Bioequivalence and Bioavailability Forum

Hi Mittyr,

❝ 1. Backups. […]

❝ For me it looks strange. What do you think about this procedure?

Outdated. We abandoned it 20 years ago. Nowadays you should start with the estimated sample size plus some additional subjects according to the expected dropout rate. Even, if you don’t add subjects the loss in power is less than most people think. See this thread; you could loose a lot of subjects until power substantially decreases.
Detlew implemented some useful functions in PowerTOST last October. Try:

library(PowerTOSt)
pa.ABE(CV=0.21, minpower=0.7)

In your example (n 22) with five (!) dropouts you still have 72% power.