The equality of REML and ANOVA estimators of variance components in unbalanced normal classification models

For completely balanced normal classification models, it is well known that REML and ANOVA estimators of variance components are equal. This paper extends this result to classes of partially balanced designs.