Research map / Federated learning and privacy / Personalized and heterogeneous FL
Federated Causal Inference on Multi-Site Observational Data via Propensity Score Aggregation
Rémi Khellaf, Aurélien Bellet, Julie Josse
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add site-level covariates Z to membership weights (ablation) | Fed-MW IPW (Logit) · Simulation K=30 sites, site effects (MSE), n_k=20 | MSE: 2.001 -> 0.427 | Improved |
| add site-level covariates Z to membership weights (ablation) | Fed-MW AIPW (Logit) · Simulation K=30 sites, site effects (MSE), n_k=20 | MSE: 2.026 -> 0.022 | Improved |
| add site-level covariates Z to membership weights (ablation) | Fed-MW IPW (NN) · Simulation K=30 sites, site effects (MSE), n_k=100 | MSE: 0.356 -> 0.206 | Improved |
| use federated propensity aggregation instead of meta-analysis (sample-size weighted) | IPW · Simulation K=30 sites, site effects (MSE), n_k=20 | MSE: 5.176 -> 0.427 | Improved |
| use federated propensity aggregation instead of meta-analysis | AIPW · Simulation K=30 sites, site effects (MSE), n_k=20 | MSE: 10.673 -> 0.022 | Improved |
| use logistic membership weights instead of Gaussian density-ratio weights | Fed-IPW · synthetic DGP B | bias | Improved |
| use neural network membership weights instead of logistic | Fed-IPW · synthetic DGP A and B | bias | Improved |
| use federated estimators instead of meta-analysis under no local overlap | Fed-IPW/AIPW · synthetic, no local overlap | bias | Improved |
| use federated estimators instead of meta-analysis under good local overlap | Fed-IPW/AIPW · synthetic, good local overlap | variance | Improved |
| use Fed-MW versus 1S-IVW without knowing shared parameters | Fed-MW · synthetic shared-parameter setting | variance | No clear effect |
| use Gaussian density-ratio weights instead of membership weights | Fed-AIPW · Traumabase | stability | Hurt |
| use federated MW instead of Meta-SW | Fed-AIPW · Traumabase | ATE estimate variance | Improved |
Closest papers by meaning
- Collaborative Heterogeneous Causal Inference Beyond Meta-analysis (ICML 2024 · similarity 0.92)
- Estimating Heterogeneous Treatment Effects by Combining Weak Instruments and Observational Data (NeurIPS 2024 · similarity 0.87)
- Do Contemporary Causal Inference Models Capture Real-World Heterogeneity? Findings from a Large-Scale Benchmark (ICLR 2025 · similarity 0.86)
- Data Fusion for Partial Identification of Causal Effects (NeurIPS 2025 · similarity 0.86)
- Doubly robust identification of treatment effects from multiple environments (ICLR 2025 · similarity 0.86)
- Federated Causal Discovery from Heterogeneous Data (ICLR 2024 · 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.
