A1289
Title: Robust mixture of linear mixed modeling via multivariate Laplace distribution
Authors: Xiuqin Bai - Kansas State University (United States) [presenting]
Abstract: The assumption of normality in random effects and regression errors is the primary cause of the lack of robustness in the maximum likelihood estimation procedure for linear mixed models. A robust method is introduced for estimating regression parameters in these models, by positing that the random effects and regression errors follow a multivariate Laplace distribution. This new methodology, implemented via an EM algorithm, is computationally more efficient compared to the existing robust t procedure in the literature. Simulation studies suggest that the performance of the proposed estimation method in finite samples either surpasses or is at least on par with the robust t procedure.
