Bivariate additive categorical regression via penalized maximum likelihood. Under a multinomial framework, the method fits bivariate models where both responses are nominal, ordinal, or a mix of the two. Partial proportional odds models are supported, with flexible (non-)uniform association structures. Various logit types and parametrizations can be specified for both marginals and the association, including Dale’s model. The association structure can be regularized using polynomial-type penalty terms. Additive effects are modeled using P-splines. Standard methods such as summary(), residuals(), and predict() are available.