[R-sig-ME] GoodnessOfFit-and-PseudoR²computingFor-glmmTMB-models

[Paul Johnson]

Hi,

My preferred approach to overdispersion is none of 1-3 but to assume it applies (it usually does, for biological data anyway), and make sure the model includes a parameter to model overdispersion. For binomial (except Bernoulli) and Poisson you can include an observation-level random effect (OLRE). You can then gauge the amount of overdispersion in the model from the size of the OLRE variance estimate. (Note the OLRE will mop up variation due to *all* sources of lack of fit, including poor model specification, e.g. fitting a straight line where a curve would fit better.)

Best wishes,
Paul


> On 6 Feb 2018, at 14:29, C. AMAL D. GLELE <altessedac2 at gmail.com> wrote:
> 
> Hi, dear all
> Please, your help for the following problem:
> when I fit a mixed model using glmmTMB (poisson family or others), how do I:
> 1) check, if the model fits well my data (goodness of fit)?
> 2) check if my model is overdisperced or not (by using sigma(model)?)
> 3) compute an pseudo R² to see the percentage of the variability of my
> response which is explained by my model?
> In advance, thanks for your answers.
> Regards,
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models