Hi friends,
I sometimes have variables in a design which actually represent an individual percentage, i.e. a value of between 0 and 100. You could calculate their mean and sd and so on. If I would have to involve those variables in hypothesis testing I think a t-test would not be applicable because at the extreme ends of the scale its distribution would not be normal (but skewed). I would use a nonparametric test. On the other hand doing sample size calculations with that kind of variables often (in any case with PASS) comes down to a ssc for a t-test with nonpar adjustment, departing from mean and sd estimations. That seems contradictory and I would like to know whether there exists a more consistent approach. Does anyone of you use transformations? While an (estimated) mean could easily be transformed likewise, would it be possible to transform an (estimated) sd for ssc? Or do you use quite another type of ssc, without the need for something like an sd?
Regards - Jim.
--
. . . . . . . . . . . . . . . . . . . . . .
Jim Groeneveld, MSc.
Biostatistician
Science Team
Vitatron B.V.
Meander 1051
6825 MJ Arnhem
P.O.Box 5227
6802 EE Arnhem
Tel: +31/0 26 376 7365
Fax: +31/0 26 376 7305
[log in to unmask]
www.vitatron.com
My computer multi-boots OS's, each of them adapting the DST twice a year
[common disclaimer]
|