The Reality-Based Community

 The deaths in the pottery barn

How many Iraqis have died up to today as a result of the US-led invasion four years ago?

Under the "pottery barn" principle of the Geneva conventions (Lancet editorial, June 2005, free registration required), it's the coalition's job to find out - essentially the US government and army. But they won't do it. It would only be about count 19 in the war crimes indictment, but obfuscation has always been essential to the project.

It's been left to a heroic band of civilians to fill the gap. Naturally their shoestring efforts are inconsistent in detail, so they can be nitpicked to media death. But their overall picture is consistent and damning.

So what is the best available central estimate of excess Iraqi deaths?

9,890 - US public

That was the median estimate given in an AP survey of the US public in February 2007. Since the reliable, downwards-biased Iraq Body Count (IBC) stood at 54,000 at the time, the American public is wildly wrong - the absurd number is an indictment of the US media.

30,000 including combatants - George W. Bush

That was Bush's answer in December 2005 to a question about all Iraqi casualties, insurgents and civilians. The estimate was too low the first time, as the IBC (maximum count) gave 30,000 at the time excluding combatants, but pardonable. In a press conference in October 2006, Bush not only smeared the Lancet researchers - "600,000 or whatever they guessed at" - but refused to update the 30.000. By then, outside Panglossia, the IBC count had risen to 48,000.

Turning to estimates based on some sort of evidence and reasoning, the next is

66,195 - Iraq Body Count of civilian deaths by violence, minimum, today

The IBC is based on reports in the English-language media, cross-checked with morgues, etc. They can't cover deaths reported only in the Arabic media, incidents that aren't reported by anybody, and non-violent deaths still attributable to the conflict, such as sick children dying from collapsing health services. The IBC is admittedly an undercount, and should be treated as a reliable lower bound only, as the Lancet says. Other surveillance-based figures from the UN are similar but don't offer the IBC's continuous series. I wish they didn't label their high count "maximum", it isn't anything of the sort.

The serious question is: how large is the IBC undercount? To get a central estimate of deaths, you have to go to the three professional surveys that have been carried out in very difficult conditions. Their totals are global, and include combatants, but it's a truism that most casualties in modem wars are civilians.

24,000 excess deaths, all causes to June 2004 - ILCS

The Iraq Living Conditions Survey (ILCS) was carried out by experienced Norwegian social scientists (Jon Pedersen et al) around April 2004 for the UNDP. It was a broad survey with a large sample in which deaths were only one of many questions. Pedersen has published, separately from the main report, a central estimate of 24,000 excess deaths at that time from all causes since the invasion (95% CI 18,000-29,000). The IBC count then stood at 8, 918. Update the ILCS number from the IBC trend, and you would have 177,000 by this month.

98,000 excess deaths, all causes to September 2004 - first Lancet study

This smaller study, carried out by leading American epidemiologists (Les Roberts et al) from Johns Hopkins, was focused entirely on deaths. Excluding as an outlier a sampling cluster in war-torn Falluja, they estimated 98,000 excess Iraqi deaths from all causes. The very large 95% CI - 8,000 to 194,000 - confused many, and both Bush and Blair rejected the core finding, in spite of strong support from most experts for the methodology. Update this central estimate by the IBC trend, and you would get 547,000 to today.

655,000 excess deaths, all causes to June 2006 - second Lancet study

Undeterred, Roberts and his team repeated the exercise in May- July 2006, with a somewhat bigger sample and a narrower confidence interval. Their new central estimate was 654,965 (95% CI: 392,979–942,636). 393k is above the upper 95% CI for the ILCS, even uprated by the IBC line, so the two findings are inconsistent. Update this central estimate by the IBC trend, and you would get a staggering 1,114,000 to today.

There's a small industry arguing about the Lancet surveys, not counting the rubbish from warbloggers. Tim Lambert's thread on this is up to 143 posts. Reading Lambert's efforts to teach denialists statistics is like watching an experiment on learning in flatworms, except that flatworms catch on faster.

From the qualified, there have been a few esoteric technical criticisms of the experimental design - an alleged "main street" bias in cluster selection - which most of the trade thinks trivial, and anyway can't affect the results much. The serious objections are from Pedersen, who's been there: (a) the prewar mortality declared by Lancet respondents is too low, in the light of thorough child mortality studies; (b) his study was bigger and more reliable. Roberts has answers to this, follow the dispute here. Responses to such surveys are clouded not only by the ordinary issues of grief, anger and fading memory, but by rationing - Iraqi parents who lose a child had and still have an incentive to conceal the death and keep using the ration card. Anger would I suppose bias death reports upwards, except where a death certificate was asked for; floating ration cards would bias downwards. This isn't like asking people about their preferred brand of toothpaste.

In any case the three surveys are consistent in direction and scale - the IBC undercount is large, Iraq is worse than you thought. I've put the four data sources together in a graph. It's a log scale, so the straight line of the IBC count corresponds to an exponential deterioration. I've left out the first few months, as the IBC changed the count afterward, and added an extra month at the other end of the timescale for legibility.

What's the black diamond in the graph? Assuming the three studies are all sound, it seems reasonable to average the ratios of their central estimates to the contemporaneous IBC counts (2.7. 8.3, and 16.9), making 9.3. Applying this to the 21 June 2007 IBC count gives:

613,000 excess deaths, all causes to June 2007, averaged extrapolation

The estimate is conservative in that it leaves out Falluja from Lancet 1, and also assumes the IBC undercount is constant. The trend line of the three surveys suggests that the IBC undercount is growing, as you would expect from the decreasing access of journalists and the growing absolute scale of the violence. I'm sure there are better ways of doing a meta-analysis, but it beats picking a number because you can live with it.

If you still can't believe this - and I have difficulties with it myself - read this interview (via Juan Cole) with a Mahdi Army gunman in Sadr city. Abu Rusil is unhinged, a psycho who wants to ethnically cleanse all Sunnis and kills other Shias en passant for dissing his chief, but he's the kind of man to whom the invasion has handed power.