Using Front Lines to Predict Deaths in the Bosnian Civil War

Abstract

What static, readily observable indicators might help us predict levels of killing in civil wars? Research using disaggregated event data and spatial models has reported strong relationships between various indicators of ethnic diversity and violent conflict during the Bosnian War (1992-1995). Bosnia is a rare case for empirical research on ethnic violence in civil wars: a census was conducted a few months before violence broke out in a country which recognized and recorded ethnic group membership, and extensively documented post-war fact-finding has recorded over 75,000 deaths and their location. Applying the resulting ethnic conflict models to other conflicts that lack such data will be difficult. This paper examines the relative fit of an alternative model based on distance to front lines and other factors that can be obtained through remote sensing, and which hence are more generalizable. Results from a spatial negative binomial count model estimated with Markov Chain Monte Carlo (MCMC) methods show that front lines model performs similarly to the ethnic conflict model in terms of fit, and both are improvements over a base model which accounts for spatial diffusion of violence and local population size only.

Keywords: Bayesian, spatial, event count, model fit, Bosnia, civilian deaths