Primary Submission Category: Randomized Designs and Analyses
Efficient Statistical Estimation for Sequential Adaptive Experiments with Implications for Adaptive Designs
Authors: Wenxin Zhang, Mark van der Laan,
Presenting Author: Wenxin Zhang*
Adaptive experimental designs are increasingly used in clinical trials and digital experiments, allowing treatment randomization probabilities to be updated based on sequentially accrued data, with objectives ranging from enhancing estimation efficiency to improving participant outcomes. However, the resulting dependence among observations induced by adaptive designs poses substantial challenges for efficient estimation and inference of causal estimands. Building on the Targeted Maximum Likelihood Estimation (TMLE) framework tailored for adaptive experiments, we introduce a new adaptive-design-likelihood-based TMLE (ADL-TMLE) for estimating a broad class of causal estimands from adaptively collected data, including the average treatment effect. We establish asymptotic normality and semiparametric efficiency of ADL-TMLE under relaxed positivity and design stabilization assumptions for adaptive experiments, while achieving improved finite-sample efficiency relative to prior TMLE approach that relies on inverse probability weighting of adaptive randomization probabilities. Simulation studies show that ADL-TMLE achieves substantial variance reduction across a range of adaptive experiments. Motivated by these results, we further propose a novel adaptive design that targets efficient estimation of causal estimands and outperforms standard efficiency-oriented adaptive designs. We further extend the proposed framework to broader settings including longitudinal structures.