Association for Public Policy Analysis & Management

The potential for risky behaviors such as alcohol consumption to have spillover effects on friends or family members has important implications for both the societal costs of such behaviors and the benefits of policy interventions. Causally interpretable estimates of these peer effects are rare since selection into the behavior and into the peer group introduces separate biases driven by the decisions of multiple individuals. I focus on a population which allows for the estimation of causally interpretable peer effects in alcohol consumption under relatively weak assumptions, sibling pairs close to the minimum legal drinking age (MLDA) in the United States.

This setting is helpful for two reasons. First, the peer group (sibling pairs) is not chosen. Second, the average alcohol consumption of young adults has been shown to increase discontinuously at the MLDA. This allows for the use of a regression discontinuity design. In short, sibling pairs in which the older sibling is just under the MLDA are compared to pairs in which the older sibling is just above the MLDA. Preliminary results based on a sample of roughly 2,000 sibling pairs from the 1997 National Longitudinal Survey of Youth suggest that the MLDA-induced increase in alcohol consumption among older siblings has either no effect or small positive effects on the alcohol consumption of the younger siblings. Sensitivity analyses considering the effect on the older sibling when the younger sibling turns 21 and the effect within various subgroups of sibling pairs (e.g. same gender pairs) produce similar results. These estimates rely on self-reports of past month consumption and are somewhat heterogeneous. However, my preferred specifications are able to rule out meaningfully large peer effects in drinking days between siblings. If unobserved factors related to the alcohol consumption of both siblings are not changing discontinuously when one sibling turns 21, then these results are causally interpretable.

Leading existing work on peer effects in alcohol consumption relies on random assignment to peer groups. Such an approach produces credible estimates of peer effects but it cannot separate “endogenous” peer effects (peer A’s alcohol consumption affects peer B’s alcohol consumption) from “contextual” peer effects (exposure to peer A affects peer B’s alcohol consumption). My work extends the literature by isolating endogenous peer effects. This is accomplished by exploiting quasi-random variation in one peer’s alcohol consumption, as opposed to variation in the peer group. Differentiating between contextual and endogenous effects is important as endogenous effects imply that the benefits of policy interventions which reduce alcohol abuse are multiplied, while contextual effects do not. If sibling relationships are similar enough to other peer relationships considered in the literature (e.g. college roommates), these results suggest that previous estimates of peer effects in alcohol consumption are primarily contextual.