Quadratic Discrimination for Multi-Level Multivariate Data with Separable Means

• Activity Number: 304
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Anuradha Roy*+ and Ricardo Leiva
• Companies: The University of Texas at San Antonio and Universidad Nacional de Cuyo
• Keywords: covariance structure ; maximum likelihood estimates ; separable means

Abstract

Under the assumption of multivariate normality we study a quadratic discriminant function of multiple m-variate observations over u-sites and over v-time points. We assume that the m-variate observations have a "jointly equicorrelated covariance" structure and a separable mean vector. A discriminant function is also developed for unstructured mean vectors. The new classification rules are very efficient in discriminating individuals when the number of observations is very small, and thus unable to estimate the unknown variance-covariance matrix. We demonstrate the classification rules on a real data set. Our result shows that our new classification rules are far better than the traditional classification rules for small to moderate sample sizes. These classification rules have plenty of applications in biomedical, medical, pharmaceutical and many other research areas.