topic Re: Test post-hoc Kruskal Wallis in Discussions

Test post-hoc Kruskal Wallis

Sarah666 — Wed, 19 Mar 2025 09:45:27 GMT

Hello,
I'm running a Kruskal Wallis test on JMP. I want to observe whether the score obtained in paragraph comprehension differs between 3 groups. The samples are small and unpaired: a group of 5, a group of 5, and a group of 11.
The Kruskal-Wallis test is significant. I want to carry out post-hoc tests to see which group differs from the central tendency.
Is the Wilcoxon pairwise test appropriate? I've also heard of the Dunn or Steel-Dwass tests, which give insignificant results...
What's more, Wilcoxon concerns two paired samples, which is not my case. A Mann-Whitney would be more appropriate, but I can't do it on JMP.
Thank you for your help,
Yours sincerely

Re: Test post-hoc Kruskal Wallis

MRB3855 — Wed, 19 Mar 2025 13:33:33 GMT

Hi @Sarah666 , and welcome: You may be confusing Wilcoxon signed-rank test with Wilcoxon rank-sum test. The former is for paired samples, the latter is not. You can choose the "Wilcoxon Each Pair" (Wilcoxon rank-sum) option to do multiple comparisons you are looking for.

Re: Test post-hoc Kruskal Wallis

Sarah666 — Fri, 21 Mar 2025 18:19:51 GMT

Bonjour,

Merci beaucoup pour votre aide.

Donc j'en conclu que l'option "Wilcoxon chaque paire" dans JMP ne nécessite pas d'appariement ? À votre avis s'agit-il d'une bonne méthode ou bien je devrais faire un test de Dunn ou de Steel-Dwass ?

Merci

Re: Test post-hoc Kruskal Wallis

MRB3855 — Mon, 24 Mar 2025 11:49:02 GMT

Hi @Sarah666 : Yes, you are correct. The language can be confusing. "Wilcoxon each pair" does not require matching; and what they mean by "each pair" is "each two group combination".

Re: Test post-hoc Kruskal Wallis

Sarah666 — Tue, 25 Mar 2025 07:51:28 GMT

merci !