Estimation of variance components in an unbalanced two-way mixed model under heteroskedasticity (Q1116239)

In an unbalanced two-way mixed model with random column effects where the error variance changes from cell to cell, the problem of estimation of variance components due to column effects and interactions is considered. Since with the increase in the size of the experiment, the number of error variances increases indefinitely, the problem of finding efficient estimates of the two variance components looks intractable. However, by grouping the cells according to cell frequencies, it is possible to construct an estimate with good asymptotic properties. A comparison with the members of a class of estimates arising from intuitive considerations, is also made.