Risky Peers and Crime

Abstract

We investigates how adolescent peer environments shape long-term criminal outcomes. Using Add Health data, we combine machine learning techniques (XGBoost with SMOTE) to impute unobserved risk preferences with a quasi-experimental design exploiting variation in peer composition across school cohorts. We find that exposure to same-gender risk-prone peers significantly increases the likelihood of arrest in adulthood, for both men and women, while opposite-gender peers have no effect. We find no impact on victimization. Results are confirmed by instrumental variables and placebo tests. The finding for women is particularly noteworthy given how little is understood about female crime. Heterogeneity analysis reveals effects are strongest among individuals from higher-income families and those with more educated mothers. Joint paper with Sebastian Garcia-Torres (HKS) and Max Steinhardt (FUB).