Hi @matti -
I’ve coded up a new model that I think might be able to implement these kinds of things. It explicitly combines beta regression within ordinal modeling. I’m currently working on comparing it to the existing models. You can see the code at this post:
Initial comparisons to the ZOIB and standard beta regression suggest it captures a larger (versus standard beta reg) and more statistically efficient (versus ZOIB) parameter on the mean of the overall distribution for a given covariate. The model only has a single linear model that predicts everything, so marginal/predicted effects over the whole distribution are fairly easy to calculate.
What is also interesting is that because it has ordered cutpoints, it could be possible also to model where the cutpoints occur–i.e., what is the point where a respondent “moves” from (0,1) to 0 or 1.
Would love to get your thoughts on the approach – @paul.buerkner too if you have time.