Why do existing studies produce conflicting results on whether violence is more prevalent in middle-range regimes? This article examines multiple violent outcomes like civil war, terrorism, and repression across various operationalizations of 'the middle.' Using a novel ensemble method combining random forest and regression/classification trees to predict conflict patterns, we identify specific conditions where the hypothesis holds for minor civil conflicts but fails regarding state repression. Our findings clarify previous ambiguity by pinpointing when regime instability leads to violence.
