I'm getting my range of premature death estimates from Margot Sanger-Katz and John Schwartz:
Volkswagen’s diesel deception unleashed tons of extra pollutants in the United States, pollutants that can harm human health. So while many commentators have been quick to say that the cheating engines are not a highway safety concern, safety — as in health — is still an issue.
Unlike the ignition defect in General Motors vehicles that caused at least 124 people to die in car crashes, Volkswagen pollution is harder to link to individual deaths. But it is clear to public health researchers that the air pollutants the cars illegally emitted damage health, and they have formulas for the number of lives lost from excess pollution in general. ...
To estimate the harm in the United States, we used two different scientific models for the effects of nitrogen oxide pollution on human health.
One comes from a sort of natural experiment, when new regulations on power plant pollution caused some counties, but not others, to cut back on nitrogen oxide pollution. The counties subject to regulation reduced their nitrogen oxides emissions by 350 tons a year.
A team of three researchers — Olivier Deschenes, Joseph S. Shapiro and Michael Greenstone — looked at the mortality rates and medical spending before and after the change. In a working paper, they found the reduced pollution was responsible for about five fewer deaths for every 100,000 people in the affected counties each year, as well as for a decrease in spending on prescription drugs. Most of the seemingly excess deaths in the higher pollution regions occurred among older Americans, though other health issues affected the young as well as the old.
The estimated Volkswagen pollution, about 46,000 tons since late 2008, is the equivalent of about 4 percent of the power plant pollution reduction they measured, meaning it could be expected to cause an estimated 106 deaths if it had similar effects.
We ran this projection by Mr. Greenstone, a professor of economics and director of the Energy Policy Institute at the University of Chicago, and a contributor to The Upshot. He said it seemed sensible as an estimate, but “the magnifying glass is really close.”
Noelle Eckley Selin, an associate professor of earth, atmospheric and planetary sciences at M.I.T., ran the numbers a slightly different way. She looked at the mortality effects of the particulate pollution produced by nitrogen oxides, using the numbers the Environmental Protection Agency uses to make health estimates. Her method brought the effects to about 40 additional deaths over the period, in addition to some other nonfatal health consequences. That probably undercounts the impact, though, since it does not consider the effects of direct nitrogen oxide pollution or smog.
Now for all the caveats. ...
via www.nytimes.com
Now for the damage estimate ... The damage per car is equal to D*VSL/VW, where D is deaths, VSL is the value of statistical life and VW is Volkswagons (VW=500,000). Without any information on what is the best estimate of lives lost I'm using a uniform distribution between 40 and 106. Similarly, without any information on what is the best estimate of the value of statistical life across averting behavior, contingent valuation and hedonic wage methods I'm using a uniform distribution between $5 million and $10 million.[*] Taking the product of 1000 random draws from both distributions, the average damage estimate is about $500,000,000. Dividing by 500,000 cars gives a damage estimate of about $1000 per car with a standard deviation of $375. So, about two-thirds of the per car damage estimates are between $625 and $1375. This range is substantially higher than Auffhammer's damage estimate of $232 per car, which is in the lower third standard deviation of my estimate.
[*] The lower end of the VSL range is about what comes out of the averting behavior and contingent valuation literatures. The upper end is from the hedonic wage (i.e., labor market) literature. Contrary to current federal government practice, I don't think it is really appropriate from a benefit-cost analysis perspective to pick a very uncertain number and run with it. The averting behavior and contingent valuation method literatures have plenty of valid estimates that suggest the VSL, in a context different from the labor market risk, is lower. Plus, the two sets of estimates tend to agree with each other (i.e., convergent validity) more than they agree with the VSL from the labor market literature. The theme of the SBCA meetings last March was that the VSL is a solid $10 million but I'm not ready to equate labor market and air quality risk.
