Conflict Mortality Surveys

These pages give assorted material directly relevant to the study of Iraq mortality published recently in The Lancet. The material also applies to conflict surveys more generally. The specific reference is to "Mortality after the 2003 invasion of Iraq: a cross-sectional cluster sample survey" by Professor Gilbert Burnham MD, Professor Riyad Lafta MD, Shannon Doocy PhD and Les Roberts PhD, The Lancet 12 October 2006. For simplicity in these pages we will refer to this as "The Lancet Study" and to the authors as "the researchers".

Guide to these pages

Updated March 2009: Bias Paper gives a paper we have written entitled "Bias in epidemiological studies of conflict mortality". This paper won the Article of the Year award from the Journal of Peace Research.

Visual Summary of Main-Street Bias reproduces the summary of the final stage of the sampling methodology given in The Lancet paper, illustrates its meaning through a schematic diagram and gives a satellite image of central Baghdad to allow readers to digest its practical implications.

Follow-up Paper links to a second paper developing the main-street bias idea that we have published in European Physics Letters.

FAQs gives the answers to some frequently asked questions.

L1 versus L2. "L1" is shorthand for the 2004 study of Iraq mortality published in the Lancet. The reference is "Mortality before and after the 2003 invasion of Iraq: cluster sample survey" by Les Roberts, Riyadh Lafta, Richard Garfield, Jamal Khudhairi and Gilbert Burnham, The Lancet, October 29, 2004. "L2" is short for the October 2006 study referred to as "the Lancet study" elsewhere on these pages. We argue that the claims of L2 authors Les Roberts and Gilbert Burnham that L1 and L2 give "nearly identical" results do not withstand scrutiny. A comparison of L1 and L2 does not rule out the possibility of substantial upward bias in L2 relative to L1.

Methods discusses the literature on the sampling methods used in The Lancet Study and other mortality studies, arguing that these are in their early stages of development as applied to conflict mortality and more research is required to understand their properties.

Urban Homicide Rates Around the World pulls together information on homicide by neighborhood from a number of different cities. The point is that violence levels can change dramatically block by block. This means that the findings of an urban mortality study can be highly sensitive to precise sampling details. Of course, an inability to reach large swathes of a territory is problematic for a mortality study but this is not our main point here. Consider a study that can reach every important zone of a city. Still, in order to ensure a representative sample, sampling must be able to penetrate to all the little tucked-away places within each zone.

Iraq Maps combines the final stage of the sampling methodology described in The Lancet paper with maps of both urban and rural areas of Iraq to display the kinds of areas that seem likely and unlikely to be sampled under this methodology. We find many spatially fragmented islands of households lying outside the survey space.

Control and Danger discusses a special aspect of the researchers' methodology: field teams had discretion to choose secondary locations for their interview clusters if they believed it was too dangerous to proceed in the originally selected location. We argue that this practice would not necessarily introduce downward bias into the mortality estimate. The bias could work in either direction.

News gives some stories from the media that we feel are pertinent for the issue of sample selection bias discussed in these pages.

This web site can be seen as an elaboration on an earlier press release. The press release still holds up fairly well, although the discussion has moved on in some respects.