Comparison of Multivariate Matching Methods That Select a Subset of Treatment and Control Observations
*Maria de los Angeles Resa, Columbia University
This paper conducts a Monte Carlo simulation study to evaluate the performance of matching methods that select a subset of treatment and control observations. The methods studied are nearest neighbor matching with propensity score calipers and the more recent methods, optimal matching of an optimally chosen subset and optimal cardinality matching. The main findings are (i) covariate balance is better with cardinality matching since it satisfies balance requirements by construction; (ii) for given levels of covariate balance, the matched samples are larger with cardinality matching than with the other methods; (iii) in terms of distances, optimal subset matching performs best; and (iv) treatment effect estimates from cardinality matching have lower RMSEs, provided strong requirements for balance. In standard practice, a matched sample is considered to be balanced if the absolute differences in means of the covariates across treatment groups are smaller than 0.1 standard deviations. However, the simulation results suggest stronger forms of balance should be pursued to remove systematic biases due to observed covariates when a simple difference in means estimator is used.