Since its establishment in 1963, the Correlates of War (COW) project has sought to build cumulative knowledge about international conflict through the application of the scientific method to the study of militarized interstate behavior. Early analyses from the COW project found substantial variation in the causal model of war across the nineteenth and twentieth centuries, but COW scholars later sought to develop a general model of war that avoided post hoc historical periodization. We use out of sample cross validation to evaluate the plausibility of assuming temporal homogeneity for statistical models of international conflict that span the nineteenth and twentieth centuries. Our results suggest that the causal model of war changes substantially across historical eras. In particular, great care should be taken in generalizing Cold War findings to other historical eras. Our findings demonstrate the importance of exploring temporal variation in the causal model of war.