Mixed Models Through The Lens of HGLM: Applications and Grand Challenges

• Activity Number: 444
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Xia Shen*+ and Lars Rönnegård and Moudud Alam
• Companies: Karolinska Institute and University of Edinburgh and Dalarna University and Dalarna University
• Keywords: R/hglm package ; hierarchical generalized linear models ; generalized linear mixed models ; high-dimensional data ; correlated random effects ; non-Gaussian random effects

Abstract

The "hglm" package is a hierarchical-likelihood-based solution for mixed models. Apart from generalized linear mixed models (GLMM), hierarchical generalized linear models (HGLM) can also solve models with non-Gaussian random effects, structured dispersion parameters, and correlated random effects. "hglm" provides a unified approach to various statistical modeling problems. We describe examples in our interdisciplinary research based on the "hglm" package, dealing with large-scale biological data, chemometrics data and geographical data. Thereafter, we discuss some big challenges that empirical scientists desire to solve using mixed models, including modeling high-dimensional interaction effects, having random effects in the mixed model dispersion parameters, joint modeling of spatial and genetic correlations, and multivariate analyses with random effects.