A comparative study of alternative estimators for the unbalanced two-way error component regression model

This paper considers the unbalanced two-way error component model studied by Wansbeek and Kapteyn (1989). Alternative analysis of variance (ANOVA), minimum norm quadratic unbiased and restricted maximum likelihood (REML) estimation procedures are proposed. The mean squared error performance of these estimators are compared using Monte Carlo experiments. Results show that for the estimates of the variance components, the computationally more demanding maximum likelihood (ML) and minimum variance quadratic unbiased (MIVQUE) estimators are recommended, especially if the unbalanced pattern is severe. However, focusing on the regression coefficient estimates, the simple ANOVA methods perform just as well as the computationally demanding ML and MIVQUE methods and are recommended. Copyright Royal Economic Society, 2002