Sample Size Determination for (Planned) Post Hoc Multiple Comparisons in One-Way ANOVA

• In Event: Saturday Roundtable Session 3:05 pm
  In Roundtable Session: Regression Modeling for Educational Research (Table 21)

Abstract

Most approaches to sample size determination in one-way ANOVA are based on expected variation in means, but these approaches provide power only for the omnibus F test. Unfortunately, sufficient sample size for the omnibus ANOVA test does not guarantee adequate statistical power for post hoc multiple comparisons that most researchers plan to perform following a significant ANOVA (that is, not many stop after reporting a significant omnibus test). This Monte Carlo study investigated the sample sizes needed for the smallest, most-interesting post hoc mean comparison expected to be performed following a significant ANOVA. We identified sample size rules that can be applied generally for Tukey-Kramer and Games-Howell MCPs across many numbers of groups and effect size of most interest conditions.

Authors

• Frank A Oppong, Ohio University

• Yuqing Liu, Ohio University

• Nina Adjanin, Northwest Missouri State University

• George A. Johanson, Ohio University

• Gordon P. Brooks, Ohio University