[R] quasi complete separation in mixed logistic

[Massimo Fenati]

Dear "R workers",
I have a question about a mixed logistic regression for 
repeated data. After fitting a traditional logistic 
regression a quasi-complete separation  with too large SE 
for a covariate (sex) was shown. In these days I read a 
lot of pubblication about the possible solutions of 
quasi-complete separation (such as exact logit, Firth 
method, delete variable, etc.). I tried with "zelig" and 
"logistf" and some improvement were seen. Now I would like 
to reduce the quasi complete separation bias for mixed 
logistic regression, but I don't known how to do it. Is 
the Bayesian inference the best choice?
Could you help me?

Thanks in advance!

Massimo Fenati


MASSIMO FENATI
-----------------------------------------
Medico Veterinario (DVM)
Via Barchessa, 19
35040 - Boara Pisani (PD)- Italy
tel: +39 0425958083
cel: +39 3392114911
fax: +39 0425958083
e-mail: massimo.fenati a infs.it