Causal Effect Estimation with TMLE: Handling Missing Data and Near-Violations of Positivity

2025-10-25

AI Summary

This study evaluates the robustness of Targeted Maximum Likelihood Estimation (TMLE) for estimating the Average Treatment Effect (ATE) under missing data and varying degrees of positivity violation. Using model- and design-based simulations—augmented with real-world data from the WASH Benefits Bangladesh trial—we systematically compare Complete-Case Analysis (CCA), Multiple Imputation (MI), and non-MI strategies across missingness mechanisms (MCAR, MAR, MNAR) and machine learning tools (e.g., Highly Adaptive Lasso [HAL], Classification and Regression Trees [CART]). We propose a novel “CCA+TMLE” framework that integrates an outcome missingness model within a non-MI setting, markedly reducing bias and enhancing robustness to positivity violations, while uncovering a fundamental bias–coverage trade-off. Results show CCA+TMLE achieves the lowest bias; CART-based MI yields the best root mean squared error (RMSE) and confidence interval coverage.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Multiagent Systems: Mechanism Design

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We evaluate the performance of targeted maximum likelihood estimation (TMLE) for estimating the average treatment effect in missing data scenarios under varying levels of positivity violations. We employ model- and design-based simulations, with the latter using undersmoothed highly adaptive lasso on the 'WASH Benefits Bangladesh' dataset to mimic real-world complexities. Five missingness-directed acyclic graphs are considered, capturing common missing data mechanisms in epidemiological research, particularly in one-point exposure studies. These mechanisms include also not-at-random missingness in the exposure, outcome, and confounders. We compare eight missing data methods in conjunction with TMLE as the analysis method, distinguishing between non-multiple imputation (non-MI) and multiple imputation (MI) approaches. The MI approaches use both parametric and machine-learning models. Results show that non-MI methods, particularly complete cases with TMLE incorporating an outcome-missingness model, exhibit lower bias compared to all other evaluated missing data methods and greater robustness against positivity violations across. In Comparison MI with classification and regression trees (CART) achieve lower root mean squared error, while often maintaining nominal coverage rates. Our findings highlight the trade-offs between bias and coverage, and we recommend using complete cases with TMLE incorporating an outcome-missingness model for bias reduction and MI CART when accurate confidence intervals are the priority.

Problem

• Estimating causal effects with TMLE under missing data conditions
• Addressing positivity violations in observational study designs
• Comparing imputation methods for exposure, outcome, and confounder missingness

Innovation

• TMLE handles missing data with outcome-missingness model
• Uses highly adaptive lasso for real-world complexity simulation
• Compares multiple imputation with machine learning methods