Modeling Repeatedly Observed Functional Data

• Activity Number: 5
• Type: Invited
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Hans-Georg G. Müller*+ and Kehui Chen
• Companies: University of California, Davis and University of Pittsburgh
• Keywords: Functional Data Analysis ; Longitudinal Studies ; Repeated Measurements ; Random Surfaces ; Behavioral Trajectories ; Demographic Trajectories

Abstract

Repeatedly observed and thus dependent functional data are encountered when random curves are recorded repeatedly for each subject in a sample. The models that for the analysis of such data that are presented in this talk are nonparametric and flexible. They provide a straightforward decomposition of the variation inherent in repeatedly observed functional data and lead to succinct descriptions for this type of potentially very large and complex data. The proposed approach is based on a two-step functional principal component analysis. The time points where functions are recorded may be irregular and sparse, as is often the case in longitudinal studies. The estimated model components are shown to be consistent across various scenarios. Illustrations include an analysis of longitudinal mortality data from repeatedly observed period lifetables for various countries and the modeling of longitudinally recorded daily behavioral profiles.