ABSTRACT
This chapter is concerned with the Bayesian statistical analysis of binary data. Such data arise in medicine, economics, and other scientific fields. A considerable literature devoted to the proper analysis of such data has emerged. The logit model, for example, is available in most statistical software packages. It is straightforward to fit by the method of maximum likelihood, and it leads to log odds that are linear in the covariates. This latter property makes it easy to interpret the effect of a given covariate on the experimental response. Albert and Chib (1993a) have developed a new approach to the analysis of binary data models (with probit and tlinks) by introducing latent random variables following a scale mixture of normal distributions. This latter approach has opened up new avenues for Bayesian inference in binary, binomial, multinomial, and longitudinal discrete data settings.
