B1691
Title: Extracting dynamic features from irregularly spaced time series
Authors: Oisin Ryan - Utrecht University (Netherlands) [presenting]
Abstract: The advent of smartphones and wearable technology has seen an explosion in research designs which involve the collection and analysis of time-series data. In time-series analysis, tools such as autocorrelation and cross-correlation functions provide the basis for understanding the dynamics underlying this data. However, the estimation of auto- and cross-correlations typically relies on the assumption that data are equally spaced in time, and this assumption is often violated in practice: For example, in social science settings, self-report measures collected through experience sampling designs often result in high irregularly spaced measurements, either by design or due to missing measurement waves. We develop and present a statistical tool, available as an R package, which allows for the estimation of auto and cross-correlations from irregularly spaced time series. Based on generalized additive models, we assess the performance of this method in comparison to both traditional approaches and confirmatory fitting of continuous-time models, the latter of which is vulnerable to problems of model misspecification and unobserved confounding, which the presented method avoids.
