Map of life expectancy at birth from Global Education Project.

Monday, November 16, 2015

Update on Death Rates

I noted in an earlier post the surprising discovery that death rates for "middle aged" (i.e. age 45-54 year old) non-Hispanic white people in the U.S. have been rising, against a background of overall decreasing death rates in all other demographic categories.

As it turns out, subsequent analysis by other researchers has modified this conclusion. Here Andrew Gelman explains that in fact, this is true only for women. The death rate for men increased until 2005, then started back down. The reason for the mistake is interesting for those of you who care about ways to lie -- or just make a mistake -- with statistics. It turns out that during the period of analysis, the age composition of white non-Hispanic people within the 45-54 year old cohort increased. In other words, more of them were near 54 at the end than at the beginning. The death rate doesn't go up a lot from 45-54, but it goes up enough to wipe out the apparent effect for men.

This is a version of what is called an error of aggregation. In any case, the result is even more puzzling than the original analysis. How would you explain it.

1 comment:

Anonymous said...

First one would have to understand what really went on with the whole group (M, F, age.) Naturally explaining the M-F difference would contribute to that! One possible avenue is speculating about what factors for M-deaths (rose and then) sagged / dropped. Where I live, candidates would be alcohol, suicide, and sports accidents. (In the sense that women die much less even hardly at all at those ages from those causes.) And why they sagged, became less important. Perhaps there was a spate of despair - it can be catching - that caused M deaths but those who didn’t succumb went on. Who knows.

My guess though is that it has to use with drugs, mostly legal ones. Don’t women in the US consume much more than men? And there would be no factors that would diminish that use, in fact probably even encouragement (med industry)?

Lastly though these kinds of raw ‘death stats’ don’t distinguish “intrinsic” e.g. medical conditions, disease, handicap - psychiatric - ‘suicidality’ - drug use etc. from living conditions (e.g. socio-economic group, no good access to docs, etc.) to “extrinsic” - being in the wrong place at the wrong time - road, work, sports accidents, getting shot, killed in a brawl, etc. That is of course hard to do because all interacts but without that effort - who knows.