If the null hypothesis in the Kruskal Wallis test is rejected, How can we perform pairwise comparisons? Can we apply Dunnett's Test for that?
1 Answer
$\begingroup$
$\endgroup$
1
You want to use a post-hoc test that is designed for the Kruskal-Wallis test. A common one is the Dunn (1964) test. This is a rank-based test, that is somewhat like performing pairwise Wilcoxon-Mann-Whitney tests, but uses the ranks from the whole Kruskal-Wallis test, not just the individual pairs.
-
1$\begingroup$ I would use a generalization of the KW test: the proportional odds ordinal logistic regression model, and do pairwise contrasts using that model. See fharrell.com/post/rpo $\endgroup$ Commented Sep 12, 2022 at 19:24