[R-sig-ME] MCMCglmm, priorR and binomial distribution

[Jarrod Hadfield]

Hi Camille,

The second prior is the correct one. The residual variance is not  
identifiable in the likelihood for binary data: I have tried to  
explain this intuitively in Section 2.6 of the CourseNotes using a  
tasteless example of hospital deaths.

Cheers,

Jarrod



Quoting Camille Madec <camille.madec at ebc.uu.se> on Thu, 31 Jan 2013  
15:21:12 +0100:

> Dear everyone,
>
> I have a model with 2 fixed factors, 1 random factor and a binary  
> response variable. I ran a MCMCglmm with family=”categorical” and  
> the prior for the residual being R=list(V=1, nu=0.002). In the  
> summary of the model I got high post.mean values (around 50 for  
> fixed effects and >1000 for random effects and sometimes up to 14000).
> I ran the same model with R=list(V=1, fix=1) which means that the  
> variance of the residual is fixed to 1, so the residual becomes a  
> fixed factor (if I understand correctly). In that case my post.mean  
> values are smaller (between zero and 24).
>
> My questions are:
> 1) Are the large values in the first case normal?
> 2) How do I know which prior is the more appropriate for the residual?
>
> Bests,
> Camille
>
> -----
> Camille Madec
> PhD student
> Plant Ecology and Evolution
> Uppsala University
> Sweden
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>



-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.