Analyzing Mixed Models Using Rank-Based Regression

• Activity Number: 180
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Yusuf Bilgic*
• Keywords: Rank-based regression ; hierarchical models ; random effects ; mixed models ; Robust ; Nonparametric

Abstract

Statistical analysis of random effects in linear models is an area lacking a robust methodology. Small sample sizes, unbalanced data, and outliers can make traditional likelihood and least squares approaches unreliable. Rank-based regression analysis for Mixed Models provides a robust nonparametric alternative. We used the R package, rlme, to get rank-based statistical analyses for two- and three-level random effects nested models. We present 5 data analyses ranging in size from 68 to 50,000 observations. We show that the maximum likelihood analysis, our goat, is the least robust.