Survey Estimates of Wartime Mortality

Many scholarly literatures require mortality rates from conflict zones, but accurate information is usually among the earliest casualties of war. While political scientists typically obtain mortality data from the news media or others, much progress has been made in demography, epidemiology, and public health conducting original surveys about the survival of siblings, friends, or others known to respondents. Unfortunately, the formal properties of estimators based on these surveys have not been established, the intuitions offered for them (and consequent data analysis strategies) are conflicting, and the statistical consequences of the political incentives of respondents in conflict zones remain unexamined. In this paper, we demonstrate the advantages of joining ongoing efforts in these other fields with insights from political science, including especially political methodology, international relations, and comparative politics. We offer the first formal proofs of the statistical properties of all existing estimators, along with simulation and empirical illustrations, to craft simple intuitions to guide best practices. We also build practical data analytic approaches, based on modern robust statistical methods, for when some respondents are suspected of intentionally biasing answers for political, military, or other strategic purposes. We offer practical advice for producing more complete and accurate mortality inferences for scholarship in our discipline and beyond. Software to implement all the methods in this paper, TrimSib, is available here.