Testing for Network Dependence in the Framingham Heart Study
share
Empirical research in public health and the social sciences often rely on observations that have been collected from members of a single social network. Sometimes, when the social network is itself of scientific interest, this is done explicitly, but often it is the result of convenience sampling of subjects who live in the same community, attend the same school, or seek medical care at the same hospital. For example, the famous Framingham Heart Study cohort includes members of extended families, residents of the common neighborhoods, and employees of common workplaces, resulting in a cohort of individuals closely intertwined in a social network. When network dependence is present, that is when social relations can engender dependence in the outcome of interest, treating such observations as independent results in invalid, anti-conservative statistical inference. We propose a test of independence among observations sampled from a single network. We demonstrate the validity and usefulness of our proposed test in simulations and apply it to observations from the Framingham Heart Study (FHS). Our results suggest that some of the many decades worth of research on coronary heart disease and other health outcomes using FHS data may be invalid due to unacknowledged network dependence.
READ FULL TEXT
