Identification and Estimation of Treatment Effects from Network Data with Confounded Social Links

We introduce a generalized propensity score (GPS) based approach to the identification and estimation of treatment effects from observational network data, wherein formation of social link between a pair of units depends on individual level characteristics. Ignoring the tie formation process, its interaction with the treatment assignment mechanism, and interference induced by the social network may lead to biased estimation of treatment effects. We propose a unified framework that addresses these challenges by jointly modeling treatment assignment and network formation. Generalized propensity score can be estimated given probabilistic models for these two processes and functional form defining effective treatment. Average potential outcomes and treatment effects are estimated with inverse probability weighting estimators. We illustrate the proposed method in several Monte Carlo studies andan empirical analysis that investigates the effect of a new political communication technology on political participation in Uganda.
Licheng Liu is a postdoctoral scholar in the Department of Political Science. He defended his Ph.D. dissertation in Political Science and Statistics at MIT in May 2024.
Licheng specializes in Political Methodology and International Political Economy. He works on proposing new methods for casual inference with longitudinal and network data using both frequentist and Bayesian approaches, and their applications in empirical studies like trade politics and comparative political behavior.