Descriptive Discriminant Analysis for Repeated Measures Data: Effects of Non-Normality on Bias and Error in Discriminant Function Coefficients

• Activity Number: 362
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: WNAR
• Author(s): Tolulope T. Sajobi*+ and Lisa M. Lix and William H. Laverty and Longhai Li
• Companies: University of Saskatchewan and University of Saskatchewan and University of Saskatchewan and University of Saskatchewan
• Keywords: discrimination ; repeated measurements ; covariance structures ; mean square error

Abstract

Discriminant analysis (DA) is a multivariate technique to classify subjects and/or describe the relative importance of variables to distinguish between groups. This study develops procedures for repeated measures (RM) data and estimates discriminant function coefficients based on parsimonious models for RM mean and covariance structures. Coefficient bias and error for normal and non-normal data are investigated using Monte Carlo simulations. Study parameters include the RM covariance structure, sample size, and mean configuration. For normal data, the absolute percentage bias was, on average, less than 7% when the covariance structure was correctly specified. Mean square error increased as the correlation among RMs increased and was about two times higher for non-normal than normal data. The proposed procedures can be used to identify the RMs that contribute to group discrimination.