A General Framework for Modeling Diurnal and Weekly Trends in Ecological Momentary Assessment Data

Authors

• Y Xiao
• KR Merikangas
• V Zipunnikov

Abstract

Ecological momentary assessment (EMA) data consist of multiple series of real-time self-reports, sampled over the course of several days or weeks. Past studies using EMA have not been known to account for the inherent diurnal and weekly trends exhibited in the responses, incorrectly assuming exchangeability across time points. Failure to properly detrend the data can (1) result in biased estimates of variability/stability and (2) introduce unwanted noise in regression models which may obscure the effects of interest. We present an example of this issue, followed by a systematic method for addressing it. Data come from the NIMH family study, a community sample (n=435) in which participants completed EMA assessments four times a day over a period of two weeks. First we show the presence of robust population-level diurnal and weekly trends in the domains of sadness, anxiousness, and energy (rated on a 1-7 likert scale). Next, we introduce a general matrix-variate framework where scores are broken down into population and subject level components. We then outline a hierarchical list of covariance structures for modeling the diurnal and weekly associations in any given domain. After selecting the appropriate structure for each domain and estimating the respective parameters, we compare the results of two different measures of stability calculated on the raw and detrended data and discuss the implications. We expand on how this system can serve as a reference for the future analysis of EMA data as well as the analysis of other similar types of longitudinal data.

Scientific Focus Area: Epidemiology

This page was last updated on Friday, March 26, 2021