Posterior Average Effects

Economists are often interested in estimating averages with respect to distributions of unobservables, such as moments individual fixed-effects, or average partial effects discrete choice models. For quantities, we propose and study posterior (PAE), where the is computed conditional on sample, spirit empirical Bayes shrinkage methods. While usefulness for prediction well-understood, a justification conditioning estimate population currently lacking. We show that PAE have minimum worst-case specification error under various forms misspecification parametric distribution unobservables. In addition, introduce measure informativeness conditioning, which quantifies relative model-based estimators. As illustrations, report estimates neighborhood U.S., permanent transitory components model income dynamics.