Effect size Wilcoxon ranksum test

[Francesco Bianchi]

Hi everybody, I'm having some issues with Wilcoxon ranksum test. I'm comparing a quantitative variabile between two groups; the variable is not normally distribuite (and not normalizable), so I have to use non parametric test. I work on Stata MP15
I have two questions:
- How can I calculate the effect size? There is a specific command?
- How can I calculate the adjusted p-value?

Thank you in advance

[Rich Goldstein]

not sure what you mean by effect size here, but maybe the "porder" option will give you what you want; see

help ranksum

what do you mean by "the adjusted p-value" here? you don't mention covariates and you appear to want only one test so ...

[Francesco Bianchi]

not sure what you mean by effect size here, but maybe the "porder" option will give you what you want; see

help ranksum

what do you mean by "the adjusted p-value" here? you don't mention covariates and you appear to want only one test so ...

for effect size I mean this https://www.stata.com/features/overview/effect-sizes/ for nonparametric test

[Phil Bromiley]

I could be wrong, but my memory of non-parametric tests is some look a lot like parametric tests based on ranks so you might be able to do the ranks yourself and apply the provided procedures.

[Mike Lacy]

One possibility is to use Somers' D. See Newson, R. 2002. Parameters behind "nonparametric" statistics: Kendall's tau, Somers' D and median differences. Stata Journal 2: 45-64.
https://www.stata-journal.com/articl...rticle=snp15_6

I can't remember all the details, but at least some forms of the Wilcoxon are argued/shown there to be tests on the Somers D statistic, about which see Newson's package: -ssc describe somersd-

The other possibility (better, in my view, as it avoids the losses associated with applying rank-based or ordinal methods to a continuous outcome) is to use -esize- with -permute- . If you like Cohen's D as an effect size measure, you could do something like:

sysuse auto
permute foreign r(d), reps(1000) : esize twosample weight, by(foreign)

[Francesco Bianchi]

thank you