Treatment Effects Accounting for Network Changes

Abstract: Networks may rewire in response to interventions. We propose a measure of the treatment effect when an intervention affects the structure of a social network. We develop a treatment-response model that incorporates dynamic peer effects and provides its identification conditions and the associated instrumental-variable strategy. We illustrate our estimation procedure using a panel data set containing information on a financial network before and after a field experiment that randomized access to savings accounts. Results show that neglecting the network change results in underestimation of the impact of the intervention and the role played by informal networks through which the intervention diffuses.