2013): “Unobserved Heterogeneity in Longitudinal Data: An Empirical Bayes Perspective,” preprint

Abstract

Abstract. Empirical Bayes methods for Gaussian and binomial compound de-cision problems involving longitudinal data are considered. A new convex opti-mization formulation of the nonparametric (Kiefer-Wolfowitz) maximum likelihood estimator for mixture models is used to construct nonparametric Bayes rules for compound decisions. The methods are illustrated with some simulation examples as well as an application to predicting baseball batting averages. Comparisons with nonparametric Bayesian methods based on Dirichlet process priors are also provided. 1.