Kramer modification, were recommended as the best procedures in most situations. For certain specialized applications, such as those requiring one-sided tests or unequal allocation of error rates, the Bonferroni method may be preferred. For the normal unequal-variance case, Welch t tests were recommended, with the Bonferroni approach used to control the experimentwise error rate. For the nonnormal case, a number of rank procedures were discussed. The performance of all pairwise Wilcoxon tests, with the experimentwise error rate controlled via the Bonferroni or Steel-Dwass approach, has certain advantages over the procedures based on joint ranking. For the case in which both normality and equality of variances are thought to be violated, the MCP based on all pairwise median tests may be a good choice. For one-factor repeated measures designs, dependent t tests with separate error terms were recommended, with error control achieved through the Bonferroni approach. Nonparametric MCPs for this design include multiple sign tests, multiple Wilcoson signed-rank tests, and MCPs based on the Friedman model. In all MCP applications, the computation of confidence intervals can provide a useful supplement to significance testing. For this reason, as well as simplicity of computation, single-step MCPs, rather than more powerful stepwise methods, were recommended.
This chapter was limited to the discussion of pairwise MCPs in one-factor designs. MCPs for complex contrasts, two-way ANOVA designs, and special situations, such as comparing experimental groups to a control group, are discussed by Hochberg and Tamhane ( 1987) and Miller ( 1981). Information about Bayesian, decision-theoretic, and robust MCPs, as well as extensive discussion of stepwise MCPs, is given in Hochberg and Tamhane ( 1987).
This chapter is a revised version of an article, "Testing Pairwise Contrasts in One-Way Analysis of Variance Designs," that appeared in 1986 in Psycho- neuroendocrinology, 11, 253-276. It appears here with the permission of Pergamon Press. The initial research was supported in part by a National Research Service Award (No. 5-T32-MH15745) from the National Institute of Mental Health to the University of North Carolina at Chapel Hill. Preparation of the current version was supported in part by Educational Testing Service. Special thanks are due to Juliet Shaffer for her valuable and detailed recommendations for improving this chapter. In addition, I would like to thank C. Clifford Attkisson, Henry Braun, and Peter Vitaliano for their comments on the original article, Charles Lewis and Jo-Ling Liang for comments on the current version, and Kay Tyberg for preparation of the manuscript.
Barcikowski, R. S., & Robey, R. R. ( 1984). "Decisions in single group repeated measures analysis: Statistical tests and three computer packages". The American Statistician, 38, 148-150.