Resolving the Identification Problem in Linear Social Interactions Models: Modeling with Between-Group Spillovers

The linear-in-means model has been a theoretical and empirical workhorse of the social interactions field. As was noted by Manski (1993), the collinearity between group-level 'contextual' and 'endogenous' effects leads to an inability to identify the structural parameters of this model. Manski called this the 'reflection' problem. This paper suggests that Manksi’s reflection problem is unique to a special case of a more general context in which agents care about multiple reference groups. Specifically, the identification problem is resolved through a model generalization to include between-group and within-group effects.