Forecasting Onsets of Mass Killing

This paper describes the first stage of a statistical analysis designed to help forecast onsets of mass-killing episodes in countries worldwide. This analysis is part of a study funded by the U.S. Holocaust Memorial Museum to assess the feasibility of building a public early-warning system for mass atrocities. The results of this preliminary analysis suggest that simple versions of familiar statistical techniques -- in this case, logistic regression -- will probably produce forecasts that are about as accurate as ones based on more complex techniques, including Bayesian model averaging and Random Forests.