Peers as treatments: Understanding contextual peer effects
Working paper, 2025
This paper develops a general framework for interpreting linear regression estimates of contextual peer effects under random peer assignment. Rather than imposing the strong assumption that peers influence individual outcomes solely and directly through specific observed characteristics, the model considers social interaction with a given peer or group as a treatment with an unknown and variable treatment effect. In this setting, a wide variety of conventional peer effect regressions are informative and can be interpreted as measuring treatment effect heterogeneity along researcher-selected dimensions of interest. These regressions can also be used to predict the consequences of counterfactual peer group assignments. The relevance of the framework and results to empirical research is illustrated using an application to measuring classroom peer effects in Project STAR.
Recommended citation: Krauth, Brian (2025). "Peers as treatments: Understanding contextual peer effects." Working paper, Simon Fraser University. https://bvkrauth.github.io/publication/peertreat