jump to navigation

Sloppy work from Les Roberts? March 9, 2011

Posted by dissident93 in Iraq mortality.
comments closed

In a new article, Les Roberts, the epidemiologist (and runner for Congress), claims that “Only 19% of the WikiLeaks reports of civilian deaths had been previously recorded by IBC [Iraq Body Count].” Roberts’s conclusion is summed up by his article’s title: ‘WikiLeaks Analysis Suggests Hundreds of Thousands of Unrecorded Iraqi Deaths’.

Two things immediately stand out:

1) The “Columbia University” study which Les Roberts cites, and on which he bases his conclusions, appears to have been conducted by Roberts himself (with the help of students), although the paper (misleadingly titled: ‘Do WikiLeaks and Iraq Body Count tell the same story? No!’) is unclear about its own authors. Roberts (in the above article) isn’t clear that it’s his own little study.

2) Roberts fails to mention that IBC worked with Wikileaks on an analysis which shows a much greater overlap between the Wikileaks and IBC data (an estimated 81% of the deaths in the Wikileaks logs matching those in IBC’s database).

What explains the difference (19% overlap claimed by Roberts*, 81% deaths overlap claimed by IBC/Wikileaks)? There are a few clues here and here which point to serious problems (if not incompetence, or worse) in Les Roberts’s study.** This episode puts me in mind of an earlier study from Roberts and his students (again designed to discredit IBC), about which Roberts made some misleading pre-publication statements (see second section at link).

Updates/Notes

* The wording of Les Roberts’s article will likely make people think he’s claiming that IBC recorded only 19% of the deaths in the Wikileaks logs. In fact his study claims a 31.6% overlap of deaths (“If this simple analysis of estimating the civilian death toll based on our estimate that the Iraq War Logs and IBC have a 31.6% overlap, the death estimate would be 324,000”page 7). This 31.6% isn’t mentioned in Roberts’s article. The 19% figure mentioned in his article refers to the percentage of “Wikileaks reports of civilian deaths” (ie logs) “previously recorded” by IBC. The deaths overlap comparison should be: 31.6% (Roberts’s claim) vs 81.2% (IBC/Wikileaks claim).

** Since I wrote the blog entry, more has come to light regarding problems with Roberts’s study. These problems include systematic failure to match Wikileaks records to IBC data, where such matches do in fact exist, and failure to account properly for morgue and other aggregate entries in IBC’s data. Please see the discussion here for details.