Minimax Estimators of the Mean Vector in Normal Mixed Linear Models

In balanced mixed linear models it is often possible to estimate the treatment means and the variance components with UMV estimators. In two instances, the mixed two-way classification model and the split-plot model, it is shown that the UMV estimator of the treatment mean vector is inadmissible with respect to quadratic loss. Shrinkage estimators are obtained which improve uniformly over the UMV estimator.