Most causal processes of interest to social researchers are likely to depend on the specifics of individuals, groups, and their social contexts. I will argue that much of this heterogeneity will be not be captured by conventional explanatory variables.The result will be unmodeled heterogeneity in causal effects. These, in turn, will lead to failures of replication and flawed tests of theoretical predictions. In this talk I lay out ways to tackle the problem using intensive longitudinal data and, in some cases, Bayesian methods of estimation.