• 统计研究中心
当前位置: 首页> 系列讲座> 正文

美国杜克大学李凡教授:Propensity score weighting in causal inference

主 题:Propensity score weighting in causal inference

讲人:美国杜克大学李凡教授

主持人:统计学院林华珍教授

时间:2022年8月31日(周三)上午9:30-10:30

直播平台及会议ID:腾讯会议,ID: 922-482-072

主办单位:统计研究中心和统计学院 科研处


主讲人简介:

李凡,美国杜克大学统计科学系的教授。主要研究兴趣是因果推断——在随机实验和观察性研究中评估治疗和干预的设计与分析,以及它们在健康研究(也称为比较有效性研究)和计算社会科学中的应用。她还致力于因果推断和机器学习之间的接口。她开发了倾向评分、临床试验、随机实验(如A/B测试)、差中差、回归不连续设计、表示学习和贝叶斯方法的方法。她还研究过丢失数据的统计方法。此外,她还做了一些关于基因组学和神经成像数据的贝叶斯图形建模的工作。


内容提要:

Covariate balance is crucial for causal inference. However, lack of balance is common in observational studies. In this talk, we overview a general class of weighting strategies for balancing covariates: the balancing weights. These weights incorporate the propensity score to weight each group to an analyst-selected target population. This class unifies existing weighting methods, including inverse-probability weights as special cases. In particular, we focus on the overlap weighting (OW) scheme, in which each unit's weight is proportional to the probability of that unit being assigned to the opposite group. The overlap weights are bounded, minimize the asymptotic variance of the weighted average treatment effect among the class of balancing weights, and also possess a desirable small-sample exact balance property. The overlap weights target at the population in clinical equipoise, or more broadly, the population with substantial overlap in baseline characteristics between two groups. We discuss several scenarios in health data science where OW is particularly relevant, such as constructing external controls with real world data.


上一条:耶鲁大学马双鸽教授:Hierarchical Cancer Heterogeneity Analysis Based On Histopathological Imaging Features

下一条:清华大学邓柯副教授:Stratification and Optimal Resampling for Sequential Monte Carlo