Apocalypse now? - Climate change and war in Africa

There is a large empirical literature trying to quantify the potentially adverse affects of climate change on the risk of violent armed conflict, which focuses almost exclusively on linking annual variation in climatic conditions to violence. A major shortcoming of this approach is that it conflates climate variability with climate change, while also implicitly assuming that adverse weather shocks will immediately trigger violent contests over scarce resources. In contrast, this study exploits changes in local climate over a longer time period; using differences in the average standardised deviation of temperature and precipitation levels between 1989-2002 and 2003-2017 across the African continent. Bayesian model averaging is used to test whether variables measuring changes in local climate contribute consistently in explaining conflict risk between 2003-17. Using disaggregated data to account for local dynamics, the reduced-form estimation shows that temperature is robustly linked to violent armed conflict: moving from low to high temperature levels corresponds to a 31% increase in conflict risk. Changes in precipitation have no discernible effect. The results are robust to changing the benchmark period for the climate variables, accounting for conflict prevalence, and considering different types of violent conflict. Examining the predictive power of the models, a leave-one-out cross-validation highlights that including information on changes in local climate improves the predictive performance of the model, as measured by the area under the precision-recall curve, by seven points, from 0.51 to 0.58; 33 points above the baseline.