Ah that makes sense on each county mapping to 1 region only.
Would that change if the higher level has multi-membership (on another example…)? I see that you favoured the (1 | b_id) + (1 | mm(c1, c2)) rather than the / in the form of (1 | b_id/mm(c1, c2)) in this post: (Specifying multi membership on level 3 in model with errorsar structure). I also see from slide 73 https://www.barelysignificant.com/slides/RGUG2019#73 that 3-level system is modelled as (1 | study) + (1 | experiment) rather than (1 | study/ experiment) - are they equivalent?
Is there much difference in performance between the linear and non-linear syntax? I was playing with several other forms of the model that would be non-linear so kept the nl = true for convenience.
And lastly, on specifying a particular prior on a specific value of the group i.e. country 1 having a different prior to country 2, how would I change the extracted stancode above to do that?
Or is there a way to do so in brmsformula and specify them in set_prior? Something like log(Fines) ~ log(Levy) + (1 | gr(region/counties, by = country) )? https://rdrr.io/cran/brms/man/brmsformula.html