Peer Effects and Multiple Equilibria in the Risky Behavior of Friends

We study social interactions in the risky behavior of best-friend pairs in the National Longitudinal Study of Adolescent Health (Add Health). Focusing on friends who had not yet initiated a particular behavior (sex, smoking, marijuana use, truancy) by the first wave of the survey, we estimate bivariate discrete choice models for their subsequent decisions that include peer effects and unobserved heterogeneity. Social interactions can lead to multiple equilibria in friends' choices: we consider simple equilibrium selection models as well as partial likelihood models that remain agnostic about the choice of equilibrium. Our identification strategy assumes that there is at least one individual characteristic (e.g., physical development) that does not directly affect a friend's propensity to engage in a risky activity. Our estimates suggest that patterns of initiation of risky behavior by adolescent friends exhibit significant interaction effects. The likelihood that one friend initiates intercourse within a year of the baseline interview increases by 4 percentage points (on a base of 14%) if the other also initiates intercourse, holding constant family and individual factors. Similar effects are also present for smoking, marijuana use, and truancy. We find larger peer effects for females and for pairs that are more likely to remain best friends after a year. We also find important asymmetries in the strength of the peer effects in non-reciprocated friendships.