Low ESS and High RHat for Random Intercept & Slope Simulation (rstan and rstanarm)

[bbbales2]

@avehtari

I guess this is related to centered vs. non-centered parameterization. It is known that centered would be better for strong likelihood (compared to prior, ie small \sigma_e, what you call small variation in sample) and non-centered would be better for weak likelihood (compared to prior, etc). rstanarm is using non-centered parameterization by default.

Yes this is what I’d try too. Here’s an interesting thread on this: Partial non-centered parametrizations in Stan

If you build your model with brms (which has similar syntax to rstanarm) you can generate code/data with make_stancode and make_standata and then do the centering yourself.