ABSTRACT
Regression models for binary and count data were introduced in Chapter 6. This chapter describes Bayesian models for more general forms of categorical or discrete data, starting with 2 × 2 tables which classify two binary variables, followed by multinomial models for single or multiple categorical outcomes and models for ordered categorical data. Regression techniques are introduced for relating multinomial and ordinal data to predictors. As in any other Bayesian application, needing to specify prior distributions
may be both an advantage and a challenge. While inferences are sometimes sensitive to the choice of prior, it can allow realistic information to be introduced and can stabilise estimates from data with small counts. The BUGS apparatus also allows models to be specified with arbitrary constraints on their parameters.
