NONPARAMETRIC T-METHOD IN TWO-WAY LAYOUTS

Abstract

In a randomized block design with one or no observation per cell, we consider a nonparametric generalization of Tukey's T-method of multiple comparisons. Our generalization is based on aligned signed ranks given for every pair of treatments, and is like Hochberg's [1] GT2. The method is compared with Sen's [5] in terms of Pitman efficiency under the situation that there are no missing values. It is suggested that our method is better than Sen's [5], when there are many missing values.