Analysis of Statistical Power Reconciles Drought-Conflict Results in Africa

Whether changes in climate substantially shape human conflict is a question of considerable recent interest to both academics and policymakers. Despite a substantial body of evidence for a strong association between climate and conflict, it remains widely claimed that large-sample empirical evidence linking climatic conditions and modern human conflict in Africa is mixed and thus any reported evidence for a strong association should be discounted. Theisen, Holterman, and Buhaug (henceforth THB) is one of the studies used to support this claim. Here we show that the results in THB are not inconsistent with earlier studies that report a substantial effect of of climate on conflict. We demonstrate using power calculations and Monte Carlo simulations that even if a large association between climate and conflict existed in the data, the approach of THB would not be able to reliably distinguish this association from a null effect, indicating that the approach taken by THB is statistically underpowered in this context. Therefore THB's analysis provides no basis for discarding earlier analyses and THB's conclusions drawn from this analysis overstate the extent to which they disagree with the literature. We also demonstrate that THB's stated advantage from using exceptionally high resolution data is unlikely to be realized in their analysis, since high resolution rainfall data was not actually collected in the majority conflict zones studied by THB. Although unremarked in the original analysis, the rainfall data in THB are interpolations of sparse and incomplete rainfall measurements.