The Lancet released the most recent version of a study that has tracked mortality in Iraq since the US invasion in March 2003 – it estimates between 393,000 and 943,000 “excess deaths” in Iraq as a consequence of the war, based on estimates of a mortality rate of 13.3 per thousand per year, up from 5.5 per thousand pre-war. (pre-war: 4.3-7.1, 95% CI; postwar: 10.9 – 16.1, 95% CI.)
An earlier study using similar methodology was released in October 2004 and caused immediate controversy. That study suggested 100,000 excess deaths attributable to the war, a far higher figure than had been estimated via any other methods. Due to small sample sizes, the study had a very wide confidence interval – there was a 95% chance that the increase in deaths was between 8,000 and 194,000 with the highest likelihood at 98,000 deaths. This wide confidence interval prompted commentator Fred Kaplan to declare, “This isn’t an estimate. It’s a dart board.”
That’s very far from the truth – the Lancet study used an epidemological method that’s widely respected and has been used to calculate the death estimates for the conflict in the Democratic Republic of Congo with very little controversy. But because the confidence interval was wide and because the timing of the study (just before 2004 US elections) was suspect, there was widespread condemnation of the study and it was ignored in many major newspapers. This American Life ran a long radio story on the study, the methodology and the conflict over the study, including an explanation by one of the study’s most quoted critics that he criticized the study without reading it and no longer considered the estimate unreasonable. Several smart bloggers also defended the accuracy of the study, walking readers in detail through proper interpretation of a “confidence interval” and explanations of why methods like the Iraq Body Count must neccesarily give an undercount of civilian deaths.
The confidence interval on the new study is much tighter than on the first study and even the low bound of the confidence interval is a stunningly large number: roughly 400,000 deaths in Iraq attributable to the conflict. (The confidence interval suggests that there’s less than a 2.5% chance that the effect is lower than 400,000 excess deaths from the conflict.) It’s interesting to me that many of the news stories on the new study don’t have the obligatory “that seems high to me” quote that reports on earlier studies included – the Washington Post story has quotes from two researchers not connected to the story who both say, basically, the research method used is sound. Whether or not this will satisfy those who believe this study is politically motivated is unclear…