Robust Discriminant Analysis Models for Multivariate Non-Normal Repeated Measures Data

• Activity Number: 236
• Type: Topic Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: SSC
• Author(s): Tolulope Sajobi* and Lisa Lix and Yukun Zhang
• Companies: University of Calgary and University of Manitoba and University of Calgary
• Keywords: discriminant analysis ; likelihood estimation ; repeated measures ; robust statistics

Abstract

Multivariate repeated measures data arise in studies in which two or more groups of subjects are repeatedly measured on several outcome variables. Discriminant analysis based on robust estimators of means and covariances have been developed for prediction in multivariate non-normal data, but not for repeated measures models. This study proposes repeated measures discriminant analysis (RMDA) based on maximum trimmed likelihood (MTL) estimators for distinguishing between two or more independent groups in multivariate non-normal repeated measures data. The misclassification errors (MER) of the RMDA procedures based on maximum likelihood (ML) and MTL estimation methods were compared in Monte Carlo study. Study parameters include population distribution, covariance structure, sample size, mean configuration, and number of outcome variables and repeated measurements. The average MERs for RMDA procedures based on MTL estimators were significantly smaller than the average MERs for RMDA procedures based on ML estimators when the data were sampled from multivariate non-normal data. Repeated measures discriminant analysis models can be used to discriminate between two or more populations