRA and CERCLA)
      d. Pesticides (FIFRA and FQPA)
      e. Toxic Substances Control Act
      f. Title VI
    2. Risk Assessment of Noncarcinogens
  B. "Individual Risk" and Agency Practice: Other Agencies
    1. The Food and Drug Administration
    2. The Occupational Safety and Health Administration
    3. The Nuclear Regulatory Commission
    4. The Consumer Product Safety Commission
  C. Beyond "Individual Risk": "Population Risk" in Agency
     Practice
III. FREQUENTIST RISK AND WELFARIST CONSEQUENTIALISM
  A. Welfarist Consequentialism: Some Clarifications
  B. The Ex Post Question: Does "Individual Risk" Degrade
     Outcomes?
  C. The Ex Ante Question: Should Frequentist Risk Play a Role in
     the Choices of the Welfare-Consequentialist Regulator?
IV. FREQUENTIST RISK AND BAYESIAN RISK: ARE THEY REALLY
    DIFFERENT?
V. BEYOND WELFARISM: FREQUENTIST RISK AND NONWELFARIST
   VIEWS
  A. Safety-Focused Consequentialism
  B. Deontological Views
  C. Contractualist Views
  D. Democratic Views
VI. RISK ASSESSMENT AND POPULATION SIZE
CONCLUSION

INTRODUCTION

EPA's decision to list a carcinogenic substance as a "hazardous waste," subject to stringent regulation under the Resource Conservation and Recovery Act, depends on the fatality risk that the substance would impose upon highly exposed individuals if discarded in unregulated landfills. If this risk exceeds 1 in 10,000, the substance is listed. (1) EPA's rule for cleanups under the Superfund statute is similarly risk-based: toxic waste dumps are to be remedied so that the lifetime fatality risk to the "maximally exposed individual" from carcinogens in the dump is within the range of 1 in 10,000 to 1 in 1 million. (2) FDA has traditionally used a 1 in 1 million threshold in determining whether carcinogenic food constituents exempt from the Delaney Clause pose a de minimis safety threat to consumers and thus should be permitted to enter or remain in the food supply. (3) OSHA, which is statutorily authorized to regulate workplace toxins that pose "significant" threats to safety, is more permissive than FDA and EPA but also focuses, in part, on individual fatality risks: the agency has generally followed the rule that carcinogens creating more than a 1 in 1000 risk for any worker are "significant," for statutory purposes, and that toxins creating a substantially smaller risk are not. (4)

In all these cases, health and safety agencies have decided to key regulatory choices to the level of "individual risk" (specifically, the "individual risk" to the maximally exposed individual or some similar construct) without any explicit statutory mandate to do so. But such mandates do exist. A salient one: when Congress in 1990 overhauled section 112 of the Clean Air Act, the section covering carcinogens and other "hazardous air pollutants," it put in place a hybrid regulatory regime that first requires polluters to use the best currently available technology for reducing emissions, and then requires EPA to consider promulgating yet more stringent emissions standards if "excess cancer risks to the individual most exposed to emissions ... [exceed] one in one million." (5) The 1 in 1 million risk level is also invoked in another provision of the amended Clean Air Act. (6) And when the legislative regime for pesticide licensing was reworked in 1996 (7)--the ban on certain carcinogenic pesticides was replaced with a "reasonable certainty [of] no harm" (8) standard both for possible carcinogens and for pesticides that might cause other toxic effects--the official House Committee report explained that this new statutory standard ought to be construed as an "individual risk" test:

   [T]he Committee expects ... that a [pesticide] tolerance will be
   considered to provide a 'reasonable certainty of no harm' if any
   increase in lifetime risk, based on quantitative risk assessment
   using conservative assumptions, will be no greater than
   'negligible.' ... [A] negligible risk [is] a one-in-a-million
   lifetime risk. (9)

In short: individual fatality risk plays a major role in our current system of health and safety regulation. Some examples have just been provided. Many more will be furnished below. In particular, "individual risk" is absolutely central to federal regulation of toxic chemicals. EPA employs an "individual risk"-based approach in administering all of its major statutes: the Clean Air Act (which addresses toxins present in air), the Clean Water Act and Safe Drinking Water Act (toxins in water), the Resource Conservation and Recovery Act and the Comprehensive Environmental Response, Compensation, and Liability Act (toxins that leach into the ground from waste sites), the Federal Insecticide, Fungicide, and Rodenticide Act (toxic pesticides), and the Toxic Substances Control Act (a backup statute authorizing EPA to take measures not authorized by the media-specific statutes). (10) FDA and OSHA follow a similar approach, as we have seen. But the focus on "individual risk" is not limited to toxins, or to federal agencies. For example, the Nuclear Regulatory Commission (NRC) has long taken the position that the ultimate safety goals governing its licensing and regulation of nuclear power plants partly concern the "individual risk" of immediate death, resulting from an accidental release of radiation, incurred by the average person living near a plant. (11) FDA sets acceptable levels of microbial contaminants in foods with reference to the "individual risk" of illness of a high-end consumer. (12) Although OSHA traditionally focuses on aggregate fatalities or lost days of work in regulating workplace conditions that cause injury (as opposed to illness), it has recently begun to consider the "individual risk" of injury--the rate at which workers in particular industries are injured by electric shock, falls, explosions, fires, and other such industrial accidents. (13) And environmental agencies in some states have followed EPA's lead and employ "individual risk" tests in regulating toxins. (14)

What accounts for this regulatory focus on "individual risk"? One answer is tempting, but wrong. The temptation is to say that regulatory agencies inevitably take the maximal level of "individual risk" as the test of safety, at least for substances and activities that cannot be removed from our lives without massive cost. Many, many chemicals cause cancer to animals at large enough doses, and can be predicted to cause some human deaths at actual doses in a sufficiently large group. (15) How else to determine which toxic exposures merit a regulatory response except by setting an "individual risk" threshold which seems very low--say, 1 in 1 million to the maximally exposed individual-and taking that as the trigger for regulatory intervention? But this response overlooks a crucial deficit in "individual risk" tests of this kind: their insensitivity to population size. Compare an isolated toxic waste dump that (under worst-case modeling) leaches contaminants to a radius of ten miles, affecting a population of 10,000; a workplace toxin employed in certain industries, to which one million workers are exposed; and a chemical in drinking water that is consumed by 100 million. For simplicity, assume that in each case every person in the exposed population incurs a 1 in 1 million risk of dying from the hazard. Then in the waste-dump case it is overwhelmingly likely that the hazard will cause no fatalities; in the workplace case it is reasonably likely that the hazard will cause one or more fatalities, with one incremental death the expected outcome; and in the drinking-water case it is overwhelmingly likely that the hazard will cause one or more fatalities, with 100 incremental deaths the expected outcome. (16)

Risk assessors typically distinguish between "individual risk"--the risk of death borne by a particular individual, either a named person or someone identified by her exposure characteristics--and "population risk." (17) "Population risk" (also sometimes called "societal risk") is the total number of fatalities resulting from a toxin, a hazardous activity, or some other threat to human life. To quote a leading textbook on risk assessment:

[Risk assessments typically] include several common measures of individual and societal risk, in particular:

* Individual risk, which is the probability of a specified individual dying prematurely as a result of exposure to the risk agents....

* Individual risk contours show the geographical distribution of individual risk....

* Maximum individual risk is the individual risk to the person experiencing the highest risk in the exposed population....

....

* Various measures of societal risk, such as ... the expected number of fatalities as a function of location or population subgroup.... (18)

Regulatory agencies might use the level of "population risk," rather than the level of "individual risk," as their measure of health and safety. (19) This is true both for agencies operating under statutes that accord high priority to the avoidance of death, illness, and injury, as opposed to other goals, as well as for agencies operating under cost-benefit statutes or other "balancing" statutes that permit a wider array of considerations to trump the goal of physical integrity. (20) Indeed, as we shall see, federal programs concerned with safety rather than health hazards generally seem to focus on "population risk" rather than "individual risk," and even health threats such as toxins, radiation, and pathogens are sometimes regulated with reference to "population risk." (21)

So the question just posed remains unanswered: why do EPA, OSHA, NHTSA, and many other agencies, federal and state, employ some variation on the "individual risk" construct--be it "individual risk" to the maximally exposed individual, to a highly exposed individual, to the median or average individual, or to some other person--in administering statutes that make human health and safety a (high-priority or ordinary-priority) regulatory goal? Why not use a fatality-based metric instead, for example one that looks at the effect of regulatory intervention in reducing the total number of deaths caused by fatal illnesses or injuries?

One plausible answer points to a seminal 1980 Supreme Court case, Industrial Union Department v. American Petroleum Institute. (22) This case, more than any other single event, triggered the rapid growth of risk assessment in the federal government. (23) And it may well have caused or at least supported the regulatory focus on "individual risk" rather than "population risk." In Industrial Union, a plurality of the Court invalidated an OSHA regulation lowering the maximum permissible workplace exposure to benzene, a carcinogen, from ten parts per million (ppm) to one ppm. The Occupational Safety and Health Act, as the plurality read it, authorized OSHA only to regulate "significant" risks--not to ban workplace chemicals or activities based on the mere possibility of an injury or fatality. (24) Note that this aspect of the Industrial Union opinion does not entail a preference for regulatory attention to "individual risk." After all, in implementing the "significant risk" threshold, OSHA could look to aggregate premature deaths resulting from the workplace toxin or activity at issue, not "individual risk" to the maximally exposed or average worker. But, in the final portion of the opinion, Justice Stevens suggested that the statutory requirement of "significant risk" be implemented through an "individual risk" test.

      Contrary to the Government's contentions, imposing a burden on
   the Agency of demonstrating a significant risk of harm will not
   strip it of its ability to regulate carcinogens, nor will it require
   the Agency to wait for deaths to occur before taking any action.
   First, the requirement that a "significant" risk be identified is
   not a mathematical straitjacket.... Some risks are plainly
   acceptable and others are plainly unacceptable. If, for example, the
   odds are one in a billion that a person will die from cancer by
   taking a drink of chlorinated water, the risk clearly could not be
   considered significant. On the other hand, if the odds are one in a
   thousand that regular inhalation of gasoline vapors that are 2%
   benzene will be fatal, a reasonable person might well consider the
   risk significant and take appropriate steps to decrease or eliminate
   it. (25)

To this day, OSHA carefully follows this dictum from Industrial Union. The agency still uses the 1 in 1000 level of "individual risk" identified by Justice Stevens as its cutoff for regulating a workplace carcinogen. (26) More generally, although EPA and FDA do not employ that cutoff--a cutoff which Stevens characterized as a reasonable construal of the Occupational Safety and Health Act, not the only acceptable construal--the Stevens dictum may well have prodded EPA, FDA, and other agencies to focus on individual, not population, risk. (27)

A second explanation for the wide use of "individual risk" tests by regulatory agencies points to the norms of risk assessment. Risk assessment is a set of techniques for quantifying health and safety threats, paradigmatically but not exclusively threats from toxic chemicals. (28) Risk assessments are very widely employed by government agencies in setting priorities and evaluating interventions, (29) and are also used in other contexts. (30) Risk assessment techniques have become quite standardized, both as a result of governmental standardization (for example, EPA's various guidelines) (31) and because of the standardization internal to the emerging professional community of risk assessors. (32) The core of risk assessment for toxins consists of two steps: drawing a dose-response curve and predicting individual exposures. As we shall see, dose-response curves and exposure assessments can be integrated to generate predictions of aggregate deaths--and sometimes are--but they are also naturally deployed to generate predictions of "individual risk." (33)

In any event, whether as a result of Industrial Union, the professionalized techniques of risk assessment, or other factors, governmental agencies in the United States, in a host of different contexts, employ the test of "individual risk" to the maximally exposed individual or some similar test as a criterion for regulatory choice. As traditional economic regulation has become less important, particularly at the federal level, an increasing proportion of regulatory activity concerns the avoidance of death and, to a lesser extent, nonfatal injury and disease. (34) In turn, "individual risk" tests have become a linchpin of government's health and safety efforts. This is misguided. In this Article, I shall launch a sustained critique of the use of "individual risk" tests by health and safety agencies. This critique does not depend on controversial normative commitments. My own commitments are welfarist and consequentialist, (35) and I have argued elsewhere in favor of cost-benefit analysis (CBA). (36) It is true that "population risk," not "individual risk," is the input to CBA as currently practiced. (37) But it emerges that normative frameworks directing agencies to accord higher priority to safety than CBA would countenance are also best specified in terms of "population risk" or cognate tests. Or so I shall argue below.

Parts I and II of the Article set the stage. Part I is a primer on risk assessment. It explains the structure of risk assessment, describes its rise to prominence as a tool for health and safety regulators, and then explores the nature of the "individual risk" numbers so central to the technique. What exactly does it mean to say that some toxin, substance, activity, or, more abstractly, some object or event imposes a 1 in x risk of death upon a particular individual? What concept of "risk" is being invoked here? The standard interpretation of the "individual risk" numbers generated by risk assessment invokes the frequentist view of risk. On the frequentist view, to say that E imposes a 1 in x risk of death upon P is to say this: over the long run, when people similar to P are exposed to events similar to E, a 1 in x fraction of those individuals will die prematurely as a result of those exposures.

Part II describes, in detail, the widespread use of "individual risk" tests by federal agencies. The practices I describe should be familiar to environmental lawyers, food and drug specialists, workplace safety scholars, and others who have specialized knowledge about EPA, FDA, OSHA, or similar agencies. These important practices will not be as familiar--I hazard to guess--for public law generalists, law and economists, legal philosophers, and other scholars who may have a deep interest in the regulatory state but have not read the latest issue of Risk Analysis or the latest version of EPA's Guidelines for Carcinogen Risk Assessment. (38) And even risk-regulation specialists might be surprised to learn just how widespread the focus on "individual risk" is. In any event, Part II seeks to show that administrative decision making across a wide swath of significant governmental programs conforms to a problematic recipe. This recipe specifies the safety of a workplace, a toxic dump, a water source, a radioactive emission, a consumer product, or some other regulatory target in terms of the frequentist "individual risk" the targeted substance, activity, or product imposes on the maximally exposed individual or some other person (with 1 in 1 million most widely used as the "safe" level).

The remainder of the Article provides a rigorous, normative critique of the frequentist "individual risk" tests described in Part II. I seek to show that "individual risk" in the frequentist sense is normatively irrelevant across a range of plausible moral theories. Part III looks at welfarist consequentialism: the moral view undergirding welfare economics and cost-benefit analysis. It argues that the kind of risk relevant to welfarist consequentialism is Bayesian risk, not frequentist risk. Bayesian risks are measures of belief, not measures of frequency. On the Bayesian view, to say that some individual has a 1 in x probability of death means that the risk analyst, the individual herself, or someone else believes to degree 1 in x that the individual will die. Part IV explores the subtle, but crucial differences between Bayesian and frequentist risk.

Part V moves beyond welfare consequentialism and examines alternative moral views: safety-focused views that accord special priority to physical integrity; deontological views that recognize the existence of moral rights (specifically, a right not to be killed and perhaps an independent right not to be put at risk of death) ; contractualist views that evaluate governmental choices by asking whether citizens in a suitable, hypothetical contracting scenario would approve the choices; and democratic views that see democratic responsiveness (including responsiveness to popular judgments about risk) as morally important. Part V argues that none of these moral views warrants regulatory attention to "individual risk" in the frequentist sense.

Part VI shifts critical focus to a different feature of the current regulatory practices described in Part II. Those practices are, in effect, doubly misguided. First, they make "individual risk," in the frequentist rather than Bayesian sense, a determinant of regulatory choice. Second, they are insensitive to population size. Whether regulators should intervene to abate some hazard depends, morally, on the number of persons at risk from the hazard. A specialty food consumed by a very small group, an industrial toxin with which many more workers come into contact, and an airborne pollutant that we all breathe might impose the very same "individual risk" on the maximally exposed, high-end, median, and average exposed individual. But the morally warranted regulatory responses in these cases will, or at least may, be very different. Part VI argues that both welfare consequentialism and alternative moral views (safety-focused, deontological, contractualist, and democratic views) demand risk assessment procedures that are sensitive to population size.

What this Article offers, in short, is a sympathetic critique of risk regulation and risk assessment. Much of the legal scholarship in this area is more radically critical. Regulation guided by risk assessment is allegedly flawed to the core--for example, because it is undemocratic, or because it is beset with uncertainties about the mechanisms of cancer, dose-response relationships, and exposure pathways. The very enterprise of quantifying safety is seen as misguided. (39) I do not believe that the very enterprise of quantifying safety is misguided. Risk assessment represents a giant leap forward for public rationality, in my view. Dose-response curves and exposure assessments are, properly, central for the regulation of toxins; parallel techniques are central for agencies that focus on other threats to life and limb. But these impressive techniques should be used to illuminate what is truly at stake in risk regulation, not to distract us with morally unimportant information. Risk regulation needs to be changed in two ways. First, it should adopt a new understanding of risk, Bayesian rather than frequentist. Second, it should adopt choice criteria that are sensitive to population size--paradigmatically, "population risk" criteria. Risk assessment, in turn, must be reworked so that it can inform regulatory choice thus revised.

I. RISK ASSESSMENT: A PRIMER

A. History and Structure

Risk assessment, generically, is a set of techniques for quantifying the fatalities or fatality risks resulting from various hazards. These techniques can also be used to quantify nonfatal illness or injury, or the risk of nonfatal illness or injury. However, because death is the central and paradigmatic harm addressed by health and safety regulators, my presentation will focus there. The best-developed variant of risk assessment, in current regulatory practice, is toxic risk assessment: a quantitative description of the fatalities and fatality risks caused by toxic chemicals. But risk assessment with respect to a much wider array of death's causes is also possible and, to some extent, practiced.

Let's start with the toxins. Toxic risk assessment is, in effect, quantitative toxicology and dates from the nineteenth century. (40) Toxic risk assessment by U.S. governmental entities is more recent than that, but still has a substantial history. (41) Toxic risk assessment at the federal level was pioneered by FDA. This agency is charged with implementing a statute that generally precludes foods containing "poisonous or deleterious" substances (42) and that imposes even stricter standards on "food additives": such additives must be "safe," (43) and carcinogenic food additives are flatly prohibited by the (in)famous "Delaney Clause." (44) FDA began systematically to engage in the risk assessment of the noncancer toxicity of food constituents in the 1950s, developing the so-called NOAEL/safety factor method which is the standard method for noncarcinogens today. (45) As for the threat of cancer: there are various escape routes around the absolutism of the Delaney Clause--for example, FDA takes the position that the Clause does not apply to the nonfunctional carcinogenic constituents of additives (46)--and in the 1970s FDA commenced a practice of quantifying the potency of certain food carcinogens. The 1 in 1 million cutoff for individual cancer risk derives from FDA practice during this period. (47)

The widespread use of toxic risk assessment at other federal agencies began in the 1980s. This development had multiple triggers, including three apparent ones: (1) the Industrial Union case, which forced OSHA to follow FDA's lead and, more generally, made clear that the courts would not permit federal agencies engaged in health and safety regulation to impose large costs on the regulated entities without some effort to justify such costs through a quantification of benefits, even in cases (as with OSHA) where the underlying statute was quite pro-safety; (48) (2) the 1983 appointment, as EPA administrator, of William Ruckleshaus, who made risk assessment his top priority and actually succeeded in infusing such techniques into administrative routines throughout the large EPA bureaucracy; (49) and (3) the publication, also in 1983, of a seminal study by the National Research Council, Risk Assessment in the Federal Government: Managing the Process (50) (the so-called Red Book), which further popularized the practice of risk assessment and, perhaps more importantly, did much to standardize it. By the 1990s, risk assessment had become such a familiar feature of the regulatory landscape that OMB, in its guidance to federal agencies regarding Executive Order 12,866, (51) instructed that the regulatory impact analysis required by this Executive Order prior to the issuance of major rules should include a "risk assessment":

      Estimating the benefits and costs of risk-reducing regulations
   [requires, inter alia] ... a risk assessment that ... characterizes
   the probabilities of occurrence of outcomes of interest....

      ... The risk assessment should generate a credible, objective,
   realistic, and scientifically balanced analysis; present information
   on hazard, dose-response, and exposure (or analogous material for
   non-health assessments); and explain the confidence in each
   assessment.... (52)

So risk assessment is now standard practice for federal agencies that regulate toxins, as well as other health and safety agencies, at the major rulemaking stage. But the practice of toxic risk assessment is really much broader than that. For example, the overwhelming majority of EPA risk assessments do not involve major rules, but other categories of administrative decision, such as clean-up decisions with respect to individual Superfund sites. (53)

The Red Book framework for toxic risk assessment has been the canonical framework (54) since its publication and runs as follows. There are four parts to toxic risk assessment: hazard identification, dose-response assessment, exposure assessment, and risk characterization. (55) Hazard identification is a preliminary step: the risk analyst verifies that the allegedly toxic substance is indeed a toxin, that there is sufficient evidence of a causal link to disease and death. (56) If so, the analysis moves on to the two central parts of the risk-assessment inquiry, namely dose-response assessment and exposure assessment. Dose-response assessment means quantifying the link between different doses of the toxin and premature death. This inquiry is, in effect, physiological: it seeks to determine how frequently the ingestion, inhalation, or dermal uptake of the toxin, into humans' bodies, leads to cancer or other fatal illnesses. This physiological inquiry is almost always grounded in two types of data--rodent bioassays, in which the differing rates of fatal illness in groups of rodents fed different doses of the toxin are measured, and human epidemiological data--and eventuates in a dose-response curve. (57) The X-axis of the curve represents human doses of the toxin; the Y-axis, the risk to a person exposed to that dose of dying prematurely as a result.

Exactly how to draw this curve, based on the animal or epidemiological evidence, is a technical (but important!) issue in risk assessment.

While dose-response assessment is physiological, exposure assessment is topographical and demographic. The aim here is to characterize the pattern of exposures to the toxin that will occur under various contingencies (for regulatory purposes, as a result of different regulatory options, including the status quo option of inaction). (58) This assessment depends on where the toxin is located; on how much toxin currently exists, and would be produced under various contingencies, at the source; on the toxin's so-called "fate and transport," i.e., how the toxin migrates through the air, the water, and other environmental pathways; and on how the human population is distributed at varying distances from the source of the toxin. What "source" means depends, of course, on the toxin and the regulatory program. It might be a single waste dump, a group of smokestacks (for example all smokestacks in factories in a given industrial category), a food type (in which case the relevant pathway is direct ingestion by food consumers), containers of hazardous workplace chemicals (in which case the relevant pathway is air transport to workers or direct worker contact with the containers), and so on. A relatively complete exposure assessment will predict the dose of the toxin that each member of the population will receive as a result of the evaluated source. Typically, members of the population will not be identified by name, but rather by their doses. That is: a relatively complete exposure assessment will produce a predicted distribution, by numbers and percentiles, of lifetime doses resulting from the analyzed source, for the status quo option of regulatory inaction and, ideally, for each regulatory contingency being assessed. (59)

To be sure, the toxic exposure assessments produced by regulatory agencies or their contractors are often not this detailed. If the agency's program focuses on risk to the maximally exposed individual, (60) then the full pattern of dosages that will occur in the status quo, or as a result of various regulatory interventions, is irrelevant. The analyst might estimate the dosage received by the maximally exposed individual by generating a full distribution of doses across the population, then using a very high percentile as maximum exposure; or she might do so more directly by maximizing the parameters underlying her exposure model and determining what dosage results. Concretely, this might mean estimating the maximum exposure to a toxic air pollutant emitted from a factory by looking at the exposure incurred by the person living closest to the factory, on the conservative assumptions that he lives there for his entire lifetime and that his inhalation rate is at the high end of the population distribution of such rates. (61)

The final stage of risk assessment, risk characterization, is the prime focus of this Article. "Risk characterization" means combining the dose-response assessment (which correlates doses and fatality risks) and the exposure assessment (which predicts doses, across the population or at least for some segment) so as to generate a prediction of the fatalities and fatality risks resulting from the toxin under various contingencies. (62) The risk assessment jargon for total fatalities, as I have already noted, is "population risk"; the jargon for the risk of death incurred by one or another individual is "individual risk." Using the dose-response assessment and exposure assessment to predict "population risk" is somewhat laborious. In general, to do that, the analyst needs a full population distribution of doses, and even then the analyst cannot simply "read" an estimate of total deaths off the dose-response curve, but instead must use probability theory to generate a probability distribution of total deaths and then a point estimate of "population risk" equaling the mean number of total deaths. (30)

Generating a prediction of "individual risk" is more straightforward. For example, if the analyst possesses a full or truncated exposure assessment showing the dosage of the toxin to the "maximally exposed" individual, then the "individual risk" to that person is simply the risk corresponding to that dosage given by the dose-response curve. And if the analyst possesses a full or truncated exposure assessment showing the exposure to the "representative" individual--the person at the median or mean of the dosage distribution--then the "individual risk" incurred by this "ordinary Joe" is the risk for his dosage predicted by our physiological graph, the dose-response curve.

I have focused, to this point, on toxic risk assessment, since the standard practices in this area readily lend themselves to predictions of "individual risk" (as I have tried to show) and since (as we shall see) many of the important cases of "individual risk"-based decision making by agencies involve toxins. But nontoxic risk assessment--quantitative assessment of the wide variety of threats to human health and safety posed by substances or activities other than toxic chemicals--is also quite important in governmental practice. A salient example here is radiation risk assessment, as pioneered at the federal level by the Nuclear Regulatory Commission. The famous Reactor Safety Study (WASH-1400), (64) commissioned by NRC's predecessor agency and published in 1975, was the first full-blown, probabilistic evaluation of the core damage accidents at nuclear reactors that could lead to dangerous releases of radiation. (65) This study, together with the Three Mile Island accident four years later, prompted NRC to make risk assessment integral to the licensing and regulation of nuclear plants. (66) Reactor risk assessment divides, very roughly, into two parts: (1) evaluating the probability of different types of releases (releases of various amounts of various radioactive isotopes); and (2) evaluating the safety threat for any given release. (67) This second component closely tracks the standard methodology for toxic risk assessment. For any given release, an exposure assessment can be performed evaluating the possible fate and transport of the released isotopes and the exposure to those substances of various members of the population; this information, when combined with a dose-response curve correlating radiation doses with an individual's risk of dying as a result of the exposure, can be used to predict the "individual risk" of death imposed on different individuals by any given release and therewith (if desired) a prediction of "population risk. (68)

With the exception of radiation, the norms of nontoxic risk assessment are less well established than for toxics. NHTSA, for example, certainly engages in risk assessment of a sort--the quantitative evaluation of motor vehicle safety (69)--but lacks a clear template analogous to the exposure/dose-response framework made canonical, for toxics regulators, by the 1983 Red Book. Extending the "exposure" construct from toxins to radiation or certain other health hazards (such as pathogens) is straightforward; extending it further, to car crashes, industrial accidents, dangerous consumer products, or other sources of bodily injury targeted by federal regulators, is not quite so easy. But some such extrapolation is often possible. As one researcher in the area of occupational injury notes:

   Injuries are acute events associated with the transfer of hazardous
   Levels of energy. A fatality only occurs when the energy source
   contacts the worker in a specific way (e.g., a tree falling on a
   logger's leg may cause a severe fracture, but probably not death,
   while the same tree striking the logger's head will usually cause
   death). Since the worker is only exposed to a potential fatal injury
   hazard for a portion of the workday, the estimation of exposure for
   traumatic injuries is complex.... (70)

Implicitly, here, "exposure" is understood as physical proximity to some machine or other workplace object that might cause death or (yet more abstractly) as the occupying of a certain spatiotemporal location in the workplace that makes some type of injury possible (falling from a high place). Some such conception of "exposure" can, in principle, provide a foundation for "exposure" assessments and "exposure"-response curves for occupational injury as well as car crashes, dangerous products, and other safety hazards. (71) Safety agencies, particularly OSHA, are just now beginning to develop risk assessment techniques along these lines. (72)

B. What Is Risk? The Frequentist Answer

What conception of "risk" is involved in risk assessment? When dose-response curves correlate an exposure amount with an "individual risk," what precisely does that risk number mean?

The answer: risk assessment trades upon a frequentist, rather than a Bayesian, conception of risk. Frequentism and Bayesianism are the two great traditions in the intellectual history of risk and probability. (73) Bayesianism has been hugely influential within economics and social science; (74) but the frequentist view of risk is the mainstream view within experimental science and, derivatively, within risk assessment, which has been dominated by toxicologists and other applied scientists. (75)

Scientific models are often probabilistic, and bedrock physical laws may be irreducibly probabilistic, as is now thought to be true of quantum mechanics. (76) Generically, scientists need to be able to attach a probability to a proposition asserting that some event will have some attribute. The Bayesian suggests that the probability of a proposition concerning an event, more generally the probability of any proposition, is simply someone's degree of belief in the proposition: the actual scientist's degree of belief, a "reasonable scientist's" degree of belief, an idealized observer's degree of belief, etc. (77) But scientists and, traditionally, statisticians have eschewed the suggestion, because it makes essential reference to minds--to beliefs--and thus has seemed too subjective for scientific purposes. (78)

Instead, following the lead of the great Austrian probabilist Richard von Mises, (79) scientists typically see probabilities as frequencies within reference classes. Consider a very simple reference class: for simplicity, a class of physical objects (the class of all rocks now existing, say) rather than a class of events, which constitute a more esoteric kind of object. What is the probability that "rocks weigh 100 pounds"? Probability numbers range from zero to one, and have some other basic characteristics: if the probability that "rocks weigh 100 pounds" is p, then the probability that "rocks do not weigh 100 pounds" is 1 - p; if the probability that "rocks weigh 50 pounds" is q, and rocks can't be both 50 pounds and 100 pounds, then the probability that "rocks weigh 50 pounds or 100 pounds" is p + q. (80) Notice that frequency numbers within the class of rocks satisfy just these mathematical characteristics. The frequency or proportion of 100-pound rocks can't be less than zero or greater than one; because rocks must be either 100 pounds or not 100 pounds, the proportion of 100-pound rocks plus the proportion of not-100-pound rocks is unity; and so on. Finally, note that the proportions here, hence the probabilities, are no more mind-dependent than the underlying objects and attributes: whether a rock exists, and has a given weight, does not depend on anyone's beliefs, and neither, then, does the proportion of 100-pound rocks.

Mises formalized the intuitive connection between frequencies and probabilities, by defining probability as the limit of the frequency of some attribute within an infinite, sequentially ordered reference class of events. (81) Imagine the infinite class of {[E.sub.1], [E.sub.2] ...}. This class has a series of segments, each of which includes the one before: {[E.sub.1]}, {[E.sub.1], [E.sub.2]}, {[E.sub.1], [E.sub2], [E.sub.3]}, and so on. For each such segment, one can calculate the proportion of the events with the attribute of interest A. If, as the segments become longer, the proportion of A-type events approaches p as a limit, then (on the Misean construct) p is the probability of events within this infinite class having attribute A. Mises's use of infinite reference classes accommodates the intuition that changes in observed frequencies in a finite series of experiments do not entail changes in the underlying probability, and has been somewhat controversial. (82) This controversy need not occupy us. The important point to understand is that Mises's construct, or some frequentist variant, is what underlies probability ascriptions within contemporary science and therewith quantitative risk assessment.

In the context of toxic risk assessment, it is pretty easy to see how this goes. The events of interest, here, are toxic exposures: the passage into a human's body of a particular dose of some toxin. For any given dose, one can imagine exposing humans to that dose, over and over again, indefinitely. This "dosing class" is an infinite class of events, specifically the infinite class of hypothetical human exposures to the particular dose. The relevant event-attribute is "causing death." Each exposure either causally contributes to, or fails to causally contribute to, the death of the person receiving that dose. For the first 1000 exposures, say, the number of subjects who die as a result of those exposures is five. For the first 10,000, say, the number is forty-eight. For the first 100,000, the number is 491. If the fractions converge, in the limit, to (say) 0.0049, then that is the frequentist, Misean probability of an event within this dosing class causing death.

To be sure, the Misean probability of death-causation within a dosing class cannot be directly observed; it is not within human capabilities to perform an infinite series of experiments. But by performing or observing a finite series of dosings, we can use statistical techniques to generate more or less precise estimates of the true frequentist probability. (83) Whatever epistemological difficulties might attend the estimation of relative frequencies, those difficulties will not be the focus of my normative critique in Parts III through V below. Rather, my focus will be conceptual. Conceptually, for the frequentist, risk is relative to reference classes. (84) Thus the ascription of risk to a particular event is arbitrary--as arbitrary as choosing a reference class, among the multiplicity of possible ones, within which to subsume a particular event. This so-called "problem of the reference class" has been much mooted by philosophers of science and probability. (85) Astonishingly, the problem is almost never mentioned within the risk assessment literature, even within scholarship that is otherwise extremely sophisticated.

Consider any particular exposure event. P ingests a 100-gram dose of the toxin. This particular event can be characterized in a multiplicity of ways. First, it can be characterized without a precise specification of P's dose. P has received a dose between fifty and 150 grams. He has, at the same time, received a dose between twenty and 500 grams. He has, at the same time, received a dose between ninety and 200 grams. Second, the event can be characterized with a precise specification of P's dose, and with further description of him. P has received a 100-gram dose, and P is forty years old. P has received a 100-gram dose, and P has a family history of cancer. P has received a 100-gram dose, and P has a family history of cancer and is a smoker. Finally, the event can be characterized with a precise specification of P's dose, and with no further description of him. Phas received a 100-gram dose. Each of these possible characterizations of the exposure event generates a different reference class: the class of all dosings between fifty and 100 grams; the class of all dosings between twenty and 500 grams; the class of all dosings between ninety and 200 grams; the class of all 100-gram dosings to forty-year-olds; the class of all 100-gram dosings to those with a family history of cancer; the class of all 100-gram dosings to smokers with a family history of cancer; and, finally, the class of all 100-gram dosings. But the frequency with which toxic exposure causes death, within these various reference classes, may well be different. (86) Those who receive a dose between fifty and 100 grams die as a result less frequently, one imagines, than those who receive a dose between ninety and 200 grams. Smokers who are exposed die more frequently than nonsmokers. And so on.

The reference class standardly used to calculate "individual risk" is the third type of class (87)--what might be called the canonical dosing class. Canonical dosing classes are composed of all exposures to humans precisely specified with respect to dose, and otherwise unspecified. Consider the exposure of P, a forty-year-old Caucasian smoker with a family history of cancer, to a 100 milligram (mg) dose of benzene. The canonical dosing class subsuming this exposure is the class of all events whereby human persons, of any age and with any other behavioral or genetic characteristics, are exposed to a 100 mg dose of benzene. Remember the form of the classic dose-response curve. This correlates exposures (defined as precisely as possible, in terms of a real number) with "risks," i.e., frequencies. And the curve is valid for all humans, not for a more precisely characterized subset: risk analysts typically use a single dose-response curve per toxin, rather than (say) one dose-response curve for forty-year-old smokers with a family history of cancer, another for forty-one-year-old smokers with a family history of cancer, another for forty-year-old nonsmokers, etc.

Thus, when EPA, OSHA, FDA, or a state agency undertakes an exposure assessment; determines that the "maximally exposed individual" or "highly exposed individual" or "representative individual" receives a particular dose; uses the generic dose-response curve to attach a 1 in 100,000 risk to this exposure; and concludes that the "maximally exposed" or "highly exposed" or "representative individual" is subjected to a 1 in 100,000 risk, what this means is the following: by characterizing this individual and exposure in the canonical way, abstracting from everything about the exposure except the dose received, we generate a class of exposures 1 in 100,000 of which result in death. But viewed another way (subsumed in a different, noncanonical dosing class), the "maximally exposed" or "highly exposed" or "representative individual" is subjected to a much lower risk. And viewed yet another way, she is subjected to a much higher one. Consider this analogy: whether some particular person's hair color is "unusual" depends on how we characterize the color ("bright red," "red," "within the red-brown range") and who we compare the person to. The very same adult male might come out as "unusual" if viewed as a bright red head and compared to all males, but not "unusual" if viewed as a person with hair in the red-brown range and compared to all persons in a particular ethnic group. The canonical "individual risk" number ascribed to the individual maximally exposed, or highly exposed, or receiving an average exposure from some toxin, is no more unique than a description of his hair color as "unusual" or "not unusual" generated in some standard way (by using the scheme for describing colors, and generating comparison classes, employed by the National Association of Barbers, for example). And once we understand this about these canonical risk ascriptions, it becomes seriously questionable why we should care about them.

II. RISK REGULATION AND "INDIVIDUAL RISK": A SURVEY OF GOVERNMENTAL PRACTICE

The preceding Part was a primer on risk assessment. In particular, it showed how the concept of "individual risk" is central to dose-response curves, a central component of risk assessment as currently structured, and it explicated the "frequentist" understanding of "individual risk." Regulatory programs relying on risk assessment as an input need not be focused on "individual risk"--dose-response curves and exposure assessments might instead be used to generate predictions of "population risk"--but the current structure of risk assessment certainly facilitates a focus on "individual risk."

This Part shows, in detail, that health and safety agencies do indeed place substantial emphasis on frequentist "individual risk." This is especially true for carcinogens, other toxic chemicals, pathogens, and radiation, but also encompasses the regulation of certain other hazards. The relevant "individual risk" is sometimes the risk to a maximally or highly exposed individual, sometimes the risk to an average individual. And these "individual risk" numbers play a range of regulatory roles. Sometimes they serve as decisional triggers: a toxin or other hazard is placed on the regulatory agenda, as it were, if the "individual risk" number is sufficiently high. Sometimes (a related idea) they serve as regulatory triggers: preexisting standards come into play if the hazard is sufficiently harmful, as measured by "individual risk" to the maximally exposed individual, the highly exposed individual, or the average individual. Finally, "individual risk" levels sometimes serve as criteria for shaping regulatory measures: rules, cleanups, or other measures should be sufficiently stringent to bring the "individual risk" (consequent upon a maximal, high-end, or average exposure) below some level. In this last role, "individual risk" might serve as a master criterion, or might instead be balanced against other criteria (for example, criteria measuring cost, technological feasibility, or "population risk").

Section A focuses on EPA, the most important health and safety agency in the United States, and the agency where attention to "individual risk" is most pervasive. Section B describes the "individual risk" based practices of other federal health and safety agencies, specifically FDA, OSHA, NRC, and the Consumer Product Safety Commission (CPSC). (88) Section C briefly discusses the role of "population risk" in regulatory choice: even at EPA, FDA, OSHA, NRC, and CPSC, "population risk" considerations do play a role, and they certainly do at other agencies, such as NHTSA.

My ambition in this Part is not to provide a comprehensive overview of governmental risk assessment practices. It is rather to present the major examples of current federal (89) health and safety programs where "individual risk" has a function in regulatory choice. Unless otherwise noted, "individual risk" means the risk of death. (90) The practices here described are the target for the revisionary, normative analysis provided in Parts III to VI below.

Two final preliminary points: first, as I have already discussed, the dominant understanding of "individual risk" within the risk assessment community is frequentist, not Bayesian. This understanding pervades the regulatory practices described in this Part. "Individual risk" is, at least implicitly, understood by EPA, FDA, OSHA, NRC, and CPSC as the frequency of death relative to a canonical dosing class or some other reference class. (91) The methodologies employed by these agencies to generate the "individual risk" numbers to which they then attach (some kind of) weight are frequentist methodologies. This point is crucial, but just because it is so general it is stated here, once and for all, and will not be belabored below.

Second, it is possible, in principle, for decisional criteria to make reference to "individual risk" yet be sensitive to population size. For example, an agency might calculate the total number of individuals incurring different levels of "individual risk" and attempt to minimize the totals within each category. (92) But, as shall emerge in the following survey of agency practice, federal agency consideration of "individual risk" typically does not take this form. Instead, the fact that the "individual risk" borne by some person in the exposure distribution lies above or below some stipulated level (be it 1 in 1000, 1 in 10,000, 1 in 100,000, or 1 in 1 million) functions as a deliberational or regulatory trigger, or an index of regulatory success or failure, regardless of the number of persons in the exposure distribution--regardless of the size of the population exposed to the hazard. This feature of current administrative practices--the use of "individual risk"-based decisional criteria which are insensitive to population size--is one way in which those practices are normatively misguided, as I shall argue in Part VI.

A. "Individual Risk" and Agency Practice: The Environmental Protection Agency

EPA is the largest health and safety agency in the federal government, and "individual risk" has a central role in this agency's decision making. Most risk assessment at EPA concerns environmental toxins, EPA's predominant regulatory target. Although EPA does also have a role in regulating pathogens and radiation, and "individual risk" considerations do come into play here, (93) this survey of EPA practice will focus on toxins. Risk assessment for toxins falls into two sub-categories: cancer risk assessment and noncancer risk assessment. EPA practices in these two areas will be described separately since there are important technical differences between cancer and noncancer dose-response curves, to be explained anon.

1. Cancer Risk Assessment and "Individual Risk"

What follows are the central examples of EPA programs where the "individual risk" of cancer is determinative, wholly or partly, of the agency's regulatory choices.

a. Air pollution (Clean Air Act section 112)

The chief provision of the Clean Air Act governing carcinogens is section 112. (94) In its original form, as enacted in 1970, section 112 required EPA to identify "hazardous air pollutants"; once a chemical was thus listed, EPA was required to regulate emissions of the pollutant so as to "protect the public health" with an "ample margin of safety." (95) The agency was slow to implement this provision, in part because of a dilemma created by the toxicology of carcinogens. Carcinogens do not have a physiological threshold below which they lack toxicity; in other words, at any nonzero dose, the incremental fatality risk (incremental frequency of cancer within that dosing class) is greater than zero. Thus EPA faced the dilemma of either banning all emissions of carcinogenic pollutants (with huge economic costs) or allowing some emissions and therewith a nonzero probability of some deaths, in apparent violation of the "ample margin of safety" language of section 112. (96)

EPA's eventual response to this dilemma was to link section 112 standard-setting to technological feasibility: for nonthreshold toxins such as carcinogens, polluters would be required to employ the "best available technology" for limiting emissions. The D.C. Circuit, in a 1987 decision involving the regulation of vinyl chloride, struck down EPA's interpretation of section 112, (97) and EPA thereupon settled on a new interpretation--one that looked to "individual risk." EPA's new test for determining a permissible emission level for carcinogenic air pollutants ran as follows: (1) the risk to the maximally exposed individual could not exceed 1 in 10,000; and (2) EPA would consider further reductions in the permissible level so as to minimize the number of individuals whose risk exceeded 1 in 1 million, but would also consider cost and feasibility considerations at this stage of the analysis. (98)

Congress quickly stepped into the fray and overhauled section 112 in 1990. The statute still applies to "hazardous air pollutants," but now provides a specific list of 188 pollutants, which EPA can revise under certain conditions. The old "ample margin of safety" test has been replaced with a more complicated structure. Specifically, EPA must set emissions limits for carcinogenic pollutants as follows: (1) the limit must be no higher than attainable using the best available technology; (99) and (2) EPA must eventually consider even lower limits if the technology-based limits "do not reduce lifetime excess cancer risks to the individual most exposed to emissions ... to less than one in one million." (100) This explicit statutory "individual risk" provision seemingly functions as a decisional trigger requiring EPA to consider lower limits, not a substantive requirement that the lower limits must meet, (101) and the statute appears to contemplate that EPA will employ its hybrid, pre-1990 test (no "individual risk" above 1 in 10,000; minimize the number of individuals incurring a risk above 1 in 1 million, taking into consideration cost and feasibility) as the substantive criterion here. (102)

b. Water pollution (Clean Water Act and Safe Drinking Water Act)

The Clean Water Act (103) requires that point sources of water pollution be licensed, either by EPA or by a state acting as EPA's delegate. License conditions must ensure both that the point source meets EPA's technology-based limitations and that discharges not result in violations of state water quality standards. These standards (which are also independently enforced by the states) are reviewed and approved by EPA. (104) Historically, EPA encouraged states to set standards for carcinogens so that an individual consuming a quantity of water (two liters a day) that was seen as somewhat above average but below the "high end" of the distribution, and eating a similar amount offish (6.5 grams a day), would incur a fatality risk in the range of 1 in 10 million to 1 in 100,000. In effect, then, EPA's review of state water quality standards focused on "individual risk" to an above-average but nonmaximal individual. (105) Several years ago, EPA revised this policy to focus on the 90th percentile consumer, rather than someone whose fish and water intake is closer to average, and to adopt the 1 in 1 million "individual risk" level as its water quality goal. (106) It should be stressed that EPA's articulated "individual risk" goals under the Clean Water Act (historically a range of 1 in 100,000 to 1 in 10 million, now 1 in 1 million) are goals rather than rigid requirements, and EPA has been willing to approve state water quality standards that are somewhat less protective. (107)

The other key water pollution statute enforced by EPA is the Safe Drinking Water Act. (108) Under this statute, EPA directly promulgates federal water standards, which apply to virtually all public water supplies. (109) These standards are set in a two-step process: first, for each drinking water contaminant EPA sets a maximum contaminant level goal (MCLG), namely that level of the contaminant "at which no known or anticipated adverse effects on the health of persons occur and which allows an adequate margin of safety." (110) Then, EPA establishes the legally permissible level--a higher level--by taking into consideration nonhealth factors. Historically, the statutory language governing this second step focused on technological feasibility but also adverted to cost considerations: EPA was enjoined to set the legal standards as close to the MCLGs as "feasible," defined to mean "feasible with the use of the best technology, treatment techniques and other means which ... are available (taking cost into consideration)." (111) In 1996, the statute was amended to permit a cost-benefit analysis in lieu of the feasibility analysis at the second step. (112)

EPA relies, in part, on "individual risk" in setting MCLGs. Where EPA has strong evidence that a substance is a carcinogen, it sets the MCLG at zero; where it has weaker evidence (say, some evidence of carcinogenicity in animal tests and no direct epidemiological evidence), EPA sets the MCLG so that the "individual risk" to the individual drinking an above-average amount of water is within the range of 1 in 100,000 to 1 in 1 million. (113)

And, historically, EPA has also relied on "individual risk" in deriving enforceable federal standards from the MCLGs. Notwithstanding the statutory "feasibility" language, EPA typically set enforceable standards so that "individual risks" to above-average individuals lay in the range of 1 in 1 million to 1 in 10,000. (114) This practice has been changed, although not radically so, by the 1996 amendments. In recent Safe Drinking Water Act rulemakings, EPA has relied both on the traditional "individual risk" test just described and on cost-benefit analyses incorporating information about aggregate deaths in setting enforceable federal drinking water standards. (115)

c. Solid waste (RCRA and CERCLA)

The Resource Conservation and Recovery Act (RCRA) (116) and EPA's implementing regulations subject solid, hazardous waste to numerous stringent requirements, covering the generation, transportation, and (most stringently) the disposal of such wastes. These requirements are generally framed in fairly specific terms--for example, requiring disposers to employ certain technologies--and not in risk terms. (117) Rather, risk assessment comes into play under RCRA in EPA's initial determination whether to "list" particular substances as "hazardous wastes," thereby subjecting those substances to the onerous requirements just mentioned. Risk assessment also has a role to play in specifying generic toxicity characteristics such that substances with these characteristics are automatically "hazardous wastes" even without being "listed" by EPA. (118)

With respect to "listing" decisions, EPA policy has been to categorize a carcinogen-containing substance as a "hazardous waste" if unregulated disposal of the substance--specifically, disposal in ordinary municipal landfills and subsequent dissemination through groundwater--would produce an "individual risk" to a highly exposed individual (an individual in the 85th or 90th percentile of exposure) greater than 1 in 10,000. The agency exercises discretion in the range between 1 in 10,000 and 1 in 1 million. Previously "listed" substances can be delisted if their "individual risk" level is below 1 in 1 million. EPA considers substances to be generically toxic if they leach certain chemicals above specified concentrations, and in setting these concentrations the agency has employed an "individual risk" level of 1 in 100,000. (119)

RCRA also empowers EPA to order remedial action at active waste-disposal sites. The general agency practice, here, is apparently to perform a detailed study of remedial options for particular sites producing more than a 1 in 1 million risk to the highly exposed individual, and then to consider cost and feasibility in choosing among these options but reject remedial options resulting in more than a 1 in 10,000 risk (again to the highly exposed individual). (120) Site-specific risk assessments may be performed, too, for hazardous waste incinerators, which under RCRA require licenses. EPA apparently once took the position that a site-specific risk assessment demonstrating an "individual risk" (to a highly exposed individual) below 1 in 100,000 was a precondition for licensure. (121) More recently, EPA in a major rulemaking determined that technology-based standards for incinerators required by the Clean Air Act would generally reduce risks from dioxin emissions to individuals in the 90th percentile of exposure below 1 in 10,000, and therefore that site-specific risk assessments were not presumptively required. (122)

While RCRA is forward-looking, the Superfund statute, the Comprehensive Environmental Response, Compensation, and Liability Act (CERCLA), (123) is backward-looking. CERCLA authorizes EPA to order clean-ups of inactive waste sites, with the expense of the clean-up borne by a range of private actors associated with the site. (124) Site-specific risk assessments are absolutely central to EPA's remedial decisions under CERCLA. And the goal of these risk assessments, like those under the Clean Air Act, the Clean Water Act, RCRA, and (with some recent exceptions) the Safe Drinking Water Act, is not to generate information about "population risk." Instead, CERCLA site-specific assessments focus on "individual risk" given a "reasonable maximum exposure." "Reasonable maximum exposure," as that construct is specified by EPA, is the "maximum exposure that is reasonably expected to occur at a site." (125) More specifically, "reasonable maximum exposure" is calculated by using:

   [some] values for exposure factors ... that are mean estimates (body
   weight) and some parameter values that are upper bounds (e.g.,
   exposure duration). For the concentration of the chemical at the
   site, the EPA guidance directs that the 95th upper confidence limit
   on the estimate of the mean concentration at the site or the maximum
   detected concentration be used, whichever is lower. (126)

EPA policy (more formalized than in the RCRA context) is to order clean-up at a CERCLA site where the individual cancer risk consequent upon "reasonable maximum exposure" exceeds 1 in 10,000; to refrain from clean-up where this risk is less than 1 in 1 million; and to exercise discretion in the range between the two levels. A clean-up, if ordered, must bring "individual risk" to the reasonably maximally exposed individual to within this same range of "individual risk" levels (1 in 10,000 to 1 in 1 million); once more, the agency has discretion, within this range, to consider factors other than "individual risk." (127)

In exercising its discretionary authority to order a clean-up and to set remedial goals under CERCLA, EPA considers factors such as cost and feasibility, but not "population risk"--at least not in any formal way, since the aggregate deaths caused by existing contamination and avoided by different remedial interventions are not quantified. (128) Some commentators have suggested that "EPA is most likely to [order a remedy within the 1 in 1 million to 1 in 10,000 range] ... when population density suggests potentially high incidence of disease" (129) and that "EPA may be more inclined to select a more stringent remedy if a large number of people may be exposed to risks from the site." (130) But it is clear that "population risk" has at most a subordinate role in EPA's administration of the CERCLA program. Hamilton and Viscusi studied a sample of 150 sites where EPA had ordered remediation and found that, although 731 cancer cases would be averted, most of these were clustered at a few sites. (131) "[A]t the majority of sites the expected number of cancer cases averted is less than 0.1 cases per site based on conservative risk parameter estimates." (132)

d. Pesticides (FIFRA and FQPA)

EPA administers two significant statutes covering pesticides. The Federal Insecticide, Fungicide, and Rodenticide Act (FIFRA) (133) is the more general statute and requires all pesticides to be "registered" (licensed) by EPA. FIFRA gives lower priority to safety than many other EPA statutes. Its central criterion is, explicitly, a balancing criterion; pesticides and pesticide uses are to be permitted absent "unreasonable risks" to human life or the environment. (134) Nonetheless, EPA in administering FIFRA has given substantial weight to "individual risk," and has employed the same range of "individual risk" levels (1 in 1 million to 1 in 10,000) as are operative in its other regulatory programs. Specifically, in evaluating pesticide licenses and license conditions, EPA considers the risks posed by pesticides to food consumers, workers, and the general public (nonworkers exposed to pesticides through pathways other than food consumption, e.g., water contamination or contact with pesticides used in the home). And it typically seeks to reduce exposures to carcinogenic pesticides, in all three categories, below an "individual risk" level of 1 in 1 million, with some tolerance for higher levels. Exposure modeling assumptions blend midrange and high-range parameter values, so the "individual risk" here is that of an individual receiving an above-average but nonmaximal exposure. (135)

The Food, Drug, and Cosmetic Act has a special section covering pesticides, (136) and these provisions are administered by EPA rather than FDA. EPA sets "tolerances," that is, maximum permissible concentrations, for pesticide residues on raw or processed food. Foods exceeding the tolerances are, legally, "adulterated" and subject to seizure. (137) Prior to 1996, EPA was required to set a zero tolerance for carcinogenic pesticide residues in processed foods, under certain conditions (by virtue of the Delaney Clause); otherwise, the statute instructed EPA to employ a balancing test in setting tolerances. In practice, for carcinogenic pesticides not covered by the zero-risk standard, EPA followed its general approach under FIFRA, namely to aim at reducing the cancer risk to an above-average food consumer below 1 in 1 million. (138)

Congress overhauled this statutory regime for pesticide tolerances with the Food Quality Protection Act of 1996, (139) which repealed the applicability of the Delaney Clause to pesticides. Tolerances for carcinogenic pesticide residues on both raw and processed foods are now to be set so that there is a "reasonable certainty that no harm will result from aggregate exposure to the pesticide." (140) Despite the switch from balancing language to language giving greater weight to safety, the House Committee report states explicitly that EPA should implement the new statutory provisions through the 1 in 1 million "individual risk" test that the agency had used (outside the Delaney context) prior to 1996. (141)

e. Toxic Substances Control Act

The Toxic Substances Control Act (TSCA) (142) authorizes EPA to control, through rulemaking, any chemical that "present[s] an unreasonable risk of injury to health or the environment." (143) This sweeping authority is hedged by a provision that makes TSCA a back-up statute--toxics are to be regulated under TSCA only if other statutes are insufficient to meet the risk (144)--and in practice has been rarely used. (145) The office within EPA responsible for administering TSCA takes the position that carcinogens expected to cause fewer than one death per year, or less than a 1 in 1 million risk to a highly exposed individual, do not warrant regulatory intervention. Conversely, in its stated justifications for the rules that have been promulgated under TSCA, EPA has invoked both "individual risk" and "population risk." (146)

f. Title VI

A very different, but significant, context where "individual risk" will likely play a role in EPA practice concerns the racial impact of state and local environmental decisions. Title VI of the Civil Rights Act of 1964 authorizes federal agencies to issue regulations prohibiting recipients of federal funds, including state actors, from activities that have a disparate racial impact. (147) EPA has promulgated Title VI regulations, and recently published an important guidance document explaining how Title VI challenges to licensing decisions by state or local environmental agencies (e.g., a decision to allow a waste dump in a minority neighborhood) will be evaluated. (148) The guidance makes clear that a disparity between the minority population affected by the licensed facility and a comparison population, not merely in overall death rates, but in "individual risks" to representative or highly exposed individuals, could constitute an illegal "disparate impact" for Title VI purposes. (149)

2. Risk Assessment of Noncarcinogens

Traditionally, noncancer and cancer risk assessment have been performed somewhat differently. Toxic effects other than cancer have been seen to have a physiological threshold. The difference, crudely, stems from the special causal mechanism for cancer--DNA damage to some cells, followed by proliferation of those cells--such that a dose so small as to be genotoxic to but a single cell might, in unfortunate circumstances, lead to fatal cancer for the organism. (150)

Dose-response evaluation is therefore performed differently for noncancer toxicity than for cancers. (151) Experiments and epidemiological studies are still used to produce dose-response data points pairing doses of the toxin with incremental risks (frequencies) of death or some other adverse effect, relative to background. But instead of fitting a linear function to these data points, or some other function without a threshold, the analyst instead identifies the so-called NOAEL ("no observed adverse effect level"): the "highest tested dose at which no statistically significant elevation over background in the incidence of the adverse effect was observed." (152) So-called safety factors are then applied to the NOAEL dose to produce a conservative estimate of the physiologically safe level. Typically, this means dividing the NOAEL dose by a factor of 10, 100, or 1000. (153) The resultant dose, termed the "reference dose" (RID) (154) by EPA, is the physiologically safe dose, to a high degree of certainty: that dose, and lower doses, do not (it can be said with great confidence) produce an incremental risk of death.

This difference between the procedure for estimating noncancer and cancer risk leads to a difference in how "individual risks" for non-carcinogens are expressed. The "individual risk" incurred by a particular person, given her exposure to a noncarcinogen, is expressed as a ratio of the dose to the RfD--not as a probability number. For example, if the RID for the toxin is a lifetime dose of 100 grams, and the exposure assessment predicts that the maximally exposed individual will receive a lifetime dose of twenty-five grams, she will be ascribed an "individual risk" index of 1/4. If the exposure assessment instead predicts a maximal exposure of 200 grams, the "individual risk" index to the maximally exposed person is two. These nonprobabilistic indices of "individual risk" are less meaningful than the probabilistic indices employed for carcinogens. All that a nonprobabilistic index number less than one means is this: to a high degree of certainty, the incremental fatality risk incurred by that person is zero. All that a nonprobabilistic index number greater than one means is the negation, namely it cannot be stated with a high degree of confidence that the individual incurs a zero incremental fatality risk.

EPA generally conducts risk assessments for noncarcinogens, and looks to "individual risk" in making regulatory decisions, in the same statutory contexts as for carcinogens. (155) For example, under the Clean Water Act, EPA encourages states to set water quality standards for noncarcinogenic toxins so that the index of "individual risk" to the above-average individual will be less than one. (156) Under the Safe Drinking Water Act, EPA sets MCLGs and (typically) federal standards so that the index of "individual risk" (again to the above-average individual) is less than one. (157) Under CERCLA, the agency looks to the "individual risk" from noncarcinogens to a person receiving a "reasonable maximum exposure" from the waste site under review, once more seeking to keep that index number below unity. (158) Under FIFRA, the agency takes into consideration the "individual risk" that those exposed to noncarcinogenic pesticides incur. (159)

Despite this procedural parallel between EPA's use of "individual risk" numbers in regulating both carcinogens and noncarcinogens, there is an important, substantive difference. Because EPA's cutoff for noncarcinogens is an "individual risk" index number equaling one, while its cutoff for carcinogens is an "individual risk" probability number ranging from 1 in 10,000 to 1 in 1 million, "individual risk"-based assessment focusing on the maximally exposed individual tracks "population risk"-based assessment for noncarcinogenic toxins but not carcinogens. Imagine that a population of 100 million is exposed to a toxin, with five million receiving maximal exposures. If the toxin is carcinogenic and EPA uses the more conservative 1 in 1 million number in regulating the toxin, seeking to bring maximal exposures to that level, it can expect a "population risk" of at least five deaths caused by the toxin after regulation. By contrast, if the toxin is noncarcinogenic and EPA uses the index number of one as its cutoff, it can expect a "population risk" of zero deaths caused by the toxin after regulation.

In other words, in some contexts involving the regulation of noncarcinogens, the advocate of risk assessment practices sensitive to "population risk" rather than "individual risk" need not be troubled by EPA's focus on "individual risk." For in these contexts the two practices are convergent. But the point should not be pressed too far. Crucially, the convergence depends on whose "individual risk" is being evaluated. Regulatory techniques for noncarcinogens that focus on the "individual risk" borne by some individual other than the maximally exposed person--the average person, say, or the person in the 90th percentile of exposure--and that seek to ensure that this individual's index number is below one do not correspond to the regulatory goal of zero "population risk." If 100 million are exposed to the toxin, five million receive maximal exposures, and EPA ensures that less exposed individuals are at zero incremental risk of fatality, then (obviously) it remains possible that maximal exposures result in a nonzero "individual risk" and hence a nonzero number of aggregate deaths.

B. "Individual Risk" and Agency Practice: Other Agencies

1. The Food and Drug Administration

FDA regulates the safety of foods, drugs, medical devices, and biologics, and employs risk assessment in all these areas. (160) "Individual risk" has long played a key role in FDA decision making, particularly with respect to food safety. (161)

The Food, Drug, and Cosmetic Act (162) requires "food additives" and "color additives" to be licensed by FDA, under a statutory standard that gives high priority to safety. (163) For food additives and color additives that may be toxic but are not carcinogenic, FDA employs the standard NOAEL/safety factor method that (as we have seen) is used by EPA for noncarcinogens. Experiments are undertaken or epidemiological data is checked, and the dose of the noncarcinogen that produces zero incremental frequency of death in the group of subjects receiving that dose, relative to background, is determined. That dose is then divided by a "safety factor" (typically 100 at FDA) to derive the "safe" level of the noncarcinogenic toxin. This is, in effect, the dose that is highly likely to involve zero "individual risk." FDA then combines that number with information about food consumption patterns to determine the maximum permissible concentration of the noncarcinogenic food or color additive in the foods to which it is added. Specifically, FDA seeks to ensure that the 90th percentile food consumer will not ingest more than the "safe" dose of the noncarcinogenic additive. (164) Because FDA focuses on the 90th percentile of the consumption distribution, not on the maximally exposed individual, this method cannot be justified as assuring zero "population risk." At least in principle, depending on consumption patterns above the 90th percentile and the size of the population, a concentration of some additive that very likely creates zero "individual risk" for the 90th percentile consumer might also be likely to cause more than zero deaths.

FDA apparently employs an analogous method in setting permissible levels of pathogenic microorganisms, including bacteria, molds, yeasts, and viruses, in foods. The degree of microbial contamination that will not sicken a high-end consumer is determined, and then divided by a safety factor. (165)

Carcinogenic food and color additives are governed by the infamous Delaney Clause: a flat ban on any additive if it is "found to induce cancer when ingested by man or animal, or if it is found, after tests which are appropriate for the evaluation of the safety of ... additives, to induce cancer in man or animal." (166) This Clause bars FDA from licensing the use of an additive, at any level, if the additive itself has been found to be carcinogenic. But FDA has long interpreted the Clause, without judicial disagreement, to include an important exception: an additive which has not itself been found to be carcinogenic, but has some nonfunctional chemical constituent that is carcinogenic, is not governed by the Delaney Clause. Instead, FDA's view is that such carcinogen-containing additives are subject to the background "safety" requirement generally applicable to additives. FDA licenses these additives by setting a maximum permissible concentration of the additive sufficiently low to ensure that the 90th-percentile food consumer incurs an "individual risk" of cancer death no greater than 1 in 1 million. (168)

The Delaney Clause applies not merely to food and color additives, but also to animal drugs and feeds. (169) The concern, of course, is that human carcinogens, if fed to animals, might accumulate in meats and other foods derived from animals. But the Food, Drug and Cosmetic Act includes an explicit exception to the proscription on carcinogenic animal drugs and feeds:

   [The Delaney Clause] shall not apply with respect to the use of a
   substance as an ingredient of feed for animals which are raised for
   food production, if [FDA] finds ... that no residue of the additive
   will be found ... in any edible portion of such animal after
   slaughter or in any food yielded by or derived from this living
   animal.... (170)

FDA's reading of this statutory "no residue" exception to the Delaney Clause has been creative. A literal reading of the exception would make it a virtual nullity, given the miniscule concentrations that are detectable with modern techniques used to analyze food contaminants. Instead, FDA has (until recently) interpreted "no residue" as "safe"; safety, in turn, has been equated with the traditional 1 in 1 million level of "individual risk." Carcinogenic residues of animal drugs and feeds have been permitted at a concentration imposing a 1 in 1 million risk on the 90th-percentile consumer. (171)

This "individual risk"-based interpretation of the Delaney Clause with respect to animal drugs and feeds was revised by FDA in 2002, in response to a determination by the Department of Justice that the "no residue" provision could not be read to countenance carcinogens that actually produced detected residues. (172) FDA's new approach is therefore more roundabout, but still incorporates an "individual risk" test: the 1 in 1 million level is used to determine how sensitive analytical methods for detecting carcinogen residues must be, and the "no residue" requirement is satisfied if none is detected above the "limit of detection" of an approved analytical method. (173)

FDA has general jurisdiction over food safety, subsuming not merely food and color additives and animal drugs, but any toxin in food--for example, environmental contaminants such as PCBs and aflatoxins. The burden of action rests on FDA--foods are not licensed by FDA, but rather are subject to seizure, regulation, and penalties if dangerous--and the underlying statutory standard is more permissive than for additives. (174) FDA decision making here is sensitive to the costs of eliminating food toxins (for example the hedonic and nutritional costs of banning food products containing the toxins, as reflected in the market price of the foods), in contrast to additive regulation, where safety is the sole acknowledged regulatory consideration. (175) Concentrations of food and color additives posing a non-de minimis risk of death or injury are flatly proscribed by the agency; that is not necessarily true for other food toxins. Still, "individual risk" plays a role in FDA regulation of such toxins (carcinogens and noncarcinogens alike) by serving to define the de minimis level. Foods containing toxins below that level--a 1 in 1 million "individual risk" for carcinogens, a zero "individual risk" for noncarcinogens--are seen as safe by FDA and therefore permissible. In effect, this de minimis, "individual risk" level serves as a trigger for regulatory analysis rather than (as with food additives) for an automatic proscription. (176)

Finally, "individual risk" is the linchpin of FDA's so-called "threshold of regulation" policy, which has been in place for nearly a decade. (177) Chemicals that are contained in food packaging materials or other food-contact articles and that migrate into food are technically "food additives," subject to FDA licensure. But the licensing process is expensive; the "threshold of regulation" policy therefore provides that chemicals in food-contact materials not currently known or suspected to be carcinogenic, which leave residues in food at a concentration below 0.5 parts per billion (ppb), are exempt from licensure. FDA arrived at the 0.5 ppb level by looking at the universe of known toxins. Noncarcinogenic chemicals, it emerges, are very unlikely to produce toxicity at that level, while for carcinogens the 0.5 ppb level corresponds to the tried-and-true "individual risk" level of I in 1 million. As the agency explained:

   [We] used potency data on a large number of known carcinogens to
   estimate the likely risk that could be expected if an unstudied
   compound were later found to be a carcinogen....

      ... FDA further restricted its analysis to the 477 animal
   carcinogens that were the subject of oral feeding studies....

      Based on the range of potencies exhibited by these 477 animal
   carcinogens, FDA has determined that most known carcinogens pose
   less than one in a million lifetime risk if present in the daily
   diet at 0.5 ppb. (178)

2. The Occupational Safety and Health Administration

OSHA regulates toxic workplace chemicals as well as other occupational hazards. For toxins, the two crucial provisions are sections 6(b)(5) and 3(8) of the Occupational Safety and Health Act. (179) Section 6(b)(5) is specific to toxins and provides:

      The Secretary, in promulgating standards dealing with toxic
   materials or harmful physical agents ... shall set the standard
   which most adequately assures, to the extent feasible, on the
   basis of the best available evidence, that no employee will
   suffer material impairment of health or functional capacity even
   if such employee has regular exposure to the hazard ... for the
   period of his working life. (180)

This language, taken alone, might be read as authorizing OSHA to regulate any chemical that is a toxin--that is toxic at some dose--and to require reductions in workplace exposures to the toxin to the lowest feasible level. (181) But section 3(8) of the Act, a generic provision describing the "standards" that OSHA is empowered to issue, defines an "occupational safety and health standard" as "a standard which requires conditions ... reasonably necessary or appropriate to provide safe or healthful employment and places of employment." (182) In Indus. trial Union Department v. American Petroleum Institute, (183) a plurality of the Supreme Court read section 3(8) as creating a "significant risk" threshold for OSHA regulation of toxins as well as other safety hazards. In effect, the Court determined that OSHA was statutorily required to recognize a de minimis level of risk; workplace toxins creating risks below that level could not be regulated.

      By empowering the Secretary to promulgate standards that are
   "reasonably necessary or appropriate to provide safe or healthful
   employment and places of employment," the Act implies that,
   before promulgating any standard, the Secretary must make a finding
   that the workplaces in question are not safe. But "safe" is not the
   equivalent of "risk-free." There are many activities that we engage
   in every day--such as driving a car or even breathing city air--that
   entail some risk of accident or material health impairment;
   nevertheless, few people would consider these activities "unsafe."
   ...

      Therefore, before he can promulgate any permanent health or safety
   standard, the Secretary is required to make a threshold finding that
   a place of employment is unsafe--in the sense that significant risks
   are present and can be eliminated or lessened by a change in
   practices.

The Court then went on to suggest, famously, that the "significant risk" requirement might be understood in terms of "individual risk." A 1 in 1000 "individual risk" was clearly significant, the Court said; a 1 in 1 billion risk was not. (185)

Industrial Union's linkage between significant risk and "individual risk" might have been rejected as dictum. An agency more self-confident than OSHA might have read the case as mandating a de minimis threshold but permitting that to be specified in "population risk" terms--as some number of premature deaths that would be caused by the workplace toxin absent OSHA intervention. (186) Instead, OSHA's practice in regulating workplace carcinogens has been to follow the letter of Industrial Union: the agency determines whether the existing concentration of a workplace carcinogen is a "significant risk," warranting OSHA intervention, by determining whether a worker exposed to that concentration for his entire working lifetime (forty-five years of exposure, five days a week, eight hours a day) would incur an "individual risk" of premature death that exceeds, or at least is not too far below, 1 in 1000. (187) Interestingly, OSHA's approach is more eclectic once it has determined that the status quo level of a workplace carcinogen poses a "significant risk." Considerations of "population risk," "individual risk," and economic and technical "feasibility" all seem to bear on the agency's decision as to what the permissible level of the carcinogen should be. (188)

What about noncarcinogens? There is an inherent tension between the standard NOAEL/safety factor method for regulating noncarcinogens--a method that seeks to ensure an "individual risk" level of zero--and Industrial Union's statement that a workplace toxin might pose an "individual risk" above zero but still be too insignificant to trigger OSHA's regulatory authority. This tension came to light in 1992, when the Eleventh Circuit struck down a rulemaking in which OSHA had used the NOAEL/safety factor method to set permissible exposure limits for a variety of noncarcinogens. (189) Since this decision, OSHA has issued only one new exposure limit for a noncarcinogen. (190)

"Individual risk" has, historically, not played a role in OSHA's regulation of workplace safety hazards as opposed to health hazards, such as falls from heights or dangerous machines. To be sure, the "significant risk" threshold created by section 3(8) of the OSHA act applies to all OSHA regulations, whether targeted at illness or injury. But "significant risk" for safety hazards has in the past been understood in "population risk" terms. OSHA officials recently stated that: "Traditionally, OSHA has based its significant risk determination for safety standards on estimates of the annual numbers of injuries and/or fatalities associated with exposure to an occupational injury hazard and the number of injuries and fatalities likely to be prevented with a new standard in place." (191)

This may be changing. As already explained, the "individual risk" of workplace injury is a perfectly coherent concept. (192) Indeed, there is an emerging subliterature, within risk assessment, that seeks to define and measure the "individual risk" of occupational injury. (193) The National Institute for Occupational Safety and Health (NIOSH), the federal government's research institute on occupational health and safety, has encouraged this research. (194) And OSHA, in the huge ergonomics rulemaking a few years ago, relied in substantial part on the "individual risk" construct in arguing that ergonomics hazards were a "significant risk" and therefore fell within OSHA's regulatory jurisdiction. (195) OSHA defined the worker's risk of musculoskeletal disorder as "the probability that a worker will experience at least one work-related musculoskeletal disorder during his or her working lifetime (45 years)," (196) and used data on workplace injuries to determine this risk for different occupations. (197)

OSHA practices in regulating pathogens, covered by the agency's broad statutory authorization to issue workplace health and safety standards, should also be mentioned. In 1991, OSHA promulgated a bloodborne pathogen standard; in 1997, it proposed a standard for tuberculosis, subsequently withdrawn. (198) In both cases the agency quantified "individual risk" and sought to show that its traditional, 1 in 1000 threshold for regulatory significance was satisfied. (199)

3. The Nuclear Regulatory Commission

Reactor safety risk assessment has been central to NRC activities since the Three Mile Island accident. The Commission has employed risk assessment to evaluate the safety of individual reactors, the generic safety of different plant designs, the advisability of new regulations for existing plants, and in other contexts. (200) The Commission formalized its commitment to risk assessment in a 1995 policy statement. (201)

Reactor safety risk assessment need not be keyed to the probability of death and illness. Instead, a probabilistic assessment (be it of an individual plant, a class of plants, a regulation, an inspection protocol, or something else) might focus on the probability of an accident--either the probability of damage to the reactor core, or the probability of what NRC calls a "large early release" of radioactivity into the environment. For example, a risk assessment of an individual reactor might conclude that the annual probability of that reactor experiencing core damage is 2 in 1 million, without tracing or ascribing probabilities to the causal paths leading from core damage to the irradiation of persons near the plant and to their deaths. Or, the annual probabilities that different amounts of radioactivity could be released as a result of core damage plus containment failure might be quantified, without deriving "population risks" or "individual risks" from these release scenarios. (202) However, the Commission in 1986 adopted a series of "safety goals" as the ultimate benchmarks for reactor risk assessment. The goals are an interesting hybrid of "individual risk" and "population risk":

The Commission has established two qualitative safety goals which are supported by two quantitative objectives. These two supporting objectives are based on the principle that nuclear risks should not be a significant addition to other societal risks....

* The qualitative safety goals are as follows:

--Individual members of the public should be provided a level of protection from the consequences of nuclear power plant operation such that individuals bear no significant additional risk to life and health.

--Societal risks to life and health from nuclear power plant operation should be comparable to or less than the risks of generating electricity by viable competing technologies and should not be a significant addition to other societal risks.

* The following quantitative objectives are to be used in determining achievement of the above safety goals:

--[Individual risk:] The risk to an average individual in the vicinity of a nuclear power plant of prompt fatalities that might result from reactor accidents should not exceed one-tenth of one percent (0.1 percent) of the sum of prompt fatality risks resulting from other accidents to which members of the U.S. population are generally exposed.

--[Population risk:] The risk to the population in the area near a nuclear power plant of cancer fatalities that might result from nuclear power plant operation should not exceed one-tenth of one percent (0.1 percent) of the sum of cancer fatality risks resulting from all other causes.

The Commission's position has been that the safety goals are to be used in evaluating NRC regulations, not in measuring the safety of particular plants. Instead, risk assessments for particular plants are to focus on the probability of core damage and "large early release" of radiation; and the general probability benchmarks regarding core damage and "large early release" are, in turn, to be shaped by the safety goals. (204) But there have been recent suggestions by the Commission and staff that the safety goals might be used on a plant specific basis. (205) In any event, the goals make "individual risk" one of several foundational, regulatory criteria for nuclear plants in the United States. Plant designs and procedures are ultimately to be evaluated by considering, inter alia, the "individual risk" of immediate death following a reactor accident incurred by the average individual living "in the vicinity" of a plant. (206) The Commission has clarified that this means "the average individual biologically (in terms of age and other risk factors) and locationally who resides within a mile from the plant site boundary." (207) The quantitative "individual risk" goal articulated by NRC--0.1 percent of the average "individual risk" of prompt fatality from other causes--translates into an annual risk of 1 in 2 million. (208) In short, reactors are (inter alia) "safe" when average individuals living very near them incur no more than a 1 in 2 million annual chance of dying immediately as a result of a reactor accident. This has been NRC's stated safety policy since a few years after Three Mile Island.

Reactor safety risk assessments focus on the probability of a reactor accident. Considerations of "individual risk" also influence federal regulation of other aspects of the safety of nuclear power generation, for example by shaping NRC criteria for maximum permissible doses of radiation received by the population or plant workers as a result of ordinary plant operation, (209) or by structuring the regulation of nuclear wastes. (210) A salient recent example involves the proposed Yucca Mountain repository for high-level nuclear waste, which is to be built and operated by the Department of Energy, and licensed by NRC, pursuant to safety standards issued by EPA. (211) Congress has enacted a number of relevant statutes, including the Energy Policy Act of 1992, (212) which suggested, without mandating, that the design of the Yucca Mountain site be focused on "individual risk" rather than "population risk." The Act stated that EPA shall promulgate "standards for protection of the public from releases from radioactive materials stored or disposed of in the repository at the Yucca Mountain site," (213) but instructed EPA to consult first with the National Academy of Sciences on various questions, including: "[W]hether a health-based standard based upon doses to individual members of the public [i.e., an individual-risk based standard] ... will provide a reasonable standard for protection of the health and safety of the general public." (214) The National Academy of Sciences thereupon produced a report, which concluded that the safety of Yucca Mountain should indeed be judged in terms of "individual risk" rather than the total deaths caused by releases from facility--specifically, "individual risk" to maximally exposed persons--and suggested that the safe level of "individual risk" might be set at 1 in 2000 or lower. (215)

EPA responded by promulgating a standard for Yucca Mountain that is framed in terms of radiation dose, not risk: persons maximally exposed to releases from the waste site must not receive more than fifteen millirem of radiation per year. (216) But EPA justified this dose-based standard by adverting to the de minimis "individual risk" imposed on the maximally exposed individual by a radiation dose at or below the level of fifteen millirem. As the agency explained:

   [W]e have based the proposed dose-based standard upon the risk of
   developing a fatal cancer as a result of that level of exposure
   based upon a linear, non-threshold, dose-response relationship....
   Dose and [individual] risk are closely related; one can be converted
   to the other simply by using the appropriate factor. (217)

A fifteen millirem annual dose, assuming a seventy-year lifetime for the maximally exposed person and the dose-response curve that EPA invoked in the Yucca Mountain rulemaking, translates into an "individual risk" of 6 in 10,000. (218) Both NRC and the Department of Energy have amended their own Yucca Mountain regulations to conform to EPA's 15 millirem/year dose limit. (219)

In the summer of 2004, the D.C. Circuit vacated EPA's Yucca Mountain regulation insofar as it mandated compliance with the fifteen millirem limit and other requirements for only 10,000 years--a compliance period that the court found to be inconsistent with the Energy Policy Act of 1992. (220) How EPA will respond to this ruling remains to be seen.

4. The Consumer Product Safety Commission

The Consumer Product Safety Commission (CPSC) administers various statutes, including the Federal Hazardous Substances Act, (221) which requires that "hazardous substances" intended for household use bear a specified label. (222) A "hazardous substance" is defined as "[a]ny substance or mixture of substances which (i) is toxic ... if such substances or mixture of substances may cause substantial personal injury or substantial illness during or as a proximate result of any customary or reasonably foreseeable handling or use." (223) A "hazardous substance" that lacks the required label is "misbranded" and cannot be introduced into interstate commerce. (224)

In 1992, CPSC issued lengthy guidelines that explain when a substance is "toxic" and that employ an "individual risk" test in defining the subset of "toxic" substances that are "hazardous" for statutory purposes. (225) Specifically, a product containing a carcinogen is "hazardous" if a consumer using the product incurs an "individual risk" exceeding 1 in 1 million. (226) The portion of the guidelines covering exposure assessment stipulates that exposure means "anticipated exposure from normal lifetime use," and that "[i] n most cases the best estimate of exposure (average exposure) is acceptable." (227) In short, the relevant "individual risk" level is that of the average, rather than maximal consumer. The "hazardous" cutoff for noncarcinogenic toxins is defined using the standard NOAEL/safety factor method. It is that product concentration resulting in a lifetime exposure to the average consumer which equals the "no observed effect level," divided by a safety factor of 10 or 100. (228)

CPSC has rulemaking authority under the Act. The agency can supplement or vary the statutorily required labeling and, under certain conditions, ban a "hazardous substance." (229) "Population risk" considerations clearly play a role in these rulemakings. (230) For example, in justifying a rule that required retail containers of charcoal to bear a label warning of the carbon monoxide risk from burning charcoal in confined spaces, CPSC explained that "there are approximately 28 deaths and 300 CO-related ... injuries associated with the use of charcoal each year." (231) From these numbers CPSC inferred both an "individual risk" of dying from charcoal-related CO poisoning (232) and an aggregate monetized cost of death and injury. (233)

C. Beyond "Individual Risk": "Population Risk" in Agency Practice

The prior two Sections described a wide range of administrative practices whereby agency choice depends, wholly or partly, on the level of frequentist "individual risk" incurred by some person exposed to a hazard, identified by her place in the exposure distribution for example, the person with an average exposure, or the person with a high-end (90th percentile, say) exposure, or the maximally exposed person. I shall argue below that these practices are normatively misguided and should be changed. This normative analysis presupposes that the practices could be different--"ought implies can"--and the best way to demonstrate this is by showing that health and safety agencies sometimes employ metrics other than "individual risk" to a particular person in evaluating hazards. In particular, federal health and safety agencies sometimes evaluate hazards by quantifying "population risk": the total deaths, illnesses, or injuries caused by the hazards and abated by intervention. What follows is not comprehensive, or even a collection of the major examples, but is rather meant to underscore that the "individual risk"-based methodologies now dominant at EPA and also employed at FDA, OSHA, NRC, and CPSC are by no means inevitable.

Even where carcinogens and other toxins are concerned, regulatory analysis sometimes focuses (at least in part) on "population risk." Various examples were interspersed in my discussion above. Since the 1996 amendments to the Safe Drinking Water Act, EPA has relied both on "individual risk" to the above-average individual, and on cost-benefit analyses incorporating information about aggregate deaths, in setting enforceable drinking water standards. (234) In regulating food contaminants that impose an "individual risk" exceeding what FDA regards as the de minimis level, the agency has taken into consideration "population risk." (235) Similarly for OSHA: once that agency has determined that a carcinogen currently found in the workplace crosses the threshold of regulability because it imposes a sufficiently high (roughly 1 in 1000) "individual risk" on the maximally exposed worker, the agency sets the permissible level of the carcinogen by attending to "population risk" as well as "individual risk" and economic and technical feasibility. (236) CPSC has pointed to the aggregate deaths and injuries caused by toxins in justifying rulemaking under the Federal Hazardous Substances Act, (237) as has EPA in its rulemakings under the Toxic Substances Control Act. (238)

A yet more compelling example, perhaps: a recent international survey of carcinogen risk assessment practice found that "European [agencies] have established the estimation of the likely incidence of cancer in the human population as the goal for risk assessment." (239) This is a "population risk" measure--or at least one much closer to "population risk" than the sorts of "individual risk" measures generally employed at EPA. (240)

My survey of agency practice, in Sections A and B above, suggests that "individual risk"-based tests are widely used to evaluate health hazards, such as toxic chemicals, radiation, and pathogens, but are much less often used for safety hazards, where the threat is injury rather than illness. One counterexample is OSHA's recent reliance on "individual risk" levels to assess wo