[R-sig-ME] estimating the "right" theta in a glmmmPQL model with neg bin error family and spatial structure
alexandre.villers at cebc.cnrs.fr
Tue Oct 13 08:36:32 CEST 2009
Good evening dear R-list members,
I would like to have your advices on the best way to estimate the theta in a spatial glmmPQL with a correlation structure (such as correlation=corSpher(form=~X+Y)) and a negative.binomial family.
Contrary to gam and the negbin family (mgcv), which can estimate theta, the negative.binomial family of MASS requires a fixed value of theta.
How am I to specify this value ? Shall I run the model on a range of values for theta and look for one that gets the scale parameter close to 1 (by the way, is it the $sigma in glmmPQL objects ?). Or is there a way to use the different functions in MASS (theta.md, theta.ml) for that purpose ?
I could not find anything on the forum pages but hope it is not a too trivial question...
79360 Beauvoir sur Niort
__________ Information from ESET Mail Security, version of virus signature database 4501 (20091012) __________
The message was checked by ESET Mail Security.
More information about the R-sig-mixed-models