EMA Statistical Modeling
(Novel Statistical Models for EMA Studies of Physical Activity)
Funded by National Heart Lung and Blood Institute 1R01HL121330 (Dunton and Hedeker, PIs)
This project will develop and test novel multilevel statistical methods to examine the effects of subject-level variances and slopes for time-varying variables in ecological momentary assessment (EMA) studies of physical activity. Low level of physical activity heightens the risk of numerous deadly diseases (e.g., heart disease, stroke, cancer, diabetes) throughout the life course. The use of EMA in physical activity research is growing rapidly, as real-time data capture methods are thought to offer significant advantages to understanding determinants of this behavior. In EMA studies, it is common to have up to thirty or forty observations per subject, and this allows us to model subject-level variances (e.g., how erratic is a subject’s mood?) and slopes (e.g., how much does a subject’s mood change across contexts?) for time-varying variables. For example, in our recent EMA work, we have found that more physically active children have greater positive and negative emotional stability than children who are less physically active. However, current multilevel modeling strategies are restricted to treating subject-level variances and slopes as outcomes. As a consequence, statistical models do not have the ability to test whether subject-level variances and slopes have predictive, mediating, and moderating effects on physical and sedentary activity. For example, we are unable to ask important research questions such as whether erratic mood mediates the effects of depression on physical activity, or whether the effects of living in a highly walkable neighborhood on physical activity are attenuated for individuals with unstable self-efficacy beliefs. This modeling restriction severely limits our ability to capitalize on the full potential of the time-varying nature of EMA data to enhance physical activity research. To address this critical methodological gap, we propose to develop multilevel modeling strategies to test subject-level effects of time-varying variables in EMA studies. We will apply these modeling strategies secondary analyses of pooled data from 5 federally- and foundation-supported EMA studies of physical activity with a combined sample size of N = 642 participants (including children and adults). The primary aims are (1) to develop innovative unified multilevel modeling strategies to test whether subject-level variances and slopes have predictive, mediating, and moderating effects on physical and sedentary activity outcomes, (2) to provide a software program that is accessible via statistical programs SAS and R, with accompanying primers to illustrate the use of the models and software, and (3) to apply these novel modeling strategies and software to examine the effects of subject-level variances and slopes for time-varying variables such as safety, traffic, negative affect, stress, fatigue, pain, self-efficacy, and intentions on physical and sedentary activity. This study has the potential to make notable methodological and substantive contributions for analysis of EMA data in physical activity research. The methods that are to be developed can easily generalize to a variety of chronic disease-relevant research areas.