Imagine a government spending program or a health or safety program, where one of the benefits is that over the population as a whole, it reduces some risk that people face and thus saves lives. Is the spending or regulation a good idea? Answering this question requires putting a monetary value on the benefits of the lives that are expected to be saved. For some years now, the Environmental Protection Agency has used a value of $7.4 million per life saved, measured in 2006 dollars, and then updated over time based on inflation and rising incomes since 2006.
Among non-economists, there is often a (perhaps natural) disinclination to be repelled by putting a monetary value on lives saved. But as economists point out, both governments and people people make decisions all the time that involve tradeoffs between costs in money or time and risks of death. Governments make decisions about what safety standards to created for cars and consumer products, or for workplace safety. The level where speed limits are set has implications for costs in terms of time on the road and a risk of lives lost. Some people take jobs with a higher mortality risk but also higher pay, or buy houses near known pollution sites at a lower price, or spend extra for cars with additional safety equipment.
Government regulators have to estimate the “Value of a Statistical Life” or VSL. The word “statistical” is meant to convey that when the government spends or regulates in an effort to save lives, it typically does not know whose life will save: for example, it does not know which traffic accidents will turn out not to be fatal because seat belt use is required. Indeed, the “Value of a Saved Life” would be more accurately described as the “Value of Reducing the Risk of a Lost Life.” The number turns out to be central in carrying out benefit-cost analyses: as an example, one study of benefits and costs of the 1990 Clean Air Act Amendments found that 80% of the benefits were from the monetary value placed on lower mortality risks leading to lives saved.
But when you look at how the value of a statistical life is estimated by the Environmental Protection Agency, you find that it relies on a group of 22 studies done between 1974 and 1991. Maureen Cropper, Emily Joiner, and Alan Krupnick argue the case for an update, based on research in the three decades since then, in in “Revisiting the Environmental Protection Agency’s Value of Statistical Life” (Resources for the Future, Working Paper 23-30, July 2023). For a short readable overview of the paper and the underlying issues, Joiner offers “Rethinking the Value of a Statistical Life” (RFF blog post, September 21, 2023).” A quick overview of this research provides a sense of what kinds of studies are done, and what types of new evidence has emerged.
But before describing the research, it’s perhaps useful to note that Cropper, Joiner, and Krupnick don’t try to provide a bottom line: that is, they don’t try to estimate whether the value of a statistical life would be higher or lower if the last three decades of research are included. This is annoying, but probably also canny. After all, you should be able to agree or disagree with the idea that it would be useful to update the VSL estimate without knowing in advance what the answer will be.
For example, of the 22 pre-1991 studies, 17 of them are “hedonic wage studies.” Basically, these studies compare the wages of workers in different jobs. They then include a bunch of other variables: age, education, marital status, race, years in the labor force, industry, occupation, and the like. They also include a variable for risk of death on the job. Thus, these studies seek to answer the question: “After adjusting for other variables that seem likely to affect wages, do those those in riskier jobs get paid more than similar workers in less-risky jobs?”
There are a variety of detailed questions one might ask about these studies. They don’t all adjust for the same variables: for example, many of the earlier studies don’t adjust for industry or occupation. Some studies also include measures of health risks other than mortality; others do not. In addition, people are not randomly assigned to jobs with different risk levels, and those who choose such jobs are not likely to be a representative sample. As a result, when you adjust for industry and occupation, you are in some ways adjusting away some of the choices made about risk.
A broader issue is that more than 90% of job-related fatalities occur for men, with a disproportionate share happening in a narrow range of jobs. Thus, it would be nice to base the value of a statistical life on other tradeoffs that people make about costs and risks.
For example, car-buyers often face a choice between paying more for a safer car, or not. The additional amount that people are willing to pay for safety can be estimated, with some effort, from this data. As one example, airbags were phased in for US cars in 1996 and 1997. Thus, one can study how much extra people were willing to pay for a car with airbags vs a very similar car from 1994 or 1995 without airbags. One can also look at different speed limits imposed on highways across the United States, and measure the implied tradeoff in terms of hours saved by travelling faster vs. greater risk of lives lost. In another example, studies have looked at people’s willingness to buy houses closer to pollution or to a Superfund site as a measure of willingness to trade off higher risk for a lower price.
I’m not claiming these kinds of studies are easy to do in a persuasive way! But non of these kinds of studies about how people value risk in different settings are included in the 22 pre-1991 studies that are the basis for estimating the value of a statistical life.
In the original 22 studies pre-1991, the other five studies are “revealed preference” studies, which are basically elaborate survey tools in which researchers ask a series of questions to get people to place a value on different risks. For example, one can survey people on how much they would be willing to pay to see their risk of death from specific causes like heart disease, stroke, cancer, or car accident reduced by a certain amount by the time they reach age 70. The responses of different people at different ages can then be woven together to build a measure of willingness to pay for risk reduction, and how this evolves over a lifetime.
As you might imagine, there is considerable controversy over the narrow question of how to construct these surveys, and also over the broader question of how much reliance to place on the choices people actually make in job, car and house purchases, and the like, vs. what people say in surveys when they don’t actually have money on the line. Here, I’ll just say that the methods for doing this kind of work have evolved since 1991, but none of the later studies are included in the current estimates of the value of a statistical life.
For me, one of the biggest benefits of the value of a statistical life concept is that it allows comparisons across different regulations and spending programs. For example, say that one regulation reduces risk at a cost of $100,000 per life saved, while another other regulation reduces risk at a cost of $100 million per life saved. With this kind of information, there is a strong case for for strengthening and enforcing the first rule, and perhaps just dropping the second rule. With such comparisons (which are not uncommon in the underlying literature), it becomes possible to save more lives at lower overall cost.
For those who would like to know more about how the idea of a value of a statistical life originated in military planning doctrine of the 1950s and the work of Thomas Schelling, as well some additional intuition about how such calculations are done, useful starting points are “Value of a Statistical Life: Where Does It Come From?” (March 27, 2020) and “The Origins of the Value of Statistical Life Concept” (Novemer 14, 2014).