"Alex" <[EMAIL PROTECTED]> writes: > Duncan, > > I think that the problem in your comparison is locatted in the > method of estimation of fixed-effects: in REML they are not > estimated maximizing a joint likelihood function including > fixed-effects and variance- covariance components as ML does. So > when it comes down to different fixed-effects specifications, they > can't be compared directly by means of the log-likelihoods.
Just to pick a small nit, you can actually get both fixed and random effects parameters by optimizing a single modified likelihood. The problem is that the modification term (see the post from I. White) depends on the design matrix for the fixed effects, so it is not invariant under reparametrization, nor under model reduction. -- O__ ---- Peter Dalgaard Blegdamsvej 3 c/ /'_ --- Dept. of Biostatistics 2200 Cph. N (*) \(*) -- University of Copenhagen Denmark Ph: (+45) 35327918 ~~~~~~~~~~ - ([EMAIL PROTECTED]) FAX: (+45) 35327907 ______________________________________________ R-help@stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
