Robust Two-Stage Modeling with Finite Mixture Conditional Distributions

• Activity Number: 297
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Tingting Zhan*+ and Inna Chervoneva and Boris Iglewicz
• Companies: Temple University and Thomas Jefferson University and Temple University
• Keywords: Robust ; Linear mixed model ; finite normal mixture

Abstract

We consider a hierarchical model with conditional distributions as finite normal mixtures to study subpopulations of collagen fibril diameter distributions. The data are collected from multiple animals with multiple microscopic fields per animal, with 200-400 observations in each field. We use a two-stage estimation. In stage 1, normal mixtures are fitted in each field using the robust minimum divergence estimator since the fibril distributions are contaminated with outliers. Then resulting field-specific means and standard deviations are modeled in stage 2 using a robust version of the linear mixed effects model to reduce the influence of outlying fields. With robust estimation in both stages, we are able to identify biologically important differences in means and standards deviations of normal components of fibril diameter distributions from wild type and decorin-deficient mice.