Causal inference under selection on observables in operations management research: Matching methods and synthetic controls

Abstract

The majority of recent empirical papers in operations management (OM) employ observational data to investigate the causal effects of a treatment, such as program or policy adoption. However, as observational data lacks the benefit of random treatment assignment, estimating causal effects poses challenges. In the specific scenario where one can reasonably assume that all confounding factors are observed—referred to as selection on observables—matching methods and synthetic controls can assist researchers to replicate a randomized experiment, the most desirable setting for drawing causal inferences. In this paper, we first present an overview of matching methods and their utilization in the OM literature. Subsequently, we establish the framework and provide pragmatic guidance for propensity score matching and coarsened exact matching, which have garnered considerable attention in recent OM studies. Following this, we conduct a comprehensive simulation study that compares diverse matching algorithms across various scenarios, providing practical insights derived from our findings. Finally, we discuss synthetic controls, a method that offers unique advantages over matching techniques in specific scenarios and is expected to become even more popular in the OM field in the near future. We hope that this paper will serve as a catalyst for promoting a more rigorous application of matching and synthetic control methodologies.