I performed an ANOVA on some test results of N groups (N = 7), which yielded a significant difference of the group means, followed by a wilcoxon signed-rank test to determine which groups are significantly different from each other. However, I did not perform the wilcoxon signed-rank test for all possible pairs as this would, after bonferroni-adjustment, reduce the power of the test considerably. Some pairs are not as interesting as others, and as such were not considered. Is this allowed? I reread some literature regarding pairwise comparisons, but they all state to perform all possible pairwise comparisons, however, without any explanation.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
You don't have to do all pairs, but, if you want to look at only some pairs, you have to decide which ones before doing the analysis.
Otherwise, you have the risk of the "Texas sharpshooter fallacy" where you shoot at a barn and then paint targets around the holes.
answered Jun 28, 2023 at 12:00
-
$\begingroup$ Alright, thank you, that was already sufficient :) $\endgroup$ Commented Jun 29, 2023 at 7:47
-
1$\begingroup$ Since you are new here, I'll let you know that, if you think an answer is "sufficient" it's usual to hit the check mark next to the answer. You don't have to, but if you think my answer is what you need, that lets other people know. $\endgroup$ Commented Jun 29, 2023 at 17:56