Bayesian simultaneous estimation for means in \(k\)-sample problems (Q1755111)

The paper under review deals with the problem of simultaneous estimation of \(k\) \(p\)-variate population means when one suspects that the means are nearly equal. As an alternative to the preliminary test estimator based on the test statistics for testing hypothesis of equal means, the authors consider Bayesian methods including hierarchical Bayes and empirical Bayes estimators. The Bayesian approach suggested here consists of two kinds of shrinkage. One is to shrink the \(k\) sample means toward a pooled mean estimator, and the other is to shrink the \(p\)-dimensional pooled estimator toward a constant like zero, which was proposed in [\textit{F. Perron}, J. Multivariate Anal. 46, No. 2, 254--261 (1993; Zbl 0780.62045)]. The former shrinkage is treated in section 2.1, and a class of minimax estimators is derived. Out of the class, the authors develop hierarchical and empirical Bayes estimators with minimaxity. The latter shrinkage is discussed in section 2.2, and it is shown that all the estimators derived by the former shrinkage can be further improved by using the latter shrinkage. In section 3, the authors derive the hierarchical Bayes and minimax estimators. Here the classes of minimax estimators are extended in order to handle the Bayes estimator, which consists of the two kinds of shrinkage, and the minimaxity of the hierarchical Bayes estimator is established. In section 4 the performance of the risk functions of the preliminary test estimator and several hierarchical Bayes and empirical Bayes estimators is investigated through simulations. Proofs are relegated to the appendix.