[R] about power of tests

[Peter Dalgaard]

rossini at blindglobe.net (A.J. Rossini) writes:

> =?iso-8859-7?q?vasilis=20pappas?= <vasileios_p at yahoo.gr> writes:
> 
> > Does anyone know how can I calculate power in wilcoxon.test,
> > kruskal. test and oneway.test in an F? I have already found
> > power.t.test for calculating power in t.test but it seems that there
> > is nothing for the others. Any answer could be useful.  Thanks for
> > your interest.
> 
> If you can explain exactly how you think power should be computed for
> a Wilcoxon, AND it is right (as opposed to a wild ball-park off the
> cuff estimate or sensitivity analysis), it'd be a nice journal
> article.

Yes.

Of course the ARE (asymp.rel.efficiency) of Wilcoxon is .95 in the
Normal case, so everyones first guesstimate is to do the power
analysis for t and multiply the power by .95 - or divide desired power
by .95 and compute N. However, this is highly dependent on
distributional assumptions, and the fact that A is for asymptotic is
really important even for Normal distributions: for small samples the
RE is nowhere near the ARE, so you'd be wise to back things up with
some simulations.

We do have power.anova.test, though.

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907