Government officials have long grappled with a question that seems like the purview of philosophers: What is the value of a human life?
Under both Democratic and Republican administrations, the answer has been in the millions of dollars. The higher the value, the more the government has required businesses to spend on their operations to prevent a single death.
But for the first time ever, at the Environmental Protection Agency the answer is effectively zero dollars.
Last week, the E.P.A. stopped estimating the monetary value of lives saved when setting limits on two of the most widespread deadly air pollutants, fine particulate matter and ozone. Instead, the agency is calculating only the costs to companies of complying with pollution regulations. . . .
OK, we talked about this the other day.
Given that the newspaper just ran a story about this last week, I don’t quite get why they’re covering it again, but whatever.
There are two things about this new article that concern me.
First, they quote “Michael Greenstone, an environmental economist at the University of Chicago” on the health risks of air pollution, but Greenstone is notorious for using statistical methods that overestimate these risks–I’m talking about that notorious air-pollution-in-China study. I’m not saying that air pollution has no effects on life expectancy, I’m just saying that no way would I trust this particular economist to give a number on this risk. You might as well ask James Heckman for a quantitative estimate of the benefits from early childhood education, or these people for an estimate of the impact of corporate sustainability on organizational processes and performance.
Second, they write:
To determine this value, government economists have turned to studies on the labor market, which show that workers demand higher wages before agreeing to perform jobs with greater risks of workplace fatalities.
Say that employers must pay lumberjacks an additional $1,000 a year to perform work that generally kills one in 1,000 workers. It follows that most Americans would forgo $1,000 a year to avoid that risk and that 1,000 Americans would collectively forgo $1 million to avoid the same risk entirely. Therefore, in this example, the value of a statistical life would be $1 million.
Nooooooo! I’ve been convinced by the work of economist Peter Dorman that, no, there is no evidence that “workers demand higher wages before agreeing to perform jobs with greater risks of workplace fatalities.” And I think that the lumberjack story is entirely made up.
The story is here: “Risk without reward: The myth of wage compensation for hazardous work.” Also some thoughts of how this literature ended up to be so bad.
Too bad the NYT reporter didn’t know about this.
Michael very serious person. tries hard. river big polynomial difficult. one mistake why no forgiveness
“Greenstone is notorious for using statistical methods that overestimate these risks–I’m talking about that notorious air-pollution-in-China study. I’m not saying that air pollution has no effects on life expectancy, I’m just saying that no way would I trust this particular economist to give a number on this risk. You might as well ask James Heckman for a quantitative estimate of the benefits from early childhood education, or these people for an estimate of the impact of corporate sustainability on organizational processes and performance.”
I think you mean something slightly different, but it sort of reads as if you’re saying that devoting one’s research career to a specific question somehow invalidates one’s perspective on that question, which would be really perverse.
Hahn:
My distrust of Heckman’s estimates on the effects of early childhood education are not from the fact that he’s worked long and hard on the problem. My distrust comes because he is using biased estimators; see discussions here and here. Heckman knows a lot more about early childhood intervention than I do, and on most aspects of the topic I trust his expertise more than mine. I just think he’s systematically overestimating the effects.
Andrew, this is a really bizarre and nihilistic view to over-index on one (not especially convincing) supposed refutation of the idea of compensating differentials. There is pretty good evidence of compensating differentials for lots of job characteristics, including risk (eg https://kurtlavetti.com/DLS_vc.pdf and https://www.aeaweb.org/articles?id=10.1257/pol.20150024)
I think the Greenstone RD paper is bad but setting up a manichaean fight between “guy who wrote an unrelated paper you like” and “guy you like because he commented on your blog” is not an admirable way to approach a literature. If Greenstone or Heckman told you not to jump off the Empire State Building, would you do it anyway out of spite?
Drive:
I’m inclined to agree with Greenstone that air pollution is bad for your health, and I’m inclined to agree with Heckman that early childhood interventions are good for kids. I just don’t trust them when they put numbers on these things. My lack of trust in their numbers is not about “spite”; it’s because those two researchers have a track record of making bold claims based on biased estimates.
Regarding compensating differentials: I read Dorman’s book on this and found it convincing, years before I ever had a blog. My take on the topic is based on my readings, including Dorman’s book and including articles by others offering different perspectives on the topic. I’ve blogged about this a few times, first back in 2004 or 2005. That said, I’m open to the possibility that Dorman and I are wrong on this. I’d have been ok with a news article that discusses the compensating differential idea and then also quotes Dorman or someone else offering the opposing position. The article just seemed very one-sided to me.
Shouldn’t we distinguish between the idea of compensating wage differentials and the measurement of such things? I find it hard to deny the existence of these compensating wage differentials (my too cursory reading of things Dorman has written suggest to me that his objections are to the measurements and not the idea). Once you constrain the analysis to comparing apples to applies, a job with higher risk will generally have higher wages to compensate for that. But the apples-to-apples comparison is not only difficult, but might be close to impossible, given the myriad other factors that determine wages. Then there are philosophical objections – how voluntary are job choices, is willingness to accept higher wages a fair measure of value (given that it is affected by a number of things, including income), etc.? Even these philosophical objections do not deny that such differentials exist – they object to how they are measured and used.
I think Dorman was right at the time that a garbage can regression from survey data with a few controls thrown in can’t possibly answer this question – and this was the state of evidence in the 90s. However, there is now well-identified work, like what drive-by linked to, showing the obvious result that risky work carries a pay premium.
Even so, there isn’t any reason to believe that whatever number comes out from well-identified work is a meaningful statement about how workers value their life. We know from psychology/behavioural economics that people routinely misunderstand small probabilities, especially when the game is only played once (you can only die once). Furthermore, any change in subjective probabilities of accidental death probably also comes from unpleasant conditions that impose less severe health hazards and will vary from case-to-case. I’d argue the value of the well-identified work is instead to check if people are making obviously irrational health trade-offs due to cognitive short-comings where government intervention may be necessary, rather than a meaningful statement on the value of a life.
I like two of the three points you make. The work by Viscusi and the others emulating him was essentially valueless. (Where was the gatekeeping?) I also agree that the leap from labor market patterns to “willingness to pay” is unwarranted. (See my comment below on cognitive dissonance avoidance below.) I’m not as impressed by the two links offered by drive-by, and, putting all else aside, there are two reasons for this.
First, I think this work, like a *lot* of other work in econ, misunderstands the point of large-n statistical analysis. If you think you’re doing science or something like it, you have to be interested in mechanisms, how things happen and not just whether. For that you need direct observation or at least very tightly controlled experiments. In econ case studies can help with this. But of course you also need to know when and how those mechanisms apply, so yes, you have to get lots of data and parse them. From this perspective, what I object to is pronouncing the existence of a mechanism when all you have are data *consistent* with it. You’ll notice a connection here to the problem with NHST, which presumes a binary between “the effect I’m looking for” and “no effect”.
Second, unless this world is way simpler than I think it is, we should expect lots of heterogeneity in relationships like working conditions and pay, and this variation is actually the subject of greatest interest. Especially if you’re a labor economist, it’s almost the whole ballgame. I want to see the heterogeneity reported, evaluated and used as a guidepost for further study.
Thanks for linking those two articles. The first I saw when it came out, and it immediately struck me that it didn’t properly distinguish between the risk of fishing in nasty weather and the general nastiness of it. Even so, I wouldn’t be surprised if there were CWD’s in this case, because the risks are so vivid and well-known, and because the labor market for capable fishing boat hands in Alaska is sufficiently tight. (I think.) The second is new to me for some reason, though I should have noticed it. I haven’t had time to look at it closely, but it looks as though no attempt has been made to measure the heterogeneity of wage responses to inspections. For instance, if it’s all about new hires (presumably there are switching costs for incumbent employees), then replacement or expansion rates should be predictive. I would also want at least a smidgeon of case study evidence, documenting a few plants in which inspections led to safety improvements and then to downward wage pressures. Actually, I would guess that OSHA (in a rational administration) would want that evidence as well to mute criticism of its inspections and mitigation demands on the part of employers.
The question is why anybody should care whether you trust the numbers or not.
Provide better ones, critically examine models, but to dismiss every actual research as not good enough is anti-science and feeds right into the hands of the current anti-science government that claims “hoax” whenever they do not trust a number, no matter how well supported it is.
You don’t have to provide a better number to show that the current ones aren’t well supported. That’s just nonsense. Sometimes, there just is no answer at the moment. That’s how science works.
Andrew likes quoting John Tukey to support this.
“The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data.”
Thanks for posting this, Andrew. About Greenstone, he’s a consumer rather than a producer of VSL estimates, so his fumbling with regression discontinuity back in the day (a fault shared with a lot of economists at the time, by the way) isn’t germane, IMO. But he’s really a true believer in cost-benefit analysis as a comprehensive decision tool. He’s the point person for efforts to identify a social cost of carbon (SCC), which in his view should guide what we do and don’t do about climate change, and of course you need a VSL – and a value for just about everything else – to compute this. If anyone’s interested, the appendix to my book on climate change (Alligators in the Arctic and How to Avoid Them, Cambridge 2022) is devoted to taking apart the SCC fixation. For economists who still say, “But how are you going to know how far to go on climate mitigation without an SCC?”, my answer is that the temperature target consensus shepherded by the IPCC, 1.5 degrees if possible, 2 max, should be our marching orders.
Greenstone has written at least two papers estimating the VSL:
Using Mandated Speed Limits to Measure the Value of a Statistical Life (https://www.journals.uchicago.edu/doi/full/10.1086/379932)
And a closely related short paper in the AEA Papers and Proceedings (https://pubs.aeaweb.org/doi/pdf/10.1257/0002828041301984)
You’re right! Those were coded “Ashenfelter” in my memory, but Greenstone was a coauthor. I remember the speed limit piece as being rather over the top. Looking again, it still is. Maybe it’s a very wry satire.
Peter:
Yeah, here’s a key bit from the abstract: “Since the states that adopted the higher speed limit must have valued the travel hours they saved more than the fatalities incurred, this institutional change provides an opportunity to estimate an upper bound on the public’s willingness to trade off wealth for a change in the probability of death,” which does seem like a parody of econ-thinking. Especially the “upper bound” bit. And the ideas that states have values. And the idea that a policy negotiated in the legislature can be directly mapped to the public’s preferences.
I remember when the 55 mph speed limit was instituted in the 1970s. The reason was not public safety (although, once they had the policy, they did promote it with the slogan “55 saves lives”) but rather to reduce fuel usage because of the oil embargo and gasoline shortage. The speed limit was considered as an alternative to other proposed policies such as gas rationing.
In my opinion, the final part of the post creates the impression that there is no wage compensation for risk at all (‘a null effect’). In many of his texts, our host has explained that he actually believes the opposite: that everything is correlated with everything, and that there are no ‘true’ null effects.
The news article suggests that workers are asking for risk compensation, either explicitly or implicitly. In some cases, this may well be true (think of astronauts). In other cases, however, this is clearly not the case (think of conflict mineral mining or child labour). Clearly, there is wage compensation in some cases, but not in others.
With this in mind, it seems that the criticism is directed at the statement which says
‘studies on the labor market, which show that workers demand higher wages before agreeing to perform jobs with greater risks of workplace fatalities.’
I am sure that child labourers or cobalt miners employed in an illegal mine do not receive hazard pay. If this assumption were to hold, the generality of the statement above would collapse.
More than the statement being false under mild assumptions, it irritates me because of its ‘wages are justified’ vibes, as if our data could explain why person X gets wage A and person Y gets wage B. Typically, such regressions would explain less than 20% of wage variation (R²).
Raphael:
Yes, good points. The relations between risk, employment, and wages are tangled, with a lot of variation. Effects are not zero; they’re just complicated.
As it happens, I’ve actually studied wage determination for child laborers — in some countries and some occupations. I didn’t get the data needed to test for compensating wage differentials, but what I found was interesting and unexpected. The heterogeneity is immense, and the wage-productivity relationship can change as the child ages.
My favorite example of wage compensation comes from “Up and Down California, 1860-1864 ” the journal of William Brewer, who was a member of an early geological expedition. About the mercury mines at New Idria, he wrote that:
“The work at the furnaces is much more unhealthy and commands the higher wages. Sulphorus acids, arsenic, vapors of mercury, etc., make a horrible atmosphere, which tell fearfully on the health of the workmen, but wages always command men, and there is no want of hands. The ore is roasted in furnaces and the vapors are condensed in great brick chambers, or ‘condensers.’ These have to be cleaned every year by workmen going into them, and many have their health ruined forever by the three or four day’s labor, and all are injured; but the wages, twenty dollars a day, always bring more victims.”
That’s quite a quote. It hasn’t come up in these comments, but an important aspect of the whole question is cognitive dissonance avoidance, a.k.a. denial. There’s a lot of qualitative evidence that denial is a significant force in a lot of dangerous work. You would expect it to reduce compensation to some extent, but cognitive dissonance is also exacerbated by the feeling of “I really need to keep this job”, which can increase with pay, right? I’ve never seen this properly sorted out or tested for prevalence.
“…And I think that the lumberjack story is entirely made up. …”
Hypothetical examples to illustrate how a method works generally are.
James:
That’s kind of my point. If it’s such a universal phenomenon, they should be able to offer a real example.
“That’s kind of my point. If it’s such a universal phenomenon, they should be able to offer a real example.”
Maybe they could but I don’t see why they should. If they are just trying to illustrate the idea a hypothetical example will generally be “cleaner” and therefore clearer. Like a physics problem that assumes no air resistance.
But if you want a real example Russian military signing bonuses would seem to be one:
“The Tambov, Krasnodar, Kurgan and Altai regions, and the republic of Tatarstan, also announced significant increases in the payments, which come on top of the monthly salary for contract soldiers fighting in Ukraine. That starts at roughly 210,000 rubles ($2,600), more than double the average Russian wage.”
James:
This is a good example and it also illustrates the point about variation: there’s no reason to think that extra amount that you might pay a soldier for risk would be comparable to the amount that a civilian would value a risk from a pollutant.
Andrew
This relates to a larger point about all studies of the “value of a statistical life.” An underlying assumption is that this value should be the same across different contexts – that the value someone places on risk of being a lumberjack can be compared with the value someone places on exposing themselves to risk when they choose to drive to work. Many studies of risk attitudes find that people evaluate risks very differently depending on context: for example, risks of airplane crashes are evaluated differently than individual risks from driving (apparently risks to a large number of people at once when they have little or no control are viewed differently than voluntary solitary acts). It remains unclear the extent to which people’s different evaluations in different contexts should matter for public policy – should we spend more per statistical life for airline safety than road safety? These are complex questions of value and it is not clear to me that there is a single “right” way to address these. But the fact remains that risk attitudes vary with context and in ways that may be irrational (or not).
There are probably examples of wage compensation, for example deep sea divers (at least that’s what I read somewhere). Of course divers are highly skilled and qualified workers and it’s hard to tell if there is a risk premium in addition to the skill premium. Also, it’s physically very hard work in often unpleasant circumstances (away from home etc) and there might be a premium for that.
It occurs to me that this type of highly skilled dangerous work is probably very well insured and there are plenty of safety measures in place. Workers who specialize in such work may do so in expectation of good pay but they also can’t easily change to a different job. These are some considerations that imho suggest that it’s very hard to reliably estimate a risk premium.
In general, the highest paid workers are definitely not the ones taking exceptional health risks or working under exceptionally adverse conditions.
People are notoriously irrational at evaluating and dealing with risks, or even understanding them. I’m not going to look up a bunch of examples, but I’ll name-check a couple here: people often care more about relative risk (and reduction of it) than about absolute risk, e.g. something that increases the risk a heart attack by 1% is considered less dreadful than something that doubles the risk of a rare brain cancer. People don’t replace the batteries in their smoke detectors, but avoid swimming at the beach because there was a shark attack 2 weeks ago and 30 miles away.
Recently there has been some news about lung disease in workers who cut artificial countertops (see https://www.npr.org/2026/01/14/nx-s1-5674884/kitchen-countertop-workers-are-dying-some-lawmakers-want-to-ban-their-lawsuits ) and this is a very, very, very rare case inasmuch as I find myself quite sympathetic to the arguments of the countertop manufacturers. They argue that if the workers cut the material as directed — use ‘wet cutting’ rather than dry, and wear a respirator — the material can be worked with quite safely. (I believe this to be true). But that article says that “over the last six months, while visiting more than a hundred fabrication shops, the safety officials did not observe any workers wearing the appropriate level of respiratory protection during high-risk cutting and polishing tasks. And even though cutting the slabs without using a stream of water to damp down the dust can be extremely dangerous, she said that her office estimates that “at least 25% of shops continue to dry-cut stone.”
If you own one of these workshops, why would you not provide respirators and insist that workers use them? And if you work in one of these shops, why would you not insist on having a respirator and wearing it? It makes no sense to me. (And I should mention that these materials are not new: the current OSHA regulations became effective in 2016 and I assume they replace an earlier set). And, if you’re OSHA, why aren’t you handing out $10K fines any time you see a worker performing a dangerous task without the safety equipment that is required by law?
Those questions above are not purely rhetorical, they’re also genuine questions, and I think they have genuine answers. The risk is due to exposure over many years, and thus, sort of like smoking cigarettes, each indiividual incident doesn’t seem like a big deal. The workers are mostly male, and there might also be an element of macho to it: what, you aren’t afraid of a little dust, are you? And I suppose the employers are so cheap (or have profit margins so low) that they don’t even want to spend the $100 per employee it cost to provide respirators…although (1) come on, that can’t be right, and (2) presumably they wouldn’t prevent and employee who insists on wearing his own respirator. Anyway I find the whole thing rather baffling; I think there is no way to explain any of this in any rational decision-making framework. In such a circumstance, we certainly should not expect ‘the market’ to automatically compensate workers for extra risk that they take.
By the way, kind of peripherally related: I occasionally do projects that involve cutting wood, and I have a very small indoor workshop with a table saw etc. It takes considerable discipline for me to wear either an N95 mask or my face shield with filtered-air supply whenever I use my table saw (and for at least ten minutes after, because small particles take a long time to settle and my workshop is not very well ventilated). I’m constantly tempted to skip the mask ‘just this one time.’
I have a friend who also wears a mask when cutting stuff, but he removes it as soon as he’s done with the cutting even though he knows that small particles remain airborne for a long time: the mask is a bit uncomfortable and the risk feels remote. He’s a professional contractor and he’s just doing what ‘everyone else’ does in his industry…actually the fact that he always wears a mask (and hearing protection, and eye protection) sets him apart even though the takes the mask off quickly.
When construction workers were doing some work for me and my wife, several years ago, I insisted that they always wear eye and ear protection when appropriate. I had to talk to the foreman about it several times to get him to crack down on his employees about the ear protection (which he also wasn’t good about himself). I told him “if you want to ruin your hearing, that’s fine, you can do that, but not while I’m paying you.”
My take-home is that most people (myself included) don’t have the right gut feelings around small risks that are frequently encountered, or small exposures with cumulative effects that are frequently experienced. Somehow we treat ‘small’ as if it’s negligible. I’ve been accused of being “too rational”, whatever that means, and even I have to really force myself to take appropriate measures.
I think these are valid points and the model of rational decision making under uncertainty becomes strained in the face of numerous behavioral examples. However, this need not destroy the idea of compensating wage differentials – it depends on the extent to which these behaviors are prevalent. Many people make more rational choices when deciding where to work and for how much. In terms of working for an “immoral” employer, I think the wage premium is a real thing (although still hard to measure properly, given the myriad factors that would need to be accounted for – but I don’t think that biased risk perceptions play a big part). I do think for some risky occupations (fishing in Bristol Bay, for example), people are attracted to the work that may have biased risk perceptions, complicating the question of whether the wages are a good measure of anything regarding the risk. So, I think it is context dependent – as with much of the behavioral economics literature, I think there is some truth in the “irrationality,” but the degree to which it predominates is unclear.
I find the idea of using imputed wage risk premiums to estimate the “value of a life” absurd, because such premiums are hard to estimate if the exist, but also because workers are in no position to understand the risks their job pose to them. The employer is in a better position – they ought to know how often workers are harmed on the job – but most obviously it’s the insurance industry that is in the position to evaluate the risks, and to know how much it costs to insure those risks.
Regarding workers’ attitude to risk, a little anecdote: as a youth I worked summer jobs in a factory. Those were well paid. The business issued safety shoes for a nominal price (a fraction of the real cost) and instructed me to always wear them on the job. I didn’t like to wear them and I observed none of the regular workers were wearing them.
My boss, a young guy, scolded me. He said it was stupid to disobey the safety rules, I had no idea how easy I could break a toe and if I got hurt he would get the blame for negligence. He said I shouldn’t compare myself to the long term workers, they were in a different position.
He was a good guy. I was stupidly and stubborn and valued comfort over safety, and that’s extremely common.
Now that I think of it, just remember COVID. People freaking out over mask wearing. No, ordinary people absolutely cannot be expected to engage in risk appropriate behavior.
Where I am, the flu is waving. Be careful and mask up!
We have to make decisions in life. Placing a value on a human life may help us make better decisions—even if that value is flawed on many dimensions.
Consider highway improvements that are designed to make roads safer. If one understood (1) how much a project would cost and (2) how many lives per year it would save, one could rank projects by lives-saved/$. Starting with the best projects and working down, fund projects until you run out of money (or out of willingness to spend money). That process would calculate a value of human life—it would be roughly the reciprocal of the lives-saved/$ of the first project that was not funded.
If you have some other projects that produce multiple benefits, this value can be a useful tool in ranking projects. It may not have to be very accurate in order to improve decision making.
For a real world example, consider the UK NICE process for determining if drugs are cost effective and should be funded by the NHS. They say
. . .means that for a medicine to be considered cost effective, it should typically generate 1 additional year of perfect health (or an equivalent combination of additional life expectancy and health-related quality of life improvements) for no more than £20,000-£30,000 over the cost of current care. It has now been agreed that NICE will apply new thresholds of £25,000 to £35,000/QALY as soon as NICE has the power to do so, following a change in regulations.”
See https://www.nice.org.uk/news/articles/changes-to-nice-s-cost-effectiveness-thresholds-confirmed.
If the NHS cannot afford to spend unlimited money on drugs, then their approach seems like a reasonable way to limit drug expenditures.
I once looked quite closely at a NICE analysis and spoke to the person in charge of that analysis. I came away feeling that the process was reasonable.
Bob,
I agree completely, for reasons discussed in this earlier post.