Comparisons with a control using normal scores and savage statistics

Abstract

Distribution-free methods for comparing t≤2 test treatments with a control or standard treatment are presented. Using normal scores statistics, one-sided tests and confidence bounds are developed for treatment versus control comparisons of medians. Using Savage statistics, analogous results are developed for comparing scale parameters of the treatments to the control with non-negative data. Probability points are computed using a union-intersection algorithm. Asymptotic relative efficiencies of these tests relative to competitors are identical to their two-sample counterparts.

Keywords:

• Many-one procedure
• multiple comparison
• treatment versus control