Identifying Peer Influence Effects in Observational Social Network Data: An Evaluation of Propensity Score Methods

There is widespread scientific and practical interest in estimating peer influence effects for a variety of behaviors, but peer influence effects are not identifiable in most observational data. Social scientists have thus had to remain largely silent on these effects — or make inconclusive or misleading arguments for their estimates of them. The state-of-the-art research has used propensity score methods to either estimate or place an upper bound on peer influence effects. The proposed research will evaluate these methods using simulations. In particular, I will demonstrate how realistic cases missing edge data leads can lead to these methods substantially underestimating peer influence effects. This means that these estimates fail as upper bounds (i.e., fail to partially identify peer influence effects). I compare these methods with regression adjustment, which may be preferrable in such conditions.