[R-sig-ME] Is it worth testing for overdispersion?
Thierry.ONKELINX at inbo.be
Tue Jun 21 14:13:15 CEST 2011
You can add an observation level random effect to the model. If there is overdisperion in your dataset, then this random effect will have a non-zero variance. You can compare models with and without this observation level random effect to test for overdisperion.
> -----Oorspronkelijk bericht-----
> Van: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed-models-
> bounces at r-project.org] Namens Iker Vaquero Alba
> Verzonden: dinsdag 21 juni 2011 13:53
> Aan: r-sig-mixed-models at r-project.org
> Onderwerp: [R-sig-ME] Is it worth testing for overdispersion?
> Dear list:
> I read some time ago in one of the posts that the option of implementing
> quasi- families in lmer had been removed. Is that right?
> I´ve been a lot of time trying to find a way to test for overdispersion in lmer
> without success. And now that I read that I won't be able to use a quasi-poisson
> distribution for my data in case the model is overdispersed assuming poisson
> errors, I wonder wether it's really worth testing for overdispersion, and what
> alternatives do I have in case my model is overdispersed.
> Thank you very much!
> Iker Vaquero
> [[alternative HTML version deleted]]
More information about the R-sig-mixed-models