Potential Outcome and Directed Acyclic Graph Approaches to Causality: Relevance for Empirical Practice in Economics

Abstract

In this essay I discuss potential outcome and graphical approaches to causality, and their relevance for empirical work in economics. I review some of the work on directed acyclic graphs, including the recent "The Book of Why," ([Pearl and Mackenzie, 2018]). I also discuss the potential outcome framework developed by Rubin and coauthors, building on work by Neyman. I then discuss the relative merits of these approaches for empirical work in economics, focusing on the questions each answer well, and why much of the the work in economics is closer in spirit to the potential outcome framework. * I am grateful for comments

Enhancing Rigor in Quantitative Entrepreneurship Research

“…Second, alternative econometric methods for causal inference are evolving rapidly (e.g., Abadie & Cattaneo, 2018; Imbens & Wooldridge, 2009), largely building on the potential outcomes (PO) framework (Rubin, 1974; 2005), which focuses on the potential outcomes of a unit of interest under alternative states of a cause (i.e., “counterfactuals”; e.g., Imbens, 2019; Imbens & Rubin, 2015; Morgan & Winship, 2007). This rapid development and the multitude of approaches require care in understanding the assumptions of the alternative techniques and in identifying appropriate technique(s) for a particular problem at hand.…”

Reflecting on “A Statistician in Medicine” in 2020

“…Would Hill take sides in the causality debate? Would he lean toward directed acyclic graphs 7 or potential outcomes, 29 or alternatives? 30,31 4.…”

“…Where would Hill fall within the raging discussion and debate around redefining statistical significance? 26,27 How would Hill alter our thinking about replicability and reproducibility in today's science? 28 Would Hill take sides in the causality debate? Would he lean toward directed acyclic graphs 7 or potential outcomes, 29 or alternatives? 30,31 Would Hill appreciate the rise of Bayesian statistics in medical applications? Would the cross‐fertilization of ideas in statistics and computer science and the emergence of the hybrid field of data science excite Hill? Would Hill caution or embrace machine learning techniques in causal inference and decision‐making? Would Hill agree that the union of multiple cultures of modeling, stochastic or algorithmic enriches our discipline? 32 Would Hill applaud the general computational advances that enable inference and prediction using large data sets? What would Hill say about the decline of experimental design in biostatistics and statistics training programs? What does Hill see as the core set of questions one must always ask when starting a medical study in 2020, either observational or experimental?…”

Too LATE for Natural Experiments: A Critique of Local Average Treatment Effects Using the Example of Angrist and Evans (1998)

“…The potential outcomes framework provides ways to think about the defining features of experiments (Imbens and Rubin 2015;Morgan and Winship 2015;Imbens 2019). This framework can easily be extended to natural experiments and instrumental variables (e.g.…”