[R-sig-ME] MCMCglmm random intercept/slope model and credible intervals

[Jarrod Hadfield]

Hi,

There is a bug in predict.MCMCglmm for random regression models. I've  
corrected it in the unreleased version, but you can change it easy  
enough:

Change

M[,which(rm.v),]<-0

on L65 to

M[,which(rm.v)]<-0


Also, this model will not give sensible results. You need to use  
something like:

R = list(V = 1, fix=1)

in the prior for the residual variance (it cannot be estimated from  
the data with categorical data)

Also,

G = list(G1 = list(V = diag(2), nu = 6)))

is pretty informative, unless you have a lot of data and replication  
at the right level.

Cheers,

Jarrod


Quoting Maya <maiski at maiski.net> on Mon, 7 Oct 2013 12:52:06 +0000 (UTC):

> Hello all,
>
> i have the following problem.
> I want to fit a GLMM with a numerical fixed effect, a factorial random
> effect, containing subject ids and a binary outcome. My code is:
>
> fam='categorical'
>
> priors1<-list(R = list(V = 1, nu = 0.002), G = list(G1 = list(V = diag(2),
> nu = 6)))
>
> modelGlmm1<- MCMCglmm(toModel ~ fixedEffect,
> random=~us(1+fixedEffect):randomEffect, data=data, verbose=FALSE,
> family=fam, pr=TRUE, pl=TRUE, nitt=nittM, thin=thinM, burnin=burninM,
> prior=priors1)
>
> mp1 = predict(modelGlmm1, interval = 'confidence')
>
>
> But as I try to get the credible intervals I get the following error:
> Error in M[, which(rm.v), ] <- 0 : incorrect number of subscripts.
>
> Could anybody give me some hints or enlighten me on the subject? I cannot
> figure it out by myself :/
>
> I have also another question: How are the credible intervals constructed:
> pointwise or simultaniously?
>
> Thanks in advance!
>
> Greets,
>
> Maya
>
> _______________________________________________
> 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.