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

David Villegas Ríos chirleu at gmail.com
Fri Feb 12 15:23:05 CET 2016

I'm trying to organize a dataset to run a multivariate mixed model in
MCMCglmm. I have several response traits and since I have unequal number of
replicates per individual for each trait, I have been suggested to arrange
the data in a long format.

I have 5 response traits. Three of them were recorded as follows:

- Day 1: trait A
- Day 2: trait A and trait B
- Day 3: trait A and trait C
- Day 4: trait A and trait B
- Day 5: trait A and trait C

During the following 6-7 days I didn' t measure anything. And then after
that, the other two traits (trait D and trait E) were recorded weekly
during 54 weeks.

I want to run a multivariate model in MCMCglmm to estimate the random
var-cov matrix and the residual var-cov matrix. The later however must have
some constrains as for example, trait B and C do not covary at the residual
(within-individual level) and traits A, B and C do not covary at the
residual  level with traits D and E. So the residual var-cov matrix should
be something like:

   A B C D E
A 1
B x 1
C x 0 1
D 0 0 0 1
E 0 0 0 x 1

Where x represents the covariances that should be estimated, and 0 the
constrained covariances.


- Is it possible to fit such a residual var-cov matrix  in MCMCglmm? Maybe
with an antedependence model?
- How can I structure the dataset for the analysis? Do I need a "time"
column? I could use "day" for the first three traits, but then what about
traits D and E?
- Does it make sense at all to run this model, or would it be more
meaningful to get a mean value of traits A, B and C and used them as fixed
effects, since I have a very low number of replicates of them? In that case
I'd of course investigate the main effects of A, C and C on E and D, and
not the covariation amontg them...but it could be ok as well.


	[[alternative HTML version deleted]]

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