Vague macrosociological theory generates uncertainty about statistical models. As a result, statistical inferences in macrosociology—where data sets are small and collinear—may carry considerably more error than is indicated by the conventional estimation and testing methods. I describe a Bayesian approach that propagates uncertainty about statistical models through to the final results. This approach is illustrated in a model of cross-national welfare state development and a pooled cross-sectional analysis of strike activity. These examples show that classical inferences can be seriously misleading when vague theory weakly guides the model specification. These ideas are considered in relation to recent methodological debates in macrosociology.

* Princeton University