Title: Adaptive spectral analysis of replicated nonstationary time series
Authors: Robert Krafty - University of Pittsburgh (United States) [presenting]
Scott Bruce - Temple University (United States)
Martica Hall - University of Pittsburgh (United States)
Abstract: A new method is discussed for analyzing associations between nonstationary time series and cross-sectional variables when data are observed from replicated independent units or subjects. The approach adaptively divides time series and values of the cross-sectional variable into approximately stationary blocks, then estimates conditional local power spectra nonparametrically through Whittle likelihood based smoothing splines. The model is formulated in a Bayesian framework and fit via reversible jump MCMC methods, which allow for the modeling of both abrupt and smoothly varying effects. The method is used to analyze data from a study of caregivers of spouses with dementia and uncovers connections between heart rate variability during sleep and quality of life.