Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2509.24259 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911183493136384 |
|---|---|
| author | Sun, Kuan Xiao, Zhiguo |
| author_facet | Sun, Kuan Xiao, Zhiguo |
| contents | This paper develops doubly robust estimators for direct (DATT) and spillover (SATT) average treatment effects on the treated in network-based difference-in-differences (DiD) designs. Unlike standard DiD methods, the proposed approach explicitly accommodates treatment spillovers and high-dimensional network confounding arising from complex inter-unit dependencies. Identification relies on a conditional parallel-trends assumption that holds after adjusting for high-dimensional network confounders. The estimators are consistent and asymptotically normal as the network size increases, and we use graph neural networks (GNNs) to estimate nuisance functions. Simulation studies and an empirical application to U.S. county-level mask mandates and their impact on COVID-19 transmission demonstrate favorable finite-sample performance, addressing limitations of conventional DiD methods that ignore network interference. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_24259 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Difference-in-Differences Under Network Interference Sun, Kuan Xiao, Zhiguo Methodology This paper develops doubly robust estimators for direct (DATT) and spillover (SATT) average treatment effects on the treated in network-based difference-in-differences (DiD) designs. Unlike standard DiD methods, the proposed approach explicitly accommodates treatment spillovers and high-dimensional network confounding arising from complex inter-unit dependencies. Identification relies on a conditional parallel-trends assumption that holds after adjusting for high-dimensional network confounders. The estimators are consistent and asymptotically normal as the network size increases, and we use graph neural networks (GNNs) to estimate nuisance functions. Simulation studies and an empirical application to U.S. county-level mask mandates and their impact on COVID-19 transmission demonstrate favorable finite-sample performance, addressing limitations of conventional DiD methods that ignore network interference. |
| title | Difference-in-Differences Under Network Interference |
| topic | Methodology |
| url | https://arxiv.org/abs/2509.24259 |