ABSTRACT
Posterior prediction plays a role in model checking. Model checking involves an assessment of model assumptions to determine if a model adequately represents the observed data. The ability to obtain draws from the posterior predictive distribution (PPD) allows to compare features of the predicted data with those of the observed data. The chapter deals with the joint distribution of the parameters and unobserved data conditioned on the observed data and integrate out the parameters. The integrand in the PPD is the product of two distributions, the second is the usual posterior distribution and the first is the full-conditional distribution of the unobserved data given the parameters and observed data. In addition to providing inference about unobserved data directly, statistical prediction and our ability to obtain a sample from the PPD is useful for other reasons.
