Sunday, January 30, 2005

# Posted 6:45 PM by Ariel David Adesnik  

LISTEN TO ME, IGNORANT WORLD: This is not a post about today's election in Iraq. Today's election is a cause for tremendous celebration. The Iraqi people have spoken, so my comments are irrelevant.

This is a post about civlian casualties in Iraq. In a passionate cri de coeur, Daniel Davies of Crooked Timber demands to know why the world has responded in silence to the fact, reported by a study in The Lancet, that 100,000 Iraqis have lost their lives because of the occupation. He writes:
The debate over whether this war worked is vitally important, because we are talking about setting a precedent for an entirely new world of international relations, and the debate is not being carried on honestly. This is quite literally madness, and also quite literally suicidal.
Although I share few of Dan's opinions, I fully share his surprise at the absence of a more forceful response to the Lancet study in the mainstream media. After its initial publication on October 29th, the study became the subject of brief articles in almost all major newspapers. But that was it. The 100,000 figure didn't become conventional wisdom. (Although by establishing the upper bounds of responsible estimate, it provided tremendous credibility to the lower, but still profoundly unreliable casualty statistics distributed by Iraq Body Count.)

Furthermore, Dan observes,
The response in the world of weblogs has been exactly the same as the rest of the media; in the immediate aftermath of the report, half-assed attempts to rubbish the survey, or links to same. Then, when this didn’t work, just pretend that it’s all been dealt with and move on. Maybe say “I’ll get back to you on that”and never do.
"I'll get back to you on that" is precisely what OxBlog said. But I never did. Why? Why has a site that has devoted so much attention in the past to the subject of civilian casualties -- in Iraq, in Afghanistan and in Kosovo -- suddenly gone silent?

It isn't hard to provide an ulterior motive for this oversight. As Dan says, those who supported the war are deserate to "protect themselves from hostile information." Of course, if that were my motive, I wouldn't know it, so I cannot confess. The question is, now that I am confronted with the issue, can I provide rational arguments in defense of my position?

But what is my position? Frankly, I don't know exactly what I think of the Lancet study. Precisely because it has not received extensive coverage from the mainstream media, I cannot rely on the expertise of others to address this highly technical issue. But I have a feeling that something is very, very wrong.

Now, some of you may remember that Fred Kaplan published a forceful refutation of the Lancet Study shortly after it emerged. Above all, Kaplan memorably observed of the 100,000 figure that, "This isn't an estimate. It's a dart board." Yet as this post demonstrates, Kaplan's remark represents a fundamental misunderstanding of what a "confidence interval" is.

Briefly, the study asserts with 95% confidence that the actual number of deaths lies somewhere in the interval between 8,000 and 194,000. Kaplan wrongly assumes that all values within this range are have an equally probability of being the actual figure. In point of fact, estimates clustered around the center of the interval are far more plausible. (NB: I refer to "deaths" rather than "casualties" because the 100,000 figure refers to both violent and non-violent deaths of both civilians and non-civilians caused by the war and occupation.)

Another objection raised with regard to the study is its dependence on a "cluster sampling" methodology. In the same post mentioned above, I think Dan explains quite well why, given the constraints inherent in conducting population surveys in a war zone, "cluster sampling" is an acceptable method. Moreover, according to multiple experts interviewed for a recent article in the Chronicle of Higher Education, the study in The Lancet relied on statistical methods that reflect the state of the art.

However, as a stranger to the art and science of statistical inferences, I am still struck by the fact that the figure of 100,000 deaths was derived from the observation of only 43 "extra" deaths. In the population clusters sampled by the survey, there were 46 deaths in the 14.6 months prior to the war and 142 in the 17.8 months thereafter. However, 63 of those deaths were observed in a single cluster in Falluja. After removing this outlier, one is left with a total of 89. 89-46=43. (All the relevant data is in Table 2 on the fourth page of the study, numbered 1860)

Now, I understand quite well that the purpose of statistics is to extrapolate significant findings from limited amounts of data. What I just can't get my head around is the degree to which limited data sets taken from chaotic war zones such as Iraq should be trusted. As the article in CHE points out, the lead author of the Lancet essay has conducted similar surveys in other warzones, such as the Congo. In those instances, his results were no less dramatic but still embraced by numerous governments including our own.

Still, I can't shake the notion that this time, something went wrong. Perhaps my suspicions have something to do with the efforts of the lead author -- Les Roberts, by name -- to demand the publication of his study before last November's presidential election. To some degree, that isn't fair, since critics should evaluate Roberts' data rather than his motives. Yet when we are dealing with such small numbers, trust begins to matter.

For example, what about the number 21? Of the 89 post-war deaths outside Falluja, 21 were the result of violence, primarily American bombing. Of the 46 deaths before the war, only 1 was the result of violence. In other words, even though human rights organizations estimate that Saddam killed something on the order of 10,000 of his own subjects per year, only one violent death was recorded in the 14.6 months before the invasion. Why?

Did the households interviewed want to protect themselves by attributing Saddam's murders to some natural cause? Or did they simply not mention the death of family members executed by the state? Or perhaps the observation of a single violent death is just a statistical anomaly. As the authors of the Lancet survey point out,
The sampling strategy somehow might not have captured the overall mortality experience in Iraq...[because] there can be a dramatic clustering of deaths in wars where many die from bombings.
For some reason, the authors seem fixated on the potential for death that results from bombing. Yet what about deaths that resulted from state-sanctioned mass murder? Perhaps these are even harder to detect in a random survey.

By the same token, one has to wonder why the only bombings that the authors seem to discuss are those initiated by American helicopters and airplanes. But what about the suicide bombings that have killed hundreds or perhaps thousands of Iraqis? According to the study in The Lancet, of all the deaths it observed, only "two were attributed to anti-coalition forces."

Again, this may just be a statistical anomaly. As noted above, cluster sampling tends to underestimate the impact of focused violence. Yet the authors don't even ask whether the focused violence they underestimate was perpetrated by the Ba'athist government and its insurgent heirs.

At the moment, because of my manifest lack of expertise and respect for the academic positions that the authors occupy, I want to dissociate myself from any explicit accusation of bias, even if analysis above intimates that it may have existed. Before reaching any sort of firm conclusion, I hope to consider the responses to this post by other bloggers with a strong interest in this subject, such as the aforementioned Mr. Davies as well as the very scholarly Tim Lambert.

What I consider most likely is that a statistical anomaly, intended by no one, is responsible for all of this confusion. In war after war, the United States has inflicted numerous casualties from the air. As a result, we have abandoned the indiscriminate carpet-bombing of the Vietnam era in favor of the precision attacks launched against Belgrade, Kandahar, and Baghdad. I find it almost impossible to believe that the methods of the post-Cold War era continue to result in casualty figures that belong to the days of Vietnam.
(0) opinions -- Add your opinion

Comments: Post a Comment