RESUMO
Bayesian methods provide direct inference in functional data analysis applications without reliance on bootstrap techniques. A major tool in functional data applications is the functional principal component analysis which decomposes the data around a common mean function and identifies leading directions of variation. Bayesian functional principal components analysis (BFPCA) provides uncertainty quantification on the estimated functional model components via the posterior samples obtained. We propose central posterior envelopes (CPEs) for BFPCA based on functional depth as a descriptive visualization tool to summarize variation in the posterior samples of the estimated functional model components, contributing to uncertainty quantification in BFPCA. The proposed BFPCA relies on a latent factor model and targets model parameters within a mixed effects modeling framework using modified multiplicative gamma process shrinkage priors on the variance components. Functional depth provides a center-outward order to a sample of functions. We utilize modified band depth and modified volume depth for ordering of a sample of functions and surfaces, respectively, to derive at CPEs of the mean and eigenfunctions within the BFPCA framework. The proposed CPEs are showcased in extensive simulations. Finally, the proposed CPEs are applied to the analysis of a sample of power spectral densities (PSD) from resting state electroencephalography (EEG) where they lead to novel insights on diagnostic group differences among children diagnosed with autism spectrum disorder and their typically developing peers across age.
RESUMO
EEG experiments yield high-dimensional event-related potential (ERP) data in response to repeatedly presented stimuli throughout the experiment. Changes in the high-dimensional ERP signal throughout the duration of an experiment (longitudinally) is the main quantity of interest in learning paradigms, where they represent the learning dynamics. Typical analysis, which can be performed in the time or the frequency domain, average the ERP waveform across all trials, leading to the loss of the potentially valuable longitudinal information in the data. Longitudinal time-frequency transformation of ERP (LTFT-ERP) is proposed to retain information from both the time and frequency domains, offering distinct but complementary information on the underlying cognitive processes evoked, while still retaining the longitudinal dynamics in the ERP waveforms. LTFT-ERP begins by time-frequency transformations of the ERP data, collected across subjects, electrodes, conditions and trials throughout the duration of the experiment, followed by a data driven multidimensional principal components analysis (PCA) approach for dimension reduction. Following projection of the data onto leading directions of variation in the time and frequency domains, longitudinal learning dynamics are modeled within a mixed effects modeling framework. Applications to a learning paradigm in autism depict distinct learning patterns throughout the experiment among children diagnosed with Autism Spectrum Disorder and their typically developing peers. LTFT-ERP time-frequency joint transformations are shown to bring an additional level of specificity to interpretations of the longitudinal learning patterns related to underlying cognitive processes, which is lacking in single domain analysis (in the time or the frequency domain only). Simulation studies show the efficacy of the proposed methodology.
