Promises and Pitfalls in the Spatial Prediction of Ethnic Violence

Despite increasing technical sophistication, the quantitative literature has made little progress in forecasting ethnic violence. Nevertheless, recent efforts in predicting the location of ethnic violence from the spatial ethnic distribution seem to be a major step forward. In 2007, Lim, Metzler, and Bar-Yam proposed an agent-based model that takes as input the ethnic map of a country and derives from it the predicted locations of ethnic violence. The model rests on the assumption that spatial group clusters of a certain critical size are most likely to display ethnic violence. Their model achieves a remarkable level of agreement between predicted and observed locations of violence. Our article scrutinizes this exercise. We show that their analysis suffers from a biased selection of groups and regions, and that the null hypothesis and unit of analysis are inadequate. The proclaimed usefulness of the model for predicting violence in new cases is made difficult by the fact that the model does not generalize from one case to another. We conclude that the model provides little advance on prior research.