Latent Civil War: Improving Inference and Forecasting with a Civil War Measurement Model

Abstract

This paper contributes to the substantive study of civil war onset and the methodological literature on handling latent dependent variables. We first introduce a new Bayesian measurement model of civil war status, using eight datasets on civil war status and a dynamic Bayesian IRT to produce measures of latent civil war status. Using the latent data, we then re-analyze canonical work on civil war onset, showing that some previous findings are sensitive to the uncertainty in civil war status. We then turn to our second methodological contribution, introducing a technique (‘posterior bagging’) for building improved forecasting models for civil war onset.