Bioequivalence and Bioavailability Forum

Dear Andrew!

❝ [...] trivial question. I blame medical training, which is short both on

❝ mathematics and logic.

Well, you already made an important step! To quote Stephen Senn:

Statistics - A subject which most statisticians find difficult
but in which nearly all physicians are expert.

❝ If I were to run a four period replicate design study in n=24, this has the

❝ statistical power of a standard two-way design in n=48 (along with the added

❝ advantage of being able to apply wider limits using SABE). If I instead

❝ choose to run a three period replicate study with equivalent power, my

❝ understanding this should include n=36.

So far, so good.

❝ However, in this case the T/R comparison involves 36 sets of data, while

❝ the 4-period design involves 48 T/R comparisons.

Ha, tricky! In the 2×2 you have 48 T/R comparisons. But in a replicate design means of the repeated administrations are compared - in the 3-way 36 and in the 4-way 24. This averaging and the modified degrees of freedom of the variances involved lead to the same power.