MCMCglmm covariance matrix specification
Hey Walid,
I had a similar problem a few months ago, I didn't want 1 trait to have any
covariance. I was not able to find a solution to it but after reading
several articles, I figured out a way.
Instead of not calculating the covariances at all (which I am not sure if
it's possible), you can set the value to 0 in the prior.
Here is my example-
I have 9 traits and I do not want my last trait to covary with any other
trait. Hence I set it to a very low value (0.001) for the within individual
covariance and use the fix command to specify which trait it is (in this
instance my 9th trait).
final_priorv1 <- list(R = list(V =diag(c(1,1,1,1,1,1,1,1,0.001),9,9), nu =
0.002, fix = 9),
G = list(G1 = list(V = diag(9), nu = 9,
alpha.mu = rep(0, 9),
alpha.V = diag(25^2,9,9))))
Remember this is within individual variance set to 0.001, you will still be
calculating the among individual covariance.
Found this solution from Dr. Houslay's MCMCGlmm Tutorial page 14. Link
here-->
https://tomhouslay.files.wordpress.com/2017/02/indivvar_mv_tutorial_mcmcglmm.pdf
Something similar from his other tutorial page 31-->
https://tomhouslay.files.wordpress.com/2017/02/indivvar_plasticity_tutorial_mcmcglmm1.pdf
With this prior, I was able to use the unstructured (us) covariance matrix
and I was able to get what I needed.
I feel what you are looking for is fairly similar, and I believe this
solution might work for you.
Happy coding!
Regards,
Srivats.
Srivats Chari
<https://sites.google.com/ucd.ie/wildl-ecol-behav-at-ucd/people#h.p_DyWP_UxHDqgq>
Post-Graduate Research Student
Twitter- @WildlifeVats <https://twitter.com/WildlifeVats>
Laboratory of Wildlife Ecology and Behaviour
<https://sites.google.com/ucd.ie/wildl-ecol-behav-at-ucd>
School of Biology and Environmental Science (SBES),
University College Dublin (UCD).
On Wed, Feb 24, 2021 at 6:29 PM Walid Crampton-Mawass <
walidmawass10 at gmail.com> wrote:
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 Qu?bec at Trois-Rivi?res 3351, boul. des Forges, C.P. 500 Trois-Rivi?res (Qu?bec) G9A 5H7 Telephone: 819-376-5011 poste 3384 [[alternative HTML version deleted]]
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models