Mortality estimation in conflict zones characterized by migration: A method for reducing survivor bias

Background: Mortality caused by a major conflict is often estimated by comparing the mortality rate of the population before and after the date the conflict began, as measured by a retrospective sample household survey. Such a design gives rise to sampling frame bias (survivor bias), since households where all members have left and been subsumed into other households are not included in the frame, and neither are households where all members have died. Our aim is to find an alternative estimation approach that will reduce survivor bias.

Method: When someone dies in a household, remaining members may seek a more secure household environment by migrating to become part of other, existing, households. New subsumed members of sample households can be identified and their past households can be included in the mortality rate calculation. The estimator is a weighted average of the mortality rates of sample households and that of in-migrants' past households. To test this new estimation methodology, we develop a simulation model in which plausible assumptions are made to reflect mortality and migration conditions in contemporary (post-war) Iraq.

Results: Frame bias can be reduced with the new method if household dissolution is mortality-dependent. An adjustment for frame multiplicity is included in the mortality rate calculation.

Discussion: The common occurrence of household dissolution in conflict zones makes prevailing estimation procedures prone to serious underestimation of mortality. A simple modification to questionnaire administration is proposed to take household dissolution into account in estimation.