Staff Report, Federal Reserve Bank of New York 504
We study a general class of models with social interactions that might display multiple equilibria. We propose an estimation procedure for these models and evaluate its efficiency and computational feasibility relative to different approaches taken to the curse of dimensionality implied by the multiplicity. Using data on smoking among teenagers, we implement the proposed estimation procedure to understand how group interactions affect health-related choices. We find that interaction effects are strong both at the school level and at the smaller friends-network level. Multiplicity of equilibria is pervasive at the estimated parameter values, and equilibrium selection accounts for about 15 percent of the observed smoking behavior. Counterfactuals show that student interactions, surprisingly, reduce smoking by approximately 70 percent with respect to the equilibrium smoking that would occur without interactions.