Sunday, January 13, 2008

Lancet is back in the news!

a new mortality study on Iraq just brought back the discussion about the Lancet papers. (Lancet 2004 and lancet 2006)

the study
"Violence-Related Mortality in Iraq from 2002 to 2006" published in the "new england journal of medicine"
is used by the usual suspect (lead by one David Kane) and in right wing editorials (WSJ) as a contradiction of the Lancet results.
the death estimate by violent cause in the new study (151000, 95% uncertainity range 104000-225000) is lower that that found in Lancet 2 (601000, confidence interval ranges from 426,000 to 793,000)

while the NJoM study finds a lower number of violent deaths, the result is still a shockingly high mortality. an interesting part of the new paper is, that security did not allow polling (mostly in 2006) in some clusters in Anbar and Baghdad.
the authors chose to reconstruct the mortality in those clusters by using the IBC numbers, a dead count based on reports in news paper articles.
the general tone of the paper tends to be more positive toward the IBC (a project using a very different method, producing definetly an undercount) and slightly sceptical of the Lancet results (a scientific study done in a very similar way).

a few talking points to notice are: (good discussion, as always, can be found on deltoid)

1. the numbers of the NJoM are in good agreement with the Lancet 1 numbers for the early period of the war.

2. while the paper finds a smaller increase in violent deaths than the Lancet 2 paper, it shows a masiive increase in the rate of non-violent deaths (doubled deathrate, some calculations lead to an estimate of 400000 total excess deads, in comparison with a total of 650000 in the lancet 2)

3. the paper does not show an increase in dathrate after the Samarra bombing and in early 2006. this is extremely strange, as the increase in violence was even registered by the US military and lead to the surge.

4. the mortality results are a small part of a huge survey about health in iraq. the questions fill about 20 pages, the relevant part being on page 16.

more soon.