Bioequivalence and Bioavailability Forum

Hello Jérôme,

❝ If we assume that alpha=5% is too stringent, with alpha=20%, 16 people for a pilot study seems a good tradeof either one or two-sided.

❝ Any opinions?

I would plainly and simply say that the higher the sample size the larger the chance that you get a good estimate of the variability and of many other things.

As for the tables, I am not entirely convinced they are a definitive source of truth. They are created on basis of assumptions re. variance distributions, I think. But we have also seen that in earlier threads that there are general potential violations of the underlying assumptions for the normal linear model as crossovers are concerned. Does that influence the variances estimates and/or their confidence intervals? I think it might but I cannot tell how.

I worhsip NLYWs and am a fan of simulations so believe it might be a good idea to use a distribution-free approach for variance CIs. Bootstrap/resampling (with sample replacement) of trial data and derive a variance CI form that, wouldn't that be an ok alternative?