Did rainfall shocks cause civil conflict in Sub-Saharan Africa? The implications of data revisions

Abstract

In their seminal paper, Miguel et al. (2004) found that negative rainfall shocks (measured as negative year-on-year rainfall growth) had caused civil conflict in sub-Saharan Africa over the 1981–1999 period. Since then, the rainfall and conflict data they used had undergone multiple revisions. We show that rainfall shocks are no longer statistically significant for civil conflict when the revised data are used. This is true whether we employ a different functional form for rainfall, extend the sample to include more recent observations, use longer lags for rainfall shocks, employ dynamic panel regression, or panel regressions that take into account of cross-sectional dependence. Using rainfall shocks as instruments for growth, we also find that growth is insignificant for civil conflict if the revised data are used. Upon further investigation, we find that updates in the rainfall and conflict data for one or a few countries may alone cause rainfall shocks to lose statistical significance.