MANOVA AND PROFILE ANALYSIS: ALTERNATIVE TO REPEATED MEASURES ANOVA AND POST-HOC PAIRED T-TESTS 1115

The purpose of this analysis was to compare ANOVA and MANOVA for repeated measures designs with respect to the assumption of sphericity, statistical significance of hypothesis tests, and meaningfulness of post-hoc analyses using paired t-tests for ANOVA and profile analysis for MANOVA. Paired t-tests determine treatment differences at each time point. Profile analysis determines the response curves of the time and time by treatment interaction, the latter reflecting differences in slope between treatments for each time interval. Statistical procedures were performed using SAS general linear models for univariate (ANOVA) and multivariate (MANOVA) repeated measures on systolic blood pressure data from 9 subjects who underwent two stress tests, exercise and pharmacologic, on different days. Repeated measures were obtained at baseline and 0, 5, 10, and 30 min recovery. ANOVA and MANOVA perform the same univariate test for treatment effect which was not significant (p =.69). ANOVA and MANOVA tests for time and time by treatment interaction were significant (ANOVA, time, p = 0.0004, time by treatment, p =.0001; MANOVA, time, p =.006, time by treatment, p =.0004). ANOVA failed the sphericity hypothesis based on Machley's test (p = 0.004). Post-hoc, paired t-tests were signficant at 0 min recovery only (p =.003). Profile analysis showed significant time and time by treatment effects between baseline and 0 min, and 0 and 5 min recovery. This comparison demonstrates that MANOVA attains similar levels of significance compared to ANOVA and may be a more appropriate statistic due to the sphericity assumption. Profile analysis may provide a more meaningful post-hoc analysis than paired t-tests when the interaction effect is significant and the treatment effect is not.