|
Abstract:
|
When observations are dependent, using statistical methods that assume independence can result in biased estimates and artificially small p-values, standard errors, and confidence intervals, and may contribute to replication crises. Here, we describe a largely unrecognized but common type of dependence due to social network connections. We propose a test for network dependence, and apply it to several published papers that use the Framingham Heart Study (FHS) data. Results suggest that some of the many decades worth of research on coronary heart disease, other health outcomes, and peer influence using FHS data may be invalid due to unacknowledged network dependence. The FHS is not likely to be unique; these problems could arise whenever subjects are recruited from one or a small number of communities, schools, hospitals, etc. As researchers in psychology, medicine, and beyond grapple with replication failures, this unacknowledged basis for invalid statistical inference should be part of the conversation.
|
