Saturday, May 07, 2005

Expert Non-Witness

Epidemiologist Debarati Guha-Sapir has repeatedly criticized high estimates of the Darfur death toll put forward by advocates such as the Coalition for International Justice and Eric Reeves. I agree with Guha-Sapir that it's of the highest importance for us to know approximately how many people have been killed, and to not let our judgement be clouded by mispresentations of the facts in this matter. So what do we know about the actual death toll in Darfur, more or less?

Reeves puts the number killed at about 400,000, based on his rather methodically laid out interpretations of available data, which he meticulously cites. One can easily criticize particular assumptions or extrapolations Reeves makes, or one can, as Guha-Sapir does, make one's criticisms more elliptically. There is the obvious point that Reeves is neither a scientist nor unbiased. Of course that does not mean that he is ill-informed, inaccurate, or inherently unreliable, but an ordinary healthy skepticism tells us that if one is unable or too lazy to do one's own analysis of the data, then his conclusions should be regarded cautiously, especially in cases where acknowledged experts disagree with them.

Unfortunately for Guha-Sapir's argument, Reeves conclusions are not especially inconsistent with the scientific data, and some experts have looked at the data and confirmed Reeves' high estimates of the death toll. Furthermore, since we have opened up the question of bias, it may be worth noting that the group of experts who express the most reticence about providing global mortality estimates tend to have a professional stake in the business of providing humanitarian aid, as does Guha-Sapir. Their reticence may be motivated by factors such as a need to demonstrate the effectiveness of an aid agency, or the need to avoid taking sides in political disputes, especially in countries like Sudan where the government is apt to expel, detain, harass or otherwise impede the work of humanitarians. Of course that does not mean that reluctance to provide estimates is not reflective of a genuinely scientific approach to the problem. However, it may be both scientific and biased, and we should therefore be cautious about accepting this opinion as the truth of the matter, especially since other experts disagree.

One study that has been cited by Reeves (and a handful of journalists) as confirming his dire view is the one done by Dr. Jan Coebergh, "Sudan: Genocide has Killed more than the Tsunami," in Parliamentary Brief, Feb 2005 (large pdf, also reposted as html without graphics here). Coebergh I don't believe is completely unbiased, but he has done us all a service in constructing three alternative models for interpreting the available data, and offering a reasoned defense of each. Regarding the death toll between February 2003 and December 2004, Coebergh concludes:

Many people have died as a result of conflict in Darfur. Figures of 218,449, 253,573 or 306,130 deaths since April 2003 are very rough estimates, but considerably more accurate than the numbers currently quoted in the media.

And which ever total you go with, it is an ever increasing one. This year looks worse than last.

Reeves' estimate for that time period was 370,000 dead, 64,000 more than Coebergh's highest figure, which is the one Coebergh favors, and 152,000 more than the lowest estimate. The differences--duly noted--seem rather huge, and yet compared to the estimates repeated in the news media at that time (variously "as many as 70,000," "up to 50,000" or "tens of thousands"), Reeves and Coebergh at least appear to be discussing the same phenomenon.

Still, I hestitate to accept Reeves' estimate as even roughly accurate without considering all of the arguments put forward by Guha-Sapir which call into question the extrapolations of Reeves, the Coalition for Justice, and also those used by Coebergh to obtain his highest estimate. A primary point of contention concerns the ratio of people killed by outright violence as opposed to starvation and disease attributable to unsanitary conditions in camps for the displaced. Guha-Sapir expresses the view held among some social scientists that normally in situations of armed conflict deaths due to starvation and disease greatly outnumber deaths due to violence. This begs the question of whether the situation in Darfur can be accurately descibed as essentially an armed conflict, much less a normal armed conflict. To support her view, Guha-Sapir adds that only 15% of the people killed in the Rwandan genocide were killed by outright violence. That's an extremely remarkable claim, as the commonly accepted estimate for the number of people butchered in the Rwandan genocide is about 800,000, and it seems unlikely Guha-Sapir is claiming that an additional 4.5 million people died in the events that followed the hundred days of mayhem (see Rwanda census puts genocide death toll at 937,000, Leave None to Tell the Story: Genocide in Rwanda: The Numbers from Human Rights Watch, Oxfam: What Happened in Rwanda in 1994?, Ghosts of Rwanda: Timeline). Reeves incidentally has cited and discussed the report from Human Rights Watch. Guha-Sapir cites no sources. On this point, then, Guha-Sapir's particular expertise may actually be preventing her from accurately assessing the facts more than it helps her to clarify them.

Nevertheless, there are a couple of respects in which I believe both Reeves and Coebergh may be unnecessarily leaping to conclusions about the number of deaths due to outright violence. In the first place, both examine the WHO retrospective mortality survey and assume that "the WHO figure did not include violent deaths." Reeves attests that Dr. Nabarro had confirmed this fact to him, and indeed Dr. Nabarro has said as much to the press. Nonetheless, my reading of the survey is that deaths due to injury or violence were in fact tallied. Thus the exercise of adding to those figures an additional figure of deaths due to outright violence would appear to be erroneous. (All of the other caveats Reeves adds to that report I agree with.) Alternatively, the report itself may be in error, or I may be missing something obvious. As it is, I cannot fully accept the interpretation offered by Reeves.

In the second place, Reeves' extrapolations from the survey by the Coalition for International Justice rely upon an assumption of average family size. As the AP reported, "projecting a precise death toll estimate from the survey is problematic because there is no certainty about the size of the group each refugee would consider to be 'family' — a key element in the calculation. Refugees included extended family — such as uncles and cousins — in their answers, said Stefanie Frease of the Washington, D.C.-based nonprofit Coalition for International Justice, which conducted the survey with the U.S. State Department." Reeves' response to that criticism misses the point, in my view, although he is certainly correct to raise the other issues. Since Reeves projects in terms of deaths per family, it makes a huge difference whether one million people represent 200,000 families as according to his figures, or 85,000 families, for example.

For these reasons, I regard the high estimate offered by Coebergh and the estimate offered by Reeves with some skepticism. Conversely, their willingness to grapple with all of the available data concerning deaths due to outright violence has shown that death tolls improperly extrapolated from limited samples are clearly inadequate to convey the magnitude of the tragedy in Sudan. But if we would do well to regard all current mortality estimates as somewhat doubtful, surely there must be some range of doubt that can be applied to keep us from sliding into pure cynicism or ignorance.

One other issue raised by Guha-Sapir concerns the dangers of extrapolating from non-representative samples. That's generally a valid concern, however, by choosing to argue on the basis of anectdotal or hypothetical evidence rather than engage with the actual data, Guha-Sapir gives the impression that those who give high estimates are not basing their inferences upon statistically significant samples. That's simply not the case. Guha-Sapir mentions the work CEDAT (Complex Emergency Database) has done to survey mortality and morbidity in Darfur, and claims that they are "piecing together more than 30 survey results." The only CEDAT analysis of mortality in Darfur (pdf) of which I'm aware presents a table of mortality data from 12 samples collected by various agencies at various locations in Darfur. This is a smaller data set than Reeves et al. work with, but it nevertheless provides a basis for drawing reasonable conclusions about the mortality rate among Darfuris who have been displaced into camps administered to by international relief agencies. The average CMR (crude mortality rate, measured in number of deaths per 10,000 people per day) for all twelve samples was 2.65, while the median CMR was 2.15. There appears to be no evidence of improvement or deterioration over time, but rather unexplained variance. Therefore, I see no compelling reason not to regard the average CMR of 2.65 as more or less representative of mortality among Darfuris residing in displacement camps. (The median value tells us about the conditions at a typical camp, but not about the overall mortality rate.)

How many Darfuris have been displaced? Conservatively, there are now more than 2 million Darfuris internally displaced, and several hundred thousand who have fled into Chad. If we assume that (a) conditions have neither substantially improved nor deteriorated over the last nine months --my sentiment is pessimistic although there are contrary indications; (b) that mortality inside camps accessible to aid workers is equivalent to mortality in areas aid workers cannot reach, which is preposterous but warranted only insofar as actual mortality in such places is unmeasured; and (c) that excess deaths among non-displaced Darfuris are not due to same causes that displaced 2 million people into camps; then based on the CEDAT data we can reasonably estimate that the displaced people inside Darfur are dying at a rate of 530 per day, or about 16,000 per month. That number would be higher if we add in the refugees in Chad, although the CMR in those camps is probably a bit lower.

In sum, Reeves' estimates of total mortality in the Darfur genocide seem a bit high to me, but because he attempts to methodically account for every single person affected by the violence, he achieves a greater verisimilitude than many who have confined their projections to the small universe of people in camps accessible to aid workers, greater than those who have wrongly taken those projections as representative of Darfur as a whole, and far greater than those who have turned their backs on strong, statistically meaningful evidence of deaths immediately caused by violent attacks. We have seen over the past year abundant evidence of episodes of extreme violence, during which many people are killed, many of whom may not be well accounted for if at all. I fear that a large portion of the number of people reported missing have in fact been murdered.

My own best sense of how many have died is 300,000, give or take 100,000. I freely admit that this estimate is based upon attitudes, prejudices, assumptions and interpretions, rather than a rigorous analysis of statistical data, although I have also looked at the data and drawn conclusions from it. I don't believe at this time that precision is entirely warranted. I do however strongly feel that serious observers have an obligation to exercize their judgement and to the best of their abilities bear witness to the atrocities that are taking place in Sudan. I commend those who have done so.

I cannot fully comprehend why an epidemiological expert such as Guha-Sapir would suggest that Robert Zoellick's ridiculously low estimate is acceptable, whereas the high estimates provided by Reeves and others are misleading and unhelpful. She offers no real estimates of her own. Perhaps the WHO Collaborating Centre for Research on Epidemiology of Disasters will see fit to put online its Darfur mortality data instead of Guha-Sapir's opinion pieces, but as that work was described by Guha-Sapir in the same terms as the work published by CEDAT, I doubt whether it would truly back up Guha-Sapir's claim. What reasons and data she has put forward might tend to justify Zoellick's highest estimate of 160,000 dead, were she to make the case. But she doesn't. Guha-Sapir rests her case on her expert authority, without fully applying her expertise to the most urgent task at hand: telling the world approximately how many people have died in Darfur.

posted by Fido the Yak at 3:27 PM.


Post a Comment

Fido the Yak front page