统计与数据科学系系列学术报告之四百六十二期

时间：2024年12月25日（星期三）10:30-11:30

地点：史带楼301室

主持人：复旦大学管理学院统计与数据科学系郁文教授

报告人：Prof. Yunxiao Chen Department of Statistics London School of Economics and Political Science

题目：Pairwise Comparisons without Stochastic Transitivity: Model, Theory and Applications

摘要：Most statistical models for pairwise comparisons, including the Bradley-Terry and Thurstone models and many extensions of them, make a relatively strong assumption of stochastic transitivity. This assumption imposes that, if the probability of player A winning B and that of B winning C are both greater than 0.5, then the probability of player A winning player is also greater than 0.5, and this transitive relationship extends to any finite set of players. However, in many real-world scenarios of pairwise comparisons, especially games involving multiple skills or strategies, the stochastic transitivity assumption does not hold, in which case models relying on this assumption can have suboptimal predictive performance. In this talk, we propose a general family of statistical models for pairwise comparison data without a stochastic transitivity assumption, which includes the BTL and Thurstone models as special cases. In this model, the pairwise probabilities are determined by an approximately low-dimensional skew-symmetric matrix. Likelihood-based estimation methods and computational algorithms are developed, which allow for sparse data with only a small proportion of observed pairs. The power of the proposed method is shown via simulation studies and real data examples. This is a joint work with Sze Ming Lee (phd student at LSE).

个人简介：Dr. Yunxiao Chen is an associate professor of statistics in the Department of Statistics at the London School of Economics and Political Science. His research focuses on statistical models and theory for complex multivariate data, with applications in education, psychology, and other social science disciplines. Dr. Chen serves as an associate editor for Psychometrika, the British Journal of Mathematical and Statistical Psychology and Psychological Methods, and as an editorial board member for the Journal of Educational and Behavioral Statistics and Applied Psychological Measurement. He was a Spencer Foundation/NAEd Postdoctoral fellow at the United States National Academy of Education and received the 2018 Brenda H. Lloyd Dissertation Award from the National Council on Measurement in Education and the 2022 Early Career Award from the Psychometric Society. His work has been published in leading journals in the fields of psychometrics, statistics, and machine learning, including JASA, Biometrika, AOAS, JMLR, and Psychometrika. Dr. Chen earned his PhD from Columbia University in 2016.