Predators, Taliban, and civilians
Peter Bergen and Katherine Tiedemann have published an analysis for the New America Foundation about US drone strikes in Pakistan. Citing previous analyses of drone strike statistics, including the recent report in The Long War Journal, Bergen and Tiedemann conclude that based on their research, between 31 and 33 percent of all casualties from drone strikes inside Pakistan have been civilians.
Bergen and Tiedemann did a fairly good job with their analysis. They debunked some of the rather outlandish estimates of civilian deaths -- including that of Andrew Exum and David Kilcullen, who implausibly claimed in a New York Times op-ed that 98 percent of the casualties from drone strikes have been civilians. Our 10 percent civilian casualty figure was, as we noted, a low-end estimate. A 30 percent civilian casualty rate is certainly plausible.
Obviously, the problem of distinguishing between civilian and militant casualties in this situation is problematic. Essentially, Bergen and Tiedemann's methodology was the inverse of ours: They counted the number of casualties specifically described in press reports as "militants" and subtracted that number from the total killed to get the number of civilian deaths. On the other hand, we counted the number of casualties specifically described as civilians, and subtracted that from the total to get the number of militants killed.
When taken together, I think our two reports provide a fairly good triangulation on what is most likely the actual number. In fact, I wouldn't be surprised if the "real" number of civilian deaths was somewhere in the middle (around 18-20 percent).