An Extension of Stein's Two-Stage Method to Pairwise Comparisons Among Dependent Groups Based on Trimmed Means

Abstract

There are some well-known, two-stage methods for controlling power or length of confidence intervals when comparing groups based on means. An issue that has received very little attention is the generalization of these methods to robust measures of location. The paper proposes a generalization of Stein's two-stage method for means when the goal is to perform all pairwise comparisons among J dependent groups based on a trimmed mean. In simulations, the most obvious generalization was found to be a bit unsatisfactory in some situations, but a simple and seemingly natural modification was found to correct this problem.

Key Words:

• Robust methods

Acknowledgments

The author wishes to thank the referees for their constructive comments provided in a timely manner.

Notes

Recommended by M. Aoshima