On some single-stage, step-down and step-up procedures for comparing three normal means.
Computational Statistics and Data Analysis, 21, (2), . (doi:10.1016/0167-9473(95)00013-5).
Full text not available from this repository.
The problem of pairwise comparisons of three normal means in the balanced one-way fixed effects anova model is considered. In order to compare the performance of a single-stage procedure, a step-down procedure and a step-up procedure which all strongly control the type I familywise error at the same level, a method of power assessment is proposed. The power function is defined to be the probability of detecting, with correct directional decision, each pair of treatments with mean values at least ?-distance away from each other. The global minimum power is then used as a criterion for comparing the three procedures. Expressions are provided allowing simple computer evaluation of the global minimum power for each of the three procedures. The findings are that the step-down procedure performs better than the step-up procedure, and both step-down and step-up procedures perform substantially better than the single-stage procedure. A table of sample sizes necessary to guarantee certain power levels for each of the three procedures is also given.
Actions (login required)