On the Robustness of Factor Analysis Models

Abstract

In this study, model robustness was examined for mainly two factor analysis models, TFA(Traditional Factor Analysis) and BCFA(Bayesian Copula Factor Analysis). There were three abnormal data scenarios, which were outlier, kurtosis, and high correlation matrix cases. Both models were applied to each of the scenario data. It was revealed that BCFA model outperforms TFA model across the scenarios: the former was superior in terms of robustness when compared to the latter. In fact, BCFA could resolve the big loading problems which arise when TFA is applied to the dataset, revealing the factor structure clearly. Additionally, some related issues are discussed in this article.