[R-sig-ME] coding multivariate + multiple membership in MCMCglmm

Jarrod Hadfield j.hadfield at ed.ac.uk
Wed Oct 12 17:50:46 CEST 2011

Hi Alex,

I'm not entirely sure exactly what you would like the model structure  
to look like. Currently you have (for the multimembership bit)

C[i,j] = U[id1[i], j]+U[id2[i], j]

where C is the nx4 response indexed by i (units)  and j (trait).

U is a mx4 matrix where m is the number of individuals indexed by id1  
or id2.

Currently you assume all elements of U have the same variance, which  
is estimated.

If it helps you can look at the design matrix:

Z<-model.matrix(~mult.memb(~trait:id1+trait:id2), data=your_data)



On 12 Oct 2011, at 15:14, Alexandre Courtiol wrote:

> Dear all,
> I am trying to fit, using MCMCglmm a multivariate mixed model  
> (gaussian)
> with an underlying multiple membership structure and I am not  
> certain about
> my code...
> Background: I have raters (N~200) who came from different origins  
> (A,B,C)
> and having different sexes (M,F), they ranked 4 sets of 20 objects  
> (sets
> 1,2,3,4) according to their preferences.
> I want to analyse wether origin and sex influence similarity in  
> rankings so
> I have computed correlations between all pairs of raters (~20,000).
> Cor1 represents the correlation observed between pairs of rater for  
> the set
> 1, cor2 represents the correlation observed between pairs of rater  
> for the
> set 2 (in the same order of pairs), cor3 for set3 and cor4 for set4.
> id1 represents the identity of one rater and id2 represents the  
> identity of
> the other rater within each pair of raters.
> I assume that the set of objects should not influence my fixed  
> effects, but
> could influence residuals and random effects.
> I coded:
>    prior <- list(R=list(V=1,nu=0.002),G=list(G1=list(V=1,nu=0.002)))
>    mod <- MCMCglmm(fixed=cbind(cor1,cor2,cor3,cor4)~sex*origin,
>                  random=~idv(mult.memb(~trait:id1+trait:id2)),
>                  rcov=~trait:units,
>                  family=rep("gaussian",4),
>                  data=data,
>                  prior=prior)
> Does it sounds good? I am particularly worried about the part
> "random=~idv(mult.memb(~trait:id1+trait:id2))".
> Thanks a lot.
> _
> alex
> 	[[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.

More information about the R-sig-mixed-models mailing list