The Causal Inference Of Observational Data Based On Propensity Score Matching

Due to the limitations of randomized control trials,causal inference in observational studies has become a popular topic in various subjects in the past three decades,especially in sociology,economics,political science,psychology and medicine.Neyman-Rubin Counterfactual Framework is the fundamental theory of the methods of solving causal inference in observational studies.Here review the content and the trace of its development systematically,and specifically elaborate the essential requirements and important assumptions of the Neyman-Rubin counterfactual framework.Propensity score matching(PSM)was first put forward in 1983 by American statisticians Rosenbaum and Rubin.It is a method that predicting the propensity score of objects of study through Logistic regression or Probit model,matching the units from treatment group with the control group by appropriate matching method to simplify multidimensional covariates to one-dimensional propensity score and estimating the causal inference.PSM is novel and simple so that it is very popular in causal inference of observational studies.After more than 30 years of development,PSM has been widely used in many fields,such as the research on treatment effect of policy in politics and the comparative effectiveness research in the medical field.In this paper,the basic principle and methodology of PSM are presented in detail.It compares PSM with regression discontinuity design(RDD)and Instrumental Variable(IV)which can also be applied in causal inference of observation studies.Here are two analyses,one is in sociology and it is about how much the college education can impact on income.The other one is in medical field and it is about the curative effect analysis of traditional Chinese medicine for hospitalized heart failure patients.It elaborates the implementation steps of PSM and summarizes its limitations,and puts forward the prospect of the future development of PSM.