I am investigating how environmental conditions influence reproductive trade-offs, especially in terms of residual covariance (allocation trade-offs). Instead of fitting a standard univariate model with one trait as the response and the other as a fixed effect in interaction with environmental characteristics, I am trying to use my previous bivariate framework and would need to estimate the covariance between traits for different levels of my environmental variable. However, in MCMCglmm, I am still unsuccessful at designing a covariance matrix structure with null residual covariances between the same trait measured in environments (not meaningful) but with non-null covariances between the two traits in a given environment. I expect the function cors could help me build a residual covariance matrix with estimated submatrices for the covariance between traits but null covariances between environments. However, I do not know and could not find which syntax to use to do this and how to specify the priors. All I manage to estimate for now is a residual matrix with null covariances: ~idh(trait:environment):units . Any help with this would be great!
Hi,
I am not specialist of MCMCglmm. On the opposite, I have been unsuccessful at using it for a particular task...
Although you are doing a different thing, I think this tutorial might help you: http://devillemereuil.legtux.org/wp-content/uploads/2012/12/tuto_en.pdf
In particular the "multi-traits model" part.
The "us" function is the analogue of the "idh" function, but it allows non null covariance between traits. Might this be the only thing you are looking for?
If you are only interested by the co-variance between traits within each environment, would it be meaningful to use a separate model for each environment?
Cheers,
Good-luck!
hugo
With some help from Jarrod Hadfield, here is the syntax I worked out:
cov = ~ us(trait:at.level(environment, 1)):units + us(trait:at.level(environment, 2)):units
adding mode terms if there more than 2 levels of environmental conditions.
The prior can be specified with a matrix for each environmental level. In my case:
R = list(R1 = list(V = diag(c(1/3,1/3)), n = 2), R2 = list(V = diag(c(1/3,1/3)), n = 2))
http://jarrod.bio.ed.ac.uk/software.html
Dear Charlotte, Could you share the full MCMCglmm prior and model syntax for this. Would be very helpful for me as I am conducting now fairly similar analysis.
@Tarmo Ketola
Here is an example, where "explanatory" is a two-level factor. However, the priors and computing details in bold will depend on your data, study system and assumptions, so please use with caution. Hope that helps!
prior
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling