A Proposed Family of Covariance Structures for Clustered Data

• Activity Number: 498
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Samantha R. Seals*+ and Charles R. Katholi and Inmaculada Aban
• Companies: University of Mississippi Medical Center and University of Alabama at Birmingham and University of Alabama at Birmingham
• Keywords: covariance structure ; clustered data ; Monte Carlo simulation

Abstract

In this project, we propose a new family of covariance structures specifically for clustered data. Originally developed to handle within-cluster spatial correlation unique to a cluster, this flexible family of covariance structures allows researchers to hand-pick the covariance structure for within-cluster and between-cluster blocks of the covariance structure. We developed a program to implement this family of covariance structures; this program obtains model estimates, the likelihood ratio test for goodness-of-fit, and fit indices. A Monte Carlo simulation study was performed to validate the program, investigate the finite-sample properties of the resulting estimators, and to compare the results to that of readily available covariance structures. As an application, we apply a covariance structure belonging to this family of covariance structures to cardiac imaging data with the goal of comparing the remodeling of the left ventricle between diabetic and non-diabetic patients who recently experienced myocardial infarction. Data observed in the left ventricle naturally clusters by major coronary artery, thus, we will demonstrate the flexibility of our proposed family of structures.