0
$\begingroup$

Does post-hoc for friedman tests or nonparametric testshave like a homogeneous subset table from SPSS? Mean doesn't represent it well so I tried using median but my data was zero-inflated so the most of the medians even in supposed homogeneous groups have lots of zeroes

Example: homogeneous subsets

Improve this question
$\endgroup$
5
  • $\begingroup$ What is a homogeneous subset table? Is it obtained from data dredging? Have you verified that your sample size is adequate for the simplest possible analysis? $\endgroup$
    – Frank Harrell
    Commented Sep 23, 2023 at 12:08
  • $\begingroup$ @FrankHarrell it's like another way of representing the post-hoc pairwise comparisons. in the example above, it's the column "1" or "2" which contain the mean of the groups which are significantly different to the mean in other columns. so that "Control group" (5.30) is significantly different than Mnemonic A (11.80) and Mnemonic B (15.60). Now the problem is the mean is not representative when the test is nonparametric $\endgroup$
    – Derf
    Commented Sep 23, 2023 at 13:10
  • $\begingroup$ Yes that's what I'd usually call data dredging, a very dangerous practice without careful multiplicity corrections. Long-term one of the better ways to think about this is getting simultaneous confidence intervals for all differences of interest. These would come from a unified model for count data, or perhaps better, from a semiparametric model such as the proportional odds model which can handle extreme clumping at zero. $\endgroup$
    – Frank Harrell
    Commented Sep 23, 2023 at 13:12
  • $\begingroup$ Thanks for the response. It's just my first time hearing "data dredging" so I did a quick informal search and I guess in a way "data dredging" concept exists in post-hoc analysis as well (???) (en.wikipedia.org/wiki/Post_hoc_analysis) $\endgroup$
    – Derf
    Commented Sep 23, 2023 at 13:37
  • $\begingroup$ It's concentrated there. $\endgroup$
    – Frank Harrell
    Commented Sep 23, 2023 at 22:55

0

Reset to default