[R-sig-ME] MCMCglmm covariance matrix specification

[Jarrod Hadfield]

Hi,

As I understand it you have 3 2x2 covariance matrices to be estimated,
one for each environment?

~us(at.level(env, 1):trait):animal+us(at.level(env,
2):trait):animal+us(at.level(env, 3):trait):animal

should work. I presume you have no shared pedigree between the
envrionments hence the cross-env covariances are not estimable? In the
computation time is long, get back to me; there are ways to
reparameterise it to make it faster but it's a bit fiddly.

For harder problems (where the covariance matrix can't be permuted such
that the estimable bits fall in blocks along the diagonal, as here) then
fixing elements to zero is probably not a good idea even if you could do
it (for example in asreml). The zero elements will force patterns in the
estimable elements to ensure positive-defitness. The antedependence
solution I posted earlier gets round this issue I believe.

Cheers,

Jarrod


On 24/02/2021 18:29, Walid Crampton-Mawass wrote:
> This email was sent to you by someone outside the University.
> You should only click on links or attachments if you are certain that the email is genuine and the content is safe.
>
> Hey all,
>
> Hope you are doing well during this time!
>
> I have been racking my brain for weeks on how to do model this issue but I
> have found nothing other than one old answer by Jarrod Hadfield (
> https://stat.ethz.ch/pipermail/r-sig-mixed-models/2015q4/024036.html) which
> recommends using an antedepedence model. Here is the issue:
>
> I have constructed a bivariate animal model (trait1, trait2) with a random
> interaction with the additive genetic random effect and the residual
> variance,i.e. (trait:env):animal. The interaction variable is a categorical
> environmental variable of 3 levels (Low, Mid, High). So my
> variance-covariance matrix has a 6x6 shape (2traitsx3env). Hence, the
> matrix would include both among-trait covariances within the same env and
> between env, and cross-env covariances for the same trait:
>
> trait1:low trait1:mid trait1:high trait2:low trait2:mid trait2:high
> 1 0 0 0 0 0
> 0 1 0 0 0 0
> 0 0 1 0 0 0
> 0 0 0 1 0 0
> 0 0 0 0 1 0
> 0 0 0 0 0 1
> (1 represent variances, 0 represent covariances)
>
> I have already run the model with both the idh() and us() specification. In
> the first case, no covariances are calculated at all, only variances are
> calculated. In the second case, all types of covariances are calculated.
>
> I need help figuring out how to specify the variance-covariance matrix in
> MCMCglmm (and prior) in a way to tell the model not to estimate the
> cross-env covariances, only the among-trait covariances should be
> estimated:
> trait1:low trait1:mid trait1:high trait2:low trait2:mid trait2:high
> 1 x x 0 x x
> x 1 x x 0 x
> x x 1 x x 0
> 0 x x 1 x x
> x 0 x x 1 x
> x x 0 x x 1
> (1 represent variances, 0 represent covariances to be estimated, x
> represent covariances fixed at 0, i.e. not estimated)
>
> any help would be appreciated!
> --
> Walid Crampton-Mawass
> Ph.D. candidate in Evolutionary Biology
> Population Genetics Laboratory
> University of Québec at Trois-Rivières
> 3351, boul. des Forges, C.P. 500
> Trois-Rivières (Québec) G9A 5H7
> Telephone: 819-376-5011 poste 3384
>
>          [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models using 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. Is e buidheann carthannais a th’ ann an Oilthigh Dhùn Èideann, clàraichte an Alba, àireamh clàraidh SC005336.