Why when we have multivariate models with different response distributions, we cannot estimate residual correlations in brms?

Modeling
1

I didn’t find the reason why you can’t do this estimation and if it’s only for students and normal

2

Not being familiar with really with multivariate models in brms, I’d have to guess it’s because when you assume your response variables are MVN, you can estimate a covariance matrix with the correlation matrix. If your responses aren’t normal, you’re not fitting an MVN and therefore there’s no correlation matrix. You’d probably have to use a copula to get the residual correlations.

2 Likes
3

Can copulas be made in brms to solve my doubt? @paul.buerkner

1 Like
4

There is an open issue about copulas for brms on github but I didn’t yet find the time to work on it.

1 Like