Simultaneous Inference with Respect to the Best Treatment in Block Designs

Abstract

In many practical situations, only inference concerning the unknown best treatment and comparisons of the individual treatments with respect to the best treatment are of primary interest. We give an explicit formulation of this problem in the framework of randomized complete block designs. A procedure based on sample means and two procedures based on signed ranks are proposed. It is seen that, for the problem considered, these procedures have a substantial advantage over what can be deduced from all-pairwise multiple comparisons procedures.