统计机器学习与因果效应评估系列学术报告（15）

报告题目：A new definition of standardized mean difference allowing for unequal variances

报告人： 石建栋 博士 （香港中文大学）

报告时间：2025年3月23日晚上：19：00-20：00

报告地点：腾讯会议980-627-761

报告摘要：The standardized mean difference (SMD) is a commonly used effect size to quantify the mean difference between the case and control groups with continuous outcomes. Under the equal variance assumption, the SMD can be naturally defined as the mean difference divided by the common standard deviation. In clinical practice, however, the equal variance assumption may be too restrictive. To allow for unequal variances, the common approach in the literature is to take an arithmetic mean of the two variances, yet by doing so, the resulting SMD cannot be unbiasedly estimated. Inspired by this, we propose a geometric approach for averaging the two variances, which subsequently yields a new and novel measure for standardizing the mean difference allowing for unequal variances. We further propose the Cohen-type and Hedges-type estimators for the new SMD, and moreover derive their statistical properties together with the confidence intervals. Simulation results show that the Hedges-type estimator performs best under a wide range of settings in terms of the bias and coverage probability, and a real data example also demonstrates that our new SMD and its estimator bring new insights to the existing literature and can also be highly recommended for practical use.

报告人简介：石建栋，香港中文大学统计系博士后研究员，主要研究方向包括：统计meta分析、医学统计、生物信息、结构性变异检测、贝叶斯统计推断等。石博士2015年及2018年于山东大学数学学院统计学专业获得学士及硕士学位，2021年博士毕业于香港浸会大学数学系，研究方向是统计meta分析。博士毕业后曾在香港科技大学作为博士后研究员研究长度长DNA测序中结构性变异检测及相关分析。目前在香港中文大学研究Metropolis-Hastings方法中的收敛诊断问题。石博士曾在高水平期刊上发表过一系列文章，包括：Statistical Methods in Medical Research、Research Synthesis Methods、Statistics and Its Interface、Communications in Mathematics and Statistics、Military Medical Research等，其中两篇文章为Web of Science热点及高被引文章。截至目前，石博士Google Scholar总引用次数为652次。