Title: Zero-inflated regime-switching stochastic differential equation models for multivariate multi-subject time-series data
Authors: Zhaohua Lu - St. Jude Children's Research Hospital (United States) [presenting]
Sy-Miin Chow - Pennsylvania State University (United States)
Nilam Ram - Pennsylvania State University (United States)
Pamela Cole - Pennsylvania State University (United States)
Abstract: Stochastic differential equation (SDE) models are widely used in the studies of human dynamics, which are often characterized by the sparse occurrences of certain behavior in some individuals. To recover the dynamics of a system with an inflation of such zero responses, we incorporate a regime (latent phase) of non-occurrence to an SDE model to account for the high proportion of non-occurrence instances and simultaneously model the multivariate dynamic processes of interest under non-zero responses. The transition between the occurrence and non-occurrence regimes is represented by a latent Markovian transition model which depends on latent regime indicators and person-specific covariates. Markov chain Monte Carlo algorithms are used for the Bayesian estimation and inference. We demonstrate the proposed zero-inflated regime-switching SDE model through a multi-subject dynamic self-regulation study for young children at 36 and 48 months.