Bioequivalence and Bioavailability Forum

Hi dixit,

I am not a SAS-user myself, so the following is said on basis of very little insight.

Your code seems to be a direct translation of the FDA code, and as such I would expect it to give a result.

Have you worked with and got meaningful results from exactly this code before?

You specify "random treatment/type = FA0(2) subject=subject G;", where a direct translation from the FDA's proposal would be "..SUB=SUBJECT G" - I am not sure this makes a difference though, but the change can be tried. Same in the other line.

Also note this page; "It will always occur if you have code which includes a REPEATED statement AND you have named an effect on the REPEATED statement AND there are two or more observations from one subject which have identical effect values." - a check of your data listing may be necessary.

It could also be that your code and data are ok, but that the optimiser does not converge. When the optimisation is done the program 'guesses' a set of initial values in the covariance matrix and then changes each of the N values (s₁, s₂, s₃ ... s_N) one by one until the likelihood of the matrix is maximised. The way these changes are done is complex, and if the functional surface of likelihood (s₁, s₂, s₃ ... s_N) is not nice and smooth, the optimiser may not converge with automated settings (pardon my French). In this case it will be necessary to manually seed the initial values, possibly even take control over how each optimisation step is done. I do not know how to do that in SAS (or in R for that matter).

Best regards
EM.