paul-buerkner/brms

Beta distribution for multivariate models

Closed this issue · 2 comments

As of now, responses' correlations can only be estimated for gaussian and student models. Could it please be extended to Beta() distributions? Many thanks!

Inference of correlations in multivariate gaussian and Student families is available due to their particular structure and doesn't generalize in a straightforward manner to other distributions.

There are ways to address that problem, but sorting that out would require a more detailed discussion of the analysis you have in mind, which should be posted in the Stan forums.

Not to hijack this thread but you can estimate correlated latent residuals for Beta-distributed observations using factor models in Stan with {mvgam} Agree though it is better to post on the Stan discourse and I can provide a reprex there to get you started