Spillover-aware synthetic control for staggered, spatially diffusing policies

Develop an “exposure mapping” synthetic control that constructs separate counterfactuals for units’ own treatment and their neighbors’ treatment intensity, allowing identification of direct effects and spillovers under staggered rollouts. Combine this with spatial permutation placebos to assess interference robustness and use DML only for nuisance estimation, keeping the causal estimand transparent. This framework explicitly models spatial diffusion and interference, addressing limitations of conventional synthetic control methods that assume no interference. It builds on documented spatial spillovers in policy pilots, synthetic control foundations, and placebo tests over space and time. The approach yields more credible estimates for real-world policies that spread beyond administrative borders and improves mechanism learning by separating own-treatment channels from neighborhood channels. The potential impact includes informing optimal rollout sequences, targeting, and coordination to maximize benefits and minimize unintended inequities in policy evaluation.

References:

1. The Explaimability of Double Machine Learning Causal Inference in Quasi-Natural Experiments—A Study Based on County Panel Sample Data. Zongxuan Chai, Tingting Zheng (2023). Automated Machine Learning.
2. A blueprint for synthetic control methodology: a causal inference tool for evaluating natural experiments in population health. Ben Barr, Xingna Zhang, M. Green, I. Buchan (2022). British medical journal.
3. A blueprint for synthetic control methodology: a causal inference tool for evaluating natural experiments in population health. Ben Barr, Xingna Zhang, M. Green, I. Buchan (2022). British medical journal.
4. A blueprint for synthetic control methodology: a causal inference tool for evaluating natural experiments in population health. Ben Barr, Xingna Zhang, M. Green, I. Buchan (2022). British medical journal.