| dc.description.abstract | This dissertation introduces a new model of peer effects in social networks with sample selection. In this model, an original exogenous network structure influences the binary choices of network members; these choices in turns shape a new network structure that subsequently influences the later continuous outcomes of network members. Additionally, the model features sample selection such that the binary choices of network members are correlated with the potential of the later outcomes, and such that later outcomes are realized only for a subset of network members. We propose an estimation method for this model which is included of two steps with control function approach in order to produce correction for bias in the estimation of peer effects of the later outcomes of network members. We suggest the use of block bootstrap routine for inference of this estimation and evaluate its performance in simulation. Finally we employ the estimation in a real data set. We estimate peer effects of high school students' subject-specific GPAs in Science and Social Studies, controlling for the potentially endogenous selection into subject courses. Besides determining whose GPAs researchers can observe in the data, sample selection can also affect which friends' GPAs students can observe and by which be influenced. Using the National Longitudinal of Adolescent and Adult Health data for high school students, while we find no evidence for endogenous selection, we find statistically significant peer effects among subject-specific GPAs. | |
