Research map / Probabilistic and scientific modeling / Treatment effects
Proximal Difference-in-Differences for Long-Term Causal Learning under Confounding and Outcome Drift
Sihyung Park, Shu Yang, Mingyang Shan, Wenyu Ye, Ilya Lipkovich
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a proximal difference-in-differences method to estimate long-term treatment effects by combining RCT data with real-world evidence under confounding and outcome drift.
Problem
Estimating long-term treatment effects when RCT follow-up is short and open-label extensions censor long-term control trajectories; existing RWE-supplementation methods are vulnerable to unobserved confounding and outcome drift.
Approach
Applies a proximal (proxy-variable) difference-in-differences identification strategy to handle unobserved confounding and temporal outcome drift when fusing RCT and real-world data for long-term causal effects.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use proximal doubly robust (PDR) estimator instead of baselines | long-term control trend · Alzheimer's plasmode | bias: -0.005 vs -0.896 (MB), -0.607 (SDID), -0.537 (MC-NNM) | Improved |
| use PDR rather than POR or PIPW under nuisance misspecification (ablation) | proximal estimators · synthetic simulation (n=500, 500 reps) | bias, coverage: POR and PIPW biased with poor coverage when a nuisance is misspecified | Improved |
| use PDR with nonlinear data (gamma=2) | PDR · synthetic simulation | bias and coverage: modest bias, slight undercoverage | Mixed |
| use proximal PDR instead of SDID, MC-NNM, MB | long-term effect estimators · synthetic simulation | bias and coverage: larger bias and lower coverage for baselines | Improved |
| strong proxy-Z violation (gamma_z 0.3 -> 1.5) (ablation) | PDR · sensitivity simulation | bias: -0.254 -> -0.866 | Hurt |
| strong proxy-W violation (gamma_w 0.05 -> 0.25) (ablation) | PDR · sensitivity simulation | bias: -0.142 -> -0.365 | Hurt |
| violate CPTL with trend acceleration alpha=0.5 (ablation) | PDR vs benchmarks · sensitivity simulation | bias: PDR -0.262 vs SDID -2.096 | Improved |
| moderate violation of W independence (ablation) | PDR vs SDID · sensitivity simulation | bias: PDR -0.365 vs SDID -0.120 | Hurt |
Experimental setup
Builds on: difference-in-differences proximal causal inference
Closest papers by meaning
- Inferring the Long-Term Causal Effects of Long-Term Treatments from Short-Term Experiments (ICML 2024 · similarity 0.88)
- Causal Identification for Complex Functional Longitudinal Studies (ICLR 2025 · similarity 0.87)
- Generalizing Causal Effects from Randomized Controlled Trials to Target Populations across Diverse Environments (ICML 2025 · similarity 0.87)
- Continuous Treatment Effects with Surrogate Outcomes (ICML 2024 · similarity 0.86)
- B-CALM: Bias-Limited Bayesian Borrowing for RCT-Anchored Treatment Effects under Covariate Mismatch (NeurIPS 2026 · similarity 0.86)
- Do Contemporary Causal Inference Models Capture Real-World Heterogeneity? Findings from a Large-Scale Benchmark (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
