In a detailed tutorial, @betanalpha makes the case for using a dirichlet prior in ordinal regression models (Ordinal Regression).

I confess, I wasn’t able to understand all that much of it besides the fact that a dirichlet prior sounds like a good idea. Is there any way to implement a weakly informative dirichlet prior on the cutpoints in brms? This was also mentioned by @paul.buerkner as a possible feature to develop (Dirichlet prior for cutpoints in ordinal models · Issue #762 · paul-buerkner/brms · GitHub), but I’m not sure if it was ultimately implemented.

I have seven response options in the ordinal scale, and am estimating responses using a cumulative probit distribution.

At the moment, I’ve had ‘success’ simply changing the extremely vague default of student_t(3, 0, 10) to something that much less favors extreme cutpoints: normal(0, 1.5), or student_t(5, 0, 1.5). Is there something wrong with this approach, if the dirichlet option is not possible? - it at least seems better than the default option to me!