Abstract

We model the observed smoking outcomes of an individual and her social contact - a spouse, friend, sibling, parent or adult child - as a simultaneous move game with complete information. We allow an individual's smoking behavior to depend on her previous behavior and carefully account for observed and unobserved heterogeneity. Our econometric model addresses simultaneity, homophily, health endogeneity, non-random attrition, and multiplicity of equilibria together. Using network data from the Framingham Heart Study, we find smaller social interaction effects among spouses and friends than Christakis and Fowler (2008) who pioneered such exploration with these data. We also find that social interaction impacts of the endogenous behaviors of siblings or a parent and an adult child are not statistically significant after disentangling them from homophily. In addition, we find that the effects of social contacts' cardiovascular disease shocks on individual smoking behavior are not statistically significant.