Generalized Levene's Test of Homoscedasticity for Correlated Data with Group Uncertainty

• Activity Number: 495
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 8:30 AM to 10:20 AM
• Sponsor: SSC
• Author(s): David Soave* and Lei Sun
• Companies: University of Toronto and University of Toronto
• Keywords: Variance heterogeneity ; heteroscedasticity ; generalized least squares ; statistical genetics ; genotype uncertainty ; family data

Abstract

To assess the homoscedasticity assumption in many statistical practices, testing for variance heterogeneity between samples is often performed. In some settings, the variance test itself can also be of primary interest. For example, in genetic association studies, it has been pointed out that a number of biologically meaningful scenarios can lead to variance heterogeneity of a quantitative trait across genotype groups. Among the many testing procedures for heteroscedasticity, Levene's test is noted for its implementation simplicity and robustness to departures from normality. However, the original test statistic is formulated for assessing variance differences between discrete groups of independent samples, limiting its use when, for example, there is genotype group uncertainty or family/pedigree data. Extending the work of Glejser [1969], and Iachine, et al. [2010], we propose an easy-to-implement extension to Levene's test that uses generalized least squares for testing variance heterogeneity associated with continuous covariates in correlated samples. We examine the performance of the method through simulations and applications.