Models of social interactions are often estimated under the strong assumption that an individual’s choices are a direct function of the average observed characteristics of his or her reference group. This paper interprets social interactions in a less restrictive potential outcomes framework in which interaction with a given peer or peer group is considered a treatment with an unknown treatment effect. In this framework, conventional peer effect regressions can be interpreted as characterizing treatment effect heterogeneity. This framework is then used to clarify identification and interpretation of commonly-used peer effect models and to suggest avenues for improving upon them.
Current version: September 2022