When reporting a Mann-Whitney-U test result, is it better to report it's effect size or the Hodges-Lehmann pseudomedian (as R give us as "difference in location") ?

  • 1
    $\begingroup$ You have asked a similar question once: stats.stackexchange.com/q/143080/3277. I said that I think it is a good idea to report the two values of HL location. Difference between them can be once of possible measures of nonparametric effect size. $\endgroup$
    – ttnphns
    Commented Jul 17, 2015 at 21:46
  • 1
    $\begingroup$ @ttnphns I was afraid that effect size and HL would be two totally different things. That's why I made this new question. And, what do you mean by "two values of HL location"? I thought that HL was a single value metric. $\endgroup$ Commented Jul 17, 2015 at 22:31
  • 1
    $\begingroup$ Pardon please. I meant: HL two-sample estimator is one value, sorry. (I doubt it might be called "pseudomedian" because the latter is one-sample HL estimator.) I think that it is one way to report the MW difference. But it is based on raw data values. When we say "effect size" we usually mean some standardized difference (independent of sample size or variance). I haven't thought how one could standardize HL 2-s estimator. Maybe z-standardize values of the raw combined sample and then compute the HL 2-s estimator? I'm not sure. $\endgroup$
    – ttnphns
    Commented Jul 17, 2015 at 22:36


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.