Getting variance components [🇷 for BE/BA]

posted by d_labes  – Berlin, Germany, 2017-02-07 12:16 (2922 d 19:29 ago) – Posting: # 17029
Views: 18,871

Dear StatR,

❝ ... More clearly, I want to extract between subject variances (sigmaBT^2, sigmaBR^2), within subject variances (sigmaWT^2, sigmaWR^2) under the test and reference formulation, also subject-by-formulation interaction (sigmaD^2 = sigmaBT^2 + sigmaBR^2 - 2*rho*sigmaBT*sigmaBR).


What is needed for that goal is to formulate a model which have these variance-covariance parameters.
Your model is one with assuming sigmaWT = sigmaWR and sigmaBT=sigmaBR, rho=1.
Thus you have only two variance-covariance parameters in your output above:
Random effects:
 Groups   Name        Std.Dev.
 Sequence (Intercept) 0.002309 # sigmaBT = sigmaBR
 Residual             0.356224 # sigmaWT = sigmaWR


Unfortunately there is currently no way to formulate a model similar to the FDA code for replicate designs within package lme4 / lmer(). See f.i. this post for more details on the why.

In the older package nlme / lme() you may formulate the model similar to the FDA code. See that post.
But nevertheless you can't get the 'correct' CI from that R solution since the degrees of freedom for the contrast T-R are different to that of SAS Proc MIXED.

Hope this helps.

Regards,

Detlew

Complete thread:

UA Flag
Activity
 Admin contact
23,376 posts in 4,912 threads, 1,667 registered users;
95 visitors (0 registered, 95 guests [including 2 identified bots]).
Forum time: 07:46 CET (Europe/Vienna)

The great tragedy of Science – the slaying
of a beautiful hypothesis by an ugly fact.    Thomas Henry Huxley

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