The content discusses a collaborative approach for causal inference with heterogeneous data across multiple sites. The key insights are:
The authors propose a sampling-selecting framework to model the heterogeneity across sites, where each site selects a biased sample from the target population.
They introduce the Clb-IPW estimator, which directly takes the weighted mean of heterogeneous propensity score functions, instead of taking the weighted mean of site-wise IPW estimators as in meta-analysis. This allows collaboration even when sites have disjoint domains.
To address the challenge of density ratio estimation, the authors incorporate outcome models using the augmented IPW (AIPW) estimator. They provide convergence rates for the nuisance models and show the asymptotic normality of the Clb-AIPW estimator.
The authors develop a federated learning algorithm to collaboratively train the outcome model while preserving privacy.
Experiments on synthetic and real-world datasets demonstrate the advantages of the proposed methods over meta-analysis approaches, especially when the heterogeneity across sites increases.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Tianyu Guo,S... at arxiv.org 04-25-2024
https://arxiv.org/pdf/2404.15746.pdfDeeper Inquiries