This article aims to demonstrate that differences in the two major datasets can significantly affect the results of predictions of mass political killing. Mass political killing (such as Hitler's killing of some six million Jews, or the Rwanda genocide of 1994) has been studied for decades with the aid of valuable datasets measuring 'democide' and 'genocide and politicide', respectively. Without attempting to take sides as to whether one or the other is a more valid measure of the phenomenon of mass political killing, the authors aim in this investigation to see what independent variables best account for the onset of mass political killing, with the state-year as the unit of analysis. The predictor variables are level of economic development; types of war and violent unrest short of war; and regime type. By using a Cox proportional hazard model, the authors find that important regime effects either appear or disappear depending on the dataset used, with regime generally having a significant effect on onset of democide, but not having a significant effect on onset of geno-politicide. It is important for the scholarly community to be aware of these dataset effects, which may be the source of some of the most important existing controversies in the literature on explaining mass political killing.
