Peculiar partial replicate [🇷 for BE/BA]

posted by d_labes  – Berlin, Germany, 2011-01-28 17:10 (4807 d 20:42 ago) – Posting: # 6510
Views: 64,019

Dear Öberster Größter Meister,

❝ With the intra-subject sWR and sWT only on the diagonal of V my alarm bells are ringing. Imagine a ref-replicated design; following your thinking we could fit an sWT on the diagonal of V even though Test isn't replicated. This sounds very wrong. Apparently others have got fits and sWT values for 2-trt, 3-seq, 3-per designs (see e.g. here). This is fishy business.


Well noticed :cool:. We had discussed this peculiarity of the partial replicate design also here in the forum if you remember. See f.i. this post and followings. But be warned: Rather lengthy and nitpickin' :-D.

As you can see there we get s2WT, but rather different values between different software. And, as I have noticed, you get also different values if you change optimization parameters in SAS Proc MIXED. Seems the REML optimisation stops by chance. Thus that value is not real, the model is over-specified, but the REML or ML estimation can't of course decide how the model has to be formulated without an explicit value for it.
It's the same as if we attempt to fit a 2x2 design with distinct intra-subject error terms for Test and Reference. We could not be successful because only (s2WT+s2WR)*0.5 is estimable.

Meanwhile I have the opinion that we can't handle this evaluation via such a detailed model of the different sources of variation. I at any rate was not successful to implement a reasonable simpler model in Proc MIXED.
Maybe this is one reason why the FDA code given in the progesterone guidance resorts to evaluation via intra-subject contrast?

BTW: Link corrected.

Regards,

Detlew

Complete thread:

UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,639 registered users;
69 visitors (0 registered, 69 guests [including 6 identified bots]).
Forum time: 13:52 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5