Essays on Microeconometrics with Applications to Social Network

This dissertation includes three chapters on microeconometrics with applications to social network. In the first chapter, we study identification and estimation of peer effects in a game theoretical social interaction model with incomplete information. We show that players’ equilibrium choice probabilities and peer effects can be identified in the presence of measurement errors in network connections by exploiting the nonparametric methodology developed for nonclassical measurement error models. Based on the identification methodology, a semiparametric estimation method is established and applied to study the peer effects on youth alcohol drinking behaviors using data of adolescents in the United States, our empirical findings show that peer effects will be significantly underestimated if measurement errors are ignored. In the second chapter, we study strategic social interaction among economic agents that are connected through the phenomena of homophily. In particular, we measure homophily effects by the differences between players’ socioeconomic characteristics. Under the symmetric equilibrium selection mechanism, we establish a nonparametric approach to identify the structural model and propose a computationally feasible two-step estimation procedure. The asymptotic properties of the two-step estimator are derived under context of “large games", i.e., the number of players going to infinity. Finally, we apply the identification and estimation methods to study the peer effects on youth smoking behaviors using data of adolescents in the United States, our empirical findings show positive and statistically significant peer effects and demonstrate the empirical importance of including homophily effect in our model. In the third chapter, we study bandwidth selection method for the smoothed maximum score estimator. The smoothed maximum score estimator is a semiparametric estimator for binary response model, which is very useful for many economics and statistics applications. The method for selecting the smoothing parameter (bandwidth) in smoothed maximum score estimator is analogous to the plug-in method in kernel density estimation. It requires initial “pilot" values of the bandwidth to obtain the optimal bandwidth. The method has the disadvantage of not being fully data-driven. In this paper, we propose a data-driven bandwidth selection method by minimizing a cross-validated criterion function. Simulation results show that our proposed method performs better than existing methods