Predicting Conflict in Space and Time

weidmann10predicting.gifNils B. Weidmann and Michael D. Ward, 2010
Journal of Conflict Resolution 54 (6)

Abstract

The prediction of conflict constitutes a challenge to social scientists. This article explores whether the incorporation of geography can help us make our forecasts of political violence more accurate. We describe a spatially and temporally autoregressive discrete regression model, following the framework of Geyer and Thompson (1992). This model is applied to geo-located data on attributes and conflict events in Bosnia over the period from March 1992 through October 1995. Results show that there is a strong spatial as well as temporal dimension to the outbreak of violence in Bosnia. We then explore the use of this model for predicting future conflict. Using a simulation approach, we contrast the predictive accuracy of the spatial-temporal model to a standard regression model that only includes time lags. Our results show that even in a difficult out-of-sample prediction task, the incorporation of space improves our forecasts of future conflict.

Additional Information

Replication data
Spatial-temporal animation of violence in Bosnia (requires Adobe Reader or Acrobat to play, click on map to start animation)