Rstanarm like covariance priors in brms?


I do like the way rstanarm writes down priors for variance components of random effects, but I don’t quite know how that can be expressed in brms. At the moment I specify for each random effect standard deviation a prior separatley, but I would like to specify a prior on the overall variance and use a dirichlet prior for allocating fractions of this to the individual components.

Is that possible in brms?

# so this is how my random effect prior often look like:
        prior(normal(0,0.5), class="sd", group="group1") +
            prior(normal(0,0.25), class="sd", group="group2")
# and a LKJ is used for the correlations

# how can I model the above through the sum of the variance of both random effects and a dirichlet to model the fraction assigned to each component?

Thanks a lot!


1 Like

This is not yet possible in brms and I want to do some more research on it to have good joint (default) priors in brms for random and fixed effects. Actually, there I will have a PhD student in Stuttgart starting soon who will work on these topics.