Predicting Mass Killings: Enhancing Theoretical Value of Models Through Complementary Techniques

This dissertation attempts to improve accuracy and utility of forecasts of mass killings by proposing two ways in which forecasts can be reconnected to theory. The first solution offered is using Confirmatory Factor Analyses (CFA) to build measures of popular predictors in forecasts. The second solution is a demonstration of four post-hoc methods for extracting insights into black box machine learning models. Results indicate that these two solutions can improve forecast accuracy as well as utility to researchers by connecting forecasts to theorized constructs.