Summary

The literature documenting a wide range of network or peer effects has blossomed in the past decade and have appeared in most major economics journals, using a variety of methods and identification strategies. Reviewing the empirics of those papers suggests a few broad classes of econometric models. The first and canonical model is the “linear-in-means,” which grows from Manski’s seminal work. More recently, it has shown that network asymmetry conditions (known as “peers-of-peers” instrument approach) can be used to instrument the endogeneity inherently present in the linear-in-means model. Moving to more recent empirical practice reveals novel and creative instrumentation strategies exploring particular empirical settings. The network identification strategies can also be combined with traditional differences-and-differences, event-study, and regression discontinuity designs. For example, under certain conditions, one can explore the variation that stems from the differential comparison of the evolution over time of well-connected versus less-well-connected individuals; or explore the variation in the network structure induced by a discontinuous change in the network. Randomized, controlled studies had substantial importance in revealing network effects in the past literature using standard methods; and, more recently, in understanding the extent to which networks can themselves be endogenous to the provision of the treatment itself. This, in turn, will present future challenges for the econometrics of networks and the identification or evaluation of treatment effects under a causal framework with endogenous interference.