时 间：2025年10月9日 14:00-15:00
地 点：普陀校区理科大楼A1514
报告人：赵彦勇 南京审计大学教授
主持人：唐炎林 华东师范大学教授
摘 要：
Mediation models reveal the underlying causal mechanisms by explaining both the direct effect of an explanatory variable on an outcome variable and the indirect effect mediated through intermediary variables. In traditional mediation models, it is assumed that the effect of the explanatory variable on the outcome variable is fixed. However, this assumption is often too restrictive in many practical applications. To capture the complex patterns of explanatory variable effects that vary with covariates, this paper proposes a high-dimensional mediation analysis method based on varying coefficient models. First, we adopt a varying coefficient model that allows the effects of the explanatory variable on both the mediator and the outcome variable to change with covariates. Second, based on B-splines, we derive an estimation method for varying coefficient high-dimensional mediation effects. To address the challenges in high-dimensional data settings, we introduce the Adaptive Lasso technique to select relevant variables and control model complexity. Additionally, we establish the convergence rate and asymptotic distribution of the obtained estimators, and propose an F-test to assess the direct effects. We demonstrate that the proposed test asymptotically follows a chi-square distribution under the null hypothesis and a non-central chi-square distribution under local alternatives, with the validity and robustness of the method verified through simulation experiments. We also propose a partial penalized Wald test for indirect effects, showing that the proposed test has a chi-square limiting distribution at zero, with its effectiveness and robustness validated through simulations. Finally, we apply this method to the ADNI anatomical MRI data to investigate potential mediating mechanisms in the factors influencing Alzheimer's disease (AD). The results indicate that the varying coefficient high-dimensional mediation model better captures the complex causal relationships, providing a new analytical tool for causal inference in high-dimensional environments. We successfully identify a set of ROI-based volumetric mediators for AD and further elucidate the influence of age on AD.
报告人简介：
南京审计大学统计与数据科学学院教授，博士生导师。2016年6月博士毕业于东南大学数学系，同年7月加入南京审计大学。主要研究方向为半参数模型、高维统计推断、分布式推断、复杂数据统计建模，主持2项国家自然科学基金、4项省部级基金（含1项重点项目）。在中国科学:数学、统计研究、JMLR、JMVA等发表论文40余篇，相关成果获全国商业科技进步奖二等奖、江苏省统计科研优秀成果奖二等奖等，担任中国现场统计研究会生存分析分会常务理事，中国现场统计研究会理事等。