Census-based multiple system estimation as an unbiased method of estimation of casualties’ undercount
Jan Zwierzchowski, Warsaw School of Economics
Ewa M. Tabeau, ICTY
There are two distinct approaches to estimating war casualties. The first one (passive surveillance) is based on merging individual level sources on war deaths and excluding overlap. It undercounts war victims, as not all deaths are recorded; undercount can be, however, estimated statistically. The second one relies on post-conflict, retrospective surveys, which are criticized as the construction of reliable sample is usually not possible. Recently, the first approach has been improved by applying a multiple system estimation (MSE) for estimating undercount on the victims’ lists. MSE requires, however, perfect matching of mortality sources. This condition is usually violated, as records collected during war contain errors, precluding reliable and full matching of sources and overestimating the undercount. This paper proposes a modification of the passive surveillance approach. Sources on war deaths are additionally matched with the pre-war population census. Death records are thus validated and split into matched (high quality) and unmatched (deficient). The overlap structure of matched records is transmitted onto the unmatched. The perfect matching assumption is then no more violated and the MSE undercount estimate is unbiased. We used the new approach to estimate the number of casualties of the 1992-1995 war in Bosnia and Herzegovina. All reliable lists of war deaths, accessible to the OTP, ICTY were integrated and the resulting lists matched with the 1991 Population Census (matching rate of 90%). Additionally, the lists were matched with each other and the overlap structure was obtained. 89,186 unique death records were extracted and the undercount was estimated at 15,546 (SD=2987), resulting in the total number of casualties of 104,732, which is consistent with the previous ICTY estimate. Has the correction of overlap not been used, the undercount estimate would equal 78,240 (SD=2876), and the total number of war related deaths - 167,426, which would be highly overestimated.
Presented in Session 71: Mortality measures and models