Research attempting to predict repression, including the policing of protest, has tended to rely on pooled time series data, which statistically produces coefficients that estimate the average relationship between each variable and the outcome across the entire pooled time period. When relationships are very stable, this statistical assumption, referred to as temporal homogeneity, is unproblematic. But, when enforced without testing, it threatens to artificially “stabilize” temporally heterogenous relationships. In terms of protest policing, this has resulted in relatively ahistorical empirical explanations of protest policing. This article imports modeling techniques from work on identifying historical periods to show how temporal moving regressions can be built to recognize and model temporal heterogeneity in the factors influencing protest policing. We present three important uses for these models: testing exhaustively for temporal heterogeneity in apparently stable findings; testing for temporal heterogeneity that may reconcile otherwise contradictory findings; and inductively combining orthogonal research lines. We demonstrate the utility of each in examinations of protest policing. More generally, we show the potential of temporal moving regressions for uncovering new insights and bringing greater historical sensitivity to research on protest and beyond.