#CatoCOVID

In response to the COVID-19 pandemic, governments have imposed social distancing policies such as school closures, bans on public gatherings, mask requirements, and lockdowns of “nonessential” businesses. These policies plausibly save lives, but they also impose economic and noneconomic costs. Society therefore faces a tradeoff between greater health versus higher gross domestic product (GDP) and greater freedom in choosing how much, and what kind of, social distancing policies to impose.

Cost-benefit analysis provides a useful framework for thinking about these tradeoffs but does not always provide definitive answers because of uncertainty about key magnitudes and the difficulty of assigning dollar values to some effects of policies. However, cost-benefit analysis allows integration of both economic and noneconomic effects in a single analysis, and it often provides suggestive bounds on what kinds of policies might be reasonable.

A key challenge for a cost-benefit analysis of COVID-19 policies is that the main benefit of such policies is most easily measured in the number and length of lives saved rather than in dollars. Thus, to compare benefits with costs, any analysis must assign a dollar value to life.

Putting a dollar figure on the value of saved lives generates controversy. Major U.S. automobile companies undertook detailed economic analyses of tradeoffs between the costs and risks associated with automobile design. This came to light during the civil trial regarding the design of the Ford Pinto gas tank that was involved in a fatal accident in 1972. Frank assessments of the risks and costs of different design possibilities led companies that conducted such analyses to be vilified in the press and penalized by juries for undertaking such safety analyses. The Occupational Safety and Health Administration (OSHA) in the 1980s also claimed that life was too sacred to value.

Despite assertions that “one cannot put a price tag on life,” many decisions by individuals implicitly accept a tradeoff between longer life and other goals. People routinely drive on the highway, ride in airplanes, climb mountains, or work in coal mines, each of which has a non-zero fatality risk. They do so because they consider the benefits of these activities to be greater than the costs imposed by the fatality risks.

Policy decisions involve similar tradeoffs. The Food and Drug Administration could insist that trials of new medications last longer and enroll more subjects, thereby reducing the risk of approving a drug with adverse side effects. But that would increase costs and delay the use of new drugs. Safety standards for seat belts or air bags could be stricter, but the resulting automobile prices would be higher. Lower speed limits would reduce the number of fatal accidents but also result in longer travel times.

Thus, public and private decisions accept a tradeoff between lives and higher incomes, increased leisure time, and liberty more broadly. The question is how to choose a specific dollar figure for the value of life that, combined with estimates of the costs of particular polices, can help determine which pandemic policies are valuable on net.

To determine how much people value reductions in the risk of death, economists use decisions observed in labor markets in which individuals accept additional mortality risk in return for higher compensation. Coal miners and construction workers, for example, are paid more than others with similar skills because they face a higher statistical risk of death on the job.

When the value of a statistical life (VSL) is calculated from labor market data, the wage in an industry is the dependent variable in a statistical analysis. The explanatory variables include measures of skill, geographic location, and other causes of wage differences. The important variable for VSL calculation purposes is the fatality rate in the occupation. The coefficient on the fatality rate is an estimate of extra compensation required for the average employee for the annual increased fatality risk, controlling for the other causes of wage differentials. If the extra compensation is $1,000 and the annual fatality risk is 1 in 10,000, then among 10,000 workers there will be one statistical life lost, and the annual extra total compensation required will be $10 million (10,000 workers x $1,000 extra for each worker). A recent estimate of VSL is $9 million (2016 dollars), or about $10 million in 2020, which implies that policies that reduce mortality risk are acceptable if the cost per statistical life saved is $10 million or less.

Regulatory economists use VSL to evaluate health and safety policies. Some interventions, such as rules mandating that lighters be childproof, only cost $100,000 per statistical life saved and are thus highly cost-effective, while OSHA regulations regarding formaldehyde exposure in the workplace cost $78 billion per statistical life saved, more than 1,000 times greater than any VSL estimate.

Although the VSL methodology relies on observed decisions in a real-world setting, and even though its numerical estimates are widely used in practice, should modifications occur for its use in a pandemic?

One argument that they should is that the COVID-19 risk is greater than the labor market risk on which standard VSL estimates are based. The VSL, as calculated from labor market studies, is the average payment required to induce acceptance of the annual increased fatality risk observed in the labor market, which ranges from 0.06/10,000 for administrative support occupations to 2.35/10,000 for transportation and material moving occupations to 2.46/10,000 for mining.

Notice that the upper end of that range is still only an annual fatality risk of 0.02 percent. Given that the infection fatality rate for the coronavirus is estimated to be more than 10 times (0.6–1 percent) the upper end of the labor market fatality rate, would the payment required to induce acceptance of such a risk be higher, lower, or the same as the current VSL of $10 million?

Extrapolations of VSL to larger risks conclude that the incremental payment required decreases as incremental risk increases, according to a May 2020 paper by James Hammitt, a prominent developer of VSL. An individual is willing to “pay” $990 (rounded to $1,000 in our earlier introductory example) by accepting lower wages to avoid a current-year mortality risk of 1 in 10,000 but $9,210 to avoid a current-year mortality risk of 1 in 1,000 (instead of the linear extrapolation of $990 multiplied by 10, which would be $9,900). Ten times the risk avoidance requires slightly less than a 10-fold increase in payment. To avoid an annual mortality risk of 1 in 100, individuals pay $48,600 (instead of the linear extrapolation of $990 multiplied by 100, which would be $99,000). Individuals would pay less than 50 times the amount that they would pay to avoid a 100-fold higher risk.

Given that the coronavirus infection fatality rate is estimated to be less than 1 percent, or 100 times the highest fatality rate in labor market data, the simple linear extrapolation of the normal VSL is too large. A more appropriate VSL would be around $5 million rather than $10 million, based on the logic above (the $48,600 payment required for inducing consent for an annual fatality risk of 1 percent multiplied by the 100 workers that would receive it). COVID-19 risk is also greater for the elderly. In 44,000 confirmed cases of COVID-19 in China, the case fatality rate was highest among older persons: 14.8 percent for people age 80 or older; 8 percent for ages 70–79; 3.6 percent for 60–69; 1.3 percent for 50–59; 0.4 percent for 40–49; and 0.2 percent for people under age 40. In early U.S. epidemiologic data, case fatality rates were similar.

Estimates of how VSL changes with age are limited to those who work, and most elderly do not work. The estimates of the ratio of VSL at age 40 to age 65 range from 1 to 2, meaning younger workers demand up to twice as much money to accept a given risk. Simulation models provide estimates of the VSL 40:70 ratio of anywhere from 2 to over 10. That is, the VSL at age 70 is only one-half to one-tenth of the standard VSL. Surveys that ask people of varying ages how much they would demand to be compensated for higher risks find inconsistent results. About half of the surveys estimate that VSL decreases with age, and the other half found no change or a larger VSL for the elderly. Hammitt concluded that “while it is plausible that VSL is smaller at older than at middle ages, neither standard models nor empirical evidence provide convincing evidence about the magnitude of the effect.”

Moreover, VSL increases with income. Thus, because of the lower income during the pandemic due to voluntary and government-imposed business declines and closures, “the appropriate VSL for evaluating responses to COVID-19 and other current hazards may be smaller than the value that was appropriate before the pandemic.” But the correct VSL is lower only if long-run income is expected to be lower. Transient reductions in income have little effect on the appropriate VSL.

Finally, in traditional analyses of health and safety regulation using VSL, such as the previously mentioned lighter and formaldehyde rules, the cost estimates include only compliance costs: the extra material and labor costs involved in complying with the rule. The cost estimates do not include any other reasons that consumers might oppose the regulation, or any broader costs more generally (such as long-run, unintended effects of new government policies).

In most traditional health and safety regulatory contexts, that omission is probably not empirically important: the airbag container makes the steering wheel less attractive for some, perhaps, but the effect is tiny. In other traditional regulatory contexts, the omission is more important but ignored: childproof caps on over-the-counter and prescription medicines are annoying for many people and especially difficult for the elderly, but most compliance cost analyses ignore such effects.

In the forced reduction of social interaction during the COVID-19 pandemic, many costs, beyond lost income, are likely important: the cost of missing your last chance to see Grandad before he dies alone, the costs to mental and physical health of isolation, the cost of lost time for single people to meet a partner in the right window of life, the cost of missing travel opportunities, or the cost of acquiring new skills or talents that require leaving the home. In economic terms, these might be dubbed the broader costs of restrictions on our nonmarket liberties, or the decline in productivity of our leisure time. If they were monetized and added to any analysis assessing lockdowns, then the cost-effectiveness of restrictions would be much reduced.

Our key point is this: a potentially important component of the cost estimate of lockdowns should consist of the monetary value of the loss of freedom to move and interact as we normally do in addition to the actual lost income from reduced market activity. Thus, the appropriate (cost-effective) policy-induced reduction in economic activity would likely be significantly smaller than currently estimated as cost-effective by many economists.

To recap, the COVID-19 fatality rate is estimated to be 100 times the highest fatality rate in labor market data, thus the normal VSL ($10 million) is too large. A more appropriate VSL is around $5 million. Evidence for the necessity of modification of VSL for age and income is varied and inconclusive. We believe that loss of liberty matters but have no actual dollar estimate to add to the measured losses of income from pandemic policy restrictions.

Now that we have a price (VSL) to value the benefits of policies that reduce social interaction (i.e., saved lives), we need estimates of the quantity (i.e., lives lost in the absence of such policies). University of Chicago economists Michael Greenstone and Vishan Nigam estimate that, relative to no private or government response, 1.76 million lives could be saved through October 1 by the implementation of various isolation measures that are less severe than those that were used for much of the country. (The suite of less restrictive measures that they examine includes 7‑day isolation for anyone showing coronavirus symptoms, a 14-day voluntary quarantine for their entire household, and dramatically reduced social contact for those over 70 years of age but no mandatory business closures or stay-at-home orders.)

MIT economist Robert Pindyck estimates that moderate social distancing policies (similar to those considered by Greenstone and Nigam) would save about 1 million lives and that strict social distancing would save 3 million if the fatality rate were 1 percent.

Using age-varying estimates of a statistical life (VSL for those under age 9 is $14.7 million, while VSL for those over age 79 is $1.5 million), Greenstone and Nigam estimate the benefits of moderate social distancing policies at slightly less than $8 trillion. The constant $5 million VSL that follows from our analysis implies benefits of $8.8 trillion ($5 million VSL x 1.76 million lives saved). Pindyck’s estimates of 1 million and 3 million lives saved imply benefits of $5 trillion and $15 trillion, respectively, using our constant $5 million VSL estimate.

In 2019, U.S. GDP was $21.4 trillion. The benefits from moderate social distancing ($5 trillion to $8.8 trillion) are thus estimated to range from 23 to 41 percent of annual national income. The estimated benefits from Pindyck’s strict social distancing scenario ($15 trillion) are 70 percent of GDP. For comparison, from 1929 through 1933 during the Great Depression, real U.S. GDP fell 29 percent over the four years.

The conclusion that some draw from this analysis is that a policy-induced reduction of national income equal to or much greater than the Great Depression would be cost-effective if the estimate of lives saved (between 1 million and 3 million) relative to nonintervention were correct.

This conclusion, however, has two major deficiencies. If the nonmarket costs of lockdowns are significant, then the market loss we should be willing to accept from lockdowns is correspondingly smaller. Valuing the nonmarket costs is difficult, but introspection suggests that these might be substantial: people routinely expend substantial time and effort to protect their leisure time or other nonmarket freedoms such as choice of religion, spouse, occupation, or country.

A second deficiency is that prominent economists have been skeptical of existing cost-benefit analyses of lockdowns. Hammitt said, “While it is conceivable a household with wealth of $60,000 [median wealth is about $95,000 and only $30,000 excluding home equity; median income is about $60,000] might be able to spend $9,000 to reduce current mortality risk by 1/1,000, spending $48,000 to eliminate the 1/100 risk seems implausible.” Even accounting for “human wealth” (the present value of future income), this still implies a large payment relative to resources.

Pindyck notes that in 2018, the total net wealth of U.S. households was $98 trillion, about $300,000 per person, quite a bit less than VSL. He suggests, “We could also look at what societies actually spend to save large numbers of lives. For example, the U.K. National Health Service (NHS) limits what they will pay for a given treatment by using a ‘Quality Adjusted Value of a Statistical Life Year’ of about $38,000, which translates to a VSL of around $1 million.” Given Pindyck’s estimate of 3 million lives saved from strict social distancing policies, his VSL would result in $3 trillion in benefits, or only 14 percent of GDP.

Another caveat to existing analyses is that they attribute both saved lives and economic disruption to policies. But voluntary social distancing seems to have caused much of the reduction in virus transmission and much of the economic loss. Restaurants and airlines lost customers well before lockdowns, and professional sports leagues canceled their schedules even before bans on large events. The crucial question for policy is any differential impact of policy on benefits versus the costs, which at this point is difficult to pin down.