Log in
Register|
Relaxation ★ Germany, 2015-11-06 11:25 (3460 d 10:34 ago) Posting: # 15623 Views: 8,165 |
|
|
Hello dear forum members. From a recent post in this forum in combination with a discussion yesterday, I was remembered that I always wanted to get the opinion of other people on a (probably small) issues. Mostly because there may be simple answers I am just not aware of and missed in the standard text books. And unfortunately, most colleagues just tell me that that is interesting (thinking "who cares"). From time to time we see development projects where multiple studies were necessary/conducted and one of these studies failed unexpectedly. As an illustrative example, BE on the low and high strength was successful, the intermediate one failed (although that particular one would have been eligible for a strength waiver anyway). From my point of view, this may simply be a misleading finding due to chance, because, staying in that example, conducting three studies at a power of 80% will result in a fifty-fifty chance to have all of them being successful (0.8^3). By implication, it might even be reasonable to "overpower" each single study in order to end with an "overall power" of 80% (0.93^3 is roughly 80%). As said, I may completely miss the point here (and failed to use the search funtion properly) and would be grateful for correction. Although I won't reject encouragment and, generally, will appreciate any input  .Best regards, Steven. |
|
ElMaestro ★★★ Denmark, 2015-11-06 11:39 (3460 d 10:20 ago) @ Relaxation Posting: # 15624 Views: 6,307 |
|
|
Steven, ❝ From my point of view, this may simply be a misleading finding due to chance, because, staying in that example, conducting three studies at a power of 80% will result in a fifty-fifty chance to have all of them being successful (0.8^3). By implication, it might even be reasonable to "overpower" each single study in order to end with an "overall power" of 80% (0.93^3 is roughly 80%). You are absolutely correct and the industry would be in a much better shape if its people were generally thinking like you. Well done  — Pass or fail! ElMaestro |
|
d_labes ★★★ Berlin, Germany, 2015-11-06 14:52 (3460 d 07:07 ago) @ Relaxation Posting: # 15625 Views: 6,246 |
|
|
Dear Steven, just 2 references which deal with that topic: Phillips, K. F. (2009) "Power for Testing Multiple Instances of the Two One-Sided Tests Procedure" The International Journal of Biostatistics: Vol. 5: Iss. 1, Article 15. Hua S. Y., Xu S., and D'Agostino Sr. R. B. "Multiplicity adjustments in testing for bioequivalence" Statistics in Medicine, Vol. 34, Issue 2, 215-231 (2015) But be warned. Rather theoretical papers with a lot of formulas  . Unfortunately the R-code given in Kem's paper (for 2 TOST simultaneously) doesn't work anymore because the numeric integration routine used isn't available on CRAN anymore, due to license problems. But there is some rumor that help is on the way ...  Thus, at moment, we can't check your belief that "overall power" is the product of the single powers. In contrast to our Ol' pirate I feel this isn't correct. But it's only a feeling  which I can not underpin at moment.— Regards, Detlew |
Ing. Helmut Schütz