Act II: Horse Tracks and Horse Races
The Effect that Heterogeneous vs. Constant Treatment Effects, Extrapolation vs. Interpolation, Balanced vs. Imbalanced Covariates on Matching and Regression
Act II: Heterogeneous Treatment Effects, Common Support, and Linear Extrapolation
Prologue
In Act I of this 3-part series, I introduced the series by talking about unconfoundedness and how it means both “no unknown confounders” and “treatment assignment is random conditional on known and quantified confounders”. The first part, I said, is almost exclusively what people associate with it, because most people who learn about unconfoundedness learn about it under a different name called “exogeneity” and just assume unconfoundedness is referring to solving the omitted variable bias problem in regression equations by including the correct covariates.
But in fact, I said, in the Rubin tradition of potential outcomes, it is a bit more than that, and that historically, people would refer both to “unconfoundedness” as more like the idea of exogeneity, and “ignorable treatment assignment” or just “ignobility” as referring to that more decision rule of “people choosing to do things without regar…
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.