1

I have 8 different groups of individuals, each with a pre- and post-treatment measurement. The data are non-normal, so I used a Jonckheere-Terpstra test to test for an ordered alternative and rejected the null. I'd like to report p-values for the comparison of pre-treatment means to post-treatment means, and I used a Wilcoxon Signed-Rank test for each age group. However, now I realize that this may not be allowed without doing a multiple comparisons test to determine which groups are significantly different. I used Dunn's test to verify that there is a significant difference between groups 4 and 5 and between 5 and 6. What does this mean for when I can perform Wilcoxon Signed-Rank to compare pre-test to post-test? Can I only do the comparisons for these two pairs? Can I compare the other groups?

kdk
  • 43
  • 5
  • I've been doing some more research, and it looks as though doing Jonckheere-Terpstra tests on both pre- and post-treatment data would be incorrect. A Friedman test might be appropriate since it seems to be the non-parametric equivalent of repeated measures ANOVA. Can anyone who understands better confirm this? If so, would I need to perform one Friedman test to test for effect of age, and another to test for effect of pre/post-measurement? Or just do the Friedman test for age, do a multiple comparisons test, and then do separate Wilcoxon Signed-Rank tests on those groups with differences? – kdk Dec 02 '14 at 00:16
  • Jonckheere-Terpstra is the ordered-alternatives version of kruskal-Wallis (one-way anova-like situation). There are similar ordered-alternative versions corresponding to Friedman as well (Page's test, one of the Match tests). If you still have an ordered alternatives, you shouldn't ignore that fact. – Glen_b Dec 02 '14 at 09:40
  • See https://stats.stackexchange.com/questions/169419/why-bother-looking-at-an-omnibus-anova-when-i-have-a-priori-hypotheses-about-gro/169452#169452 – kjetil b halvorsen May 11 '21 at 15:55

0 Answers0