Bayesian Identifiability
Diana J. Cole · 2020
This chapter considers a more applied approach to demonstrate that identifiability can still be a issue when a Bayesian approach is used, and therefore must be considered. It starts with a brief introduction to Bayesian statistics and gives the definitions of identifiability. The chapter discusses the problems associated with identifiability when using a Bayesian approach and examines the overlap between the posterior and prior distribution. It uses data cloning to investigate identifiability and examines how the symbolic method can be used to investigate Bayesian identifiability. Bayesian identifiability can be considered to be equivalent to identifiability of the likelihood, therefore any appropriate exhaustive summary derived from the log-likelihood exhaustive summary could be used. Results on Bayesian identifiability would be identical to the classical formation of the same model.