This is an old version, view current version.

21 Problematic Posteriors

Mathematically speaking, with a proper posterior, one can do Bayesian inference and that’s that. There is not even a need to require a finite variance or even a finite mean—all that’s needed is a finite integral. Nevertheless, modeling is a tricky business and even experienced modelers sometimes code models that lead to improper priors. Furthermore, some posteriors are mathematically sound, but ill-behaved in practice. This chapter discusses issues in models that create problematic posterior inferences, either in general for Bayesian inference or in practice for Stan.