统计与数据科学系系列学术报告之四百八十二期

时间： 2025年8月12日（周二）15:00-16:00

主持人：复旦大学管理学院统计与数据科学系冯项楠副教授

地点：史带楼503室

报告人：黄薇教授 The University of Melbourne

题目：General Treatment Effect Inference using Generalised Empirical Likelihood Weights for Complex Data

摘要：Identifying and estimating the causal effect of a treatment or policy from observational studies is of great interest to economics, social science, and public health researchers. There, confounding issues usually exist (i.e. individual characteristics are related to both the treatment selection and the potential outcome), making the causal effect not directly identifiable from the data. We identify the general treatment effects by a weighted conditional expectation under uncounfoundedness assumption. The weights can be nonparametrically estimated by a generalised empirical likelihood subject to an expanding set of moment equations. The dual solution of the estimator enjoys fast computation and stable results. We have applied the framework to a broad range of data types, including discrete, continuous, and functional treatment variable, high-dimensional confounding variable, and data with measurement errors. For discrete treatments, the proposed estimator attains the semiparametric efficiency bounds. For continuous treatments, the estimator is more efficient than that constructed from the true weights. For high-dimentional confounders, a doubly-robust general treatment effect estimator can also be obtained through the generalised empirical likelihood from a deep neural network weight. In this talk, I will give an introduction to the framework and its application in some of the aforementioned complex data contexts.

个人简介：Senior Lecturer at the University of Melbourne (tenured, aka Associate Professor in North America), and Australian Research Council (ARC) Discovery Early Career Research Awards 2025 (DECRA25) recipient. She obtained her Bachelor of Science and Master of Philosophy degrees from the Chinese University of Hong Kong. She then pursued a PhD under the supervision of Prof Aurore Delaigle in Mathematical Statistics at the School of Mathematics and Statistics, University of Melbourne, where she was awarded her degree in 2019 and was appointed as a continuing Lecturer (Research and Teaching) in the same year. Her research focuses on methodological and theoretical statistics, with particular interests in nonparametric statistics, causal inference, measurement error, and functional data analysis. Her publications are mostly on JASA, JRSSB, JMLR, JBES and top conferences including NeurIPS, ICML, and ICLR.