A Simulation Study of a Bayesian Hierarchical Changepoint Model with Covariates

This paper presents a simulation study to investigate the behavior of estimates of covariate effects in a Bayesian hierarchical changepoint model. The model is a segmented linear regression model with one changepoint within a fully Bayesian framework. We introduce covariate effects into the model in three ways: as an affect on the interaction term, an affect on the slope after the changepoint, or as an affect on the timing of the changepoint. We estimate all parameters using Markov chain Monte Carlo and compute bias, relative bias and average mean square error of the posterior means of the parameters capturing the covariate effcts. We also investigate the effects of model misspecification on the characteristics of the effect of using an inappropriate fitted model.