In this dissertation I explicitly model peer network formation and explore different channels through which peer effects influence student achievement. Using unique data from a survey "Secondary Education Survey in Hong Kong" (SESHK), I estimate an econometric model to show how peer connections are formed and to quantify peer effects. I focus on four separate types of peer networks: friends, studymates, emotional supporters, and seatmates. Additionally, students can be affected by their peers through three different channels, namely cognitive abilities, personality traits, and behavioral spillovers.;I estimate the magnitude of these peer effects on academic performance through all three channels and all four networks. Peer effects are identified through a combination of explicitly modeling peer selection and additional set of instruments available in my data. Empirical results show that friends and studymates are endogenously formed, which lead to overestimation of peer effects in traditional exogenous peer formation models. All investigated peer types show positive peer effects but the effect for seating proximity is relatively weaker. Peers are also found to have significant effects through particular cognitive abilities or personality traits. Smart studymates and conscientious friends positively affect a student's mathematics score, while conscientious studymates and smart friends do not have such an effect. These results show that understanding the formation of different peer types is important in peer effect estimation.;Finally, I apply the estimates obtained from the peer effect model in policy analysis. Randomizing peers decreases inequality while tracking increases inequality among students. Restricting a certain proportion of students from social interactions improves students' overall academic performance. The optimal proportion is different for different subjects.
展开▼