统计与数据科学系系列学术报告之四百七十四期

时间： 2025年7月1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刚博文博士

地点：史带楼301室

报告人：李少博博士 Dr. Shaobo Li University of Kansas

题目：Measuring partial association between mixed data

摘要：Partial association refers to the relationship between variables while adjusting for a set of covariates . When the variables of interests are not all continuous, limited approaches are available to assess partial association. We propose a new framework for studying partial associations between mixed-type data including continuous, binary, ordinal and multinomial, by using a novel unified residual, which is motivated by the surrogate residuals introduced by Liu and Zhang (2018). The idea is to map the residual randomness to the same continuous scale, regardless of the original scales of outcome variables. It applies to virtually all commonly used models for covariate adjustments. We justify that conditional on , are independent to each other if and only if their corresponding unified residual variables are independent. Based on this result, we develop a general measure to quantify the strength of partial association. In the presence of multinomial outcome, the measure ranges from 0 to 1, which purely reflects the strength without direction due to the nominal nature. The measure also permits visualization tools to further identify the shape of association if the variable of interests has orders. Moreover, the measure gives rise to a general procedure for testing the hypothesis of partial independence. Three real data analyses are conducted to demonstrate the utility of our proposed methods. Among these examples, an analysis of college students’ wellbeing data under the COVID-19 phenomenon is of particular interests. Our analysis reveals (i) significant moderation effects (i.e., the difference between partial and marginal associations) of some key risk factors; and (ii) an elevated moderation effect of physical health, loneliness, and accommodation after the onset of COVID-19.

个人简介：Dr. Shaobo Li is an Associate Professor of Business Analytics at the University of Kansas. Dr. Li’s research interests lie broadly in statistical methodologies including high-dimensional quantile and robust mean regressions, discrete data analysis and semiparametric regression. His research also spans business domains including marketing, finance and information systems, in which he is particularly interested in marketing data privacy, corporate bankruptcy and equity premium prediction. Dr. Li’s work has been published in prestigious business and statistics journals such as Marketing Science, Information Systems Research, JASA and JBES.