At 11:17 2004-03-24, you wrote:
>I'm working with count data following over-dispersed poisson
distribution
>and have to work with mixed-models on them (like proc GENMOD on SAS sys.).
>I'm still not to sure about what function to use.
This is confusing: Proc GENMOD fits generalized linear models (GLM) and
handles modelling of overdispersed Poisson data using quasi-likelihood
(e.g. SCALE=P) or the negative binomial distribution (DIST=NEGBIN), but not
generalized linear mixed models (GLMM).
In R a GLM is fitted using the glm function, and specifying
family=quasipoisson is the equivalent of SCALE=P in SAS. The negative
binomial case is handled either by the negative.binomial family function
(when the shape parameter is known) or the glm.nb function (if you want to
estimate the shape by ML). Both negative.binomial and glm.nb are found in
Venables and Ripley's MASS package.
>It seems to me that a glmmPQL will do the job I want,
If GLMM is what you want, the functions glmmPQL (from MASS) or GLMM (from
lme4, by Bates and co-workers) will most likely handle your needs.
>but I'll be glad if people who worked on this type of data can share
what
>they learned. Thanks for your time.
If you describe your data and what you'd like to do, someone may be able to
help you.
The MASS functions, and much more, are described in Venables and Ripley
"Modern Applied Statistics with S." GLMMs are described in e.g. Brown
and
Prescott "Applied Mixed Models in Medicine", but it's very SAS
oriented.
//Henric