Saturday, February 19, 2005

Lancing a boil (with update 2/23) 

Commenting on my post about Bastiat's window, in which I considered the Lancet study an example of the fallacy, Tim Lambert thinks I'm mistaken.

that is what the study measured: the change in the death rate. If it had gone down, the study would have found a net benefit.

...Roughly half of the 60,000 violent deaths were due to the skyrocketing murder rate in Iraq, not military or terrorist operations. And the 15,000 dead reported in the press will only include some of the civilian deaths, so it is absurd to suggest that all of the difference were insurgents. The Lancet study indicates that at most 5% of the excess deaths were of insurgents. (Though with a subsample this small there is enormous room for sampling error.)

I appreciate Tim trying to think this through, but the Lancet study (you need a subscription to actually read it) doesn't do quite what he says. As Tim points out, "there is enormous room for sampling error," because the actual error range around the death total (the original 100,000 -- I don't see a note on the error band on the 60,000) is +/-94,000 on the total figure. That sampling error comes from how the study is done, which bears some notice. This study was done by sampling a set of 30 or so cities in Iraq and then extrapolating. Thus the sampling error.

But this misses my larger point, which is that those who want to rely on the Lancet study have to assume that the reporting of people about recent deaths and reporting on deaths during the Saddam period were the same. Many of Saddam's dead were not murdered in the presence of witnesses; there is no indication that the authors of the study charged Saddam with a death for a missing person. It was noted in the IHT that the authors sought death certificates to verify the interviewees' memories, but eventually felt it was too insulting in many cases. Which did they believe and which did they not? Also, information was collected during a period when the success of the war against the insurgents (vis-a-vis the war against Saddam); since some were uncertain that America would stay and see through the mission -- thank you, John Kerry -- deaths caused by Ba'athists was probably suppressed by fear of recrimination. Third, about 37,000 of the deaths the Lancet study uses come from a count by anti-occupation groups. (REDACTED: See below.) They also chose both to change their list of randomly sampled areas so they didn't have to drive as much; this meant they stayed close to Baghdad and the Sunni triangle, probably oversampling high-casualty areas. Last, comparing an equal length period between Saddam's reign and occupation assumes that both death rates were at "steady state" levels. As I said in the original post last month, in the immediate aftermath of any war comes a period of heightened violence. Comparing a period under a long-term brutal dictator to a transition phase in the development hopefully of a democracy doesn't meet the test of ceteris paribus needed to validate the study.

A proper study would estimate the steady state rate of democide under Saddam versus the transitional death rate increase in the postwar period. That is the test of what is seen versus what is not seen. My conclusion that the latter is less than the former is speculative, but the Lancet study does not dissuade me, and Lambert's claims focus only on that which is seen. Thank you, Tim, for providing us yet another example of the wisdom of Bastiat.

UPDATE (2/23): Lambert and his drone indicate in comments that they think most of this is mistaken. I have put in the comments page references to the parts they contend are wrong to show that I am referring to the article itself. In the process I found that I had misread a point I initially made about the use of the Kifah numbers by the authors. That's what I get for writing under the influence of Nyquil late on a Saturday night. I have struck through the line that was wrong, and I regret the error, and thank Lambert for pointing it out. On the rest, however, I think he's wrong and that he hasn't brought facts to the debate, only his biases.
The Lancet article includes only one violent death of the 7438 preinvasion individuals they interview (see Table 3). That fact alone should indicate the lack of measurement of prison/torture chamber deaths under Saddam.