报告人：Mengsi Gao（University of California, Berkeley）
时间：2024-09-19 15:10-17:00
地点：Zoom（ID:927 3523 8747；Passcode:123456）
Abstract: This paper investigates the identification and inference of treatment effects in randomized controlled trials with network interference. Two key features characterize my setting: first, the intervention may influence the network structure, which mediates the treatment effect on outcomes. Second, latent variables could drive both network formation and outcomes, introducing potential endogeneity, particularly when spillover effects are of interest. I make three primary contributions. First, I define the parameters of interest within a post-treatment network framework, separating the direct treatment effect from the indirect effect mediated by changes in the network. Within a linear and additive outcome model for both treatment and network mediators, I provide a causal interpretation of the linear model coefficients. Second, I establish the consistency and asymptotic normality of ordinary least squares (OLS) estimators when no unobserved confounders are present. Third, to address endogeneity arising from unobserved confounders, I employ shift-share instrumental variables (SSIV). I show that IV estimates with SSIV are consistent and asymptotically normal when networks are relatively sparse. For denser networks where IV estimates may be inconsistent, I propose a modification to SSIV that restores consistency. I also provide consistent variance estimators for both OLS and IV estimates, ensuring valid inference. Simulation evidence suggests that our theory performs well. Finally, I revisited Prina (2015) for empirical illustration and found that treatment may affect various outcomes through different channels.
Bio: Mengsi Gao is a 6th-year Ph.D. student at Berkeley Economics, advised by Bryan Graham and Peng Ding. Her research focuses on econometrics, especially causal inference and networks. Before Berkeley, Mengsi received her B.A. (Economics) from Zhejiang University and M.A. (Economics) from Duke University.
