A Comparison of Propensity Score Matching Methods in R with the MatchIt Package: A Simulation Study.
Jiaqi Zhang Β· OhioLink ETD Center (Ohio Library and Information Network) Β· 2013
Propensity score matching (πππ) methods are becoming increasingly popular in non-experimental and observational studies to reduce selection bias through balancing measured covariates.This process has been developed into a relatively systematic and scientific branch of matching methods.πππ‘πβπΌπ‘ is a package in the statistical programming software π that allows for matching using several methods, including nearest neighbor, caliper, stratification, and full matching in order to find cases balanced on the propensity score between the treatment and the control group and achieve causal inference.Choosing which of those options to implement can be confusing for researchers.In this present study, these different methods are explained and a simulation study is conducted using example data to illustrate differences in these methods.The generated data is assigned based on a function of observed covariates and randomness, simulating selection bias, and analyzed to examine whether any of five popular propensity score matching methods perform more effectively in balancing covariates and reducing the selection bias within a given sample size.This study shows that each propensity score matching method -Nearest Neighbor (1:1), Nearest Neighbor (2:1), Caliper, Stratification, and Full matching methods -performs well in matching, and they all provide strong evidence to make casual inferences.This is particularly true for Caliper and Full matching.π code, detailed results and suggestions for future study are also provided.