Learning Conflict Duration: Insights from Predictive Modelling

• In Event: New Methods in the Study of Conflict Dynamics

Abstract

Why do some conflicts last longer than others? Previous work on conflict duration posits information asymmetries and credible commitment problems can cause protracted civil wars. The bifurcated nature of conflict studies, based on the notion that civil and interstate wars are qualitatively different, has so far prevented studies from including both types of conflict in the same dataset. Thus, empirical evidence is lacking as to whether the explanations apply to both types of conflict, or they are indeed separate phenomena.

This paper expands on the Cunningham and Lemke (2013) study on combining civil and interstate wars by including a large number of predictors taken from the rich civil war literature. The proposed framework unpacks the bargaining failure framework into three components governing power projection over distance, which I argue to be the main determinant of duration: material capability, politics, and geography. In doing so, I do not discriminate between the ‘types’ of war and provide a general theory of conflict duration.

I empirically test the general theory using a diverse set of advanced quantitative methods. More specifically, I employ predictive modelling techniques such as machine learning, deep learning, and ensemble methods to demonstrate that the majority of predictive covariates of war duration are indeed common to both civil and interstate wars. Further, in most cases, the direction of the effect holds across types, suggesting that the underlying mechanism operates in a similar fashion. Taken together, I contribute to the rationalist literature by providing novel empirical evidence showing that a unified model can explain the duration of both types of war.

Author

• Gokhan Ciflikli, London School of Economics