On Improved Extensions of theT-Method of Multiple Comparisons for Unbalanced Designs

Abstract
Some improvements and clarifications of extended T-type procedures of multiple comparisons for unbalanced designs are given. In view of the general discussion, we obtain a procedure particularly tailored for one-way layouts with only two different sample sizes. In all such designs, the given procedure yields confidence intervals which are not longer than those obtained by Spjøtvoll and Stoline's (1973) method for all pairwise comparisons and are strictly shorter for some of them. For moderately unbalanced designs, the new procedure is also superior to Hochberg's GT2 method using the average confidence interval length for all pairwise contrasts as criterion.