Teenage dropouts and drug use: Does the specification of peer group structure matter?

Four alternative structures of peer groups are compared in an empirical analysis of teenage dropouts and recent drug use. In general, individual-specific covariates remain robust regardless of group structure specification in dropout models, but lose significance in models of drug-use. Estimates of correlated school effects depend on the specification of group structure. Contextual group effects have no influence on the probability that an individual uses drugs, but demonstrate some statistical significance, albeit ambiguous and strongly dependent on the specification of group structure. Endogenous peer effects do not influence the probability of dropping-out of school, but exhibit positive complementarities with respect to recent drug-use. Modeling the probabilities of leaving school and recent drug-use within a jointly distributed empirical framework indicates that unobserved attributes bridging the two types of behavior demonstrate positive correlation.