BizDoc

A few thoughts on just about everything.

Name:
Location: East Coast University, United States

Wednesday, October 11, 2006

Back Talk: About that New Study in Lancet

Back Talk: About that New Study in Lancet

Here's some more discussion. Note that according the this, the Lancet itself reported 567,000 child deaths from sanctions, which seem to have slipped the minds of the current study's authors

The Lancet

OK, it's been a while, but this one had to be answered.
Where to begin. First of all, the study massively underreports pre-2003 deaths. According to several sources, approximately 90,000 people per year died as a direct result of economic sanctions against the country. While these figures are certainly disputable, the current Lancet study indicates a much lower death rate than that implied by the sanctions studies. So it appears that they are starting from an inaccurate beginning position.
Also of great concern is their sampling method. Although they claimed to use random sampling, a reading of the methodology indicates that the sampling was not particularly random. They confined the study to urban areas, and proceeded along streets until they found somebody home, then proceeded to the next 40 homes where somebody was home. Given that conflicts — and the related deaths — tend to be localized, it is not surprising that if they found one conflict related death in a cluster, they found several. This means that the researchers were effectively selecting on the dependent variable. This is a type of research error that should not be made by a first-year graduate student.
Alternatively, each cluster could be considered as a single data point — but then, the sample size would be 47, nowhere near enough to statistically extrapolate anything meaningful. Of course, the starting point of major urban areas is also a form of selection bias which skews the results.
Most of these errors should have been caught if the article had been peer reviewed, but it wasn't. The first cite in the references is dated Oct 2, 2006, less than 10 days before publication of the article. If the article had been peer reviewed, there would be at least several months between the submission of the article and publication, during which other academics would review it for errors, consistency, etc. This clearly did not happen; the authors were still collecting data in July of 2006.
Finally, as a minor point, there is a statement at the end of the article that "We declare that we have no conflict of interest." Yet the principle author of the article admitted that a previous study of Iraqi deaths was released with timing that was specifically selected to influence the 2004 presidential election. There are other types of conflict of interest besides financial.
I'm embarrassed, sometimes, that I'm an academic.