Bioequivalence and Bioavailability Forum

Hi Bears

❝ Snot <- lm(lnY~Trt+Subj:Seq+Per+Seq) for a 2x2x2 ABE. It

❝ works o.k..

I guess you are right, but I think this is fishy. I could be wrong, but the colon as far as I understand from the manual is used for interactions?! Try and look at the model matrix with your syntax, where you will see columns filled with zeros. For example, with these data you will see a column called "Subj2:Seq2" which consists of all zeros, simply because Subj2 is assgined to Seq1 (and therefore the column called "Subj2:Seq1" has a few ones in it). A zero clumn is meaningless as it conveys no info to the fitting.

In my naïvity I just thought Subj%in%Seq would be better, and I just tried it, but that seems to result in the same phenomenon anyway. In both cases if you compare e.g. the column "Subj2:Seq1" with the "Subj2" column of the original fit, they are identical (and so forth). Therefore, in this case I think the explicit declaration of a nested factor in the model is not necessary.

EM.