Transporting Causal Effects Across Populations Using Structural Causal Modeling: The Example of Work-From-Home Productivity

Abstract

Transportability is a structural causal modeling approach aimed at “transporting” a causal effect from a randomized experimental study in one population, to a different population, where only observational data are available. It offers a way to overcome the practical constraints in inferring causal relationships, such as endogeneity concerns in observational data and the infeasibility of replicating certain experiments. While transportability holds significant promise for research and practice, it has thus far seldom been implemented in practice, likely due to the lack of practical guidelines for application of transportability theory, or lack of guidance on handling the statistical challenges that might arise. Using a practical problem as an illustration—estimating the effect of telecommuting on worker productivity—we attempt to bridge the theory-practice gap and delineate some challenges faced when putting transportability theory to practice. In this research note, we offer a detailed procedure for transporting a causal effect across different populations; and we discuss some practical considerations for its implementation, including how to conceptualize causal diagrams, determine the feasibility of transport, select an appropriate diagram, and evaluate its credibility. We also discuss the current limitations, challenges, and opportunities for future research on transportability that would make it more amenable for broad practical use.

Keywords: transportability, causal inference, causal diagrams, work-from-home productivity