Implement zero inflated multinomial distribution

[ABWerks]

Multinomial regression is possible in brms using the multinomial family and it works very well (thank you)! However there is currently no way to account for 0 count across all possible choices. This happens often when modeling clinic use in health care settings where patients have the opportunity to engage in care across multiple modalities (in person, telehealth, nurse hotline, etc..) in any year, but often don't visit at all. Having the ability to account for the latent probability of no visits would (I think better) model the ebb and flow of care.
Koslovsky has done some work here and I have used that package, but it lacks the flexibility of brms:
Paper
Repo

[paul-buerkner]

Thanks! I will take a look when I have time :-)