报告题目:Finite-Sample Inference for Composite Null Hypotheses:A New Method and Pharmacogenomics Studies

报告人:王铭(美国堪萨斯大学)

报告摘要：

This presentation bridges applied pharmacogenomics (PGx) research with methodological developments in statistical inference, motivated by the challenges of evaluating complex, multidimensional outcomes in small samples. The work consists of three applied studies followed by the development and validation of a finite-sample testing framework for composite null hypotheses based on pointwise rejection. The proposed framework evaluates each point in the null region at a modified significance level, ensuring finite-sample control of the Type I error uniformly over the null set. Unlike traditional likelihood ratio tests that rely on asymptotic approximations, the method achieves improved reliability in small samples and in settings involving boundary points or nuisance parameters. Through examples in one to higher dimensions and extensive simulation studies, the pointwise approach demonstrates accurate size control and competitive or superior power compared to likelihood ratio tests, Bonferroni adjustments, and universal data-splitting methods. The framework is further extended to nonparametric settings through testing problems involving distributional functionals, including median-based hypotheses. These results show that the pointwise-rejection principle can be implemented without parametric likelihood assumptions while maintaining finite-sample error control under appropriate calibration.

报告人简介：

王铭，堪萨斯大学讲师，2001年，本科毕业于北京师范大学，2009年，硕士毕业于加州大学洛杉矶分校，2026年，获得堪萨斯大学博士学位，研究方向为统计与概率，其研究内容为复合零假设的有限样本推断及药物基因组学研究。本人已发表三篇文章分别在 Currents in Pharmacy Teaching and Learning，Journal of Pharmacy Practice 和 Current Pharmacogenomics and Personalized Medicine。与他人合作，另有两篇数学文章预印本，其研究方向是调和分析和偏微分方程。

报告时间: 2026.6.17(周三) 上午 10:30-11:30

报告地点: 逸夫楼1537