Model-data fit evaluation: posterior checks and Bayesian model selection

Abstract

Listen

Translate article

The goal of statistical modeling for researchers to construct a model or series of models to help describe and make inferences about complex processes. Before researchers can make inferences from a Bayesian model, the model must be evaluated for its adequacy. Two dominate frameworks exist in Bayesian model evaluation. Absolute fit, evaluating one model's fit to the data, uses the posterior predictive model check approach. Relative fit, comparing two competing models, relies on Bayes Factors and information criteria approaches, among others.