讲座人：刘诗尧 北京大学

讲座时间：2026年5月13日上午10:30-12:00

讲座地点：北京大学国家发展研究院承泽园校区249教室

讲座题目：To Impute or Not? A Causal Diagram Perspective

内容简介：Missing data are common in social sciences. Empirical researchers often face a dilemma: they must choose between imputing missing values or excluding incomplete observations. Both options have costs. Sometimes, imputation offers no meaningful statistical improvement, while complete-case analysis does—such as when the predictor is not missing at random (Arel-Bundock and Pelc, 2018). In other cases, the situation reverses.

We argue that the decision to impute or not should be treated as a causal identification problem: rather than simply specifying whether the missing-data mechanism is MAR, as previously suggested (Arel-Bundock and Pelc, 2018; Honaker and King, 2010; Lall, 2016; Marbach, 2022; Pepinsky, 2018), researchers should clarify what causes observations to be missing. More importantly, researchers need to identify variables that produce missingness and explain how these variables are causally related to each other and to the missingness pattern.

We propose that researchers incorporate missingness mechanisms into their causal structure using Directed Acyclic Graphs (DAGs). They should also consider whether to impute missing values as part of the overall causal inference process. This approach integrates prior recommendations into a unified framework, resolves inconsistencies in the literature, and extends to generalized linear models. Finally, we demonstrate our approach with an empirical example examining how personality traits affect voter attitudes in Israel (Rivlin-Angert et al., 2025).

主讲人简介：刘诗尧，北京大学国家发展研究院助理教授，主要研究领域包括政治方法论与比较政治行为学，研究成果发表在Journal of Politics, Contemporary Educational Psychology等国际权威期刊，曾获得政治方法学会、国际传播学会等组织的多项国际学术奖项。刘诗尧毕业于美国麻省理工学院与北京大学，分别获政治学博士，经济学学士（政治学、经济学与哲学）、理学学士（统计学双学位）。

组织者：

（国家发展研究院 ）李力行、席天扬、王轩、徐化愚、易君健、黄清扬

（经济学院）刘冲、吴群峰、曹光宇、年永威、施新政

（光华管理学院）张晓波、仇心诚