Sandwich - Simsalabim [🇷 for BE/BA]

posted by d_labes  – Berlin, Germany, 2010-04-28 12:58 (5111 d 02:42 ago) – Posting: # 5253
Views: 56,124

Dear Helmut,

----- Sandwich -----

❝ you have already outed yourself as a passionate wheel-reinventer, but why don't you simply go with the Welch-test (orwhatsover for unequal variances)?


To cite myself near the beginning of this thread: "I would go for a parallel groups study with exactly 2 groups with the 'simple' t-test (Welch variant ...)". Simple was here meant with respect to the lme() model used in bear.

But I'm seeking for something that can be generalized easily to more then 2 groups. Pairwise tests I'm not so happy with.

Eventually this (recently discovered) could be a way:
Herberich E, Sikorski J, Hothorn T (2010)
"A Robust Procedure for Comparing Multiple Means under Heteroscedasticity in Unbalanced Designs"
PLoS ONE 5(3): e9788. doi:10.1371/journal.pone.0009788
online resource

Easy to implement in R, only a couple of code lines.
But I was not a sandwich gourmet up to now :-D.

----- Simsalabim -----
<nitpicking>

❝ ...

❝ lnPKT <- rnorm(n=nT,mean=MeanT,sd=CVT*MeanT)

❝ lnPKR <- rnorm(n=nR,mean=MeanR,sd=CVR*MeanR)

❝ ...

:ponder:

If you are really interested in dealing with log-transformed metrics as lnPK suggests :-D I would suggest you the following modification:
# short hand function
   CV2sd <- function(CV) return(sqrt(log(1.0 + CV^2)))
...
     lnPKT  <- rlnorm(n=nT, mean=log(MeanT), sd=CV2sd(CVT))
     lnPKR  <- rlnorm(n=nR, mean=log(MeanR), sd=CV2sd(CVR))


Of course this will not affect the comparison between t-test with equal variances and Welch t-test I think.
</nitpicking>

Regards,

Detlew

Complete thread:

UA Flag
Activity
 Admin contact
22,993 posts in 4,828 threads, 1,653 registered users;
96 visitors (0 registered, 96 guests [including 5 identified bots]).
Forum time: 15:40 CEST (Europe/Vienna)

Never never never never use Excel.
Not even for calculation of arithmetic means.    Martin Wolfsegger

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5