Detection of Outliers in Bayesian Factor Analysis Using Information Complexity

• Activity Number: 202
• Type: Topic Contributed
• Date/Time: Monday, August 3, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Hamparsum Bozdogan*+
• Companies: The University of Tennessee
• Keywords: Detection of Outliers ; Bayesian Factor Analysis ; Information Complexity ; Posterior Probabilities

Abstract

In this paper we study the problem of detection of outliers in Bayesian Factor Analysis (BAYFA) of Press and Shigemasu (1989) (PS89), Bozdogan and Shigemasu (1998) (BS98), and Press (2003). Information theoretic measure of complexity (ICOMP) criterion of Bozdogan (1990, 1994, 2000, 2004) are derived along with Akaike's (1973) AIC in the context of BAYFA model as a decision rule in simultaneously choosing the number of factors, m, and best approximating factor pattern structure. A case deletion diagnostics is used to determine the outliers in BAYFA model using ICOMP ratios of the deleted observations over the full data ICOMP. These are compared with the log Posterior Probability ratios on Monte Carlo simulated data with varying sample sizes and the number of outliers present. The results obtained show the accuracy of the new approach in recovering the true outliers in BAYFA.