Tech Central Station 0, Lancet 2

Tech Central Station has published Tim Worstall's admission that his critique of the Lancet Iraq study was completely wrong:

Further to my article of Friday on this subject. I'm afraid I mangled the statistical argument. My inadequate knowledge of the subject led me to make an argument that is incorrect. I stand by my contention that there is something fishy about this study (leaving aside the politically motivated timing of its publication, something the author has been clear about himself) yet have to admit that I have not found it, leaving me with nothing but personal prejudice upon which to stand my argument. I would also like to make clear that this subject was not "assigned" to me, the idea, research, argument and errors were all my own, as was my request for this clarification. Just in case you are wondering, being fact checked by the Pajamahaddin and being found in error does hurt and I hope that future writings will be, where necessary, so corrected."

On his blog Worstall thanks Daniel Davies and me for the correction. Such decent behaviour is unfortunately not common at Tech Central Station. Authors like John Lott and Iain Murray just repeat their false claims, while Glenn Reynolds posts a correction but does not acknowledge the source.

Tech Central Station didn't just post Worstall's correction by itself. They have had a second attempt at debunking the Lancet study, posting an article by Michael Fumento. Fumento argues:

the researchers didn't feel themselves bound by anything official, like death certificates. Interviews were just fine. "In the Iraqi culture it was unlikely for respondents to fabricate deaths," they wrote.

Unfortunately, Fumento seems to have missed the immediately preceding sentences in the Lancet paper, where they noted that, when asked, 81% confirmed with death certificates:

In 63 of 78 (81%) households where confirmations were attempted, respondents were able to produce the death certificate for the decedent. When households could not produce the death certificate, interviewers felt in all cases that the explanation offered was reasonable eg, the death had been very recent, the certificate was locked away and only the husband who was not home had the key. We think it is unlikely that deaths were falsely recorded.

Fumento's "killer" argument is:

Cluster sampling can be valid if it uses reliable data, rather than on inherently unreliable self-reporting. But it can also be easily skewed by picking out hotspots -- like determining how much of a nation's population wears dentures by surveying only nursing homes.

In fact, intentionally or otherwise, that's pretty much what The Lancet did. Most of the clusters had no deaths whatsoever. But here's the real bombshell: "Two-thirds of all violent deaths were reported in one cluster in the city of Falluja," the journal reported. That's it; game over; report worthless.

Trouble is, Fumento has once more been extraordinarily careless in his reading of the study. Here are the two sentences in the report that follow the one he quoted:

If we exclude the Falluja data, the risk of death is 1.5-fold (1.1-2.3) higher after the invasion. We estimate that 98 000 more deaths than expected (8000-194 000) happened after the invasion outside of Falluja and far more if the outlier Falluja cluster is included.

That's right, they properly excluded the outlier Falluja in their estimate of 98,000 and Fumento didn't notice this fact. That's it; game over; Fumento article worthless.

I'm starting to feel embarrassed for Tech Central Station. Do you think they'll have a third go at the Lancet?

Also of interest is Chris Lightfoot's demolition of more lame critiques of the study.

More like this

I get it, so taking somebody's word for it 19 percent of the time is irrelevent. And here's a new one for you. The alleged 100,000 deaths were those above the pre-war baseline. That baseline was predicated on a figure of 5.0 deaths per 1,000. BUT the figure for the US at the time was 8.3 deaths per 1,000. Obviously Iraq was one of the safest countries on the face of earth prior to the Yankee imperialist invasion. In fairness, the CIA Worldbook uses a 5.6 per 1,000 figure for Iraq but what was it's source? Saddam's ultra-trustworthy government, of course. Thus The Lancet is using figures even lower than the Land of Baghdad Bob was putting out.

Oh god, this is embarrassing. Michael, if you ask Tim politely, he might delete that comment and we will all pretend you didn't dig this hole for yourself.

In the first place, you are not apparently questioning the accusation that you misinterpreted the treatment of the Fallujah data point. That is enough on its own to make you a laughing stock.

But then you go on. The reason why the Iraqi death rate listed in the CIA factbook is lower than that of the USA is that Iraq is a younger population than the USA, and people in general die of causes which are more linked to age than to violence. A death rate estimated on its own as a snapshot is a nearly meaningless number; for any real-world application, what matters is either the change in death rates from year to year, or the by-cohort death rate. I'm not going to explain what a "by-cohort death rate" is, because you badly need practice in looking things up.

Your statement about "taking somebody's word for it 19% of the time" is equally laughable. Quite apart from the barefaced cheek of trying to bluff your way out of the fact that you were talking rubbish about death certificates, the 81% figure quite clearly shows that there is good reason to believe that (adult) deaths were reported accurately. There is a separate issue with respect to infant deaths, which is noted in the paper, but I advise you not to get involved in that debate, because, if I may be blunt, you haven't got the skills for it.

Tim Worstall has gone substantially up in my estimation for immediately recognising he'd pulled a howler and going hands-up as soon as he realised it. You have responded to someone pointing out two devastating errors in your piece by saying "Here's a new one for you". You're a disgrace to the regiment, and Tech Central Station is hardly a prestigious regiment.

Tim, TCS WILL have a third go at it because their audience doesn't read the source material.

They HAVE to keep going at it because 1) they have to keep sponsoring the war and 2) they have to keep bashing science.


Fumento - from 'Mythbuster' to 'Featherduster'

When I see a study with error bars as wide as on this one, I ask: Why did anyone pay attention to this study. So there's a 95% probability that the number killed was between 8000 and 190,000 (or thereabouts). Such a "study" is meaningless. It has no predictive value.

When I see that the author submitted on the requirement that it be published before the election, I get more than a little bit suspicious of the objectivity.

The sampling requires trusting the Iraqi interviewers, one or more of whom may have a conflict of interest. Furthermore, it is biased towards the high side by human nature - nobody is going to conceal a death, especially since the US military compensates for civilian deaths, but they may very well exaggerate deaths for a variety of reasons.

When NGOs who are not friendly to the war, but who have experience in estimating civilian casualties, have much lower figures, it's time to throw the study away.

By John Moore (not verified) on 12 Nov 2004 #permalink

John Moore,
The fact is, that the DoD did not want to "do body counts" of civilians. Tell me why? Wouldn't you say, that they were afraid of what might transpire?

"study" is meaningless.

No, it is the best estimate that we've got. You are free to sponsor your own study.

It has no predictive value.

It's called 'estimation'. 'Prediction' is a different area.