Leveraging intensive longitudinal data to better understand health behaviors

摘要

Behavioral scientists have historically relied on static modeling methodologies. The rise in mobile and wearable sensors has made intensive longitudinal data (ILD) — behavioral data measured frequently over time — increasingly available. Consequently, analytical frameworks are emerging that seek to reliably quantify dynamics reflected in these data. Employing an input-output perspective, dynamical systems models from engineering can characterize time-varying behaviors as processes of change. Specifically, ILD and parameter estimation routines from system identification can be leveraged together to offer parsimonious and quantitative descriptions of dynamic behavioral constructs. The utility of this approach for facilitating a better understanding of health behaviors is illustrated with two examples. In the first example, dynamical systems models are developed for Social Cognitive Theory (SCT), a prominent concept in behavioral science that considers interrelationships between personal factors, the environment, and behaviors. Estimated models are then obtained that explore the role of SCT in a physical activity intervention. The second example uses ILD to model day-to-day changes in smoking levels as a craving-mediated process of behavior change.