Brms ordinal regression vs ordered_logistic/ordered_probit

Helping a colleague with a brms-specified ordinal regression model and I’m a little confused by the terminology. For ordinal regression, I’m most familiar with the ordered_logistic/ordered_probit specification with parameters for the latent mean and cutpoints. In brms I see generated stancode is using the cumulative_logit_lpmf/cumulative_probit_lpmf and intercepts of some sort. Is intercept simply another name for cutpoints?

This is a useful paper: https://psyarxiv.com/x8swp/ which says

Note that there is no intercept in the predictor term, because the thresholds
τk replace the model’s intercept as both are not identified at the same time.

Ok, so in the terminology of that paper, ordered_logistic with cutpoints is the same as a cumulative model with thresholds, yes?

Yes. See also my ordinal tutorial paper.

1 Like