Yang, Muzhe (2008). Identification and estimation of social interaction-based models: A changes-in-changes approach with an application to adolescent substance use. 2008 Add Health Users Conference.
It is acknowledged that teenage behaviors and their health outcomes will not be well understood without considering social interactions during this rapid transition period. Finding a strong correlation in behavior among peers cannot justify the existence of peer effects. The impact of friends on an individual's behavior will be confounded by their mutual influences, individual's self selection into peer groups and peer's shared unobserved environmental factors. To isolate peer effects from other factors, I adopt a linear in means modeling strategy which rests upon a spatial autoregressive model. This distinguishes my study from current empirical studies on social interactions, most of which are based on linear in expectations models. Using Add Health, I identify and estimate peer effects of adolescents' health related behaviors: substance use through difference in differences (DID), using time differencing to exclude individual "fixed effect" and treatment control differencing to exclude peer group level unobserved heterogeneities. To accommodate possible behaviors of sorting into treatment (the group with higher expected gain receiving the treatment), I estimate such heterogeneous treatment effects based on changes in changes (CIC), a generalized version of DID. This strategy allows for identifying not only peer effects, but also the treatment effect corresponding to a hypothetical policy intervention of removing a drug user friend from his or her own peer group. Identification of these two parameters ultimately leads to constructing a "social multiplier," which bears rich policy implication, especially seen from an economic perspective.
2008 Add Health Users Conference
City of Publication