At last, empirical elicitations of the magnitudes of those risks (and costs!) too small to matter and those too large to abide

Abstract

AbstractScholars, decision-makers, and pundits have long tossed around words and numbers meant to demarcate those health/safety risks that are too small to matter at all (called, variously, “trivial” or “de minimis” or “insignificant”), or risks that are so large as to be abhorrent (“unacceptable,” “intolerable,” and the like). But there has been nothing approaching a consensus about how large or small these important probabilities are, and worse, there have been no serious attempts to explore these concepts via any carefully-elicited preferences from laypeople themselves. Perhaps even more concerning, the dominant paradigm—cost-benefit analysis (CBA)—requires balancing of (monetized) risks against the costs of controlling them, but the parallel concepts of de minimis costs and intolerable costs have never been formally considered. As a result of these lapses, all attempts to consider equity or justice in the distribution of how risks and costs are borne are an absurdity, despite claims to the contrary.This article reports on a pair of elicitations involving 1350 laypeople. In one elicitation, a careful set of background and contextual explanations and a structured series of questions probed where each subject perceived lifetime excess mortality risk (from involuntary exposure) to become “so small you would not spend a moment thinking about it,” and where it becomes so large that “you would strongly object, as an ethical matter, to ever having to face it.” With a separate sample, we asked the parallel questions with regard to regulatory costs imposed by the government in the name of risk reduction.Although there was significant interindividual variation in responses, the central tendency of each of these four science-policy-values “goalposts” was as follows: (1) de minimis risk—a lifetime excess chance of premature death of 1 in 770,000; (2) intolerable risk—1 chance in 570; (3) de minimis cost—$25 per year; and (4) intolerable cost—$775 per year. These results offer, for the first time, a principled and evidence-based starting point for further refinement of which risks or costs to individuals can/should be disregarded in setting policy, and which must dominate policies regardless of how total “monetized lives saved” and total dollar costs compare to each other when viewed simplistically.We offer several prescriptions for how CBA might change, and how environmental/health/safety policy might improve, to take account of these fundamental thresholds in how risks and costs actually impose burdens on citizens. Both those interventions that impose intolerable costs while only reducing diffuse and trivially small risks, and those failures to intervene (sufficiently) that perpetuate intolerable risks for the sake of avoiding trivial and eminently manageable costs, are inefficient, unjust, and are thus unethical.Keywords: Risk assessmentstated preferencecost-benefit analysisde minimis riskenvironmental equity AcknowledgementsWe gratefully acknowledge the financial support provided by the U.S. National Science Foundation (Decision, Risk, and Management Sciences Program), under Grant # 1629287. We thank Marcus Mayorga for computer programming of the surveys, and a small group of colleagues and students at the University of Michigan School of Public Health who served as a focus group to help refine and shorten the risk experiment. We especially thank two anonymous reviewers for their helpful comments.Both psychometric surveys were approved as exempt by the Institutional Review Board (IRB00011706) of Decision Science Research Institute, Inc. The cost experiment was approved on Dec. 2, 2019, and the probability experiment on Feb. 26, 2020. The IRB concluded that the survey questions posed minimal risk to subjects, and that all information was obtained so that the identity of the human subjects cannot readily be ascertained, directly or through identifiers linked to the subjects.Author contributionsBoth authors jointly developed, tested, and refined the psychometric instruments, and analyzed the data collected. AMF wrote the first draft of this ms. and BBJ commented on it.Disclosure statementThe authors report there are no competing interests to declare.Data availability statementData and instruments for all our NSF project studies, including the one reported here, are archived at Johnson, Branden, and Finkel, Adam. Estimating the Net Benefits of Environmental, Public Health and Safety Regulations. Ann Arbor, MI: Interuniversity Consortium for Political and Social Research [distributor], 2022-07-30. https://doi.org/10.3886/E1756 61V1.Notes1 “De minimis” is Latin for “about trifles,” as in the hallmark aphorism that “the law is not concerned with trifles” (de minimis non curat lex). Therefore, we expect that the judicial system will refuse (in a principled way) to take up matters where the harm is so trivial as to be indistinguishable from no harm at all, or, at worst, from the kinds of harms humans silently bear incessantly.2 Eliciting citizen views directly is not the only way to learn about them. “Revealed preference” studies can sometimes infer these views by observing behavior, as when economists assert that the value of life can be inferred by estimating the extra wages paid to workers who “demand” them to compensate for specific excess risks of death. We encourage others to see if there exist any analogous ways to infer what risks citizens deem de minimis or intolerable by observing which risks they apparently pay zero attention to or seek to avoid at all costs. But this “stated-preference” experiment is how we’ve chosen to begin the effort.3 In effect, “intolerable cost” is often currently defined as “whenever a special interest successfully argues that he/she/they shouldn’t have to pay that much,” which is of course (worse than) no definition at all.4 In the specific but rare case of perfectly equal distribution, the “body count” and the individual risk are just scalar multiples of each other, akin to measuring weight in milligrams versus kilograms.5 This quote is attributed to pioneering asbestos researcher Dr. Irving Selikoff.6 In this article, we are concerned only about de minimis risks when the consequences are maximally severe (premature death). There is a separate literature about trivial consequences (e.g., what amount of cognitive decrement from lead exposure is too small to concern even those affected?).7 We note particularly three excellent books published between 1976 and 1987—Lowrance Citation1976, Fischhoff et al. Citation1981, and Whipple Citation1987—devoted in part to discussions of de minimis risk. However, none of the chapters in any of these books attempt to estimate RD using elicitations from laypeople, as we do here.8 Besides these single studies we cite, there are several excellent reviews of multiple studies here, notably Mumpower Citation1986, Wilson Citation1988, Hrudey and Krewski Citation1995, Hunter and Fewtrell Citation2001, Rae Citation2007, and Roy and Kshirsagar Citation2021.9 FDA has described certain very small risks as de minimis, which of course is consistent with using a higher threshold (1/million) as an upper limit for this concept. Blumenthal (Citation1990) notes that the estimated cancer risk for D&C Orange #17 was one in 19 billion, which FDA described as de minimis.10 Using January 2022 data from Oxford’s Global Change Data Lab (https://ourworldindata.org/grapher/united-states-rates-of-covid-19-deaths-by-vaccination-status), the risk of death for a boosted individual was about 0.3 per 100,000 per week. Using a 95% reduction to approximate the effectiveness of a well-fitted N95 respirator, this equals roughly a 7.5 × 10-6 per year fatality risk. It’s hard to extrapolate this mid-pandemic annual risk to a lifetime risk, but it is clearly far higher than 1/million/lifetime, so Allen is favoring a quite permissive definition of “not much else in life as low as that.”11 To be more precise, the language from the “Benzene” case referring to 10-9 used the example of “taking a drink of chlorinated water.” So it is possible that upon converting this one-time tiny probability to a lifetime one (say, by multiplying by 8 drinks/day and 25,500 days per 70 years), this de minimis pronouncement actually represented a lifetime risk of 2 × 10-4—although this would be suspiciously close to the other risk value (10-3) that Court used in the same sentence as unacceptably high. It’s probably more likely that the Court meant a 10-9 lifetime risk, in apposition to a 10-3 lifetime risk, and just explained the former example poorly.12 This background rate is roughly 2 × 10-4/yr in the U.S. and E.U., so 5% of that annual risk, converted to a 70-year lifetime, would imply a rather high estimate of RD, at 7 × 10-4.13 As Richard Wilson wrote in Citation1988 (supporting a value of 7 × 10-5 as de minimis because it is about 1/10th of the variability in the natural background exposure to ionizing radiation), “I know of no one in the world who alters, or even chooses, his place of residence or his lifestyle to keep his natural radiation dose level low.”14 A related approach was discussed by Hunter and Fewtrell (Citation2001), who suggested that society could look at all cases of a particular disease (they used gastroenteritis as an example), deeming as de minimis any exposure that contributed to less than some percentage (say, 5%) of all cases.15 Lundgren and Stefansson make two other unusual arguments that we disagree with. First, they say that because “one person’s sure death cannot be de minimis,” the entire concept is flawed. We respond that RD is a probability for all, not a certainty for any one. Second, they say that if the world population was 1020, a risk of 10-12 might not seem so negligible. We agree, but point out that the actual world population is 8 × 109, with obvious implications for RD.16 The claim that the probability of a given harm could even have its own “disutility function” is controversial. Some observers (Pratt and Zeckhauser Citation1987) claim that this is nonsensical—that only the outcome (in this article, death) can have disutility, but the probability of the outcome cannot. By this view, a 1-in-X chance of harm has exactly 1/Xth the disutility of the harm itself, for all X; different people can have different views about (different willingnesses to pay for) the harm, but not about the probability (see EPA Citation2011, p. 6: “it seems unreasonable… that an individual would value a second risk reduction of Δp much less than an initial reduction of Δp”). This view, of course, would not allow even for the very notion of de minimis risk (a very small probability of a given harm is just a very small fraction of whatever the disutility of the bad outcome is), or of intolerable risk (Shrader-Frechette Citation1985). It also reflects an expected-utility view of human perception, a view that has been waning for nearly a century (Allais Citation1953) and that continues to be supplanted by richer depictions such as regret theory (Bell Citation1982) and prospect theory (Thaler Citation1980; Tversky and Kahneman Citation1992; Booij and van de Kuilen Citation2009).17 One similar invocation of citizen preferences is much older than Byrd and Lave’s article. According to Hey et al. (Citation2010), the Comte de Buffon argued in 1777, as an attempt to use RD to “resolve” the “St Petersburg Paradox,” that a rational person would offer no more than $13 to play this infinite-expected-value game. Buffon argued that since a 56-year-old man “ignores the possibility of dying in the next 24 hours,” which had a probability of 1/10,189 according to mortality tables of the time, events with less than 1/10,000 probability could be ignored.18 The upper end of the Supreme Court’s pronouncement in the “Benzene” case is 10-3 per lifetime; although there is very little case law suggesting that OSHA has an absolute duty to regulate risks above 1/1000, by leaving it to OSHA to determine where below 1/1000 the boundary between “significant” and “insignificant” risk must fall, the Court may have implied that its upper benchmark for risk is prima facie “intolerable.”19 For example, the OSHA PEL for trichloroethylene (TCE) is 100 ppm. EPA’s inhalation unit risk factor for TCE is 2.3 × 10-5 per (part per billion).20 Since 1996, U.S. federal regulatory agencies must separately analyze the economic effects of their rules on small businesses, and three major agencies (EPA, OSHA, and the Consumer Financial Protection Bureau) must invite a panel of 15 small business representatives to jointly develop a report recommending special accommodations for this constituency. This requirement gives “intolerable cost” to one subgroup a favored (early) place in the regulatory development process, and there is no parallel consideration extended to those who bear intolerable risk (perhaps caused by the harm-promoting exemptions given to small business!); see Finkel Citation2011, also Bromberg and Finkel Citation2023, for a proposal in this regard.21 A reviewer helpfully pointed out that some subjects may not see a lawnmower as at all useful in their life, while others might regard a power mower as necessary and a push mower as an unacceptable though safer alternative. See Section 4.7 (“Limitations”).22 Another important implication of the way we designed this scenario is that all of the data herein should be interpreted as pertaining to a single narrow matter like the costs and benefits of one particular product, hazard, or exposure. We are not seeking estimates of “de minimis cumulative risk from all air pollutants considered together,” or “what aggregate risk from all EPA-regulated substances is intolerable?”, or “how much should a citizen never be expected to pay for all public policies combined?” Future research could certainly define the “four corners” in more expansive ways.23 We asked each subject to provide an alias last name for themselves, just to personalize the script (while keeping their real last name hidden). So if someone said their last name was “Johnson,” and indicated they were female, a photo of a doctor talking to a woman would appear on-screen, and the script would start with “I’m so very sorry, Mrs. Johnson, to have to tell you this awful news, but…”24 Specifically, we told subjects that “if you name a very large probability (that is, 1 in ‘extremely small number’), you are in effect saying that to gain the moderate benefits of this consumer product you are willing to consider almost any probability of harm as tolerable. If you name a very small probability (1 in ‘extremely large number’), you are in effect saying that almost any probability of harm is too intolerable to justify the benefits of this consumer product, and you are willing to give it up for yourself and others (e.g., a ban).”25 Similarly, a few subjects used “something positive that’s never happened to them or anyone they know,” like getting the grand prize in a lottery, to estimate D by analogy.26 We conducted the survey very early in the U.S. experience of Covid-19, and a few subjects cited the Covid death rate of 2.3% (although this was at the time the case fatality rate, not a risk) as helpful to their answer.27 Not every respondent imagined mortality in the numerator; one subject chose 1/35 as intolerable because of the probability of “being called on in class when you don’t know the answer.”28 As this article was in final preparation post-review, an interesting controversy began to influence the future of regulatory cost-benefit analysis. The White House Office of Information and Regulatory Affairs proposed sweeping revisions to Circular A-4, the guidance governing CBA review in the federal system, and OIRA’s new-found enthusiasm for agencies to apply “distributional weights” to costs affecting poor citizens versus rich ones dominated the comments received (Xie et al. Citation2023). Some commenters (Sullivan Citation2023) argued that these weights are inappropriate and policy-laden, despite OIRA’s clear (and we believe, correct) argument in the revisions that agencies have always been applying weights, merely implicit and equal ones. But others (see, e.g., Adler Citation2023; Finkel Citation2023) explicitly argued that such weights are the only ways to correct the long-standing error in CBA that fails to transform units of dollar cost into units of welfare decrement, as the welfare toll of a given dollar cost is clearly larger for a poor person than a rich one.29 Situs inversus is a benign genetic condition in which all the internal organs in the chest and abdomen develop in a mirror-image (left to right) position to the “normal” arrangement.Additional informationFundingU.S. National Science Foundation (Decision, Risk, and Management Sciences Program), under Grant # 1629287.