@cluhmann thanks a lot for the quick reply. I"d be very glad for any pointers w.r.t a possible reparametrisation for the model. Please note, I am using the hierarchical model in non-centered form from the example directly, and need to increase the acceptance probability to e.g 0.999 when I go from 5 to 50 groups.
Also the example sets the acceptance probability to 0.99 already for 5 groups. This doesn’t seem to match wit your expectation, i.e. this specific example (using the recommended non centered parametrisation) already warns about divergences for the default 0.9 acceptance threshold.
I am probably overlooking sonething here. I am just getting started…