Improving control over unobservables with network data

TL;DR

This paper introduces a network-based method for causal inference that accounts for unobserved confounders by leveraging homophily, improving estimates of treatment effects in social network data.

Contribution

It develops a model accommodating key network features and proposes a robust estimator for treatment effects that addresses unobserved confounding, with applications to educational data.

Findings

Estimator recovers a larger effect of parental involvement than OLS.

Method is robust to unobserved confounders in network settings.

Applicable to both sparse and dense networks.

Abstract

This paper develops a method to conduct causal inference in the presence of unobserved confounders by leveraging networks with homophily, a frequently observed tendency to form edges with similar nodes. I introduce a concept of asymptotic homophily, according to which individuals' selectivity scales with the size of the potential connection pool. It contributes to the network formation literature with a model that can accommodate common empirical features such as homophily, degree heterogeneity, sparsity, and clustering, and provides a framework to obtain consistent estimators of treatment effects that are robust to selection on unobservables. I also consider an alternative setting that accommodates dense networks and show how selecting linked individuals whose observed characteristics made such a connection less likely delivers an estimator with similar properties. In an application, I recover an estimate of the effect of parental involvement on students' test scores that is greater than that of OLS, arguably due to the estimator's ability to account for unobserved ability.