学术讲座
时间：2015年4月8日（星期三）下午3：00-5：00
地点：心理学院201室
题目：Propensity Score Methods and Its Application in
Social and Medical Research
（倾向性得分方法及其在社会科学和医学研究中的应用）
主讲人： 王晓非 副教授
（美国杜克大学医学院 生物统计和生物信息学系）
报告人简介：
王晓非副教授，早年于华东师范大学心理系获学士学位（1990），1993年于北京大学心理系获硕士学位并留校任教，1994年赴美国北卡罗琳娜大学（University of North Carolina
at Chapel Hill） 攻读博士学位，2003年获生物统计学博士学位，之后至今先后任职于杜克大学（Duke University）医学院生物统计和生物信息学系（Department of Biostatistics and Bioinformatics） 助理教授、副教授。同时，目前他还是杜克大学癌症研究所的生物统计学家、肿瘤学联盟临床试验委员会（Alliance for Clinical Trials in Oncology Group）的高级统计学家、统计学和数据中心联盟（Alliance Statistics and Data Center）的统计学副主任。他主要的研究兴趣是癌症临床试验和转化研究的研究设计和分析工作，是多项NIH（美国国家健康研究院）研究项目的主持人，已在相关领域的国际重要学术期刊发表学术论文30余篇。
讲座内容提要：
Propensity Score Methods and Its Application in Social and
Medical Research
Observational (nonrandomized) studies are increasingly being used to evaluate treatment effect in social and medical research. Because of the bias associated with treatment selection and the imbalance of baseline covariates between treatment groups in observational, standard statistical methods may yield biased statistical inference on treatment effect. In this talk, we review statistical methods that have been proposed to address treatment selection bias in social and medical research.These methods include multivariable regression and four propensity score based methods. The propensity score is the probability of treatment assignment conditional on observed baseline characteristics. Under certain assumptions, the propensity score allows one to analyze an observational (nonrandomized) study so that it mimics some features of a randomized controlled trial. Four propensity methods are reviewed,including matching on the propensity score, stratification on the propensity score, inverse probability of treatment weighting using the propensity score, and covariate adjustment using the propensity score. We illustrate these methods with a lung cancer surgery study, in which long term and short term clinical outcomes between two surgical procedures - VATS and Open lobectomy – are compared. In the end, we discuss several new advances in treatment effect estimation with observational studies, including a moment-based empirical likelihood method, which does not require a specification of propensity score model, and the instrumental variables method, which does not require all covariates affecting treatment selection are measured.
（倾向性得分方法简要说明：当前，人文社会科学领域中一个重要的方法论突破就在于开始关注如何通过严格的统计技术进行因果推论，因果推论背景下的倾向性得分方法是近年在国外发展起来的重要因果分析方法，是排除背景变量混杂干扰，通过模拟随机抽样，使用抽样调查数据计算干预效果的有效方法。并提供三篇相关论文参考。）