Reducing bias in Observational Studies: An Empirical Comparison of Propensity Score Matching Methods

Abstract

This paper aims to compare the performance of four widely used propensity score matching (PSM) methods, namely; Nearest neighbor matching, Caliper matching, Mahalanobis metric matching including the propensity score, and Stratification matching, in terms of bias reduction on observational data from which the treatment effects are intended to be assessed. Material and Methods: The selection bias, standardized bias and percent bias reduction are evaluated for each of the PSM methods using empirical data drawn from the Nigeria Demographic Health Survey of 2013. Factors that are associated with Ideal family size determination were extracted. The women were then divided into two groups: those who have at least a secondary school education, subsequently regarded as 'treated' group, and those who have no form of formal education, regarded as 'control group. Results: The balance metrics adopted showed a high level of imbalance between the two groups of interest for the unmatched data. Caliper matching was shown to have outperformed the other three methods in the task of bias reduction and achieving balance between the two treatment groups. Conclusion: Results from this study can help medical and health researchers to choose appropriate propensity matching methods to estimate treatment effect in the presence of confounding variables. (English) [ABSTRACT FROM AUTHOR]