[R-sig-ME] weights = varIdent - differences even if "weighted factor" is not in the model

[Friso Muijsers]

Dear mixed-models mailing list users,

I have come across a strange behaviour of the "varIdent" function in the 
nlme package, that makes me wonder if my understanding of this function 
is completely wrong:

By accident I found the following: if I have a model using a varIdent 
function for a certain factor, that is not included in the final model 
(for whatever reason) then it still changes the model estimates. I 
thought that, if a model does not contain a factor, than leaving it in a 
weights-argument should not change the fitting process.

Small example:

fm1 = lme(distance~Sex, data = Orthodont, random = ~1, weights = 
varIdent(form = ~1|age))
fm2 = lme(distance~Sex, data = Orthodont, random = ~1)

anova(fm1,fm2)
summary(fm1)
summary(fm2)

I have the feeling that I should simply remove the weights argument of 
factors that are not in the model, but I am now un-shure if that is the 
right thing to do, and if there is something important here, that I am 
missing!?

Can someone help me back on course? I have to admit that I am not very 
firm with the whole theory of the different weights-methods, allthough I 
have read the respective chapter of Pinheiro & Bates  book


Thanks for reading!

Friso

-- 
Friso Muijsers

Institute for Chemistry and Biology of the Marine Environment (ICBM)
Carl-von-Ossietzky University Oldenburg
Schleusenstrasse 1
26382 Wilhemshaven