The Lancet released a study suggesting that ~100,000 Iraqi civilians have been killed that otherwise would not have since the occupation of Iraq began. Then someone at Slate who doesn’t know much about statistics attempted to debunk the Lancet article. Two problems with said debunking:
1) The estimated number of dead is 100,000 with a confidence interval of (8,000-194,000). The critic fails to realize that while it is possible that the number of dead could be 8,000 (or 194,000), it is highly unlikely. It is most probable that the extra number of deaths is between 50,000-150,000 (give or take).
2) The arguments regarding the ‘clumpiness of data’ and cluster analysis are specious. Imagine Fallujia has 1000 houses, each with 10 people in them. We blow up ten houses, killing all ten people in each house (totals: 100/10000 dead). Now we pick ten houses (including “former”, blown-up houses), and ask how many have died (which is essentially what the Lancet authors did). The odds of not picking a single house with causalities are ~89.5%. To use high-fallutin’ language: spatially heterogenous data cause cluster analysis to underestimate a given phenomenon. In other words, the cluster analysis is yielding a conservative estimate.
Rather than argue whether removing a brutal dictator is worth 100,000 lives, I would like to focus on the consequences of being blamed for so many deaths–many of which, according to the study, have resulted from airstrikes. Given that Iraq is a heavily armed, traditional, and tribal society (with a reputation for fierceness, at least in the Arab world), does anyone think that the insurgency is still largely foreign-based? Wouldn’t you conclude that the foot-soldiers, at least, are Iraqi?
Note: I just found a really good, albeit detailed, rebuttal of the arguments against the Lancet article.