[R-sig-ME] residual covariance structure and long format data in MCMCglmm

[David Villegas Ríos]

Hi all,

I'r trying to run a multivariate MCMCglmm with 3 traits. I was suggested to
use the long format since I have unequal number of replicates per trait.
Trait 1 and 2 were replicated twice, trait 3 was replicated five times.
Traits are gaussian.

The way I measured the trait for each individual is as follows:

day 1: trait1
day 2: trait1 and trait 2
day 3: trait1 and trait 3
day 4: trait1 and trait 2
day 5: trait1 and trait 3

>From the model, I'm interested in extracting the between-individual
variances/covariances and if possible, the within-individual
variances/covariances.

This is my attemp so far. Letter identifies the trait.

mod1=MCMCglmm(value~(letter-1), random=~us(letter):id,
rcov=~idh(letter):xxxx, family=c("gaussian"), data=ALL)

My questions are about the rcov bit, the residual variances/covariances...

- First I don't know if with my experimental design, it makes sense to
estimate residual covariances ("us" structure) or constrain them as in the
model above ("idh" structure)

- Second, I don't know how to define the xxxx variable according to my
experimental design.

I guess both questions are related.

Any advise will be appreciated.

Thanks


David

	[[alternative HTML version deleted]]