Lancet Links

*[Comment deleted. Seixon, you have your own thread why you can criticize the Lancet's statistics. You are not allowed to comment about the Lancet statistics on other threads. Tim]*

The people best placed to do more studies of Iraqi mortality and improve the resolution of the Lancet study are the US military...

Why haven't they done this? Answer: the last thing they want to do is to improve the 8000-190000 95% confidence interval to something like 40000-140000.

And if there were *methodological* problems with the study, whey aren't serious epidemiologists pointing them out? (If Seixon and assorted innumerate bloggers are the leading critics, that says a lot.)

*[Comment deleted. Seixon, you have been told before that if you violated my comment policy and made personal attacks on other commentors you would banned for 24 hours. So now you are banned for 24 hours. Either you wanted this to happen so that you could play the victim, or you can't control yourself. Tim]*

Seixon has been banned for 24 hours for repeated violation of my comment policy. I've deleted his comments and all the ones criticizing him from this thread. Any discussion of Seixon should go in [the Seixon thread](http://scienceblogs.com/deltoid/2005/10/elephants.php), not here.

Monbiot's column is excellent, with the exception of the headline and the final paragraph. It is too much of a leap to equate the Lancet's measurements with the "extent of US and UK war crimes". The Lancet is simply attempting to measure deaths, without elevating them to the (legal? ethical?) standard of "war crime". Monbiot undermines his own case for rigorous and accurate reporting by implying the two are equivalent.

It is too much of a leap to equate the Lancet's measurements with the "extent of US and UK war crimes"

In what way? The invaders bear the responsibility for each and every death caused by the war. Aggressive war is a crime, its perpetrators are war criminals.

What fun! Let's list the ways in which the Lancet Darfur study is better than the Lancet Iraq study!

Questionnaire available?
Iraq: No , Darfur: Yes

More detailed report available
Iraq: No , Darfur: Yes

"Household" well Defined?
Iraq: Unknow, but appears Poor. Darfur: Fair (when absent counted) Poor (otherwise)

Sampling Proportional to Population?
Iraq: Fair (spatial sampling in large areas). Darfur: Good ("pen-spinning" in small areas)

Confirmatory data reported?
Iraq: No, Darfur: Yes (sex ratio)

Sample Size?
Iraq: too small (~1000). Darfur: too small (~3200)

Design effect?

Iraq: 29, Darfur: 3.5-11.3

I listed the "design effect," because it is a measure of the reduced precision due to clustered sampling (lower is better). One key methodological conclusion of the Lancet Darfur study is that the 30x30 sampling plan -- which was developed for immunization surveys -- simply isn't big enough for this kind of study, where the outcome (violent death) is so clustered.

P.S., Hitchens isn't really citing the Lancet Darfur study: his figure of 400,000 deaths in Darfur appears to be complete bunk, and does not appear in the Lancet study.

None of which would justify Hitchens calling one one study a "crazed fabrication" and the other a "reliable estimate", even if was aware of these differences. Which I'm sure he isn't.

Furthermore, the Iraq study only has a design effect of 29 if Falluja is included. Exclude it and the d.e. is 2, less than Darfur. Also, deaths in the Iraq study were confirmed with death certificates while in Darfur they were not.

The number of clusters is what jumped out at me (the complete amateur) as the main problem with the Lancet study when I first started reading the discussions about it and thought about it for a bit. It's fine if most violent deaths are distributed more or less evenly, at least within a given province. The Fallujah cluster demonstrates the problem with this assumption. If 1 percent of Iraq was hit very hard it would only be a fluke if you'd catch one of these places with just 30 clusters. And we know that there has been very intense fighting in certain areas. So maybe some other sampling method should be tried, not that I know what that would be, unless it simply means many more clusters. I'm not blaming the Lancet team, which did the best it could with limited resources.

I don't mean to defend Hitchens. The Lancet Iraq study wasn't a "crazed fabrication" and he should be citing the UN Darfur figures (which are plenty bad).

But to say that the Lancet Iraq DE isn't so bad if you exclude Falluja isn't much of a defense. The whole point is that you have to exclude Falluja if you want meaningful numbers. But Falluja is where most of the violent death in their sample occured.

One reason why the Darfur DEs are much smaller is because those studies were conducted in much smaller areas, rather than the whole country (as in Iraq).

In fact, it seems to be common in humanitarian crises to conduct one Iraq-style study per refugee camp, rather than one per country. A recent JAMA article about Ethiopian studies counted 125 such surveys conducted over a year (with 6 being of higher quality than the Iraq study and probably dozens of equal quality). The 4 Darfur studies reported in the Lancet paper are just a fraction of the data collected in Darfur.

It's not Roberts' fault that he's trying to make a single 30x30 study substitute for what usually takes at least a dozen similar surveys, but there you have it.

I didn't realize that we were up to 125 surveys in Ethopia as compared to just one in Iraq. The only explanation I can think of for this discrepency is that the coalition is trying to make sure that such surveys don't happen in Iraq.

*[Comment deleted. Seixon, you have your own thread where you can write about the Lancet study. Please put any comment that mentions the study in that thread. Tim]*

Oh, so everyone else can talk about the study, except me. How genius. How about you delete the portions in the comments that you feel are about the Lancet study that shouldn't be here instead of deleting the whole thing?

I have saved the comments, and I am looking at it right now, and all of the comments are on topic for this thread.

In fact, I talked about things very similar to what has already been said by others here. Now are you going to stop being Orwellian and let me take part in discussions?

Here's a comment entirely without mentioning it! Enjoy!

I didn't realize that we were up to 125 surveys in Ethopia as compared to just one in Iraq. The only explanation I can think of for this discrepency is that the coalition is trying to make sure that such surveys don't happen in Iraq.

Yeah. Either that or the fact that there hasn't been any war in Ethiopia since 2000. Or terrorists blowing people up. Insurgents killing random people. Not eliminating possible explanations again to favor a predetermined conclusion now are we Mr. Lambert?

Yes Seixon, everybody else can discuss the Lancet study in this thread, but you have to do it in your own thread. This is to allow non-Seixonian discussion to proceed. If you don't like these rules, tough.

Yes, Lambert. I know. You don't like scrutiny. The emperor hath no clothes. Yadda, yadda. Even if I am staying on topic, you will delete because I scrutinize things you hold dear. You only make yourself look weak.

Pathological numeracy:

Apparently, by sex and age adjusting the IRBC figures, we can determine that only 7,976 Iraqi civilians have been killed (http://www.logictimes.com/civilian.htm)!! Which they say themselves, in triumph mind you, makes living in Iraq no more risky than driving a car in the US!!

But wait, there's more good news! Only 3.8% of the fatalities are civilians killed by the US during the occupancy, for a grand total of only 948! (Although it seems to me in the "details" of the calculation presented in the colorful table of Appendix A, we seem to be multiplying by the % of total mortality ascribed to US cause, b variable, and yet again by the % of occupation-period-only mortality ascribed to US cause, d variable, which doesn't seem right; but what do I know, in light of the confirmatory estimate of 1,000-1,500 independently derived by one of the esteemed commentors).

"we seem to be multiplying by the % of total mortality ascribed to US cause, b variable, and yet again by the % of occupation-period-only mortality ascribed to US cause, d variable, which doesn't seem right; but what do I know, in light of the confirmatory estimate"

Oh, never mind, these are being done as parallel estimates.

"...there hasn't been any war in Ethiopia since 2000."

Other than the one with Eritrea.

There's an article by Seymour Hersh in the current issue of the New Yorker which is relevant to the Lancet paper. According to Hersh, there's been a largely unreported air war going on in Iraq. He cites one Marine press release in November 2004 which stated that Marine airplanes (I don't know what fraction of US airpower in Iraq this would represent) had dropped 500,000 tons of ordnance since the beginning of the war. Which is about the total dropped on Cambodia. 7 million tons for the entire Vietnam War.

I'm wondering if the 500,000 tons is a misprint. It's a staggering figure if true. What difference does all this smart bomb technology make you can drop 500,000 tons of ordnance and not wipe out this relatively small insurgency? And how much collateral damage would this inflict? Depends on how many bombs are dropped in cities, no doubt, but still, I suspect you can't drop this many bombs without killing a lot of civilians. Laser guidance or no laser guidance.

Here's the link.

http://www.newyorker.com/fact/content/articles/051205fa_fact