5

What post hoc test can I do after Kruskal-Wallis?

Is there a post hoc test for non-parametric test? Some says there's none.

I ran Kruskal-Wallis test and as a result the null hypothesis was rejected. Is there a post hoc test that I can use to pinpoint which group/s is/are different?

Ferdi
  • 5,179
sammyyy
  • 51

2 Answers2

2

I think probably the most common post-hoc test for Kruskal-Wallis is Dunn test. This article from 2015 has some discussion about its popularity. It has implementations in R, and I imagine most other common software packages.

Sal Mangiafico
  • 11,330
  • 2
  • 15
  • 35
1

I once used several Wilcoxon t-tests with a Bonferroni correction. Bonferroni correction will account for the error usually found by performing several different tests on the same data sets. By using this test you will be able to see which groups are significantly different.

User25678
  • 11
  • 5
    No, individual Wilcoxon tests re-rank the data after removing the non-tested groups. The most general solution is to recognize that the Kruskal-Wallis test is a special case of the proportional odds ordinal logistic model. Being a regression model, it admits all contrasts of interest in a consistent fashion and allows for simultaneous confidence intervals for odds ratios. I prefer simultaneous CLs to p-value corrections. – Frank Harrell Oct 28 '17 at 12:28