![[2602.12379] Deep Doubly Debiased Longitudinal Effect Estimation with ICE G-Computation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.12379] Deep Doubly Debiased Longitudinal Effect Estimation with ICE G-Computation
Summary
This paper presents D3-Net, a novel framework for estimating longitudinal treatment effects using ICE G-computation, addressing error propagation and enhancing robustness in outcomes.
Why It Matters
Estimating treatment effects accurately is crucial for effective decision-making in various fields, including healthcare and social sciences. The proposed D3-Net framework improves upon existing methods by reducing bias and variance, making it a significant advancement in longitudinal data analysis.
Key Takeaways
- D3-Net mitigates error propagation in ICE G-computation through Sequential Doubly Robust pseudo-outcomes.
- The framework employs a multi-task Transformer for auxiliary supervision, enhancing model stability.
- Final estimates are derived using Longitudinal Targeted Minimum Loss-Based Estimation for optimal properties.
- Comprehensive experiments show D3-Net outperforms existing ICE-based estimators in bias and variance reduction.
- This research contributes to more reliable longitudinal effect estimation, benefiting various applications.
Computer Science > Machine Learning arXiv:2602.12379 (cs) [Submitted on 12 Feb 2026] Title:Deep Doubly Debiased Longitudinal Effect Estimation with ICE G-Computation Authors:Wenxin Chen, Weishen Pan, Kyra Gan, Fei Wang View a PDF of the paper titled Deep Doubly Debiased Longitudinal Effect Estimation with ICE G-Computation, by Wenxin Chen and 3 other authors View PDF HTML (experimental) Abstract:Estimating longitudinal treatment effects is essential for sequential decision-making but is challenging due to treatment-confounder feedback. While Iterative Conditional Expectation (ICE) G-computation offers a principled approach, its recursive structure suffers from error propagation, corrupting the learned outcome regression models. We propose D3-Net, a framework that mitigates error propagation in ICE training and then applies a robust final correction. First, to interrupt error propagation during learning, we train the ICE sequence using Sequential Doubly Robust (SDR) pseudo-outcomes, which provide bias-corrected targets for each regression. Second, we employ a multi-task Transformer with a covariate simulator head for auxiliary supervision, regularizing representations against corruption by noisy pseudo-outcomes, and a target network to stabilize training dynamics. For the final estimate, we discard the SDR correction and instead use the uncorrected nuisance models to perform Longitudinal Targeted Minimum Loss-Based Estimation (LTMLE) on the original outcomes. This second-st...
