Say I have categorical data where each category falls along the same spectrum. Perhaps a bad example, but stars on Amazon, Lyft, etc. One approach might be using a regression model to predict star number, an alternate approach might classify a driver/seller according to the most likely star-number category.
I believe that I've read about some middle ground where classes are returned (not just integers predictions but fixed to a certain set of classes) where information is shared among classes (so if 4 is the most likely class, then we'd expect 3 and 5 to have more confidence than 1 or 2.) I think this is called "Ordered Categorical Variables" but when I google this term I'm not so sure that it's what I'm describing.
So my questions are twofold. (A) What area of stats answers what I'm describing and how would one approach this sort of problem? And (B) If I'm wrong about what Ordered Categorical variables are... what are they in plain English?
For specific modeling, I typically follow a Bayesian framework, using tools like PyMC3 and Stan; I add this info as it might (or might not) be useful in framing your answer.
Edit: I found this Q/A which addressed my question pretty well and includes some hints as to how to implement from a Bayesian perspective.
