Determinants of Civil War and Excess Zeroes

This paper considers the determinants of civil conflict, using a zero-inflated modelling approach that deals with the problem of excess zero observations, which we argue are related to two distinct data generation processes. Despite their continued use in the literature, traditional probit and logit models have limited capacity in dealing with this issue and can create misleading results. This is illustrated by estimating the model in Elbadawi and Sambanis (2002) using their data and a zero-inflated modelling procedure, which leads to results that suggest a role for the grievance variables in contrast to the original article. A general greed-grievance model is then estimated on a sample of 134 countries, over 54 years. Again, while the standard probit model results tend to emphasise opportunity variables, as found in other studies, the zero-inflated model gives more support for grievance effects. In particular, polity, ethnicity and inequality are found to play a significant role in contrast to earlier studies.