Research map / Probabilistic and scientific modeling / Treatment effects
Collaborative Heterogeneous Causal Inference Beyond Meta-analysis
Read the paperCheck your idea against 108,967 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 Clb-IPW and Clb-AIPW estimators that collaboratively combine heterogeneous sites by weighting propensity models jointly, avoiding meta-analysis, with a federated outcome-model training algorithm.
Problem
Causal inference (ATE estimation) across heterogeneous data sites whose covariate distributions differ from the target population, where existing methods reduce to meta-analysis of per-site estimates and fail when a site cannot cover the full population.
Approach
Instead of re-weighting each site to the target and then meta-analysing, directly use the weighted mean of all sites' selection propensity models for each individual's inverse weight, so each site's estimator is inconsistent alone but the sum across sites is consistent; extends to a decoupled AIPW with a public census covariate set and a federated FedAvg outcome model.
Key results
- On synthetic data, Clb-IPW and Clb-AIPW stay stable as mean KL heterogeneity rises from 0 to 4, while Meta-IPW's MSE grows with heterogeneity.
- Clb-IPW shows tighter confidence intervals than Meta-IPW in all four propensity/outcome misspecification combinations at mean KL 3.
- On the COVID-19 fake-news real data, the estimators give similar effects (about 0.034 vs 0.015 for the two studies), with Clb-IPW and AIPW showing slightly larger effect sizes.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use collaborative IPW instead of Meta-IPW as site heterogeneity increases | ATE estimation · Synthetic data | MSE | Helped |
| Use Clb-IPW instead of Meta-IPW | ATE estimation · Synthetic data | variance | Helped |
| Use AIPW with a correctly specified outcome model when the propensity score model is misspecified | AIPW estimators · Synthetic data | consistency | Helped |
| Use AIPW with a correctly specified propensity score model when the outcome model is misspecified | AIPW estimators · Synthetic data | consistency | Helped |
| Use AIPW when both propensity score and outcome models are misspecified | AIPW estimators · Synthetic data | consistency | Hurt |
| Use Clb-IPW or AIPW instead of Meta-IPW or Meta-AIPW | ATE estimation · Two fake-news studies | estimated effect size | Mixed |
Limitations
- Requires assumptions of homogeneous treatment effects and unconfoundedness across sites (Assumption on homogeneity and unconfoundedness).
- Density ratio estimation suffers curse of dimensionality; K-NN method only practical for d up to about 8.
- Optimal communication-efficient weighting of propensity models not solved.
Experimental setup
Datasets: synthetic Gaussian covariate data COVID-19 fake news sharing nudge studies
Metrics: mean squared error 95% confidence interval coverage/width
Baselines: Meta-IPW Meta-AIPW exponential tilting density ratio estimation K-NN density ratio estimation
Builds on: Meta-analysis (inverse-variance weighting) augmented inverse propensity weighting double machine learning FedAvg
Closest papers by meaning
- Inverse-Variance Weighting for Estimation of Heterogeneous Treatment Effects (ICML 2024 · similarity 0.87)
- Do Contemporary Causal Inference Models Capture Real-World Heterogeneity? Findings from a Large-Scale Benchmark (ICLR 2025 · similarity 0.86)
- Hierarchical and Density-based Causal Clustering (NeurIPS 2024 · similarity 0.85)
- A Meta-learner for Heterogeneous Effects in Difference-in-Differences (ICML 2025 · similarity 0.85)
- Covariate-Adjusted Deep Causal Learning for Heterogeneous Panel Data Models (NeurIPS 2026 · similarity 0.85)
- Data Fusion for Partial Identification of Causal Effects (NeurIPS 2025 · similarity 0.85)
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.