Robust Estimation of Distributional Mixed-Effects Model with Application to Tendon Fibrilogenesis Data

• Activity Number: 18
• Type: Topic Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Tingting Zhan*+ and Inna Chervoneva and Boris Iglewicz
• Companies: Thomas Jefferson University and Thomas Jefferson University and Temple University
• Keywords: Distributional mixed e?ects model ; Joint estimation ; Finite normal mixture ; Maximum likelihood ; Two-stage estimation

Abstract

A new robust statistical framework is developed for comprehensive modeling of hierarchically clustered non-Gaussian distributions. A distributional mixed effects model (DME) with conditional distributions from any parametric family is proposed as general framework for accommodating variable non-Gaussian conditional distributions and modeling their parameters as dependent on fixed and random effects. We develop a new divergence-based methodology and computational algorithm for robust joint estimation of DME model. Performances of the proposed robust joint, maximum likelihood joint and two-stage approaches are compared for analyzing fibril diameter distributions in real animal data and in simulated data with structures similar to fibril diameter distributions. Overall, numerical studies indicate superior efficiency of joint estimation as compared to previously considered two-stage approaches, and superior accuracy of robust joint estimation for contaminated data.