时 间：2026-05-08 14:30 - 15:30
地 点：普陀校区理科大楼A1514
报告人：李杰 中国人民大学副教授
主持人：张心雨 华东师范大学副教授
摘 要：
<img class="akeylayout_img"rstanding time-varying causal effects in functional data is increasingly important across many scientific domains. Traditional methods largely target scalar or vector outcomes, but functional responses measured repeatedly over time demand a new methodological framework. In such settings, treatment assignments (or exposure levels) may also vary over time, and ignorability given observed covariates is often untenable due to unmeasured confounding. We introduce a joint modeling framework that incorporates latent confounders into both the treatment and functional outcome models, and propose a new EM algorithm, FJ-MCGEM, for parameter estimation. The procedure yields unbiased estimates of both average and heterogeneous treatment-effect functions. The method is robust to irregular and sparse temporal measurements, making it well-suited to real-world functional data. We apply the method to the Study of Women’s Health Across the Nation (SWAN) to assess how depressive symptoms affect follicle-stimulating hormone (FSH) trajectories during the menopausal transition. The analysis reveals dynamic and evolving effects over time, illustrating the framework’s flexibility and value for causal inference with functional data in the presence of latent confounding.
报告人简介：
李杰，中国人民大学统计学院副教授，中国人民大学青年英才。2017年本科毕业于北京师范大学数学科学学院，2022年于清华大学获得统计学博士学位。主要研究方向为函数型数据分析、时间序列、非参数统计等。目前主持国家自然科学基金青年项目，且在SCI期刊发表论文多篇。