Dear Allstaters,
suppose we are using a Poisson model instead of using a Cox model.
The potential presence of an extra-Poisson variation would affect the significance of the covariates in the model (inflation/deflation of betas standard errors).
Is the assessment of extra-Poisson dispersion always necessary for a correct model specification?
Is there any connection with a possible misspecification of the equivalent Cox model?
In my case, had I not decided to use Poisson instead of Cox, I wouldn't have taken this aspect into account.
(I'm using Poisson via GLIMMIX macro as a practical device to build further models that will also account for clustering.)
Many thanks,
Fabio
Fabio Pellegrini, Biostatistician
Laboratory of Clinical Epidemiology of Diabetes and Cancer,
Department of Clinical Pharmacology and Epidemiology
Consorzio Mario Negri Sud
via Nazionale 8/A
66030 Santa Maria Imbaro (Chieti), Italy
Phone: ++39 0872 570264
Fax: ++39 0872 578263
e-mail: <[log in to unmask]>
http://www.negrisud.it
|