Scholarlink

<p>Randomized controlled trials are the gold standard for studying causal relationships. However, datasets collected from randomized experiments are often limited in sample size due to limitations in time and budget, which can lead to high-variance treatment effect estimates by the simple difference-of-means estimator. To reduce the variance of the estimated treatment effect and improve the estimation accuracy, we propose a novel model-averaging-based covariate adjustment method. We use model averaging to combine group-specific working regressions for comparing treatment group and control group, and then adjust the difference-of-means estimator. We study the proposed model-averaging-based treatment effect estimator under the Neyman–Rubin model for randomized controlled trials and provide theoretical justification for its asymptotic normality, and provide a conservative estimator of the asymptotic variance, which can yield tighter confidence intervals than the difference-of-means estimator. Moreover, numerical simulations as well as an analysis of real-world data from a primary biliary cirrhosis randomized trial show that model-averaging-based adjustment can be advantageous in reducing estimation error and providing effective confidence intervals.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Average treatment effect estimation in randomized controlled trials under the Rubin causal model via model averaging

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Huihang Liu,
  • Ziwen Gao,
  • Xinyu Zhang

摘要

Randomized controlled trials are the gold standard for studying causal relationships. However, datasets collected from randomized experiments are often limited in sample size due to limitations in time and budget, which can lead to high-variance treatment effect estimates by the simple difference-of-means estimator. To reduce the variance of the estimated treatment effect and improve the estimation accuracy, we propose a novel model-averaging-based covariate adjustment method. We use model averaging to combine group-specific working regressions for comparing treatment group and control group, and then adjust the difference-of-means estimator. We study the proposed model-averaging-based treatment effect estimator under the Neyman–Rubin model for randomized controlled trials and provide theoretical justification for its asymptotic normality, and provide a conservative estimator of the asymptotic variance, which can yield tighter confidence intervals than the difference-of-means estimator. Moreover, numerical simulations as well as an analysis of real-world data from a primary biliary cirrhosis randomized trial show that model-averaging-based adjustment can be advantageous in reducing estimation error and providing effective confidence intervals.