Heterogeneity and Causal Moderation

Chapter 7 — Heterogeneity and Causal Moderation Average treatment effects tell you what happened on average—but heterogeneity reveals who was affected, how much, and why. This chapter develops tools for estimating conditional average treatment effects (CATEs) using both classical regression methods and modern causal forests. It distinguishes descriptive heterogeneity analysis from causal moderation—where the goal is to identify the precise causal source of differential effects—and addresses the intensive and extensive margins of treatment response, including bounding approaches and the Tobit model.

1. In the presence of potential moderation of treatment effects through predetermined covariates, researchers can estimate heterogeneous treatment effects by covariates using classical and modern tools.
2. Causal moderation analysis differs from conventional heterogeneity analysis in that the goal is to pinpoint the precise, causal source of the heterogeneity.
3. The limits of ex post moderation analysis highlight the value of considering theory and identifying potential moderators in the design stage.
4. For outcomes with extensive and intensive margins, a fundamental internal validity concern can be alleviated only by invoking stronger assumptions.