THE EQUALITY–EQUITY DILEMMA IN COST–BENEFIT ANALYSIS Comment on Daniel Farber’s Inequality and Regulation: Designing Rules to Address Race, Poverty, and Environmental Justice

“Equality” and “equity” are chameleonic terms: they take on different meanings and shades depending upon speaker and context. As Martha Minow has observed, the terms are often deployed as “weapons in polarized political arguments rather than analytical tools.”¹² To use these terms as analytical tools, we must first settle on provisional definitions of each.

“Equality,” in the sense that I invoke it here, refers specifically to VSL equality—the use of equal-dollar VSLs across individuals and groups. “Equity,” for its part, will serve as shorthand for the goal of making members of disadvantaged groups “better off” in the standard welfare sense: policy makers advance equity when they achieve outcomes that members of disadvantaged groups rationally prefer over the status quo. My argument here is that to determine whether policies make members of disadvantaged groups better off in a standard welfare sense, we will need to set strict adherence to VSL equality to the side.

To be sure, the particular conceptions of equality and equity that I invoke here are far from the only possible operationalizations of those terms. Elsewhere in this issue, David Weisbach notes that instead of VSL equality, we might conceive of equality in terms of respect. Equality of respect—or “respect egalitarianism”—“respects all individuals … equally by taking their views into account, including their risk versus return tradeoffs as measured empirically by their VSLs.”¹³ So conceived, equality of respect would require policy makers to honor the risk trade-offs that individuals rationally make in the real world. The use of equal-dollar VSLs would violate the equality-of-respect principle “because it implies, indeed explicitly states, that [lower income individuals’] well-informed views on matters about their own life are wrong.”¹⁴ Weisbach presents the equality-of-respect concept not to advocate for it but to illustrate that appeals to “equality” do not resolve the debate of VSL differentiation. Ultimately, all claims of equality must answer the question posed by Amartya Sen in the title of his seminal Tanner Lecture: “equality of what?”¹⁵ Rather than try to answer that question here, I will try to show that Farber’s own operationalization of equality—the use of equal-dollar VSLs—collides with other normative objectives that distributional cost–benefit analysis aims to advance.

“Equity,” for its part, encounters a similar Sen-like equity-of-what question. Following Farber, I will take as a given that agencies, when making regulatory decisions, ought to give additional weight, formally or informally, to the welfare of disadvantaged individuals and groups. Also like Farber, this comment will interpret disadvantage in both economic and racial/ethnic terms. The examples below leverage sources that report relevant data for African Americans and Latinos, but my focus on those groups is not meant to suggest that the aims of distributional cost–benefit analysis are limited to those groups. The same analysis would apply to any group with average incomes below the population average (e.g., Native Americans).

In line with the overall cost–benefit analysis enterprise, I will interpret welfare in terms of preferences: an individual’s welfare is higher under outcome A than under outcome B if the individual would prefer A to B (i.e., would give up something of nonzero value to switch from B to A). A preference-focused measure of welfare can be framed in monetary willingness-to-pay terms, but the use of a money metric does not mean that money is all that matters. As Cass Sunstein has observed, a willingness-to-pay measure of welfare “honors qualitatively diverse goods that people care about for diverse reasons.”¹⁶ Below, I will consider an alternative measure of welfare that gets at the same concept without using money as the metric.¹⁷ For now, the key point is that “equity” in the equality–equity dilemma refers to the goal of making members of disadvantaged groups better off in conventional welfare terms.

Again, this is not the only possible way to define “equity.” We could, among other alternatives, define “equity” in terms that refer only one side of the cost–benefit equation. For example, we might imagine someone saying that a motor vehicle safety regulation advances racial equity if it reduces African American vehicle fatalities—or if it reduces the relative or absolute disparity between African American and white vehicle fatality rates. However, cost–benefit analysis reflects the sensible premise that costs and benefits ought to be weighed against each other—not all regulations that provide benefits remain “worth it” once we account for costs. And if that is true at the population level, it would seem to be true at the subgroup level as well: If we genuinely care about the well-being of disadvantaged groups, we should evaluate policies based on both the costs and the benefits accruing to members of those disadvantaged groups. Otherwise, we might end up—in the ostensible name of equity—adopting policies that leave members of those groups worse off.

The conflict between equality and equity—“equality” again referring to VSL equality and “equity” referring to the goal of making members of disadvantaged groups better off in conventional welfare terms—arises in the context of regulatory decisions that require trade-offs between lives and dollars. This dollars-versus-lives question is ubiquitous in federal regulation: according to one estimate, seventy percent of the total benefits of major federal regulations are “directly attributable to the monetized value of reducing early mortality.”¹⁸ For air quality standards promulgated by the EPA, the proportion of benefits attributable to lives saved is even higher: the EPA estimates that eighty-five percent of the monetized benefits of Clean Air Act regulations from 1990 to 2020 stem from reductions in premature mortality related to one type of pollutant—particulate matter.¹⁹ In very broad brushstrokes, saving lives is what most high-cost federal regulations do.²⁰

When federal agencies are deciding whether to promulgate regulations that impose economic costs in pursuit of lifesaving benefits, they need a mechanism for comparing dollars to lives. Across the regulatory state, the “value of a statistical life” serves as that mechanism. The VSL is, formally, the amount of money that individuals are willing to exchange for a small change in mortality risk divided by the change in mortality risk. For example, if an individual is willing to pay $10 to avoid a 1-in-1-million risk of death, or willing to accept $10 for a 1-in-1-million increase in her risk of death, then her VSL is $10 million.²¹ Federal agencies derive their VSL figures primarily from wage-risk studies, which seek to estimate how much workers are compensated for taking higher-fatality-risk jobs. As of 2023, most federal agencies use a VSL of around $12 million when conducting cost–benefit analyses of lifesaving rules.²²

Standard microeconomic theory suggests—and empirical analysis confirms²³—that willingness to pay for mortality risk reduction increases with an individual’s income. The relationship between VSL and income arises not primarily because high-income individuals value their lives more but because they value their dollars less. Due to the diminishing marginal utility of income, an additional dollar of income for Jeff Bezos contributes much less to his well-being than an additional dollar of income contributes to your and my well-being. Thus, if Jeff Bezos derives the same utility from his life as you and I derive from our lives, but Jeff Bezos derives much less utility from an additional dollar than you and I derive from an additional dollar, then Jeff Bezos should demand more dollars in exchange for a given increase to his mortality risk than you and I would.

The relationship between income and VSL is reflected by a parameter known as the “income elasticity of the value of a statistical life.”²⁴ The income elasticity of the VSL is the percent change in the VSL for a percent change in income. An income elasticity of 1 implies that when income rises by one percent, VSL rises by one percent (and similarly, when income doubles, VSL doubles). While federal agencies do not use the income elasticity of the VSL to differentiate among deaths averted in any given year, they rely on the income elasticity of the VSL to update the overall VSL from year to year.

Most federal agencies have converged on an income elasticity of the VSL equal to 1.²⁵ An income elasticity of 1 corresponds to a condition known as “log utility,” in which the utility of income is the natural logarithm of total income. Because the first derivative of the natural logarithm of x is 1/x, log utility implies that the marginal utility of income is the inverse of total income. Thus, when income doubles, the marginal utility of income decreases by half. If someone with $50,000 of income values her life the same in utility terms as someone with $100,000 of income, but the person with $100,000 of income has half the marginal utility of income as the person with $50,000 of income, then the person with $100,000 of income should have twice the willingness to pay for mortality risk reductions (because she gets the same marginal utility from an additional $2 as the person with $50,000 of income gets from an additional $1).²⁶

The income elasticity of the VSL is a critically important variable for estimating the distribution of regulatory benefits. Consider the example of passenger vehicle safety standards. African Americans experience a disproportionate share of passenger vehicle occupant fatalities in the United States: the passenger vehicle occupant fatality rate for African Americans in 2019 was approximately 8.1 per 100,000, versus 6.3 per 100,000 for non-Hispanic whites (see Table 1). Using the Department of Transportation’s 2019 VSL of $10.9 million,²⁷ the monetized burden of passenger vehicle occupant fatalities for African Americans was approximately $881 per person, versus $691 per person for non-Hispanic whites. Thus, a regulation that reduced passenger vehicle occupant fatalities across the board would disproportionately benefit African Americans relative to whites. But this doesn’t account for the fact that mean household income for African Americans in 2019 was thirty-eight percent lower than mean household income for non-Hispanic whites. Adjusting VSL for income with an income elasticity of 1, the monetized burden of passenger vehicle occupant fatalities was higher for whites ($752 per person) than for African Americans ($598 per person).²⁸ Using income-elastic VSLs, a regulation that reduced passenger vehicle fatalities across the board would disproportionately benefit whites relative to African Americans.

To be clear, the only compelling reason to think that average VSLs vary across racial and ethnic groups is that average income varies across racial and ethnic groups. While some studies have considered whether VSL varies by race and ethnicity, the findings are inconsistent. For example, Kip Viscusi reported in a 2003 study that African American workers receive less additional compensation for performing high-fatality risk jobs than white workers, resulting in implied VSLs of $7 million to $9 million for African Americans versus $13 million to $15 million for whites.²⁹ However, Viscusi concluded that it would be “inappropriate to attribute the observed differences to a greater willingness by black workers to bear risk.”³⁰ Rather, Viscusi suggested that a “more likely” explanation lay in “structurally different labor market opportunities by race.”³¹ By contrast, from a 2010 study, James Hammitt and Kevin Haninger found that after controlling for income, African American survey respondents were willing to pay more for mortality risk reductions than non-Hispanic whites.³² However, Hammitt and Haninger’s experiment did not involve real money at stake, raising concerns about “hypothetical bias” (a phenomenon in which individuals report a higher willingness to pay in survey settings than in real-world settings).³³ All in all, the evidence is too scant and inconsistent to conclude that there are any racial and ethnic differences in willingness to pay for mortality risk reduction except for differences attributable to income.

So far, we have focused only on the benefits side of the ledger, but a full distributional analysis requires consideration of the cost distribution as well. Even if African Americans receive a disproportionately small share of a regulation’s total benefits, they still may be among the regulation’s largest net beneficiaries if they also bear a disproportionately small share of regulatory costs. In the vehicle safety case, competing intuitions suggest that the share of costs borne by African Americans may be disproportionately low or disproportionately high relative to other groups. The average number of vehicles is lower for African American–headed households (1.38 as of 2009) than for households headed by non-Hispanic whites (1.99),³⁴ so insofar as the costs of complying with safety regulations are passed through dollar-for-dollar to vehicle owners, African Americans are likely to bear less than their population-proportionate share of costs. On the other hand, higher-priced car models often incorporate new safety features before the National Highway Traffic Safety Administration (NHTSA) mandates fleetwide adoption. Therefore, lower-priced models generally are the ones most affected by new vehicle safety standards. Insofar as African Americans have lower average incomes and tend to buy lower-priced cars, they may bear more than their population-proportionate share of regulatory costs. And insofar as costs are passed through to shareholders and/or employees of automakers, African Americans may bear a disproportionately small share of costs (because they tend to hold less stock³⁵) or a disproportionately large share (because they account for eighteen percent of the car manufacturing workforce versus just thirteen percent of the workforce writ large).³⁶

Uncertainty about the distribution of costs remains even when (indeed, especially when) costs are borne in the first instance by the government. Imagine, for example, a safety-motivated roadway redesign project or an effort to install crash cushions on roadside objects that will reduce deaths in vehicle crashes while imposing a fiscal cost. If the expenditure is financed by higher income taxes on the rich, then most racial and ethnic minority groups will likely bear less than their population-proportionate share of the costs because most minority groups are underrepresented in the top income brackets. If the expenditure is financed by cuts to government programs that disproportionately benefit racial and ethnic minorities (e.g., Medicaid³⁷), then racial and ethnic minorities will likely bear more than their population-proportionate share of the costs.³⁸

Although ambiguity about the distribution of costs makes it difficult to say whether any particular vehicle safety regulation has a disparate impact on racial or ethnic minorities, the example of passenger vehicle safety serves to underscore two points. First, using a higher VSL for a particular group does not necessarily redound to the benefit of group members. For an area such as passenger vehicle safety, where the lives saved by more stringent standards are likely to come disproportionately from lower-income groups, using population-average VSLs rather than income-adjusted VSLs will predictably lead to higher benefit estimates and therefore stronger regulations. But stronger regulations do not necessarily benefit members of lower income groups.

This point is perhaps most easily illustrated through a simple numerical example. For expositional ease, I will use a hypothetical regulation with made-up numbers for fatality reductions and per-person costs but real-world numbers for VSL and income. For a real-world regulatory example in which assigning higher VSLs to lower income groups makes members of those groups worse off, interested readers can consult my analysis of NHTSA’s 2014 rear visibility rule.³⁹

Our hypothetical regulation will reduce passenger vehicle fatalities by one percent across the board. Costs of $7.25 per person will be distributed evenly across the population (for example, because higher-income households buy more cars but lower-income households are more likely to buy cars affected by the safety regulation). Using the Department of Transportation’s actual population average VSL for 2019 ($10.9 million),⁴⁰ the regulation would narrowly pass a cost–benefit test (benefit of $7.43 per person). Now imagine that we take into account the distribution of vehicle occupant deaths and household income across racial and ethnic groups and we use income-adjusted VSLs with an income elasticity of 1. In that case, the regulation would fail a cost–benefit test (benefit of $7.16 per person). Non-Hispanic whites would benefit, on average, from the regulation (benefit of $7.52 per person versus cost of $7.25). African Americans would be harmed, on average, from the regulation (benefit of $5.98 per person versus cost of $7.25). Here, a switch from income-adjusted VSLs to population-average VSLs would push the regulation over the breakeven threshold, and if the regulation were then adopted, non-Hispanic whites would benefit on net and African Americans would be harmed on net.

The example is, to be sure, highly stylized. One can imagine other cases in which a switch from income-adjusted VSLs to population-average VSLs might benefit African Americans.⁴¹ But as the example illustrates, Farber’s blanket statement that the use of population-average VSLs “amounts to a degree of redistribution in … favor”⁴² of lower income groups is not necessarily true. Indeed, we cannot even say that it is true in a more-likely-than-not sense.⁴³ Theoretically, the effect of higher VSLs on lower-income groups is ambiguous, and the set of empirical studies that systematically assess the distributional effects of higher VSLs for lower-income groups is empty.⁴⁴

Another way of framing the same point is this⁴⁵: Most high-dollar-cost federal regulations—air quality standards, fuel economy standards, and vehicle safety standards⁴⁶—entail a choice between more stringent and less stringent national rules. Agencies don’t, for example, have the statutory authority to adopt different air quality standards for predominantly African American and predominantly white neighborhoods: the Clean Air Act envisions that standards will be the same nationwide.⁴⁷ Likewise, federal motor vehicle safety standards don’t depend upon the race or ethnicity of a vehicle’s driver (and it’s unclear how, as a practical matter, they even could, unless the federal government were to ban interracial sales of used cars, which would surely be both unconstitutional and undesirable). In the context of national standards, higher VSLs lead to higher estimates of total benefits and thus weigh in favor of more stringent rules, whether those higher values are assigned to higher-income or lower-income groups. Thus, in the motor vehicle safety example in Table 1, where the ratio of total African American deaths to total white deaths is 0.263 to 1,⁴⁸ raising the VSL for African Americans by $1 has the same effect on estimated total benefits as raising the VSL for whites by 26.3 cents. In that context, it makes little more sense to say that “higher VSLs for African Americans redistribute in favor of African-Americans” than to say that “higher VSLs for whites redistribute in favor of African Americans.” Again, the practical effect of raising the African American VSL by $1 is the same as the practical effect of raising the white VSL by 26.3 cents. Either adjustment raises the total benefit estimate by the exact same amount.

Second, and of particular importance for the purposes of this comment, sensible distributional analysis requires income-adjusted VSLs. If the income elasticity of the VSL reflected irrational behavior by members of lower-income groups, perhaps paternalism would justify the use of equal-dollar VSLs. But there is nothing irrational about having a higher marginal utility of income when your income is low, and someone with a high marginal utility of income will generally be less willing to sacrifice income for risk reduction. Indeed, it would be quite puzzling if members of lower-income groups had the same willingness to pay for fatality risk reduction as higher income groups: it would call into question the bedrock assumption of declining marginal utility of income that underlies much of positive and normative economics. Moreover, as the vehicle-safety example illustrates, equal-dollar VSLs are not serviceable substitutes for income-adjusted VSLs in distributional analysis when average incomes differ significantly across groups. The use of equal-dollar VSLs rather than income-adjusted VSLs may mislead us into concluding that regulations benefit groups whom they harm.

As far as I know, no scholar seriously advocates the use of equal-dollar VSLs in distributional analysis (at least, not without other adjustments that offset the effects of VSL equality).⁴⁹ Farber himself says he is ready to “abandon the use of fixed-dollar values on life” in the context of “equity weighting,” a methodology in which analysts calculate costs and benefits separately for each distributionally relevant group and then apply a greater numerical weight to effects on worse-off groups when aggregating results.⁵⁰ As the discussion above makes clear, less comprehensive forms of distributional analysis, such as the Hippocratic criterion, likewise require the abandonment of equal-dollar VSLs. Without income-adjusted VSLs, the Hippocratic criterion would potentially allow some actions that leave lower-income groups worse off while exerting a veto on some actions that make lower-income groups better off. The actual effects on the welfare of disadvantaged groups would be scattershot.

If everyone understood what Farber understands—and what readers who have reached this point also understand—then the equality–equity dilemma might be no dilemma at all: regulators evaluating the distributional impacts of proposed regulations could use income-adjusted VSLs uncomplicatedly. But the limited available evidence suggests that the public reaction to income-adjusted VSLs may differ rather dramatically from Farber’s, even when income-adjusted VSLs are used to advance equity objectives. That evidence comes from two sources: first, the academic literature on income-adjusted VSLs, and second, the “senior death discount” debacle at the EPA in the early 2000s.

First, a close reading of the academic literature on equity and cost–benefit analysis reveals that even very sophisticated scholars lapse into using language that suggests that lower-income groups benefit when policy makers assign them VSLs that exceed their willingness to pay. For example, Thomas Kniesner and Kip Viscusi—two of the world’s leading experts on the VSL—wrote in a recent working paper that “application of an equitable VSL provides a bonus subsidy to Blacks relative to whites of about 20% for the median worker.”⁵¹ By “bonus subsidy,” the authors simply mean that the median African American worker makes less than the median white worker, and therefore—as compared to the alternative of income-adjusted VSLs—use of equal-dollar VSLs boosts the VSL for the median African American worker relative to the median white worker. But the fact that even preeminent scholars describe higher VSLs as a “bonus subsidy” for individuals and groups to whom those higher values are assigned goes to show just how hard it will be to shake the idea that higher VSLs necessarily benefit—and lower VSLs necessarily harm—economically disadvantaged groups.

To be clear, Kniesner and Viscusi—who comprehend the intricacies of VSL estimates better than possibly anyone else alive, and who have contributed enormously and generously to my own understanding of VSL-related issues—certainly know that using higher VSLs for lower-income groups does not always provide a “bonus subsidy” to those groups in a welfare sense. Indeed, Kniesner and Viscusi stated that “when policy costs are borne privately, requiring levels of safety that are reflective of a VSL that is greater than that of affected groups will reduce their expected welfare levels.”⁵² And Kniesner and Viscusi are not the only scholars who (a) understand that higher VSLs don’t necessarily benefit lower-income groups but (b) use language that suggests otherwise. Farber, for his part, does the same. For example, Farber writes that “[a]gencies … elevate the interests of the poor by deviating from theory and treating the value of life as constant regardless of wealth.”⁵³ Again, there is zero evidence to support that statement: no study shows that lower-income groups fare better when regulators apply equal-dollar VSLs rather than income-adjusted VSLs. By asking the general public to accept the abandonment of equal-dollar VSLs in distributional analysis, Farber is asking laypeople to understand that the use of income-adjusted VSLs does not operate to the detriment of lower-income groups in that context when Farber himself suggests (without evidence) that the use of equal-dollar VSLs operates to the benefit of lower-income groups in a related context.

To be sure, these two ideas—(a) that the use of equal-dollar VSLs doesn’t necessarily benefit lower-income groups in standard cost–benefit analysis and (b) that the use of equal-dollar VSLs wouldn’t necessarily benefit lower-income groups in distributional analysis—are distinct. I am critiquing Farber for disregarding (a), whereas Farber is asking laypeople to accept (b). It may be that (b) is much easier to comprehend, such that we can expect laypeople to understand (b) even when sophisticated scholars of regulation seem to ignore (a) in their academic writing. This would be asking a lot of the American people, but perhaps we can be Kennedy-esque about it and ask our fellow countrymen for a lot.

Unfortunately, the second source of evidence regarding the fallout from using different values for different groups—the “senior death discount” debacle at the EPA—does not lend itself to optimism about the public’s capacity to comprehend the nuances of the VSL. In the fall of 2002, EPA published two cost–benefit analyses that sought to monetize the benefits of reducing air pollution: a September 2002 “technical addendum” to the Bush administration’s “Clear Skies” legislative proposal⁵⁴ and a November 2002 regulatory-impact analysis of a proposed rule regulating engine emissions.⁵⁵ Both analyses included a “base estimate” using equal-dollar VSLs along with an “alternative estimate” that monetized mortality benefits using a “value of statistical life year” (VSLY) approach.⁵⁶ Although the EPA’s alternative estimate did not differentiate on the basis of income, the public reaction to the EPA’s alternative estimate serves as a cautionary tale regarding future efforts to use income-adjusted VSLs.

In both the “Clear Skies” and engine-emissions analyses, the EPA observed that the people whose deaths are accelerated by exposure to particulate matter (PM) often suffer from other ailments, such as chronic obstructive pulmonary disease (COPD) and cardiovascular disease. In its alternative estimates, the EPA assumed that when reductions in air pollution prevent premature deaths, the individuals whose deaths are averted will live an additional five years (or six months if they suffer from COPD). To estimate values of a statistical life year for those individuals, the EPA took two tacks. First, for individuals under the age of 65, the EPA used a VSLY of $163,000. It arrived at the $163,000 figure by assuming that the VSL for a 40-year-old worker is $3.7 million in 1999 dollars and that the $3.7 million figure corresponds to 35 remaining years of life at a 3% discount rate. Second, for individuals over age 65, the EPA used a VSLY of $258,000. It arrived at that figure by assuming that the VSL for a 70-year-old is $2.3 million in 1999 dollars (a figure based on a study estimating that the VSL for a 70-year-old is 37% lower than the VSL for a 40-year-old) and that the $2.3 million figure corresponds to 10 years of remaining life at a 3% discount rate. Projecting forward to 2020 and updating values to reflect expected income growth, the EPA estimated that the value of delaying the death of a non-COPD sufferer under the age of 65 would be $1 million, while the value of delaying the death of a non-COPD sufferer over 65 would be $1.6 million. “Since we assume that there is a 5-year loss in life years for PM related mortality, regardless of the age of person dying, this necessarily leads to a lower VSL for younger populations,” the EPA stated.⁵⁷

Notwithstanding the fact that the EPA’s alternative estimate implied that the value of averting a premature death for an individual over the age of 65 was 60% greater than the value for someone under 65, the Natural Resources Defense Council fired off an audacious press release charging that EPA’s alternative estimate “calculated the value of saving the life of a 65-year-old senior as being only worth 63 percent of a younger person’s life.”⁵⁸ The popular press quickly picked up this highly misleading claim. “How could the Environmental Protection Agency decide that the life of someone over 70 years old is worth $1.4 million less than the life of a younger person?” a New York Times news analysis asked.⁵⁹ EPA officials predictably ran into protests at hearings in California, Florida, Iowa, Pennsylvania, and Texas, where demonstrators sported stickers reading “Senior Discount: 37 Percent Off.” Responding to that backlash, EPA Administrator Christine Todd Whitman announced in May 2003 that the agency would drop the alternative estimate from future analyses.⁶⁰

What can the senior death discount debacle tell us about the use of income-adjusted VSLs in distributional analysis? In the senior death discount case, the EPA used lower VSLs for seniors at an intermediate step in a calculation that ultimately put greater weight on seniors than on younger individuals whose lives would be extended by reduced exposure to air pollutants. Activists seized on the lower VSL for seniors, ignoring subsequent steps of the analysis, and media accounts failed to add context or nuance. Analogously, Farber and others propose to use lower VSLs for African Americans and other disadvantaged minorities at an intermediate step in a distributional analysis that will ultimately give greater weight to those groups’ interests. The gamble that Farber and fellow advocates for distributional analysis are making is that the use of lower VSLs for seniors and the use of lower VSLs for disadvantaged minorities will elicit dramatically different reactions.

Although the two cases are distinguishable, the distinctions would seem to suggest that the use of lower VSLs for disadvantaged minorities in the distributional context would be more—not less—problematic from a public perception perspective. In the senior death discount case, the EPA had an easy response to the allegation that it was placing less weight on seniors’ lives than on younger adults’ lives: it wasn’t doing that. Premature deaths due to particulate matter exposure counted for 60% more in the agency’s analysis if the person in question was over age 65 rather than under. If reporters had bothered to read the relevant agency documents, they would have seen that the “senior death discount” accusation was a “four Pinocchios” claim.⁶¹ In the distributional context, by contrast, agencies really would be assigning lower average VSLs to African Americans than to non-Hispanic whites when analyzing subgroup-specific effects. Realizing that these lower VSLs would be in service of ensuring that regulations don’t impose disparate impacts on disadvantaged minorities would require a degree of sophistication far beyond what we saw in the senior death discount debate.

Moreover, the United States’ history of racial discrimination heightens the expressive stakes of race-differential VSLs. While the United States has a long history of age discrimination as well as racial discrimination,⁶² only one of these two forms of discrimination constitutes “America’s original sin.”⁶³ We did not maintain a system of age-based slavery for two-and-a-half centuries, we did not lynch people because of their age, and we did not force older people to ride at the back of the bus (unless they were both older and darker-skinned). Perhaps most significantly, our Constitution never counted elderly people as being worth three-fifths as much as younger adults for purposes of apportioning representatives and direct taxes. The United States’ ugly record on race makes the pursuit of racial justice through cost–benefit analysis all the more urgent but also all the more fraught. Consider that the use of income-adjusted VSLs with an income elasticity of 1 would result in an average VSL for African Americans that is 62% of the VSL for non-Hispanic whites. It does not require hypersensitivity to historical resonance for one to recoil at the idea of assigning a value to African American lives that is roughly three-fifths the value for whites.

Importantly, the concern about using different average VSLs for different racial and ethnic groups is not simply a public relations worry. As I have argued elsewhere in the context of income-adjusted VSLs, “it is a bad thing if millions of Americans think that the federal government values their interests less than the interests of other, richer Americans—and bad for reasons beyond the fact that agency officials may endure a few difficult news cycles.”⁶⁴ Replace “richer” with “whiter” and the claim becomes even stronger. It is bad because faith in government may decline—and individuals may be discouraged from participating in civic activities—if they think that their government literally values their lives less. And it is bad because disrespect—whether intended or not—imposes its own distinct harms on the individuals who experience it. Ben Eidelson wrote of the senior death discount episode, “[T]he suggestion that some people’s lives were less worth saving was understandably heard to say that the people themselves were worth less.”⁶⁵ And if it was understandable there (where it was straightforwardly refutable by facts), it is all the more understandable here (where refutation would require many additional analytical steps).

Summing up so far: the equality–equity dilemma in cost–benefit analysis appears to be a genuine dilemma. Maintaining equal-dollar VSLs in distributional analysis would produce results that do not necessarily benefit—and may in fact harm—members of disadvantaged minority groups, but the abandonment of equal-dollar VSLs raises serious concerns that go well beyond “bad PR.” The challenge for distributional analysis is how to account, in a responsible way, for the fact that willingness to pay for fatality risk reduction varies by income without inadvertently (though predictably) conveying a message to lower-income groups and disadvantaged minorities that their lives are valued less than the lives of their fellow citizens.

How should regulatory cost–benefit analysis respond to the equality–equity dilemma? This section sketches out four possible approaches: one that makes VSL equality the priority, a second that gives up on the goal of using cost–benefit analysis to advance equity, a third that seeks to smooth over the dilemma through arithmetic acrobatics, and a fourth that seeks to enlist trusted voices in validating distributional methodologies.

First, and least attractively, regulators could charge full steam ahead on distributional analysis while maintaining equal-dollar VSLs across income, racial, and ethnic groups. By now, the problem with this approach should be clear: regulators would be calculating costs and benefits by subgroup using VSLs that—in the case of African Americans—are roughly 1.5 times the average willingness to pay for fatality risk reduction that members of that group would rationally have, assuming logarithmic utility functions (1.3 times for Latinos). We would, in effect, be congratulating ourselves for advancing racial equity through regulations that in many cases may leave African Americans and Latinos worse off in the aggregate. This approach would not just be haphazard—it would amount to Enron-style accounting with respect to real historical debts toward racial and ethnic minorities. I see nothing to recommend this approach, and I know of no one who advocates it.

Second, upon confronting the equality–equity dilemma, policy makers could give up hope of incorporating racial justice into cost–benefit analysis—at least for rules that reduce fatality risk (which is to say, most high-dollar-cost federal rules). This would not be an entirely unreasonable reaction, even for policy makers who are deeply committed to racial justice. In many cases, data on the distribution of benefits and costs will be too sparse for agencies to say with any confidence whether a regulation benefits or burdens disadvantaged minorities on net. And requiring agencies to assess distributional effects of proposed rules would add an extra step in the regulatory process—a step that would potentially slow down the promulgation of lifesaving policies.⁶⁶ On top of that, the use of disparate-impact analysis in regulatory decision-making could expose regulations to legal challenges—for example, on grounds that the relevant statute does not authorize the agency to consider distributional consequences. As Farber notes, some broadly worded statutes seem to leave room for regulators to consider distributional effects along racial and socioeconomic dimensions,⁶⁷ but in other cases, the argument that the agency has authority to consider distributional effects under the relevant statute “may be too much of a stretch.”⁶⁸ Regulators could conclude—quite rationally—that the uncertain distributional payoffs of disparate-impact analysis do not justify slowing down the regulatory process and placing lifesaving rules in potential legal jeopardy, especially considering the additional risk of expressive harms if disparate-impact analysis abandons equal-dollar VSLs.

Moreover, even if agencies decline to consider racial equity when crafting lifesaving regulations, those same agencies still can seek to advance racial justice in a myriad of other ways. For example, the EPA could continue to make disadvantaged communities top priorities for project funding under its quarter-century-old Environmental Workforce Development and Job Training Program.⁶⁹ The Department of Transportation could use its discretionary authority to channel infrastructure investments—to the extent permitted by the relevant statutes—toward transit projects in underserved communities.⁷⁰ Agencies could continue to review and update their procurement policies to ensure that minority-owned businesses can compete successfully for federal contracts. In short, while lifesaving regulations represent the lion’s share of high-dollar-cost federal agency rules, rulemaking is itself just one function of the administrative state. Agencies that face up to the equality–equity dilemma in rulemaking may decide—quite understandably—to focus their equity-promotion efforts on non-rulemaking areas.

Yet I expect that many readers will not be ready to accept this resigned conclusion. The very phenomena that give rise to the equality–equity dilemma—the stark differences in income and risk exposure across racial and ethnic lines—cry out for a robust policy response that doesn’t leave any available tools on the table. In line with this view, President Biden committed his administration on day one to a “whole-of-government equity agenda,”⁷¹ and a whole-of-government effort would necessarily include regulation as a component. At the very least, we should consider more creative responses to the equality–equity dilemma before giving up on the goal of advancing racial equity in the risk-regulation domain.

Accordingly, a third approach to the equality–equity dilemma deserves attention and reflection. This third approach would seek to modify the arithmetic terms of analysis to make the use of differential VSLs less salient. The Indian economist E. Somanathan has suggested one method for accomplishing that goal. In Somanathan’s proposal, “[l]ives, rather than money, are used as the unit of account.”⁷² Everyone’s life could be valued at one life-unit, but the ratio between life-units and dollars could differ by income. For example, say that the median African American worker’s income is two-thirds the population median, the population-median VSL is $12 million, and the income elasticity of the VSL is 1. One dollar would be worth 1/12 million life-units for the median U.S. worker and 1/8 million life-units for the median African American worker. By valuing different people’s dollars differently in life-unit terms, Somanathan’s approach sidesteps the need to value different people’s lives differently in dollar terms.⁷³

The life-unit approach could be used as a complement to or a substitute for standard cost–benefit analysis with equal-dollar VSLs. If it were used as a complement, regulators would first compute population-wide costs and benefits as under the status quo and then analyze effects on each subgroup in life-unit terms. The results of the subgroup analyses would be directionally the same as if regulators had used income-adjusted VSLs: any policy that produces a net benefit for a subgroup in dollar terms also would produce a net benefit for that subgroup in life-unit terms. But regulators never would have to assign different dollar values to different people’s lives. Alternatively, the life-unit approach could replace standard cost–benefit analysis altogether: rather than evaluating policies based on whether they generate net benefits in dollar terms, regulators would evaluate policies based on whether they generate net life-unit gains. In this iteration, life-unit analysis would be a form of equity weighting (since pecuniary costs and benefits for lower-income individuals and groups would count for more in the total tally).⁷⁴

The life-unit approach has the virtue of being clever. But is it too clever by half? For one thing, expressing costs and benefits in unfamiliar life-unit terms would undermine an oft-cited advantage of cost–benefit analysis: transparency.⁷⁵ The current practice of cost–benefit analysis, which displays results in terms of dollars, arguably allows laypeople to better understand and critique regulatory choices. Replacing dollars with life-units would render the results of cost–benefit analysis significantly less legible to nonexperts.

For another, the life-unit approach would not entirely inoculate distributional analysis against the charge that regulators are assigning lower VSLs to disadvantaged minorities than to non-Hispanic whites. Whenever an agency rule relied on a distributional analysis that used different life-unit-to-dollar ratios for different groups, opponents of the rule still would have an incentive to point out that, once one cuts through the fractions, the life-unit approach has the effect of assigning lower-dollar VSLs to members of lower-income groups. And they would not be wrong. By avoiding any explicit statement that the VSL is lower for lower-income groups, agencies might leave themselves less vulnerable to “death discount” charges, but they would not be entirely invulnerable.

For that reason, the life-unit approach—if pursued—should be pursued in tandem with a fourth approach to the equality–equity dilemma: enlisting trusted voices to validate distributional methodologies. Before an agency uses income-adjusted VSLs in distributional analysis—either explicitly or through the conceit of life-units—it should seek the support in advance of leaders who enjoy credibility among members of the groups to whom lower VSLs will be assigned. When opponents of an agency rule opportunistically and inevitably level the “death discount” charge, those leaders should be prepared to counter the opponents’ narrative. They should be prepared to explain—in language far more straightforward than this comment’s academic prose—that the use of income-adjusted VSLs doesn’t devalue members of disadvantaged groups but rather advances those groups’ interests.

Who exactly those trusted leaders might be will vary based on the relevant regulations and the affected groups. Elected officials outside the executive branch—for example, senators and congressmembers from disadvantaged minority groups—seem like natural candidates, but the pantheon might also include clergy members, labor union leaders, heads of well-respected civil rights organizations, and prominent advocates within the environmental justice movement. To obtain support, agencies will likely need to reach out early on and integrate potential validators into the regulatory decision-making process. And, of course, racial and ethnic groups are not political monoliths: not all members of a given group will place their trust in any given validator. However, “death discount” charges are much less likely to scar if they encounter immediate and informed resistance from recognized and respected voices.

The analysis here is admittedly speculative. We have no data that demonstrates how income-adjusted VSLs will be received by the public when those adjustments are submerged in opaque arithmetic and/or defended by trusted voices—and the reason we have no data is that no one has tried these approaches. But it is even more speculative, to the point of being wishful thinking, to believe that public acceptance of differential VSLs will come easily. If there is a lesson to be learned from the senior death discount experience, it is that the use of different VSLs for different groups presents challenges that will require careful planning, creativity, and coalition building.

The equality–equity dilemma in cost–benefit analysis warrants attention in its own right because it presents a first-order challenge to efforts to incorporate racial and economic justice into regulatory policy making. But it also merits focus because of what it can potentially teach us about the broader enterprise of cost–benefit analysis, which now suffuses the regulatory state and supplies the justificatory language for life-and-death decisions.

First, the equality–equity dilemma highlights the reality that regulatory cost–benefit analysis operates, if not always in the public eye, at least in the public’s peripheral vision. Occasionally, agencies’ methodological choices leap from the fine print of technical documents into the public’s consciousness. A recent example involved an EPA proposal in late 2022 to impose new and more stringent standards on methane emissions from crude oil and natural gas facilities.⁷⁶ Buried on page 129 of a technical supplement to the proposal was the agency’s explanation of its module for monetizing the mortality benefits of a reduction in greenhouse gases.⁷⁷ According to the agency, “[v]aluation of mortality risk changes outside the U.S. is based on an extrapolation of the EPA value that equalizes willingness-to-pay as a percentage of per capita income across all countries (i.e., using an assumed income elasticity of 1).”⁷⁸

Commentators quickly picked up on what those words meant: for purposes of the methane rule, EPA was valuing foreigners’ lives in proportion to foreigners’ incomes. Thus, as Vox’s Dylan Matthews adroitly observed,

Importantly, Matthews was careful to stress that these controversial judgments are “implicit” in EPA’s analysis, which never mentioned Qataris, Burundians, or any other specific foreign nationality. And he was right that the EPA’s analysis implied vastly different weights on the lives of individuals in different countries, even though the EPA never explicitly said as much. The story soon hit the NPR airwaves,⁸⁰ though it has not become the debacle for the agency that the senior death discount was. Perhaps that is because differential VSLs for foreigners don’t elicit the same negative reaction as differential VSLs for Americans; perhaps it is because the Biden EPA enjoys more goodwill than the Bush administration with the environmental groups that seeded the “senior death discount” narrative twenty years ago.⁸¹ In any event, the episode is another reminder that what happens in cost–benefit analysis doesn’t always stay in cost–benefit analysis: agencies should be prepared for the technical details of their regulatory assessments to reach nontechnical audiences, sometimes in a transmogrified form.

Second, the equality–equity dilemma underscores the need for agencies to consider—as part of a comprehensive assessment of regulatory costs and benefits—the potential discursive harms that flow from their own articulation of their methodological choices. This, again, is more than a matter of political expedience or bureaucratic self-preservation: accounting for all the costs and benefits of a regulatory proposal requires an agency to consider the consequences of potentially sending a message to some individuals that their lives are worth less. And this is doubly true for distributional analysis, where the discursive harms of differential VSLs—if not carefully managed—will fall disproportionately on members of already disadvantaged groups.

Finally, reflecting on the equality–equity dilemma offers us a richer perspective on cost–benefit analysis’s institutional dimensions. Over the last quarter century, scholars have come to understand cost–benefit analysis as a mechanism that allows the president and his advisers to monitor agencies across the executive branch. Regulation involves trade-offs, and the discipline of cost–benefit analysis makes those trade-offs transparent to White House officials.⁸² As Jonathan Masur and Eric Posner put it, “quantification forces regulators to put their decisionmaking into a format that can be evaluated by generalist superiors.”⁸³ Scholars have argued that cost–benefit analysis can and should play a parallel function vis-à-vis Congress and the courts, enabling lawmakers and judges to comprehend and critique choices made by expert agencies.⁸⁴

President Biden’s day-one memorandum can be viewed through this institutional lens: distributional analysis is being used as a technique of presidential administration that seeks to align agency action with the White House’s high-level racial justice objectives. And more generally, the institutionalist perspective has contributed hugely to our understanding of cost–benefit analysis across the regulatory state, highlighting the fact that cost–benefit analysis functions as a management tool as much as a normative decision-making framework. But viewing cost–benefit analysis only in intra-executive or interbranch terms would leave out a critical element of the context in which cost–benefit analysis operates. Cost–benefit analysis renders regulatory decision-making legible not only to actors inside the government but also to outsiders. The equality–equity dilemma arises precisely because the language that regulators use to communicate their decisions to other actors inside government also conveys messages—not always intended—to the broader public.

The equality–equity dilemma thus allows us to see, in stark terms, the opportunities and constraints generated by cost–benefit analysis’s multiple audiences. That cost–benefit analysis must address multiple audiences at once is both a virtue and a vulnerability. It is a virtue insofar as transparency is a virtue: cost–benefit analysis enables citizens to see, in clear terms, how regulators are navigating life-and-death choices that affect us all. It is a vulnerability insofar as it gives rise to the risk that—as in the “senior death discount” case—quantitative inputs may be construed or misconstrued as devaluing certain individuals and groups. In the end, the fact that agencies must simulcast to these multiple audiences may be a net asset or a net liability for the “cost–benefit state.”⁸⁵ Either way, it is a stubborn reality with which regulators—if they are to use cost–benefit analysis of lifesaving rules as a tool to advance racial justice—must ultimately reckon.