Skip to content
DraconiticResearch
Search Sources API
203,234 studies31 sourcesUpdated just now
Peer-reviewedCC BYIEEE Transactions on Information Theory

The Decaying Missing-at-Random Framework: Model Doubly Robust Causal Inference With Partially Labeled Data

  • Yuqian Zhang1
  • Abhishek Chakrabortty2
  • Jelena Bradić3
  1. 1Renmin University of China
  2. 2Texas A&M University
  3. 3Cornell University
Open at sourceDOICite

Abstract

In modern large-scale observational studies, data collection constraints often result in partially labeled datasets, posing challenges for reliable causal inference, especially due to potential labeling bias and relatively small size of the labeled data. This paper introduces a decaying missing-at-random (decaying MAR) framework and associated approaches for doubly robust causal inference on treatment effects in such semi-supervised (SS) settings. This simultaneously addresses selection bias in the labeling mechanism and the extreme imbalance between labeled and unlabeled groups, bridging the gap between the standard SS and missing data literatures, while throughout allowing for confounded treatment assignment and high-dimensional confounders under appropriate sparsity conditions. To ensure robust causal conclusions, we propose a bias-reduced SS (BRSS) estimator for the average treatment effect, a type of 'model doubly robust' estimator appropriate for such settings, establishing asymptotic normality at the appropriate rate under decaying labeling propensity scores, provided that at least one nuisance model is correctly specified. Our approach also relaxes sparsity conditions beyond those required in existing methods, including standard supervised approaches. Recognizing the asymmetry between labeling and treatment mechanisms, we further introduce a de-coupled BRSS (DC-BRSS) estimator, which integrates inverse probability weighting (IPW) with bias-reducing techniques in nuisance estimation. This refinement further weakens model specification and sparsity requirements. Numerical experiments confirm the effectiveness and adaptability of our estimators in addressing labeling bias and model misspecification.

Related

  1. Inference for Directions in Canonical Correlation Analysis with Partially Paired Data
    Chau Tran, Xiaodong Li
    arXivPreprintarXiv licenceFull textStatisticsarXiv
  2. Bayesian Inference in Partially Observed Critical Systems
    Mihika Singhania
    arXivPreprintarXiv licenceFull textProbabilityStatistical & nonlinear physicsarXiv
  3. Modular Bayesian Inference for Partially Misspecified Mixtures with Uncertain Class Labels
    Maximilian Autenrieth, Konstantin Karchev, David A. van Dyk +2
    arXivPreprintarXiv licenceFull textCosmologyStatisticsarXiv
  4. Bias-Corrected Empirical-Likelihood Inference for Single-Index Expectile Regression Under Missing-at-Random Responses
    Pingping Liu, Shuanghua Luo, Junxiang Lu
    AxiomsPeer-reviewedCC BY 4.0General mathematicsCrossref
  5. Doubly robust target inference for generalized linear regression with completely missing covariates
    Huali Zhao, Ke Deng
    arXivPreprintarXiv licenceFull textStatisticsarXiv

Cite

BibTeXRIS
BibTeX
@article{zhang2026decaying,
  title = {{The Decaying Missing-at-Random Framework: Model Doubly Robust Causal Inference With Partially Labeled Data}},
  author = {Yuqian Zhang and Abhishek Chakrabortty and Jelena Bradić},
  journal = {IEEE Transactions on Information Theory},
  year = {2026},
  month = jul,
  doi = {10.1109/tit.2026.3713863},
  url = {https://doi.org/10.1109/tit.2026.3713863},
  note = {CC BY},
}
RIS
TY  - JOUR
TI  - The Decaying Missing-at-Random Framework: Model Doubly Robust Causal Inference With Partially Labeled Data
AU  - Yuqian Zhang
AU  - Abhishek Chakrabortty
AU  - Jelena Bradić
PY  - 2026
DA  - 2026/07/15
JO  - IEEE Transactions on Information Theory
DO  - 10.1109/tit.2026.3713863
UR  - https://doi.org/10.1109/tit.2026.3713863
AB  - In modern large-scale observational studies, data collection constraints often result in partially labeled datasets, posing challenges for reliable causal inference, especially due to potential labeling bias and relatively small size of the labeled data. This paper introduces a decaying missing-at-random (decaying MAR) framework and associated approaches for doubly robust causal inference on treatment effects in such semi-supervised (SS) settings. This simultaneously addresses selection bias in the labeling mechanism and the extreme imbalance between labeled and unlabeled groups, bridging the gap between the standard SS and missing data literatures, while throughout allowing for confounded treatment assignment and high-dimensional confounders under appropriate sparsity conditions. To ensure robust causal conclusions, we propose a bias-reduced SS (BRSS) estimator for the average treatment effect, a type of 'model doubly robust' estimator appropriate for such settings, establishing asymptotic normality at the appropriate rate under decaying labeling propensity scores, provided that at least one nuisance model is correctly specified. Our approach also relaxes sparsity conditions beyond those required in existing methods, including standard supervised approaches. Recognizing the asymmetry between labeling and treatment mechanisms, we further introduce a de-coupled BRSS (DC-BRSS) estimator, which integrates inverse probability weighting (IPW) with bias-reducing techniques in nuisance estimation. This refinement further weakens model specification and sparsity requirements. Numerical experiments confirm the effectiveness and adaptability of our estimators in addressing labeling bias and model misspecification.
SN  - 0018-9448
SN  - 1557-9654
ER  -