Tuesday, May 17, 2005

New Estimates of Iraqi War Deaths

I have some posts linked together from seveval weeks ago talking about the problems with the Lancet/Johns Hopkins study below. I have written thousands of words and listed about 24 specific statistical issues with this study. Jim Lindgren is talking about it and wonders how to reconcile the Lancet estimate of 98,000 with the new estimate of 24,000 from the new Iraq Living Conditions Survey. I haven't read the new study yet, but I can answer that question. The 98,000 estimate is vapor. It sticks in people's minds, but you have to remember, at best, at very best, this number represents a broad range, a confidence interval, from 8,000 to 194,000. The shape of the probability distribution within that range is in dispute, but a dart thrown at it could easily land on 24,000. There is no reason, without having read the second study, to suppose that the 24,000 estimate conflicts in any way with the first estimate. Statistically, it is the same number.

There is more discussion by Tim Lambert and Tim Worstall. Frank Warner mentioned the issue in one of his threads. According to Tim Worstall, the confidence interval quoted in the new study is 18,000 to 29,000, a much more satisfactory range. 12% were less than age 18. I recollect that it was much higher in the Lancet study, which points to its lack of precision. This study seems to have had a lot more resources, but I suspect is suffers from a lot of the same statistical problems. There is still a confusion of civilian vs. combatant deaths, which I think we have to be careful of.

Once again, note that I don't necessarily disagree with the numbers. I would like to point out, however, that both are historically very low for what you would expect from this kind of invasion.

0 Comments:

Post a Comment

<< Home