Bayesian Regression Models for Big Spatially or Longitudinally Correlated Functional Data

• Activity Number: 452 - Advancements in Complex Functional Data Analysis
• Type: Invited
• Date/Time: Wednesday, August 1, 2018 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Jeffrey S Morris* and Lin Zhang and Hongxiao Zhu and Veera Baladandayuthapani and Hojin Yang and Wonyul Lee and Michelle Miranda and Philip Rausch
• Companies: The University of Texas M.D. Anderson Cancer Center and University of Minnesota and Virginia Tech University and UT MD Anderson Cancer Center and The University of Texas M.D. Anderson Cancer Center and Food and Drug Administration and The University of Texas M.D. Anderson Cancer Center and Humboldt University
• Keywords: Bayesian modeling; functional data analysis; functional regression; spatially correlated data; longitudinal data; quantiles

Abstract

A series of Bayesian regression modeling strategies that can be used for high-dimensional spatially- or longitudinally correlated functional data will be described. Intrafunctional correlation is handled through basis function modeling, while interfunctional correlation is captured by one of three approaches: (1) parametric or nonparametric random effect functions, (2) separable or non-separable spatial (or temporal) inter-functional processes, or (3) tensor-basis function modeling. Rigorous Bayesian inference is done in such a way that adjusts for any potential multiple testing issues. We will describe these general approaches and illustrate them on a series of complex, high-dimensional, spatially and longitudinally correlated functional data sets coming from strain tensor data from a glaucoma study, bladder cancer genomic maps, event-related potential data from a smoking cessation study. We will also discuss recent work in which we have developed spatiotemporal quantile functional regression approaches that we are applying to model temporal climate change in terms of intraseasonal temperature and precipitation distributions.