| 234 | 1 | 65 |
| 下载次数 | 被引频次 | 阅读次数 |
倾向值方法 （加权或者匹配）在社会科学量化研究中得到越来越多的应用，但是经由倾向值方法处理的样本并不必然能够达成理想的混淆因素平衡性。混淆因素的不平衡性问题可以从理论与操作层面进行辨析。从理论上讲，传统倾向值方法依据的是等比例误差削减分析框架。这个框架虽然有其吸引力，但背后有一系列难以满足的假设条件。正因如此，倾向值方法有时无法很好地平衡混淆因素。与之相比，一个更加适配社会科学经验研究的倾向值分析框架是单调性不平衡划界框架。在操作层面上，与单调性不平衡划界分析框架一致，有三种新兴的分析方法 （粗粒度精确匹配、熵平衡法与混淆因素平衡倾向值法）可以确保混淆因素在实验组与控制组之间的平衡。
Abstract：The propensity score method(whether through weighting or matching)is increasingly applied in quantitative research in the social sciences. However,samples processed using the propensity score method do not necessarily achieve ideal covariate balance. The problem of covariate imbalance can be analyzed from both theoretical and practical perspectives. Theoretically,traditional propensity score methods are based on the proportional reduction of error framework. While this framework has its appeal,it relies on a series of assumptions that are often difficult to meet. Consequently,propensity score methods sometimes fail to adequately balance covariates. In contrast,a more suitable framework for propensity score analysis in social science research is the monotonic imbalance bounding framework.On the practical side,consistent with the monotonic imbalance bounding framework,three emerging analytical methods—coarse exact matching,entropy balancing,and covariate-balancing propensity scores—can ensure covariate balance between treatment and control groups. The methodological advantages of these approaches are demonstrated through two empirical examples.
1.胡安宁：《教育能否让我们更健康--基于2010年中国综合社会调查的城乡比较分析》，《中国社会科学》2014年第5期。
2.胡安宁：《倾向值匹配与因果推论：方法论述评》，《社会学研究》2012年第1期。
3.胡安宁：《统计模型的“不确定性”问题：与倾向值方法》，《社会》2017年第1期。
4.Abadie,A.,Diamond,A.&Hainmueller,J.，“Comparative Politics and the Synthetic Control Method”,American Journal of Political Science,2015,59(2):495-510.
5.Angrist,J.D.，“Lifetime Earnings and the Vietnam Era Draft Lottery:Evidence from Social Security Administrative Records”,The American Economic Review,1990,80(3):313-336.
6.Bernard,K.,Sainburg,T.,Geraldo Bastías,Jiang,S.,Yizhou Sun&Jacob G.Foster.，“A Primer on Deep Learning for Causal Inference”,Sociological Methods&Research,2024,00491241241234866.
7.Black,B.S.,Lalkiya,P.&Lerner,J.Y.，“The Trouble with Coarsened Exact Matching”,Northwestern Law&Econ Research Paper Forthcoming,Available at SSRN:https：//ssrn.com/abstract=3694749 or http：//dx.doi.org/10.2139/ssrn.3694749.
8.Fong,C.,Hazlett,C.&Imai,K.，“Covariate Balancing Propensity Score for a Continuous Treatment:Application to the Efficacy of Political Advertisements”,The Annals of Applied Statistics,2018,12:156-177.
9.Hainmueller J.，“Entropy Balancing for Causal Effects:A Multivariate Reweighting Method to Produce Balanced Samples in Observational Studies”,Political Analysis,2012,20(1):25-46.
10.Ho,D.E.,Imai,K.,King,G.&Stuart,E.A.，“Matching as Nonparametric Preprocessing for Reducing Model Dependence in Parametric Causal Inference”,Political Analysis,2007,15:199-236.
11.Iacus,S.,King,G.&Porro,G.，“Causal Inference without Balance Checking:Coarsened Exact Matching”,Political Analysis,2012,20:1-24.
12.Iacus,S.,King,G.&Porro,G.，“Multivariate Matching Methods That Are Monotonic Imbalance Bounding”,Journal of the American Statistical Association,2011,106:345-361.
13.Imai,K.&Ratkovic,M.，“Covariate Balancing Propensity Score”,Journal of the Royal Statistical Society:Series B(Statistical Methodology),2014,76:243-263.
14.Kennedy,Edward H.，“Semiparametric Theory and Empirical Processes in Causal Inference”,Statistical Causal Inferences and Their Applications in Public Health Research,Edited by Hua He,Pan Wu,and Ding-Geng(Din)Chen,NY:Springer,2016.
15.Koch,B.J.,Sainburg,T.,Geraldo B.P.,Jiang,S.,Yizhou,S.&Foster,J.G.，“A Primer on Deep Learning for Causal Inference”,Sociological Methods&Research,2025,54(2):397-447.
16.Lalonde,R.J.，“Evaluating the Econometric Evaluations of Training Programs with Experimental Data”,The American Economic Review,1986,76:604-620.
17.Lee,B.K.,Lessler,J.,&Stuart,E.A.，“Improving Propensity Score Weighting Using Machine Learning”,Statistics in Medicine,2010,29(3):337-346.
18.Mátyás,László，ed.,Generalized Method of Moments Estimation,Cambridge:Cambridge University Press,1999.
19.Naimi,A.I.,Moodie,E.E.,Auger,N.,&Kaufman,J.S.，“Constructing Inverse Probability Weights for Continuous Exposures:A Comparison of Methods”,Epidemiology,2014,25:292-299.
20.Raftery,A.E.，“Statistics In Sociology,1950-2000:A Selective Review”,Sociological Methodology,2001,31(1):1-45.
21.Rajeev,D.&Wahba,S.，“Causal Effects in Non-Experimental Studies:Re-Evaluating the Evaluation of Training Programs”,Journal of the American Statistical Association,1999,94(448):1053-1062.
22.Reshetnyak,E.,Systematic Evaluation and Comparison of Entropy Balancing and Covariate Balancing Propensity Score Methods(Doctoral dissertation),Fordham University,2017.
23.Rubin,D.B.&Thomas,N.，“Affinely Invariant Matching Methods with Ellipsoidal Distributions”,Annals of Statistics,1992,20:1079-1093.
24.Rubin,D.B.，“Multivariate Matching Methods that are Equal Percent Bias Reducing,I:Some Examples”,Biometrics,1976a,32:109-120.
25.Rubin,D.B.，“Multivariate Matching Methods that are Equal Percent Bias Reducing,II:Maximums on Bias Reduction for Fixed Sampled sizes”,Biometrics,1976b,32:121-132.
(1)所谓混淆因素，是指为处理变量和结果变量之间关系带来混淆效应的变量。例如，混淆是否上大学和收入之间关系的一个混淆因素是个人的能力。
(1)通过加权，实验组和控制组中那些彼此相似的个体获取更大的权重，因此经过加权后的数据构成了一个伪总体，其中实验组和控制组之间的相似度（或者说平衡性）得以加强。与之相比，通过匹配，实验组和控制组中彼此相似的个体被抽离出来形成一个子样本。和加权的结果近似，这个子样本中实验组和控制组的相似度也得到了加强。
(1)例如，在实验组中的一个个体的倾向值得分为0.6，我们以0.05为卡尺，则控制组中所有倾向值取值为0.55~0.65的个体都可以用来进行匹配。显然，卡尺设置得越大，我们在实验组中能够用来匹配的人数就越多，但弊端是，相应的匹配效果（即混淆因素在实验组和控制组之间的平衡性）就会越差。
(2)当然，我们在不断放宽卡尺的同时，也可以对不平衡性有更高的要求。此时作为参数的卡尺大小与不平衡性上限之间呈现反向单调关联。
(1)随着混淆因素的增多，这样的交互分类就会是多维的，但无论维度如何，特定单元格内部的个体在已经考虑到的混淆因素上还是很相像的。
(1)如果一开始涉及的混淆变量涉及分类变量（比如性别或者地区），这时候做精确匹配是一个非常直接的精确匹配过程。比如，东部地区男性和东部地区的男性匹配。
基本信息:
DOI：10.14167/j.zjss.2025.06.015
中图分类号:C91-03
引用信息:
[1]胡安宁,袁野.提升混淆因素的平衡性：倾向值方法的新进展[J].浙江社会科学,2025,No.346(06):58-71+85+158.DOI:10.14167/j.zjss.2025.06.015.
2025-06-15
2025-06-15