Continuous-time hidden Markov models enable researchers to study the relations between risk factors and outcomes repeatedly measured at unbalanced, unequally spaced assessment times while accommodating measurement error. Despite extensive implementation of variable selection methods designed to identify potential relations in exploratory settings for hypothesis generation, currently none have been applied to this class of models commonly found in psychological research. To fill this gap, we develop a Bayesian continuous-time hidden Markov model with variable selection priors and provide a flexible R package to facilitate the application of our method in practice. We showcase the variable selection performance and estimation accuracy of our method on simulated data and apply it to intensive longitudinal data collected in a smoking cessation trial to identify potential risk factors associated with smoking behaviors after a quit attempt.
Here’s a link to the pdf.