B0481
Title: Generalized multilevel functional principal component analysis
Authors: Xinkai Zhou - Johns Hopkins University (United States) [presenting]
Julia Wrobel - Emory University (United States)
Ciprian Crainiceanu - Johns Hopkins University (United States)
Andrew Leroux - Colorado School of Public Health (United States)
Abstract: A new generalized multilevel functional principal component analysis method is proposed for non-Gaussian multilevel functional data. The method consists of (1) binning the data along the functional domain; (2) fitting local multilevel generalized linear mixed effects models in every bin to obtain initial estimates of the functional linear predictors at each level; (3) using fast multilevel functional principal component analysis to smooth the linear predictors and obtain their eigenfunctions; and (4) estimating the global model conditional on the eigenfunctions of the linear predictors. Extensive simulation studies show that the method provides accurate estimation of the eigenfunctions and scores, is computationally stable, and is scalable in the number of study participants, visits, and observations within visits. Methods were motivated by and applied to a study of active/inactive physical activity profiles derived from wearable accelerometers in the NHANES 2011-2014 study. The essential, accompanied components of the R software are also described.
