For the cumulative family, brms doesn’t support category-specific effects. However, in the Stan code that can be saved when using brm() with “family = cumulative()”, I found the functions cumulative_logit_lpmf() and cumulative_logit_cs_lpmf(), the second one giving a hint that category-specific effects could be possible for the cumulative family when replacing the call to ordered_logistic_lpmf() in the model block by a call to cumulative_logit_cs_lpmf(). So my question is: Why are category-specific effects not supported for the cumulative family? I guess there is a proper reason but I can’t see it yet.
cumulative models with category specific effects can imply negative probabilities.