Correlated Random Effects as a Tool for Inferential Social Science

Partly aided by the increasing availability of multilevel modeling software, a generation of social scientists has learned that random effects beneficially account for clustered data while conservatively adjusting the estimates of predictor variables. Less commonly, scholars can also interpret the variances and covariances of the random effects to gain additional insights into their data. I share two examples of the use of correlated random effects for testing novel research questions. The first example focuses on social network data and the evolutionary hypothesis that generosity should be directed toward individuals who are generous themselves. This question can be investigated using a multilevel parameterization of the Social Relations Model. The second example employs multinomial logistic regression to examine individual-level behavioral variation. The use of correlated random effects potentially reveals the extent to which individuals who regularly engage in one behavior also spend more or less time in other behaviors. The development of new statistical methods can spur theorizing, and correlated random effects provide opportunities for social scientists to ask and answer novel research questions.