1

I just ran across something that I thought was interesting but don't fully understand. I have some categorical data that I am comparing in R. I use a Mann-Whitney U test when there are just two groups and Dunn's test when I have more than two groups.

What makes the significance change between the two tests?

For example: There is no significance between two groups with a U test. But in a Dunn's test they become significant if I add a third group. Is this because of how rank summing works?

Chris
  • 121
  • 3
  • 3
    Explained here: http://stats.stackexchange.com/a/111891/3277. Dunn's post-hoc test uses ranks obtained in Kruskal-Wallis, so it "knows" that there were more than 2 groups. Mann-Whitney does not "know" it. So the results need not to coinside. – ttnphns Jun 27 '16 at 17:21

0 Answers0