Act I: Matching vs Regression: In Defense of the ATT
Behavioral implications of unconfoundedness and exogeneity
Prologue
Imagine a racetrack where matching and regression are two thoroughbreds racing one another in a battle for causal inference supremacy. Both powerful contenders have their detractors and advocates, but since each appears unbiased in similar situations, why should either consistently outperform the other? This is what today’s substack essay is about.
My dive into this horserace between matching and regression began earlier this semester while rewriting the Mixtape for Yale University Press—an unexpectedly challenging but exhilarating process. As some of my beliefs about causal inference, as a topic and field, have changed since first writing the book in 2018 and 2020, the rewrite has pushed me to rethink the book’s organization, the topics to include and exclude, each chapter’s framing, and my own pedagogical approach. When I began focusing on my selection on observables chapter (previously entitled “Subclassification and Matching”), my dissatisfaction with the chapter left me…
Keep reading with a 7-day free trial
Subscribe to Scott's Substack to keep reading this post and get 7 days of free access to the full post archives.