Dunn's post-hoc test
Margaret
I like to use the Mann-Whitney U-test with one of the less conservative
Bonferroni-type corrections when carrying out post-hoc multiple
comparisons further to an application of the Kruskal-Wallis test. I am
not a user of GraphPad Prism. However, I am aware that this package
encourages the user to select Dunn's test as the appropriate post-hoc
test for the Kruskal-Wallis test, presumably with the Bonferroni
adjustments already applied.
I would be most grateful for thoughts on the relative merits of either
of the above two approaches when deciding to do pairwise comparisons as
post-hoc tests to the Kruskal-Wallis test.
Thank you very much in advance for your kind contributions.
Best wishes
Margaret
Thom
I think Dunn's procedure is just another name for the Bonferroni
correction (Dunn's conribution was to provide tables of critical values
before computers allowed us to calculate observed p directly).
Thom
Thom
What correction to use depends on the numbers of groups. I think there
is an argument to use Fisher's LSD with only 3 groups (effectively
relying on only on the omnibus Kruskall-Wallis test for protection),
while with greater than 3 groups I'd suggest a more powerful
alternative to Bonferroni such as the Holm procedure (or more
controversailly the Hochberg procedure).
If you rank the raw scores and then run parametric ANOVA that would
also give you the full range of ANOVA post hoc tests and virtually
identical results to Kruskall-Wallis.
Thom