RJ Rummel vs Lancet study, round 2

R.J. Rummel has a response to my earlier post on the Lancet study. Unfortunately he still does not understand what the researchers did.

In his original post Rummel claimed the pre-invasion statistics came from Saddam's Ministry of Health. In fact, they come from the survey the researchers conducted. Despite my explanation, Rummel now argues:

However, then there is Figure 1, which is unreadable except for its description that lists the data as crude mortality per year before and after invasion. For the before, I can only guess that the Ministry of Health's statistics were used. No source indicated for the figure.

No, the before comes from the survey they conducted. If the source was something other than their own survey they would have said so.

Rummel also stated that they had not carried out tests of significance. I pointed out that they gave 95% confidence intervals for their results. Rummel replies:

Anyway, there are confidence intervals of 95 percent on all major survey statistics, except, so far as I could find, on the difference in proportions.

From the study: "If we exclude the Falluja data, the risk of death is 1.5-fold ([95% CI] 1.1-2.3) higher after the invasion." Since 1 is outside the confidence interval the increase is statistically significant. Rummel continues:

I'm not comfortable that a difference of 46 vs. 90 deaths should be the basis for proclaiming to the world that over 100,000 excess deaths occurred since the invasion. Would you take a medicine that showed only a .4 percent (100x((90/7868)-(46/7438)) improvement over taking a placebo, especially when there were side effects (in the case of Iraq, the claim of over 100,000 excess deaths provided support to the anti-American left, the terrorists, and the Ted Kennedy Vietnam-all-over-again sayers)?

The increase was not 0.4%, but 0.4 percentage points or 50%. Killing 0.4% of the population of Iraq is actually rather a large effect. Admitting the existence of a problem is the first step to do something about it.

Rummel writes:

I assume that Lambert accepts that on which he did not comment.

This is an incorrect assumption. I addressed the most serious flaws in Rummel's post. It does not follow that I agree with the rest. Briefly,

1. In the survey, Death certificates were required of death claims in at least 2 out of 30 households, which leaves a lot of room for false claims

This is misleading. Most of the 30 households did not have any deaths so in those households there is no possibility of a false death claim. The relevant statistic is that 81% of the time that they asked for a death certificate they got one, which does not leave much room for false claims.

2. ... how did the fear of a family about the interviewers (when life is at stake, can one really trust what an interviewer promises about secrecy) bias the results?

The interviewers, who are in the best position to know, did not think that the results were biased by such fears.

3. There was no attempt to determine if a family supported the terrorists, or was part of the previous tyranny.

Sure, but so what? Are Saddam supporters going to instantly forge death certificates to inflate the numbers?

4. The survey found that there was no evidence of "widespread wrongdoing on the part of individual soldiers on the ground." In fact, it found no wrongdoing, that is, what could be violations of the Geneva Conventions. However, again, this is never noted in media reports of the survey.

I'm afraid that bad news is more newsworthy than good news.