Large, Sparse Optimal Matching with R package rebalance

Abstract

A new R package for matching in observational studies, rcbalance, is presented. rcbalance is designed to exploit sparsity among potential treated-control pairings and can conduct matches on a very large scale at low computational cost. Unlike existing packages, it also supports refined covariate balance constraints, which use prioritized lists of nominal co-variates to induce high degrees of balance on the covariates and their interactions, even when it is difficult to find individual pairs that are similar on many covariates. Matching with rcbalance is demonstrated using data from an observational study of right heart catheterization.