pooled median variance estimation

Traditional meta-analyses of median differences often exclude studies that do not report measures of dispersion such as interquartile range or range, potentially introducing selection bias. This work proposes a Direct Variance Estimation (DiVE) method that constructs a variance estimator for the pooled effect using only the reported median differences and sample sizes from each study, without requiring any dispersion statistics. The approach is grounded in asymptotic theory and validated through extensive simulations across diverse distributional settings. Results demonstrate that DiVE performs comparably to or better than conventional two-stage methods, particularly in small-sample scenarios and under various underlying distributions. By enabling inclusion of studies previously excluded due to missing dispersion information, DiVE enhances the completeness and reliability of evidence synthesis in meta-analysis.

This study addresses the substantial bias often introduced in meta-analyses when estimating standard deviations solely from the five-number summary—specifically, the minimum, maximum, and median—due to insufficient information, which can compromise inferential reliability. To mitigate this issue, the authors propose a novel estimation method based on a scaled Beta distribution that incorporates data shape characteristics to improve accuracy. A comprehensive sensitivity analysis is systematically conducted to quantify estimation uncertainty. Through extensive simulation studies and real-data applications, the proposed approach demonstrates markedly superior performance over conventional estimators across a variety of underlying distributions. Additionally, the authors provide an interactive web tool to facilitate practical implementation, enabling researchers to readily assess and correct potential bias in standard deviation estimates, thereby enhancing the robustness of meta-analytic findings.

该研究探讨了SDR方法在小域中估计方差时存在的偏差问题，通过理论公式和模拟分析表明SDR方法会导致小域内方差估计值的膨胀，并提出了实施SDR时循环次数D应至少为3但不超过5以控制估计值变异性。

Traditional simulation studies often rely on mean and standard deviation to assess the quality of asymptotic approximations; however, the existence and convergence of moments are not guaranteed by distributional convergence alone, and such approaches inadequately characterize near-normal approximations contaminated by outliers. This work proposes replacing conventional moment-based summaries with quantile-based summary statistics, specifically employing the median, median absolute deviation, and empirical confidence interval coverage as robust and universally applicable evaluation metrics. By shifting away from the mean-centered paradigm, this approach overcomes both theoretical and practical limitations inherent in moment-based methods and provides a more interpretable and reliable criterion for evaluating simulation results.

This paper addresses the challenge of design-based inference for the average treatment effect (ATE) in finely stratified randomized experiments—particularly under the extreme stratification regime where each stratum contains only one treated or one control unit. We propose a novel pairwise-differenced-mean variance estimator that pairs adjacent, similar strata. Unlike existing estimators, ours remains well-defined and upwardly biased with controllable magnitude even in the single-unit-per-stratum limit. Under a similarity assumption on adjacent strata, we prove analytically that our estimator exhibits reduced bias and is asymptotically superior to state-of-the-art alternatives. Finite-population bias analysis and i.i.d. superpopulation modeling, corroborated by Monte Carlo simulations, demonstrate that under high-quality stratification, our method yields substantially narrower confidence intervals and improved inferential accuracy. Our key contribution is the first variance estimation framework that simultaneously ensures theoretical rigor—via finite-sample bias characterization and asymptotic dominance—and practical robustness across realistic stratification scenarios.

本文解决了在主体内实验设计中如何准确估计处理效应的问题，通过构建潜在结果框架和敏感性分析方法来评估并改进现有方法。

This study addresses the challenge of variance estimation in local pivotal methods arising from the computational intractability of second-order inclusion probabilities. To overcome this limitation, two solutions are proposed: Monte Carlo simulation-based correction and model-assisted bootstrapping. Methodologically, Gaussian processes are introduced to construct synthetic populations, while pairwise independence tests are incorporated to enhance estimator stability. Experimental results demonstrate that, across various noise levels and sampling configurations, the model-assisted bootstrap approach exhibits superior robustness compared to conventional correction techniques. Ultimately, this work effectively improves the precision of statistical inference under spatially balanced sampling designs.

该研究通过将训练子样本视为第二阶段抽样，解决了模型辅助估计中的不确定性量化问题，并提出了两种方差估计方法。

This study addresses the substantial bias often incurred by conventional variance estimators—such as those based on Taylor linearization—for the generalized regression (GREG) estimator in high-dimensional auxiliary variable settings, which undermines inferential reliability. The paper provides the first systematic characterization of the asymptotic bias of GREG variance estimation under high-dimensional regimes and introduces a novel cross-validation–based approach to construct an unbiased variance estimator. Under mild distributional assumptions on the covariates, the proposed method is shown to be asymptotically unbiased. By integrating high-dimensional asymptotic theory with the model-assisted estimation framework, this work establishes a rigorous theoretical foundation for variance estimation in high-dimensional GREG settings and demonstrates through numerical experiments that the method exhibits excellent finite-sample performance.

This study addresses the challenges of estimating causal effects under cross-temporal trends in platform trials, where ambiguous target definitions and biases arising from pooling multi-period data remain unresolved. To tackle these issues, the authors systematically delineate conditional and marginal estimands and comprehensively evaluate the bias-variance trade-offs of model-based approaches, G-computation, and augmented inverse probability weighting (AIPW) estimators under complex temporal dynamics. The work reveals how estimator performance evolves with varying target populations and data selection strategies, thereby clarifying the applicability boundaries of each methodological approach. Ultimately, these findings provide a robust statistical foundation for regulatory decision-making in adaptive platform trial settings.