Predicting State Failure: A Classification Tree Approach

Abstract

We deviate from standard practice in the literature on state failure and instead use classification tree methods to predict state failure. We argue that the rarity of state failure, along with classification tree’s simplicity of use and interpretation makes this approach more attractive than existing methods. Our dataset includes a comprehensive list of the correlates of conflict variables according to the current literature. We identify an objective and measurable hierarchy of a subset of these variables as necessary for predicting state failure. Further, our methodology suggests that different types of states can have very different reasons for failing, reasons that straightforward regression techniques, with their focus on marginal analysis, simply cannot identify. There are different pathways to state failure. In fact, our approach rank orders the likelihood that a particular pathway will lead to state failure. Last, we suggest that our methodology is more accurate than current attempts at predicting state failure. There are many roads into the abyss. We provide a road map.