A Structural Model of Segregation in Social Networks

In this paper, I develop and estimate a dynamic model of strategicnetwork formation with heterogeneous agents. The main theoretical resultis the existence of a unique stationary equilibrium, which characterizesthe probability of observing a specific network in the data. As aconsequence, the structural parameters can be estimated using only oneobservation of the network at a single point in time. The estimation ischallenging, since the exact evaluation of the likelihood function iscomputationally infeasible even for very small networks. To overcomethis problem, I propose a Bayesian Markov Chain Monte Carlo algorithmthat avoids the direct evaluation of the likelihood. This methoddrastically reduces the computational burden of estimating the posteriordistribution and allows inference in high dimensional models. I presentan application to the study of segregation in school friendshipnetworks, using data from Add Health. The latter contains the actualsocial network of each student in a representative sample of US schools.My results suggest that for White students, the value of a same-racefriend decreases with the fraction of whites in the school. Thisrelationship is of opposite sign for African American students. Themodel is used to study how different desegregation policies may affectthe structure of the network in equilibrium. I find an inverted U-shaperelationship between the fraction of students belonging to a racialgroup and the expected equilibrium segregation levels. These resultssuggests that these policies should be carefully designed in order to be effective.