If you haven’t read my previous forty posts on the Lancet study, here is a handy index. All right, let’s go.

First up, via Glenn Reynolds we have Andy S, who critiques the Lancet study despite not having read the thing. This is not a good idea, especially since he is relying on Kaplan’s flawed Slate article. Andy has three comments:

Firstly the use of a scientific publication to make POLITICAL POINTS is reprehensible. This Lancet study was published just before the American Presidential Election and was clearly an attempt to make President Bush look bad just before he faced the electorate.
The authors did hope to influence public policy—they had found an alarming number of deaths from coalition air strikes and wanted the US government to take action to reduce this. I think it was naive of them to think that their study would be considered fairly in the heat of the election campaign. In any case, this is irrelevant to the question of whether the study is sound or not.

His second comment is about the exclusion of Falluja from the estimate:

I don’t want to get technical but under this methodology Fulluja should have been included! The original Sample Design should have ensured that it was selected because it was known to be atypical! The original sample design would have accounted for the fact that the Falluja cluster was an atypical cluster and the analysis of the results would reflect this. There are sound and well established mathematical techniques to accomplish this.
Well, if they’d known exactly how atypical the Falluja cluster was they could have accounted for it. But that would have required data that they did not have. Excluding Falluja biases the results downwards. If you insist on including Falluja the number is 300,000 excess deaths. I think that number is too high because the Falluja cluster was atypical.

His third comment is:

As I understand the Lancet survey, interviewers went to selected households and asked about family deaths in the period between the invasion of Iraq and the day of the interview. To say none ( as most people probably did) is kind of boring so some respondents would be tempted to invent a dead relative. If the interviewee had an anti American agenda, as some surely did the temptation to lie would be even greater.
If Andy had read the study he would now that they did a spot check by asking to see a death certificate. In 81% of the cases checked they got to see a death certificate. In the remaining cases the reasons given for not being able to prove the death seemed plausible. It’s possible that some of these were invented, but this is not enough to make a big difference.

Glenn Reynolds has another post on the Lancet study, with comments from readers and bloggers.

First, he links to Brian Crouch who reckons that Andrew Bolt’s critique discredits the Lancet study. No it doesn’t. I don’t want to seem harsh here, but if you haven’t even studied basic statistics and you criticize the statistics of a study that has been peer reviewed by professional statisticians you are likely to end up looking pretty silly.

Second, he has a comment from reader Dave Ujeio who defends the study, having studied it one one of his stats courses. Ujeio gets it right—he had a good teacher and I give him an A.

Third, Hugh Thorner

There’s no need to debunk the 100,000 civilian casualty figure being cited so often by war opponents. In progressive circles it’s an article of faith that pre-war sanctions killed 5000 Iraqis per month. Cost of the war two years later? 20,000 Iraqi civilians saved! And counting…
Sorry, but the 100,000 is excess deaths. It’s the increase in deaths due to the war. Deaths from the sanctions are included in the prewar death rate. Note the pre-war death rate in the study was for 2002 when the oil-for-food program had greatly reduced malnutrition and child mortality.

Next, Aron Spencer

1) the distribution of probable dead is not normal. It actually probably resembles a Poisson distribution.
Binomial, I would think, but both are reasonably approximated by a normal distribution in this case.
2) the study distribution’s 95% confidence range covers so much of the possible range as to be a nearly flat distribution (at least relatively speaking).
Let’s see, a 67% confidence interval goes from 50,000 to 150,000. I wouldn’t call it that flat.
3) even if the statistics were acceptable, there are serious questions about the sampling, as pointed out in the original debunking.
Kaplan’s original debunking got a ridiculous number of things wrong in his description of the sampling. See here.
4) the author of the original study is known to have biases related to the research.
Ad hominem. Are we supposed to disqualify everyone with an opinion on the war, one way or the other, from researching into casualties?

Then, John M:

Are we honestly to believe that twice as many non-combatants have died as a result of the liberation of Iraq as were American combatants in 8 years of VietNam? In a war designed and fought to minimize civilian casualties with things like GPS guided bombs?
If you want to compare the deaths with Vietnam the relevant comparison is with the number of Vietnamese deaths. Which was 1-2 million. Maybe GPS guided bombs successfully minimize civilian casualties and maybe they don’t. The way to find out is to actually count the number of deaths.

And Craig Bond:

Based on this information, is it technically incorrect to claim that 8000 or 194,000 would be “rare” events. Instead, the correct conclusion, as in the “debunking” article by Kaplan, is that we can be 95% confident that the true number of casualties lies between the bounds. It says nothing of the probability of any of these outcomes.
Yes, the 95% confidence interval by itself doesn’t tell us what the probabilities are. But this doesn’t mean that each value is equally likely. We can also construct other confidence intervals. We can be 67% confident that the number is between 50,000 and 150,000. In this sense the end points of the 95% CI are less likely and the middle is most likely.