Bayesian hierarchical robust factor analysis models for partially observed sample-selection data

This paper introduces a class of scale mixtures of normal selection factor (SMNSF) analysis models which are robust against departures from normality and designed to correct sample-selection bias. Various properties of this class of models are established, including a stochastic representation, a distributional hierarchy, and a quantification of sample-selection bias. A hierarchical Bayesian methodology is also developed for estimation purposes. It involves a simple and computationally feasible Markov Chain Monte Carlo algorithm that avoids analytical or numerical derivatives of the log-likelihood function. Results from simulation studies attest to the good finite-sample performance of the new model in terms of sample-selection bias reduction and robustness against outliers. A data illustration is included.