We Are Over-Preventing Covid

In the Oct. 12 JAMA, David Cutler and Lawrence Summers estimate the total costs that the U.S. will suffer from the covid pandemic. Here is their key table:

So their numbers imply that we will lose $6.95T due to covid harms (deaths and impairments), and $9.17T from efforts to prevent those harms (lost income and mental impairment). Note that Cutler and Summers didn’t divide these costs into prevention versus harm prevented; that’s my division. And note that while they included the main covid harms, they missed some big costs from prevention, such as children getting worse schooling, less socializing, and a general dislike of wearing masks.

If we just look at the costs that Cutler and Summers consider, we get a prevention to harm cost ratio of 1.32, which is optimal if on average 1.0% more prevention effort cuts health harms by about 1.3%. I don’t know if that’s right, but at least it doesn’t seem crazy.

However, those virus harm estimates come from assuming a $7M value for each of these lives lost, and that I say does seem crazy. (They also assume 625K total virus deaths, 2 people impaired by 35% per death, and $7.6T income lost over the next ten years.)

As I’ve discussed before, it makes much more sense to value life-years lost, relative to income. And as we’ve so far seen 9.2 life-years lost on average per covid death (in U.S. through July 11), this $7M value estimate translates to a life-year lost being worth 12.1 years of income. And that’s crazy high.

For such crazy high estimates, we have no one to thank more than Kip Viscusi. He arguably started the literature estimating the value of life, and he’s been a big influence on it ever since, via co-authoring a huge number of papers (a big % of papers he cites in his review articles are by him), and editing a major journal that publishes stuff on the topic (J. of Risk & Uncertainty). And Viscusi is proud to present himself as a crusader for higher life-value estimates:

The watershed event that led to the adoption of the [Value of Statistical Life] was the 1982 conflict between the US Department of Labor and the Office of Management and Budget (OMB) over the proposed hazard communication regulation. The agency had valued the reduced mortality risks by the ‘cost of death’, leading to relatively modest benefit values. In its review of the regulatory proposal, the OMB concluded that the proposed regulation failed the required test that benefits must exceed costs. The Department of Labor appealed the decision to then Vice President George H.W. Bush, and I was asked to resolve the dispute. The only change I made to the OMB’s critique is that I introduced my VSL estimates into the agency’s analysis. Use of my VSL figure of US$7.4 million (in 2015 dollars) increased the estimated benefits of the regulation by a factor of 10. (more)

As I’ve noted before, since a lot of government regulation is justified as protecting citizen lives, those who seek to justify more regulation tend to seek higher value of life estimates. Just a year before, in 1981, Viscusi had just published an estimate that each U.S life was worth $17.9M in 1976$, implying a life-year lost was worth 20.3 years of income! That’s much higher than most of the estimates he’s published since, and his estimates tend to be higher than those of other authors. Also, Viscusi has long been skeptical of the idea of adjusting the value of life for the number of life years left – he says the value of life doesn’t depend on age.

A few months ago I went through five other papers, mostly lit reviews, on the value of life, and I translated the favored median estimate in each paper into a median life-year-to-income ratio:

  • A 2002 review of 33 job papers found 1.76.
  • A 2003 review of 30 road safety papers found 2.92.
  • A 2011 review of 850 stated preference estimates found 2.98.
  • A 2003 study of 76 US federal regulations (the ones Viscusi is proud to have influenced) found 3.34.
  • A 2003 Viscusi review found 4.53 re 30 US labor market papers, 5.94 re 22 non-US labor market papers, and 1.88 re 11 U.S. housing and product market papers.

Setting aside the estimates Viscusi influenced, I’m comfortable with accepting an estimate of 3.0. But if we apply that to the covid estimates Culter and Summers used, we’d get only $1.08T and $0.64T for covid death and impairment harms, which gives a prevention to harm cost ratio of 5.31. (Which is close to the 5.2 median estimate from my most recent poll.)

And its crazy to think that on average we are getting a 5.31% cut in covid harm for each 1% increase in prevention cost we pay! But that’s what it would take to justify this level of prevention spending relative to harms prevented. In fact, in most recent poll, the median estimate is that we’d instead get only a 0.10% cut in covid harm from 1% more prevention. Over a factor of 50 weaker than needed. And that’s why I say we are way over-preventing covid. (A scenario I warned against back in March.)

Note that I’m not blaming all this excess prevention on government; private choices clearly drive much of it. I’m talking about the U.S.; prevention elsewhere may have been more effective. I’m not saying all prevention efforts are equally effective. And I’m not weighing in here on exactly what are the best prevention strategies, nor on what is their maximum possible effectiveness. Sure, the best probably are far more effective than what we are and have been doing. But, alas, doing less than what we have been doing seems to me a far more politically feasible option than proposing to identify and switch to far better strategies. (Like maybe trying variolation or vouching.)

I’ve long said that we spend way too much on medicine, because the marginal value is far below our marginal costs.  (A topic on which Cutler and I sparred before.) Seems that sort of thing continues to hold in a pandemic. And governments, instead of correcting for this problem, mainly just make it worse.

Added 26Oct: Tyler says this sort of marginal analysis is irrelevant due to “non-linearities”. Doesn’t really say more than that. But most econ systems are not linear, yet that doesn’t stop econ elsewhere. He could say because of “non-convexities”, but this requires that either we are at a boundary or that there be big moves from the status quo that beat all small moves. Yet for that to be a relevant critique here, Tyler needs to be able to point to big changes that we should adopt soon, and likely will adopt soon. Absent that, critiques of small changes to the status quo remain quite relevant.

GD Star Rating
Tagged as:
Trackback URL: