[R] GEE models

[Bonenfant Christophe]

Hi there,

I'm trying to fit GEE models on binomial data. My problem is how to "select" a model if ever it is possible and how I decide a factor has to be drop out or not. More specifically:

1) Are QIC (Pan 2001) reliable and can be used in a model selection framework (Burhnam and Anderson 1998)? If yes, how can the QIC be estimated from available information in a gee object?

2) If not, can R compute an ANOVA table like for GLM (seems that SAS does it)? The problem is to know whether a factor with more than 2 levels is significant or not. Can I consider such a factor as significant if at least one modality (using a Wald-test)  is significant?

3) Some models can be fitted with a gml but not with a gee? Is there any explanation for that?

Any answer would be greatly appreciated,

Christophe



------------------------------------------------------------------------
Bonenfant Christophe

Laboratoire de Biométrie et Biologie Evolutive
UMR CNRS 5558 - Université Claude Bernard Lyon 1
43, Boulevard du 11 Novembre 1918
F-69622 Villeurbanne Cedex

Tél: 04-72-44-81-11
Mobile: 06-62-11-20-00
Courriel: bonenfan at biomserv.univ-lyon1.fr
In Tartiflette we trust...