统计机器学习与因果效应评估系列学术报告（16）

报告题目：Semiparametric efficient estimation for random-effects meta-analysis

报告人： 魏嘉金 博士 （香港浸会大学）

报告时间：2025年3月23日晚上20：00-21：00

报告地点：腾讯会议980-627-761

报告摘要：Meta-analysis is a commonly used statistical method for synthesizing evidence from multiple scientific studies. When heterogeneity exists between the studies, the random-effects model is routinely applied that assumes, by virtue of convenience, a normal distribution for the study-specific effect sizes. Such a normality assumption can be questionable, and in fact, it has raised increasing concerns in the recent literature. Inspired by this, we propose, for the first time, a semiparametric framework for learning the random-effects model without imposing the normality constraint on the between-study population. Specifically, by solving an estimating equation composed of the efficient score functions, we derive new estimators for the mean effect size and the between-study variance, and moreover study their asymptotic properties including the consistency, efficiency, and asymptotic normality. We further show that the maximum likelihood estimators of the two parameters under the normality assumption are the special cases of our new estimators, and more surprisingly, they are able to remain consistent and asymptotically normal even if the model is misspecified. Simulation studies show that our new estimators are efficient and robust, and they perform better than the existing methods in the presence of considerable heterogeneity. Lastly, two meta-analyses are also reanalyzed, where our new methods alter the conclusions from the previous literature.

报告人简介：魏嘉金，香港浸会大学数学系博士后研究员，主要研究方向包括：Meta分析，半参数模型，测量误差模型，机器学习方法在化学图谱数据中的应用等。魏嘉金博士2017年本科毕业于北京师范大学-香港浸会大学联合国际学院统计系，2020年硕士毕业于香港浸会大学数学系，2024年博士毕业于香港浸会大学数学系，研究方向是Meta分析和半参数随机效应模型。博士毕业后，魏嘉金在香港浸会大学数学系从事博士后研究工作，研究方向包括控制风险回归和发表偏倚。魏博士曾在International Statistical Review, Chemometrics and Intelligent Laboratory Systems, Military Medical Research等高水平期刊上发表过一系列文章。