[R-sig-ME] Negative binomial mixed effects modeling

David Duffy davidD at qimr.edu.au
Wed May 12 05:33:12 CEST 2010

On Tue, 11 May 2010, David Zajanc wrote:

> Hi,
> I am looking into ways to fit a negative binomial mixed effects model to
> our data, but have not found any obvious methods in searching through
> the archives of this list.  I noticed suggestions for using the
> quasipoisson distribution, but our dataset appears better suited for the
> negative binomial model, based on past analyses using GLM.  It appears
> that "glmer" of package "lme4" is not currently capable of fitting a
> negative binomial model, though please correct me if I'm wrong about
> this.  Any feedback or suggestions would be welcomed.

If you have only a "single level" of random effects eg simple clusters, 
then gamm() [mgcv] or spm() [SemiPar].  If a negative binomial GLM fits, 
then an "ordinary" poisson 
GLMM may be appopriate (the REs have to come from a gamma to give a 
negative binomial IIRC, in which case maybe one could use the survival 
package with gamma frailty???).  You could also go to 
(other) semiparametric mixed models (see refs in Diao and Lin Am J Hum 
Genet 2005l 77:97-111).

In one of my analyses of total body mole count, diagnostics showed the 
negative binomial and cube-root-transforms were equally nice for mixed 
model, FWIW.

| David Duffy (MBBS PhD)                                         ,-_|\
| email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
| Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v

More information about the R-sig-mixed-models mailing list