Alejandro Sanchez Becerra
Spillovers, Homophily and Selection into Treatment: The Network Propensity Score
Propensity score matching (PSM) is a well-known method to estimate treatment effects when there is selection on observables. PSM fails to identify any relevant causal effects, however, when there are spillovers amongst friends and individuals befriend similar people (homophily). I propose a novel network propensity score matching (NPS) approach that identifies both treatment effects and spillovers amongst friends when there is selection on observables and homophily on observables. I show that the proposed NPS -a three dimensional vector of probabilities- can be used to identify causal effects for individuals with similar observables, analogous to the propensity score. I then propose estimators that are consistent and asymptotically normal for settings with multiple large networks. I evaluate my methodology on an information intervention to increase microfinance adoption in Southern India, where selection and homophily are particularly salient, finding positive treatment effects but limited spillover effects. In the extensions I show how to conduct robustness checks, extend the NPS to settings with relationships intensity, and how interpret the NPS in stratified multi-stage experiments.
Econometrics, Applied Microeconomics, Networks, Causal Inference
Francis J. DiTraglia