2026年第55期(总第1199期)

演讲主题：Hard Labels, Soft Scores: Bias and Remedies for Machine-Learned Covariates

主讲人：彭景 美国康涅狄格大学商学院运营与信息管理系副教授

主持人：李建斌 供应链管理与系统工程系教授、副院长

活动时间：2026年07月07日（周二）10:30-12:00

活动地址：管院大楼107教室

主讲人简介：

Jing Peng is an Associate Professor and the PhD Coordinator in the Department of Operations and Information Management at the University of Connecticut. He received his PhD from the Wharton School of the University of Pennsylvania.

His research lies at the intersection of econometrics, digital platforms, and human-AI interaction. His research outputs have been published in top business journals including Information Systems Research, Journal of Marketing Research, Management Science, and MIS Quarterly.

He has won multiple best paper awards and received prestigious academic honors, such as the INFORMS Information Systems Society Gordon B. Davis Young Scholar Award and the Sandra A. Slaughter Early Career Award.

活动简介：

Machine learning and artificial intelligence increasingly generate covariates for downstream regressions, offering researchers powerful new tools for empirical analysis. Yet researchers often discretize continuous prediction scores into hard labels without clear guidance, a common practice that can introduce systematic bias. While these approaches expand the scope of feasible analyses, such discretization can systematically attenuate estimates relative to both soft scores and true labels, though attenuation is not automatic and soft scores are not generally consistent either. This paper shows how validation data can correct scale bias through orthogonalization and proposes a weighted least squares hybrid strategy for settings with partially observed true labels.