Working Paper: NBER ID: w23826
Authors: Isaiah Andrews; Emily Oster
Abstract: We develop a simple approximation that relates the total external validity bias in randomized trials to (i) bias from selection on observables and (ii) a measure for the role of treatment effect heterogeneity in driving selection into the experimental sample.
Keywords: external validity; treatment effects; randomized trials
JEL Codes: C1; C18; C21
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Volunteers systematically differing from non-volunteers in their treatment effects (C90) | External Validity Bias (C52) |
| Covariance between individual-level treatment effects and weights derived from observable characteristics (C21) | Total External Validity Bias (C52) |
| Ratio of total bias to bias from participation on observables (C46) | External Validity Bias (C52) |
| Greater private information about treatment effects (C21) | External Validity Bias (C52) |
| Participation influenced by unobservables (C92) | Reweighting to match the target population (C83) |
| Degree of private information regarding treatment effects (C90) | External Validity Bias (C52) |
| Average Treatment Effect estimated from a randomized trial (C90) | External Validity Bias (C52) |