[R-sig-ME] Random effect on only one of the response variables in bivariate LMM (MCMCglmm)

[Jarrod Hadfield]

Dear Nicolas,

Is replicate 1 for trait 1 the same as replicate 1 for trait 2: do you  
expect trait 1 and trait 2 to be correlated if they are in the same  
replicate, over and above that due to strain effects?

If not then

random = ~ us(trait):Strain+us(at.level(trait,2)):replicate,
rcov = ~idh(trait):units

is appropriate, and you do not need to fix any variances.

Cheers,

Jarrod

Quoting Nicolas Rode <nicolas.o.rode at gmail.com> on Wed, 30 Jul 2014  
16:35:46 -0400:

>  <r-sig-mixed-models at r-project.org>
>
> Dear all,
>
>
>
> I'm trying to analyze the genetic correlation between two traits using a
> bivariate normal distribution in MCMCglmm.
>
> I have 32 strains with 3 replicates and 1 observation per replicate for
> Trait1 and 3 replicates and 2 observations per replicate for Trait2. I
> addition to the genetic variance-covariance matrix for Trait1 and Trait2, I
> would like to get the residual variance for Trait1 and Trait2, as well as
> the between replicate variance for Trait2.
>
>
> Would anyone know if it would be possible to fit a between replicate random
> effect to only one of the two traits in MCMCglmm?
>
>
> I detailed 3 different methods I’ve been using below but none of them
> appears satisfactory.
>
>
>
> 1/Fixing the between replicate variance to 1 for Trait1 (‘priorvarRep ‘ and
> fix argument below) also results in fixed residual variances for Trait1 and
> Trait2.
>
>
>
> priorvarRep<-priorvar
>
> diag(priorvarRep)<-c(1,diag(priorvarRep)[2])
>
>
>
> priorm1 <-list(G=list(G1=list(V= priorvar,n=2), G2=list(V=
> priorvarRep,n=2,fix=c(1,0))), R=list(V= priorvar,n=2))
>
>
>
> m1<-MCMCglmm(Value~-1+Trait, random = ~ us(Trait):Strain, rcov = ~
> idh(Trait):units, family = "gaussian", ,prior=priorm1, data = data)
>
>
>
> 2/ Fixing the residual variance of Trait1 to 1 results in very weird
> results for the residual variance of Trait2.
>
>
>
> Priorm2 <-list(G=list(G1=list(V= priorvar,n=2), G2=list(V= priorvar,n=2)),
> R=list(V= priorvarRep,n=2,fix=c(1,0)))
>
>
>
> 3/Out of curiosity, I’ve also tried a model without fixing any variance.
> The between replicate variance fitted for Trait1 is then pretty close to
> the residual variance of Trait1, but this might be an artifact.
>
> Priorm3 <-list(G=list(G1=list(V= priorvar,n=2), G2=list(V= priorvar,n=2)),
> R=list(V= priorvar,n=2,fix=c(1,0)))
>
>
>
>
>
> Thank you very much for your help.
>
>
>
>
>
> Best regards,
>
>
>
> Nicolas
>
> 	[[alternative HTML version deleted]]
>
>



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