[R-sig-ME] Monte Carlo simulations for lmer with binomial link

[Jarrod Hadfield]

Hi Marco,

The newest version of MCMCglmm calculates p-values when you call  
summary. These should be reasonably close to what should be obtained  
under mcmcsamp/pvals.func unless there is a lot of over-dispersion. In  
this case the p-values will be larger  - as they should be.

For testing groups of effects as in anova, I recently recommended to  
someone that they calculate the posterior covariance matrix and pass  
it to Wald.test in the aod package. I am not advocating this, but I  
would be interested in other people's thoughts. If the posterior  
distribution of the effects is close to multivariate normal (its  
multivariate-t for Gaussian models) I can't see a problem, but perhaps  
there are some issues....

Cheers,

Jarrod



Cheers,

Jarrod


Quoting Marco van de Ven <m.a.m.vande.ven at gmail.com>:

> Hello,
>
> I fitted a linear mixed effects model (lmer) with the binomial link
> function (family = binomial). I tried to obtain pMCMC values for this
> regression model by using pvals.fnc, but this does not work. Similarly,
> mcmcsamp does not seem to work with binomial dependent variables. Are
> there alternative methods for obtaining p-values for these lmer models
> with Monte Carlo (or other) simulations? Many thanks in advance!
>
> Cheers,
>
> Marco van de Ven
> MPI Nijmegen
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> 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.