[R-sig-ME] Generalized Poisson Regression (Underdispersion)

Ben Bolker bbolker at gmail.com
Tue Jan 19 21:32:24 CET 2016


   Depends whether you mean in a standard (extended) GLM with only fixed 
effects (in which case the question is a little bit off-topic here ...) 
or in a GLMM.

   In the GLM case you can use the quasipoisson distribution in glm(); I 
think you have lots of other choices (including the ComPoissonReg 
package).  library("sos"); ?findFn is your friend.

   In the GLMM case your options (that I know of) are a lot more 
limited. You could use MASS::glmmPQL with a quasi-Poisson response (but 
I would definitely test it on simulated data with known characteristic 
first!).  You could use a GLMM with a Poisson response and adjust the 
standard errors yourself.  Other distributions could be built into any 
of the 'toolbox' packages (glmmADMB, glmmTMB, brms), but you'd have to 
do it yourself or convince someone else ...
    You might get somewhere with the hglm (hierarchical GLM) package as 
well, although I don't immediately see an underdispersed Poisson family.
   Ordinal response models (e.g. the ordinal package) are another 
alternative ...

   Looking forward to hearing other people's answers

   cheers
     Ben Bolker


On 16-01-19 02:27 PM, Timothy Lau wrote:
> Hello,
>
> I noticed that STATA has some functionality for handling
> underdispersion with a theta estimate (similar to glmer.nb ?).
> http://www.stata-journal.com/sjpdf.html?articlenum=st0279
>
> What are some options that are available to me in R and the GLMM
> particularly for similar situations?
>
>
>
> Best,
> Timothy
>
>
>
> "The will to win means nothing without the will to prepare."
> http://en.wikipedia.org/wiki/Juma_Ikangaa
>
> _______________________________________________
> 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