We will be implementing estimation for ordinal data soon.
We need to lock down an interface for the user to specify what is necessary for categorical estimation.
Issues I can think of right away include:
(1) Specifying data cast. That is, which columns are ordinal? If the observed data is an R dataframe, then this information is already included. Do we insist on dataframes in order to estimate ordinal columns? Or do we add another option somewhere?
(2) Specify high and low thresholds. Highest and lowest thresholds may not occur in every group. We need to give the user a way to supply these.
(3) Is there an additional model/parameter specific component, or is all user-supplied specification attached to a data object?