Multivariate limited translation hierarchical Bayes estimators

Based on the notion of predictive influence functions, the paper develops multivariate limited translation hierarchical Bayes estimators of the normal mean vector which serve as a compromise between the hierarchical Bayes and maximum likelihood estimators. The paper demonstrates the superiority of the limited translation estimators over the usual hierarchical Bayes estimators in terms of the frequentist risks when the true parameter to be estimated departs widely from the grand average of all the parameters.