Bioequivalence and Bioavailability Forum

Hi ElMaestro,

❝ In perspective, could this methodology then be extended to a 3-treatment parallel study (e.g. US Ref, EU Ref, Test)?

Intuitively – if we walk on EMA’s stony ground of ‘leave one out’ – I would say so.* Duno what will happen if we keep all treatments in the model.

❝ I am thinking we'd just need another sigma squared on the diagonal and then be good to go?

I feel tempted to give you all statisticians’ standard answer on any arbitrary question: “Positively maybe!”

❝ If that's the case then the WNL approach –however counterintuitive and retarded it might seem– could provide a solution to a problem which cannot be resolved by the standard t-test approach?

See above. Any ideas how to code that in R without setting the Silly-o-Meter on fire?

❝ Oh and by the way, I think this would imply that the denominator DFs would have to be calculated from treatment A, B as well as C, when a 90% CI for A/B is constructed - that's perhaps also worthy of a heated debate further down the road...?

Sure. We are moving in a circle. Remember the sim’s Detlew performed in Dec 2011?

---

• If I erred, I am prepared to accept discipline up to level three.
  Punishment of a higher order will be considered unjustified
  and rejected with prejudice. David Brin (Startide Rising, p30, 1983)