Bayesian approach to outlier detection in multivariate normal samples and linear models

by Alexandre Varbanov
Technical Report No. 614
School of Statistics
University of Minnesota
June 26, 1996


Abstract

Chaloner and Brant (1988) propose a Bayesian method for identifying outliers in univariate linear models. This paper presents an approach generalizing their idea to multivariate normal samples and multivariate linear models. The posterior distribution of the squared norm of the realized errors is used for outlier identification. Bayes factors are used for examining whether or not an observation is an outlier.

Some key words: Bayes factors; Multivariate linear models; Outlier detection; Posterior distribution.


Click here to download the complete PostScript technical report.