Bioequivalence and Bioavailability Forum

Thanks for your reply!
Sorry, I didn't know that this is off topic, because I assumed this falls into the category "bioavailability".

Yes, of course I am questioning my program. But on the other hand I tried to check as good as possible. In the article from Kupper and Hafner there are tables (unfortunately for the one sample and the two sample case only) which show the final sample size given an initial sample size. I checked my program according to this table - it works out fine. Also, I checked my results via FARTSSIE now (thanks for the link). In this spreadsheet only the case without coverage probability is considered. These values coincide with the "initial sample size" from my program (the first while loop) - at least for the test cases I used. nQuery also has a module for calculating such a sample size (i.e. without coverage probability) but here nQuery uses z-quantiles (I again get the same results by changing the t-quantiles to z-quantiles).

Some more rough results (incorporating the coverage probability)

1. for the parallel design the program above gives about 2 subjects more per group than nQuery's MTC.
   
2. for the 2x2 design the program above gives 1-2 subjects more (total n) than nQuery's MOC. However, the weird thing here is that I plugged in the same value for the standard deviation to get these similarities. That is, for instance sigma=0.2 as input for my program and also 0.2 for the standard deviation of differences in nQuery. But the latter one should actually be 0.2*sqrt(2)... When comparing the sample size obtained with input 0.2 (my program) to 0.2*sqrt(2) in nQuery then I can't see a pattern so far (sometimes only 1-2 off, but sometimes more than 10 off). Besides that, the problem that the final sample size is smaller than the initial one (obtained by the first while, w >= quantile*sqrt(Var(S))) remains. Shouldn't one at least consider taking the maximum?

❝ I'm not quite sure if the design constants are specific for the TOST procedure and make sense for your problem. Check the underlying formulas for each design you will implement with respect to these constants. Hope this helps.

Me either, but the same argument applies to CVfromCI? CVfromCI isn't TOST specific, is it?
I have to learn more about the design constant, and I will check out the references you gave me.
Thanks again for all the comments.

Best regards,
Ben