[R-sig-ME] lme4 - GLMM dispersion parameter?

John Maindonald john.maindonald at anu.edu.au
Thu Aug 2 00:39:54 CEST 2012

If Liam is wanting the (approximate) equivalent of a dispersion 
parameter, then put in an observation level random effect.
Create a factor that has one level for each observation, and 
include this as a random effect (maybe +~1|obs)

Genuine poisson variation is uncommon in such contexts --
typically one can expect some clustering.  One should
accordingly check the fitting of an observation level random

John Maindonald             email: john.maindonald at anu.edu.au
phone : +61 2 (6125)3473    fax  : +61 2(6125)5549
Centre for Mathematics & Its Applications, Room 1194,
John Dedman Mathematical Sciences Building (Building 27)
Australian National University, Canberra ACT 0200.

On 01/08/2012, at 10:56 PM, Ben Bolker wrote:

> Liam Crowther (BIO <L.Crowther at ...> writes:
>> Dear list users, I'm using lme4 to model the densities of several
>> bee species in response to landscape gradients. For some species
>> I've used just a random intercept and for others I've allowed a
>> random effect of forage quality, this is determined by comparing
>> maximal models with the different random components before refining
>> the fixed effects. The dependent variable is a count at a
>> transect of which there are repeated measures so I'm using a GLMM
>> with Poisson errors, examples of final models below:
>> hy25<-glmer(Bh~+DATE+bees$X250PCURB+bees$X250PCOSR+bees$X250PCWOO+ 
>> (1|TRANSECT), data = bees,
>> family =poisson)
>  It may work for now but it's ugly and maybe eventually problematic to use
> bees$ inside the formula:
> hy25<-glmer(Bh~DATE+X250PCURB+X250PCOSR+X250PCWOO+(1|TRANSECT), 
>  data = bees, family =poisson)
> would be clearer.
> [snip]
>> In total there are 338 observations of 42 subjects, 
>> is there a general method for extracting a dispersion
>> parameter from models such as these 
>> (there are models for 7 spp. at 3 different scales) where there are
>> differing numbers of predictors and random effects?
>  Don't quite understand the question.  Poisson models don't
> have dispersion parameters.  Do you want to extract the random-effects
> variances and covariances (?VarCorr) ? Compute an estimate
> of overdispersion (sum(residuals(model,type="pearson")^2), and
> and see http://glmm.wikidot.com/faq ?
>  Ben Bolker
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

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