Thanks Paul.
I just tried it on one of my datasets, and it works like a dream :-)
In DirichletReg it is possible to specify different models for each component in the proportion, is something similar possible in brms? And If I wanted a more robust regression, would it make sense to use a horseshoe prior or something similar?
In the documentation phi appears to have a log link, but in the model summary it states that the identity link was used, this even if I set dirichlet(…, link_phi = “log”), is this an error?
Thank you
/Jannik