How to Deal With Temporal Relationships Between Biopsychosocial Variables: A Practical Guide to Time Series Analysis

Longitudinal data allow for conclusions about the temporal order of events and interactional dynamics between several processes. The aim of this article is to provide a concise and pragmatic description of time series analyses (TSAs) of patient samples with numerous (or daily) repeated biological, behavioral, or psychological measurements. In addition, the article demonstrates how to implement the described analyses with the software R.

To illustrate the concrete application of the time series method, we use two case series of patients with anorexia nervosa. Upon awakening, the patients collected salivary cortisol on a daily basis and answered several questions on a handheld computer (electronic diary) regarding psychosocial variables at the time of salivary collection.

Basic concepts of time series analysis such as stationarity, auto- and cross-correlation, Granger causality, impulse response function, and variance decomposition are presented. In addition, we demonstrate vector autoregressive analyses with three variables. For Patient 1, we demonstrate how TSA is used to detect cortisol and anxiety decreases during inpatient treatment and also how TSA can be used to show that an increase in cortisol is followed by a next-day increase in anxiety. For Patient 2, TSA was used to show higher salivary cortisol upon awakening on the days the patient was weighed compared with other days. In addition, we show how interdependencies of depressive feelings, positive anticipations, and cortisol values can be quantified using TSA.

Time series designs enable modeling of temporal relationships and bidirectional associations between biopsychosocial variables within individuals. These individual patterns cannot be derived from traditional group-based statistical analyses. This article provides accessible research tools for conducting TSA relevant to psychosomatic and biobehavioral research.