SAULEAU Erik-Andr? wrote:> Dear R-list,
>
> I have to design the validation of a score (ordinal values between 0 and 6)
reputed to separate 4 groups of patients with known frequencies in the
population. I think the more accurate is to calculate sample size under median
test. Is there a function for that in R (not in the pwr package)?
>
> Thanks in advance, with all my best, erik S.
>
>
>
>
> ================================> Erik-Andr? SAULEAU
>
> SEAIM
> Centre Hospitalier
> 87, Avenue d'Altkirch
> BP 1070
> 68051 Mulhouse C?dex
> Tel: 03 89 64 67 53
> Mel: sauleauea at ch-mulhouse.fr
> Web: www.ch-mulhouse.fr
The median test has low power and should be avoided. For a validation
study a hypothesis test is seldom of interest and you might base the
sample size on the needed precision (margin of error; confidence
interval width) of some estimate.
--
Frank E Harrell Jr Professor and Chair School of Medicine
Department of Biostatistics Vanderbilt University