0

I am doing pairwise comparisons between 5 groups. The Dunn test with Bonferroni adjustment only came out significant for two groups (say A and B). However, when I added 5 data points to one of the groups (say C), the significant difference between A and B disappeared. total N= 148 (A= 28, B= 43, C= 13). Dunn's calculation does not seem to care to total sample size or overall mean. I do not understand what made the significance disappear.

Arti Thakur
  • 1
  • 1
    It might use some common standard deviation. But in general, trying out many different things until the desired significances pop up is unscientific. – Michael M Sep 13 '23 at 17:08
  • Dunn's test or better say Dunn's approach for nonparametric pairwise post-hoc testing uses information about all k groups in each comparison, rather than only about the two currently compared – ttnphns Sep 13 '23 at 19:05
  • The z-statistics computation in Dunn does not seem to use common SD. I am not p- fishing and using Dunn for post-hoc comparisons. Reviewers asked to add these 5 data points and significance dissapeared between two groups, though the points were added to the third group. I need some explanation for this behavior. Upon checking Dunn’s computation, I am not able to explain this development. Thus asking the community. – Arti Thakur Sep 13 '23 at 19:08
  • @ttnphns could you share the reference for this? – Arti Thakur Sep 13 '23 at 19:09
  • First, define what you mean by the Dunn's test. (What I mean is the post-hoc pairwise comparison after Kruskal-Wallis.) – ttnphns Sep 13 '23 at 19:14
  • Thanks for correction. I mean the same. Posthoc pairwise comparison after Kruskal-Wallis. – Arti Thakur Sep 13 '23 at 19:30
  • 1
    Then, please get the formula of the test. You'll see that information from all k groups is utilized in each pair compared – ttnphns Sep 13 '23 at 19:43

0 Answers0