I am using the Friedman test and the Dunn test to compare the results of a few algorithms, yet, despite having the Friedman test showing me a significant difference (pvalue=4.579491803168107e-08), the Dunn test (adjusted with Bonferroni method) shows all 1s for all pairwise comparisons. The Dunn test's results show differences when whenever I discard the adjustment, but I don't know whether it's safe to publish the Dunn test's results without applying any adjustment to them (A reviewer has asked me to include the test results in my paper. I don't know how I should respond him).
Something seems wrong with your computations.i believe you have corrected too many times.Dunn test is usually one method of multiple comparisons and you have followed it by bonferroni which gives. overall hlas two rounds of corrections. IMO use only bonferroni. Dunn is about means and Friedman's test is non parametric test so bonferroni should apply .see the attachment for details. Only one division is required for bonferroni . David Booth
Yes, of course.
Dunn's test has very low power. And yes, an adjustment for multiple comparison is required.
The selection of the statistical tool depends on the type of data distribution. You can test the distribution of data parametric or nonparametric.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics