I submit part of a message I received from Ana.
The between-subject variance of test, sig_BT^2, the between-subject variance of reference, sig_BR^2, and the G matrix for the random model, can be computed as follows. This still assumes that the test formulation precedes the reference. The ‘_11’ notation on the lambda parameters (which indicates that they are for the first random model and first group) has been dropped to improve readability:
G(1,1) = sig_BT^2 = lambda(1,1)^2
G(2,2) = sig_BR^2 = lambda(2,1)^2+lambda(2,2)^2
G(1,2) = rhosig_BTsig_BR = lambda(1,1)*lambda(2,1) (= between subject covariance for test and reference)
The subject-by-formulation interaction variance in general is:
sig_D^2 = sig_BT^2 + sig_BR^2 - rhosig_BTsig_BR , and can be computed from the above equations as:
sig_D^2=G(1,1)+G(2,2)-2*G(1,2)
Using above formulas and the data from the variance tab of the progesterone Full RSABE ABAB template (attached) I have calculated sig_D^2 to be ~ -0.80391.
I have G (1,1) =0.85299
G(2,2)=0.000000833919
G(1,2)= 2X 0.82840678
So sig_D^2 = -0.8038
I do not see how it can be negative, but is my calculation correct? Or is it a function of the imperfection of the simulated data used by the RSABE template. I find this puzzling,
Any thoughts,
Thanks,
Angus
FDA RSABE Project template- Full Replicate Executed v1.4.phxproj (3.08 MB)