Sunday, January 29, 2006
Not content with the Lancet's extrapolated estimate (as of October 2004) that 100,000 Iraqi civilians in excess of the usual rate had died since the beginning of Operation Iraqi Freedom, Attaturk (one of Atrios' co-bloggers) is now saying that the Lancet study underestimated civilian deaths by 60%. The supporting "analysis" is here.
The argument is obscure, but it turns on the Lancet's exclusion of casualties from Fallujah from the original estimate of 100,000 dead. This clown argues that if the Fallujah data were included, the total number would rise to 250,000. Suffice it to say that he does not explain his reasoning or his arithmetic, but if you try to derive his math from the accompanying tables you still only get to around 130,000, not the 250,000 claimed by Rising Hegemon. And, I believe (but will not waste my time verifying) that the original Lancet study took into account the exclusion of the Fallujah data in arriving at its headline estimate for the whole country.
Both claims are sharply at odds with even the current Iraq Body Count estimates, although the two are measuring somewhat different things (IBC measures casualties from combat violence, whereas the Lancet purported to measure excess deaths from all causes). More tellingly, the new lefty number seems to require that 150,000 civilians died in Fallujah (if anybody else can explain the math better, please do). Even Islam Online -- not exactly one to sugarcoat the depredations of the war -- reported that 70% of the city's pre-war population of 300,000 had fled before the city was retaken in November 2004. So even if American soldiers ruthlessly slaughtered everybody remaining in the city -- an absurd and libelous accusation -- the new lefty number is laughably high.
No doubt, the purveyors of this "social science" also claim that they "support the troops." Well, maybe not the Canadian guy.
The 100,000 estimate in the Lancet study excluded Falluja. That's why they gave the number as "100,000 or more". 250,000 or so is what the survey implies if you include Falluja. It doesn't require you to assume that all those 150,000 dead were in Falluja -- they could be in other high mortality neighbourhoods around the country.
You've missed the most important difference between the IBC and a survey based estimate -- not all deaths are reported in the media so the IBC is guarenteed to be a huge undercount.
Hmm. I didn't follow the transition between your third sentence and your fourth sentence. If only Fallujah samples were excluded, then how do you assume that the incremental 150,000 dead came from places other than Fallujah? And if even most of those deaths come from Fallujah, they cannot add up to anything like 150,000 additional.
I appreciate that the IBC relies on press accounts and that does guarantee, as you say, an undercount, but I think if you do the math the biggest difference between IBC and Lancet is that IBC only counts violent deaths. Lancet counts all alleged incremental deaths. I could be wrong, though -- I admit I haven't dug in to the Lancet data in a very long time, and even when I did I am no expert.
Setting aside the statistics, doesn't the Lancet study require complete data about deaths in pre-war Iraq in order to establish the incremental deaths since the war? Does it do that effectively?
I wish I had a link for you to a radio story I heard about the man who led the Lancet survey to come up with the 100,000 number.
His method was based on statistical sampling in various neighborhoods in over a hundred locations across Iraq. He then extrapolated this data to the entire nation. We do something similar when we poll people.
Since he couldn't sample in Fallujah, due to the ongoing security situation, this segment of the sample was excluded. Had it been included, then the extrapolated numbers would have been exponentially higher.
Did that make sense?
The Lancet study was based on a laughably small sample size (under 1000 households, only 6 cities) which meant that the only truthful claim they could make was that it was 95% probable that the total number of casualties (all causes) was between 8,000 and 198,000. Hence the widely advertised 100,000 figure. It was the mid point of a huge and indeed meaningless range. The Lancet study leader has also systematically lied about whether their "excess deaths" included military deaths, claiming they were "all civilians" on the BBC, when as IBC points out, his methodology does not allow him to distinguish civilian from military.
It is worth noting that whilst the Lancet study excluded the Fallujah data in respect of total deaths they did not in relation to their conclusions on cause of death. This was important as it enabled them to claim a mid point 100,000 deaths with almost all caused by Coalition airstrikes. Without the Fallujah data (reported by those in the city whilst it was under terrorist control) included this claim fails (I wonder why ?). That was why it was included by the study. This is of course wildly at odds with IBC figures which show 85% of post invasion deaths caused by terrorism or criminal activity, even with theinflated numbers on Fallujah used (IBC data - under 10,000 deaths caused by Coalition forces to date, 70% in initial invasion, 70% of remainder in Fallujah).
Also worth noting that in press comment the study leader repeatedly made the claim that 100,00 was a minimum figure (contradicting his own methodology) and that with Fallujah data in the the total deaths were 300,000. Also worth noting that this higher number would mean that relative to the scale of the Iraqi population the Iraqis would have suffered more dead than either the Japanese or Germans in WWII from air bombardment by fleets of thousands of US and British bombers flattening every city in their respective countries with conventional and (for Japan) nuclear weapons.
There is a rival much larger study (25 times as many households interviewed, total geographic coverage as v 6 selected cities) which was carried out by the UN Development Agency at the same time as the Lancet study which gives a 95% probability of all deaths being between 18,000 and 29,000 - mid point 24,000,near enough. This is consistent with IBC estimates if you assume that the 7250 "civilians" killed by Coalition forces in the invasion were actually almost all military/militia fighting out of uniform (the absence of excess child/female deaths in the UN data strongly supports this).
The Lancet study is an utterly worthless piece of political agitprop, statistically invalid in every respect. The UN study was a seriously conducted exercise which produced much lower numbers consistent with the press reports and hospital body counts analysed by IBC and which has accordingly been completely ignored by the press as not making good enough headlines. Taking UN and IBC data together you can get to a fairly sensible analysis of the Iraq war civilian and total deaths which indicates totals in the 30-40,000 range and very low totals caused by the Coalition forces - perhaps 10% of this number.
A worthwhile discussion can be had on this matter using worthy sources. The Lancet study is not a serious source.
Actually, one of the strange things about the Lancet paper is that they did not sample large parts of the country because of the security situation -- including areas with little or not insurgency but that were a long way from Baghdad -- but they did sample Fallujah. They just extracted the Fallujah data from the "headline" result.
Whatever the validity of the original study, the extrapolation about which I posted proves, if nothing else, the wisdom of excluding it in the study results. If the inclusion of the Fallujah sample results in an increase in casualties that would have been physically impossible (which seems to be what happened), then you have to wonder if the either the sample or the method were screwed up at some level.
Tim Lambert, who has made a blogname for himself defending the Lancet study, says in the first comment that it all makes sense, but not precisely enough for me to understand. If he persuades me, fine.
In any case, there have been a lot of cogent blog posts parsing the Lancet study generally, both pro and con. I've concluded either that my statistics are too weak to figure out which is right, or that I haven't tried hard enough. No matter.
The backstory of the Lancet published study is that the author, an American leftist, insisted it be published at a specific time without peer review.
The timing was related to the 2004 US elections.
That a prestigeous medical journal chose to prostitute its name in such a way lessens its credibility.
One thing that I have learned in my years in medical technology: the article referees for medical journals are particularly weak in statistics. We regularly see unbelievably bad statistics in medical journals -- I am guessing that the Lancet paper would have been much more heavily scrubbed had it been submitted to a well-regarded social science (political science or economics) journal.
Some journalist at "Canada's new socially progressive and cross-cultural national newspaper" tried to add back in the lurking variable that was specifically removed from an already questionable study.
Kinda reminds me of how the light reflected from the moon now disproves greenhouse effect global warming. (sorry, couldn't resist)