A note on the interpretation of results in Deschênes and Greenstone (2011) by Dell, Jones, and Olken (2014)
Deschênes and Greenstone (2011) use interannual weather fluctuations to identify the effect of temperature on mortality. They find that days with temperature above 90 F sharply and significantly increase the mortality rate. Note that days with mean temperature above 90 F are very rare. In several regions the number of days is close to 0.1 (one day every ten years on average).
Deshênes and Greenstone divide the sample in nine regions and repeat the panel estimate for each of them. (Dell, Jones, and Olken 2014)regress the nine regional coefficients of temperature above 90 F on the average number of days in which temperature above 90 F is observed in each region.
One would expect a significant negative relationship, indicating that regions with more extreme temperature events have adapted at the extensive margin to reduce mortality. However, they do not find a significant relationship and this is taken as evidence of lack of adaptation. This conclusion is questionable.
The regional regressions reveal that days with temperature above 90 F are significantly harmful only in regions where the extreme temperatures are observed with some frequency. In the other regions the estimates are not precise and sometimes the coefficients are negative, which is a counter-intuitive result. The estimates of six of out nine coefficients are thus not precise. It is not a surprise that Dell, Jones, and Olken (2014) do not find a significant relationship and this should not be taken as evidence that hottest regions do not adapt to the extreme temperatures.
For a more detailed discussion see here.
