Peer Group Effects on Student Outcomes: Evidence from Randomized Lotteries

Abstract

This study examines both school and classroom level peer group effects on middle school student academic and behavioral outcomes in a mid-size urban district in the South. Identification of peer effects faces three major methodological challenges: simultaneity bias, selection bias and correlated omitted variable bias. This study circumvents the simultaneity bias by using pre-determined peer characteristics and lagged values of peer behaviors. In order to eliminate the endogeneity bias from self-selectivity and correlated omitted variables, I exploit the magnet school admission lotteries to form instruments for the endogenous peer variable. A magnet school admission lottery randomly assigns students to the magnet school (treatment group) or neighborhood schools (control group); meanwhile, conditional on the attendance zone, the lottery also randomly assigns students to the peers who they will encounter in either the magnet school or the neighborhood school. Therefore, magnet school lotteries bring exogenous variation in peer characteristics and will be exploited to overcome the critical issue of selection bias in identifying peer effects. Three research questions have been investigated in this study: the average peer group effect, the effects of peer heterogeneity, and the heterogeneous peer effects. This study finds substantial classroom peer effects on both student academic achievement and disciplinary infractions. For example, students score higher and behave better when the class has more high performers, well behaved peers, or peers from high income families. Percent black students is found reducing individual academic achievement and increasing student disciplinary infractions from both level analyses. This study does not find any evidence that the dispersion of peer characteristics impacts individual academic achievement or behavioral records. Significant heterogeneous peer effects are found on some subgroups of students. For instance, female students are less impacted by peers, and black and low income students tend to have more behavioral problems if surrounded by high proportion of same group peers.