If I have a omnibus statistical test with unpaired assumption (e.g.Kruskall-Wallis), could I ever use the opposite (paired) post-hoc analysis? (e.g. Wilcoxon signed-rank for post-hoc analysis of Kruskall-Wallis), or rather I should stick with the same paired/unpaired logic?

  • $\begingroup$ If the data is unpaired, why would you use a paired test? $\endgroup$ – user2974951 Jul 22 at 6:28
  • $\begingroup$ I've seen several highly-cited papers doing that. That's the reason for my question. $\endgroup$ – MrT77 Jul 22 at 12:57
  • $\begingroup$ You should stick with the same paired/unpaired logic. I don't know why the people in these papers that you mention did not follow this logic, but it could be a case of p-hacking. You may want to ask a new question with some specific examples with sources, so that we may better assess the situation. $\endgroup$ – user2974951 Jul 22 at 13:32
  • $\begingroup$ Thanks, I will follow your advice and post a new question with the references. $\endgroup$ – MrT77 Jul 22 at 13:40
  • $\begingroup$ @user2974951 New question here: stats.stackexchange.com/questions/418626/… $\endgroup$ – MrT77 Jul 22 at 14:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.