Holm-Bonferroni sequential corrected p-values for Mann-Whitney-U test

When performing the Kruskal-Wallis analysis of variance on k samples, up until now, you can choose between Dunn, Conover-Iman and Steel-Dwass-Critchlow-Fligner with the addition of Bonferroni correction (for the first two mentioned, respectively) as post hoc tests. However, I often read and see scientists using multiple Mann-Whitney-U tests with Holm-Bonferroni sequential corrected p-values* as and alternative post hoc (a method intended to control the familywise error rate for multiple comparisons). Fortunately, in XLSTAT I am able to perform Mann-Whitney-U for multiple groups, however I can't add Holm-Bonferroni sequential correction so I have to calculate it myself. Given it is a valid alternative, it would be great to add it to Kruskall-Wallis as a post-hoc alternative. Please correct me if that approach has any flaws compared to e.g. Dunn-Bonferroni.

*Holm, S. (1979). A simple sequentially rejective multiple test procedure. Scandinavian journal of statistics, 65-70.

  • Stefan B
  • Aug 10 2018
  • New ideas
  • Attach files