Gross Distortions, Sloppy Methodology and Tendentious Reporting
How the New England Journal of Medicine Undercounted Iraqi Civilian Deaths
By ANDREW COCKBURN
Almost five years into the destruction of Iraq, the orthodox rule of thumb for assessing statistical tabulations of the civilian death toll is becoming clear: any figure will do so long as it is substantially lower than that computed by the Johns Hopkins researchers in their 2004 and 2006 studies. Their findings, based on the most orthodox sampling methodology and published in the Lancet after extensive peer review, estimated the post-invasion death toll by 2006 at about 655,000. Predictably, this shocking assessment drew howls of ignorant abuse from self-interested parties, including George Bush (“not credible”) and Tony Blair.
Now we have a new result complied by the Iraqi Ministry of Health under the sponsorship of the World Health Organization and published in the once reputable New England Journal of Medicine, (NEJM) estimating the number of Iraqis murdered, directly or indirectly, by George Bush and his willing executioners at 151,000–far less than the most recent Johns Hopkins estimate. Due to its adherence to the rule cited above, this figure has been greeted with respectful attention in press reports, along with swipes at the Hopkins effort as having, as the New York Times had to remind readers, “come under criticism for its methodology.”
However, as a careful and informed reading makes clear, it is the new report that guilty of sloppy methodology and tendentious reporting — evidently inspired by the desire to discredit the horrifying Hopkins findings, which, the NEJM study triumphantly concludes “considerably overestimated the number of violent deaths.” In particular, while Johns Hopkins reported that the majority of post invasion deaths were due to violence, the NEJM serves up the comforting assessment that only one sixth of deaths in this period have been due to violence.
Among the many obfuscations in this new report, the most fundamental is the blurred distinction between it and the survey it sets out to discredit. The Johns Hopkins project sought to enumerate the number of excess deaths due to all causes in the period following the March 2003 invasion as compared with the death rate prior to the invasion, thus giving a number of people who died because Bush invaded. Post hoc, propter hoc. This new study, on the other hand, explicitly sought to analyze only deaths by violence, imposing a measure of subjectivity on the findings from the outset. For example, does the child who dies because the local health clinic has been looted in the aftermath of the invasion count as a casualty of the war, or not? As CounterPunch’s statistical consultant Pierre Sprey reacted after reading the full NEJM paper, “They don’t say they are comparing entirely different death rates. That’s not science, it’s politics.”
Superficially at least, both the Hopkins team and the new study followed the same methodology in conducting their surveys: interviewing a random sample of households drawn from randomly selected “clusters” of houses around the country, in which the head of the household was interviewed. While the Johns Hopkins team demanded death certificates as confirmation of deaths and their cause, the NEJM study had no such requirement. That survey was based on a sample of 9345 households, while the 2006 Johns Hopkins report drew on a sample of 1849 households. In reports on the NEJM study, much respectful attention was paid the fact that their sample was bigger, which, uninformed reporters assumed, had to mean that it was more accurate. In fact, as their papers’ own pollsters could have told them, beyond a certain point the size of a sample makes less and less difference to the accuracy of the information, with accuracy increasing as a factor of the square root of the ratio between the two.
Far, far more important than the size of the sample, however, is the degree to which the overall sample is truly random, that is, trule representative of the population sampled and here is where the first of many serious questions about the NEJM effort arise. As the authors themselves admit, they did not visit a significant proportion of the original designated clusters: “Of the 1086 originally selected clusters, 115 (10.6%) were not visited because of problems with security,” meaning they were inconveniently situated in Anbar province, Baghdad, and two other areas that were dangerous to visit, (especially for Iraqi government employees from a Shia-controlled ministry.) While such reluctance is understandable–one of those involved was indeed killed during the survey–it also meant that areas with very high death tolls were excluded from the survey.
To fill the gap, the surveyors reached for the numbers advanced by the Iraqi Body Count, (IBC) a U.K. based entity that relies entirely on newspaper reports of Iraqi deaths to compile their figures. Due to IBC’s policy of posting minimum and maximum figures, currently standing at 80,419 and 87,834, their numbers carry a misleading air of scientific precision. As the group itself readily concedes, the estimate must be incomplete, since it omits deaths that do not make it into the papers, a number that is likely to be high in a society as violently chaotic as today’s Baghdad, and higher still outside Baghdad where it is even harder for journalists to operate.
Nevertheless, the NEJM study happily adopted a formula in which they compared the ratio between their figures from a province they did visit to the IBC number for that province, and then used that ratio to adjust their own figures for places they did not dare go. Interestingly, the last line of the table on page 8 of the Supplementary Appendix to the report, “adjustment for missing clusters using IBC data,” reveals that in using the Body Count’s dubious figures to fill the holes in their Baghdad data, the formula they employ actually revises downward the rate of violent deaths on what they label “low mortality provinces.”
A paragraph in the published abstract of the report, blandly titled “Adjustment for Reporting Bias” contains an implicit confession of the subjectivity with which the authors reached their conclusions. As Sprey points out, “they say ‘the level of completeness in reporting of death was 62%,’ but they give no real explanation of how they arrive at that figure.” Les Roberts, one of the principal authors of the Johns Hopkins studies, has commented: “We confirmed our deaths with death certificates, they did not. As the NEJM study’s interviewers worked for one side in this conflict, [the U.S.- sponsored government] it is likely that people would be unwilling to admit violent deaths to the study workers.”
The NEJM does cite an effort to check information given by the household heads by also interviewing household daughters about any deaths among their siblings. Again, this data is rife with inconsistencies, particularly in the siblings’ reports of pre- and post- invasion deaths, incocnistencies egregious enough thatthese interview results were not folded into the calculations used to determine the report’s conclusions.
Further evidence of tendentious assessment surfaces in the section blandly titled “Response Rates” in which the authors report that “Of the households that did not respond, 0.7% were absent for an extended period of time, and 1.1% of households were vacant dwellings.” Given current Iraqi conditions, houses are likely to be vacant, or their owners absent for long periods, because something nasty happened there–i.e. higher death rates. Yet, as Sprey points out, there is no effort by the authors to account for this in their conclusions.
As a statistician, Sprey is most affronted by the enormities committed under the heading “Statistical Analysis” in the NEJM paper, where it is stated, “Robust confidence intervals were estimated with the use of the jackknife procedure.” The “confidence interval” cited in the report is 104,000 to 223,000 with a 95% uncertainty range. This does not mean, as many laypeople assume, that there is an 85% chance that the “true number” lies somewhere between those two figures.
Sprey explains its true meaning this way:
“If you went out and did the same study using the same methods and the same size sample, but with different households, a thousand times, then 950 of those studies would come up with a figure between 104,000 and 223,000. But the ‘jackknife’ they refer to is simply a procedure by which you don’t rely on data to estimate the confidence interval.
They admit in “Statistical Analysis” that their confidence interval is simply a calculation based on their numerical guesses quantifying the unknown size of the three key uncertainties in their survey: the real death rates in clusters they didn’t dare visit; the percentage of unreported deaths in their sample; the proportion of the Iraq population that has fled since the invasion. So this is a computerized guess on what the confidence interval might be without using any of the actual data scatter from their survey. This in sharp contrast to the Johns Hopkins team, who rigorously used the data of their survey to estimate their confidence interval. To call it ‘robust’ is simply a disgrace. It’s not robust, it’s simply speculation.”
If any further confirmation of the essential worthlessness of the NEJM effort, it comes in the bizarre conclusion that violent deaths in the Iraqi population have not increased over the course of the occupation. As Iraq has descended into a bloody civil war during that time, it should seem obvious to the meanest intelligence that violent deaths have to have increased. Indeed, even Iraq Body Count tracks the same rate of increase as the Hopkins survey, while NEJM settles for a mere 7% in recent years. As Roberts points out: “They roughly found a steady rate of violence from 2003 – 2006. Baghdad morgue data, Najaf burial data, Pentagon attack data, and our data all show a dramatic increase over 2005 and 2006.”
These distortions come as less of a surprise on examination of page 6 of the supplementary appendix, an instructive table that reveals that the 279 men and women engaged in collecting data for the survey labored under the supervision of no fewer than 128 local, field, and central supervisors and “editors.” Senior supervisors were shipped to Amman for a training course, though why the Iraqi government should want to send its own officials abroad for training is unexplained, unless of course some other government wanted a hand in the matter.
Finally, there is the matter of the New England Journal of Medicine lending its imprimatur to this farrago. Once upon a time, under the great editor Marsha Angell, this was an organ unafraid to cock a snoot at power. In particular, Angell refused to pander to the mendacities of the drug companies, thereby earning their undying enmity. Much has evidently changed, as the recruiting ad for the U.S. Army on the home page of the current New England Journal reminds us.
Andrew Cockburn is the author of Rumsfeld: His Rise, Fall and Catastrophic Legacy.