Using Genes as Instrumental Variables in Analyses of Social Network Data

• Activity Number: 383
• Type: Invited
• Date/Time: Tuesday, August 3, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): A. James O'Malley and J. Niels Rosenquist*+ and Alan Zaslavsky and Nicholas A. Christakis
• Companies: Harvard Medical School and Harvard Medical School and Harvard University
• Keywords: Genes ; Instrumental variable ; Longitudinal observations ; Peer effect ; Social network ; Two-stage least squares

Abstract

We describe methodology for evaluating peer effects in a social network with respect to health behavior using an "instrumental variable" approach. The key idea is that a person (an "ego") may have peers (or "alters") who are randomly "assigned" genes predisposing the alters to certain health behaviors, and that this random assignment can be seen as a kind of natural experiment, exposing the ego to peers who either exhibit or do not exhibit the pertinent behaviors. Using non-parametric two-stage IV estimation and alternative structural equations models, we will assess whether obesity and smoking in an ego's alters (e.g., friends, siblings), are causally related to similar behaviors in the ego and if so test the more complex hypothesis that similarity effects will vary according to the closeness of the relationship between the ego and the alter.