Matching Methods for Large Observational Studies

As technologies have rapidly developed in recent decades, data sets have grown in size and become more accessible for analysis, e.g., electronic health records, medical claims data, educational databases, and social media data. This chapter focuses on size-scalable matching techniques for the common case of optimal pair matching. It is devoted to incorporating other matching techniques when the sample size is large. The chapter discusses several different matching designs – matching with multiple controls, matching without pairing, full matching, and coarsened exact matching. The idea of splitting into subpopulations based on observed covariates is practical and reasonable, but it can restrict the possible matches in undesirable ways. Caliper matching is a common matching technique to improve the balance of a covariate. Several important matching techniques can be adapted for use in sparsified networks. The first class of techniques focuses on excluding the candidate matches that are far apart and do not deserved to be included in the matched data.