The Duration of Civil Wars: Reassessing the Evidence
Recent studies of the determinants of conflict duration have produced widely
divergent results. We argue that this divergence can be traced to (i.)
different assumptions about time dependence inherent in conventional
statistical models of duration (ii.) the general lack of comparability across
datasets employed in the study of conflict length. In this paper, we seek to
reconcile previous findings by addressing both of these issues. On the
methodological side, we implement a generalized gamma model, which encompasses
most traditional duration models as special cases, without making strong
assumptions about the time dependence of conflict. We also extend the dataset
of Collier et al. (2004) to include key variables proposed by other leading
studies of conflict duration. Our results are generally compatible with earlier
findings. However, using simulation, we find the magnitude of the effects to be
quite different.