Estimation of the variance and covariance matrices of the random effects and the residuals is considered. The Multivariate Analysis of Variance (MANOVA) and the Restricted Maximum Likelihood (REML) estimators are examined. A simple procedure for adjusting the MANOVA estimator for nonnegativeness is suggested. For both balanced and unbalanced designs, biases and mean square errors (MSEs) of the different procedures are compared through simulation studies.
展开▼