This note proposes a new decomposition of average treatment effects on nonnegative outcomes. It represents the total effect as a population-weighted sum of the effects for two groups: those induced to participate by the treatment and those participating regardless of it. The usual decomposition into extensive and intensive margins used in the literature is generally incompatible with such a causal interpretation. The difference between decompositions can be substantial and yield diametrically opposed results.
This content is only available as a PDF.
© 2014 The President and Fellows of Harvard College and the Massachusetts Institute of Technology