The CATT SATT on the MATT: semiparametric inference for sample treatment effects on the treated
8 February 2024
Studies sample variants of average treatment effect on treated
Derives distributions and variance estimators for each variant
Sample ATT can have larger variance than population ATT
Proposes 'mixed' ATT to address this issue
Also studies weighting all individuals by treatment propensity
Simplified treatment effect on the treated
This paper studies modified versions of a common causal estimand, the average treatment effect on the treated (ATT). It finds that replacing population parameters with sample versions changes statistical properties in complex ways. A key finding is that the common sample ATT can be estimated with more uncertainty than the original ATT, motivating an improved variant. The analysis also introduces weighting individuals by likelihood of treatment receipt.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper