Power Analysis & Mann-Whitney test

[JulijaM]

I need help in deciding which responce function to use in (post-hoc) power calculation for Mann-Whitney test: normal, laplace, logistic etc. Is there any theoretical resource that could explain in which case which fonction is most appropriate? Thank you in advance!

[sAura]

You can always run the test for a sample dataset and check for the Real time and CPU time from the SAS log for all the functions. Based on the timings, you can easily decide, as to which approach will be more speedy for your dataset you are working with.

[Doc_Duke]

I would typically use ordinal.  The Mann-Whitney test is a rank-based test, so the results are invariant under monotonic transformations.  If you have some idea of the underlying distribution, then you could choose that.  However, ordinal will always work.

There is a huge literature on why one should NOT do post hoc power analysis.  That doesn't keep my clients from asking for it.