A Note on the Cowles-EM Algorithm for Bayesian Ordinal Probit Models

We describe a novel procedure for Bayesian cutpoint simulation in ordered probit models. Our approach capitalizes upon a series of auxiliary binary probit models implied by the ordinal probit to craft tailored proposal distributions for Metropolis-Hastings (M-H) sampling. Our scheme for posterior simulation, termed the Cowles-EM algorithm, demonstrates strong mixing properties and produces more iid-equivalent simulations per computation period than alternatives often used in the literature. An illustrative application involving determinants of Body Mass Index (BMI) is also provided.