Protected versus Unprotected Multiple Comparison Procedures.

摘要: Conventional wisdom suggests the omnibus F-test needs to be significant before conducting post-hoc pairwise multiple comparisons. However, there is little empirical evidence supporting this practice. Protected tests are conducted only after a while unprotected without regard significance of F-test. Monte Carlo methods were used generate replications expected provide 0.95 confidence intervals +/0.001 around nominal alphas 0.10, 0.05, and 0.01 for 42 combinations "n" (5, 10, 15, 20, 30, 60, 100) numbers groups (3, 4, 5, 6, 8, 10) . Unprotected protected using Dunn-Bonferroni, Dunn-Sidak, Holm, Tukey's Honestly Significant Differences (HSD) procedures. Means standard deviations observed per-experiment Type I errors rates percentages error falling below, within, above determined total number errors. in sample size was minimal. differences control among four comparison procedures when as or unprotected. The Dunn-Bonferroni had best an test with 96.0% within interval 87.3% fell below test, thus being very conservative. As tests, Dunn-Sidak Holm tended liberal, but conservative tests. HSD liberal both situations, much more so test. These results, combined ease suggest method may most accurate easiest mode. (Contains 4 tables 13 references.) (Author/SLD) ******************************************************************************** Reproductions supplied by EDRS that can made from original document. versus Multiple Comparison Procedures J. Jackson Barnette University Iowa