Please visit the new home of Majikthise at

« Right wing rhetoric on North Korea | Main | Tim Blair and the web of belief »

October 13, 2006

Interview with co-author of Lancet study of Iraqi deaths

Democracy Now has an excellent interview with Les Roberts, co-author of  Mortality after the 2003 invasion of Iraq: a cross-sectional cluster sample survey. Roberts and his colleagues estimate that between 400,000 and 800,000 more Iraqis have died since the US invasion than would have died if pre-invasion death rates had held steady.

Some ill-informed partisans, including the president of the United States, have mistakenly claimed that the method of cluster analysis has been discredited.

Roberts on cluster analysis:

You know, I don't want to sort of stoop to that level and start saying general slurs, but I just want to say that what we did, this cluster survey approach, is the standard way of measuring mortality in very poor countries where the government isn’t very functional or in times of war. And when UNICEF goes out and measures mortality in any developing country, this is what they do. When the U.S. government went at the end of the war in Kosovo or went at the end of the war in Afghanistan and the U.S. government measured the death rate, this is how they did it. And most ironically, the U.S. government has been spending millions of dollars per year, through something called the Smart Initiative, to train NGOs and UN workers to do cluster surveys to measure mortality in times of wars and disasters.

So, I think we used a very standard method. I think our results are couched appropriately in the relative imprecision of [inaudible]. It could conceivably be as few as 400,000 deaths. So we’re upfront about that. We don’t know the exact number. We just know the range, and we’re very, very confident about both the method and the results.

Roberts also describes in detail how the researchers distributed the clusters across the country.

Most of the critics of the study are focused on why the estimates might be too high. They argue that the investigators just happened to have surveyed a lot of unusually violent places. This is always possible, however, that objection cuts both ways. Unless there's some reason why the authors were more likely to stumble upon the hotspots, you've got to admit that it's equally likely that this sample missed most of the hotspots by chance.

HT: Deltoid.


TrackBack URL for this entry:

Listed below are links to weblogs that reference Interview with co-author of Lancet study of Iraqi deaths:

» Some Good Lancet Article Posts By Lindsay from Mike the Mad Biologist
Over at Majikthise, Lindsay has been doing some really good debunking of the critics of the Lancet article that indicates roughly 655,000 Iraqis have died as a result of the Iraq War and Occupation. [Read More]


I'm not so sure I'd describe partisan claims that the method of cluster analysis has been discredited as "mistaken."

That seems to imply that they care whether the claim is true or not. They don't. (Thus the claim falls in the category "bullshit," as described by Harry Frankfurt.)

We don't call people "mistaken" when we believe they are lying, and we shouldn't call them "mistaken" when they are bullshitting.

(This is, of course, pure semantic pedantry, and does not detract from Lindsay's main point.)

The idea that cluster sampling as such is controversial is really funny. A simple random sample of 10000 people, say, is prohibitively expensive even in western countries. Hence, practically every really big survey is clustered, usually at several levels. It's a real pain because it makes standard errors very hard to compute; there are entire books on how to.
Compared to the GSOEP, say, their clusters are large-ish though, but it's understandable given the constraints.

So much more on Iraq and N Korea not reported in US media

I'm not an epidemiologist, but I've read a helluva lot of journal papers in my time--often in fields beyond my expertise. It's seldom difficult to spot the holes and BS in a bad paper.

When I read the first Lancet study, I was surprised to learn: (a) that their 100,000 estimate excluded the Falluja cluster; and (b) why it excluded that cluster. If you have some stupid secret agenda that hinges on reaching a big number at the bottom line, you don't throw out the crazy-big outlier--you downplay everything but the crazy-big outlier.

The Lancet authors didn't pull any funny business in that first paper. They handled the Falluja cluster responsibly. There's no way they were fudging anything. I haven't read their second paper yet, but I'll be really surprised if, when I do read it, I find that it feels in any way fishy.

Anyway. Thanks for following this story, Majikthise.

IIRC, Rummy praised UNICEF's reportage on Saddam's killing of children prior to the time we liberated invaded. The sampling metholology is the same.



I have a question if clinton did this what would you say. Would you say we invaded Iraq, or Clinton lied. Would you question the US media into not reporting everything. Do you hate Bush because your liberal ideas are not heard from republicans. If republicans protected the envirnoment and were socialist republicans and let gays marry and had FREE government health care and just went crazy would you still complain about every little thing in IRAQ or any war for that matter. I remember Clinton bombing IRAQ and going into Somila I didn't hear from you then. It's only a just war unless fits your agenda to leftist communist thinking. You know thru all this I don't hear any democrats solutions. All I hear is false reports, lies and no solutions to terrorist. Also terrorists who even admit they are terrorists who are caught in middle east and shipped to any jail are not US citizens and when they caught I wouldn't waste anything and just shoot them on the spot. Along with Saddam.

If republicans protected the envirnoment and were socialist republicans and let gays marry and had FREE government health care and just went crazy would you still complain about every little thing in IRAQ or any war for that matter.

Of course I would. What a silly question. Even socialist paradises aren't allowed to slaughter people in other countries for no reason.

As I posted on another thread, cluster sampling is a valid technique, but the Hopkins team did not apply it wel.

Just some of the Hopkins errors:

1. Results depend upon choice of baseline, with wide variation described here:

2. Failure to randomise within clusters, a particular problem when estimating a high-variance variable.

3. Dependency on Sunni staff.

4. Households not independent and likely self-selecting.

Far more reliably in statistcal method was the ILCS household survey, which yielded more deaths than Iraq Body Count but less than half that of Hopkins.

The comments to this entry are closed.