ABSTRACT
This chapter focuses on models with multiple latent variables. Bayesian networks (BNs) model multivariate distributions of discrete variables. BNs are not as frequently used in the psychometric community as CFA, IRT, and LCA models. A BN is a statistical model that structures the joint distribution for a set of discrete variables by recursively specifying conditional distributions. The chapter discusses the terms that do not immediately map onto latent variable psychometric models. BNs can be employed as psychometric models by specifying observable variables as dependent on latent variables for an examinee. The chapter focuses on how BNs can be leveraged as psychometric models and discusses several examples of other psychometric models that can be framed as BNs. Dynamic BNs extend the basic BN to model changes over time and related longitudinal data structures, and so can be gainfully employed in psychometric modeling of change during the course of an assessment.
