题目：Time-varying Treatment Effects of Functional Data with Latent Confounders: Application to Sleep Heart Health Studies
报告人：李杰
报告时间：2024年11月15日10：00-11：30
地点： 综合楼644会议室
报告人简介：
李杰，中国人民大学太阳成集团tyc7111cc讲师，中国人民大学青年英才。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、时间序列分析、非参数统计等。曾荣获2021年国际统计学会简·丁伯根奖一等奖，目前主持国家自然科学基金青年项目。
报告摘要：
Exploring the causal effect between variables is an important issue in lots of scientific research. Existing literature on causal inference mainly studies one-dimensional or multi-dimensional data, but functional data with repeated observations per individual frequently appears in a wide variety of applications. In functional data, treatment design may change across time, and its treatment effect is a time-varying function. Besides, most methods for treatment effect estimation based on observational data rely on the ignorability assumption that treatment assignment is independent of the potential outcomes given the observable covariates. This assumption can be violated when unobserved latent covariates are involved. We propose a novel method for unbiased treatment effect estimation with unobserved latent covariates for functional data. We propose to solve this challenging problem using a joint likelihood method with a Monte Carlo EM algorithm. Moreover, our proposed method is flexible to estimate both heterogeneous treatment effect of individuals and average treatment effect, providing a reliable inferential tool in making treatment decisions. It can also be applied to the irregular and sparse data. The method leads to meaningful discoveries when applied to investigate the dynamic effect of sleep quality on heart rate variability.
| 上一条： 没有了 |
| 下一条： 没有了 |
