报告时间：2019 / 11 / 20（周三） 15:00 - 16:00
报告地点：明德主楼 1016 会议室
报告主题：Demystifying a Class of Multiply Robust Estimators
报告摘要：When estimating the population mean of a response variable subject to ignorable missingness, a new class of methods called the multiply robust procedures has been proposed. The advantage of the multiply robust procedures over the traditional doubly robust methods is that the former permit the use of multiple candidate models for both the propensity score and the outcome regression, and the multiply robust estimators are consistent if any one of the multiple models is correctly specified. Such a property is termed multiple robustness. Somewhat surprisingly, we show that these multiply robust estimators are special cases of the doubly robust estimators where the final propensity score and outcome regression models are certain combinations of the candidate models. To further improve model specifications in the doubly robust estimators, we adapt a model mixing procedure as an alternative method to combine multiple candidate models. We show that the multiple robustness property and asymptotic normality can be also achieved by our mixing-based doubly robust estimator. In addition, our estimator and the established theoretical properties are not confined to parametric models. Numerical examples further demonstrate that our proposed estimator is comparable to or can outperform existing multiply robust estimators.
报告人简介：李伟，北京大学数学科学学院博雅博士后，2018年于北京大学获得概率论与数理统计博士学位，导师周晓华教授和耿直教授，博士期间曾经访问美国西雅图华盛顿大学生物统计系一年。主要研究方向为因果推断，因果网络，缺失数据，高维数据分析，不确定性人工智能等。目前已在包括Journal of Econometrics, Biometrika, Statistics in Medicine等国际著名统计期刊上发表学术论文6篇。主持中国博士后科学基金第66批面上项目一项。