Nearly 2/3 of a million people. This was calculated from comparing prewar and current death rates. The statistics are certainly complicated, but if this is true then about 2.5% of the population of Iraq has died as a result of the invasion and subsequent breakdown of all institutions.
It would be interesting to read through the full paper and try to evaluate it. Since the final number of 655 000 additional deaths comes from multiplying a small number (derived death rate) by a large number (the population of Iraq), relatively small errors in the former introduce large numbers of missed or excess deaths. Very similiar to estimates of Chernobyl casaulties---small number (chance of death) multiplied by large number (exposed people). In physics, these are called Fermi Problems referring to Enrico's habit of asking students the (easily estimatable) question of how many piano tuners there are in Chicago.
In any case, there's some evidence here that the total number of deaths is very high. Perhaps more than a factor of ten above the documented death totals. Though we do need to be careful since these are measuring different numbers. The 655k number is "how many people died in Iraq since the US invaded minus how many people would have died in Iraq if the death rate were unchanged" during the same time. The 50k number is "media reported deaths of civilians due to US, anti-US, or sectarian forces committing violent acts."
You can read the full report online. I just now realized that these are the same folks who prepared the Darfur analysis that US State Department uses to condemn the Sudanese regime. 200 000 deaths in Sudan. That's genocide according to most reasonable observers (and even the state department). What is 655 000 deaths in Iraq? It's a civil war at least.
By the way, you can see a rebuttal of supposed problems in the earlier cluster analysis story and continued here. It's worth noting that cluster sampling tends to UNDERestimate rare (like death) effects, not overestimate them. But read the article for yourself.Posted by duver001 at October 11, 2006 12:03 PM