AI

The Spread of Physical Activity Through Social Networks

Abstract

We study the evolution of daily physical activity of 44.5K users on the Fitbit social network over a period of eight months. A time-aggregated analysis shows that average alter activity, gender, and body mass index (BMI) are significantly predictive of ego activity when controlling for ego BMI, gender, and number of friends. The direction and effect size of the associations surfaced vary when considering chronic conditions self-reported by a large portion of the users including diabetes, dyslipidemia, hypertension and depression. When considering the co-evolution of activity and friendship on a month by month basis in a within-subject analysis, we show via fixed effects modeling that the fluctuations in average alter activity significantly predict fluctuations in ego activity. Finally, we investigate the causal factors that may drive change of physical activity over time. We leverage a class of novel non-parametric statistical tests to rule out homophily as the sole source of dependence in activity, even in the presence of unobserved individual traits.