时 间：2026年10月9日（周五）10：00-11：00
地 点：普陀校区理科大楼A1514室
报告人：杨玥含 中央财经大学教授
主持人：刘玉坤 华东师范大学统计学院
摘 要：
Factorial experiments are widely used in economics and business to study multiple treatments, but many outcomes are recorded on ordered categorical scales. Classical average factorial effects are difficult to interpret for ordinal outcomes because the estimands depend on the scores assigned to ordered categories. We instead define probability-based main and interaction effects in randomized factorial experiments that compare potential outcomes through their ordering. These estimands depend on the unobserved joint distribution of potential outcomes, which makes identification challenging. Using only observable marginal distributions, we derive sharp bounds for main effects without parametric or distributional assumptions. Interaction effects involve the joint influence of multiple factors, and the treatment combinations associated with different factors overlap, introducing additional identification difficulties. We provide valid bounds accordingly. The proposed theory allows for an arbitrary number of ordered categories and an arbitrary number of factors. We further incorporate a monotonicity assumption into the framework, tightening the identification regions. We construct nonparametric bootstrap confidence intervals for all identification regions, and illustrate the methods with a field experiment on education subsidies and an HIV curriculum in Kenya and a clinical trial on residual cancer burden.
报告人简介：
杨玥含，中央财经大学统计与数学学院教授，博导，北京大学博士。中央财经大学青年英才、龙马学者青年学者。主要从事因果推断、迁移学习、复杂数据分析等研究。在JASA、Biometrika、JBES、《中国科学：数学》等国内外期刊发表论文50余篇。