Late Again with Defiers

Pré-publication, Document de travail: We show that the Wald statistic still identifies a causal effect if instrument monotonicity is replaced by a weaker condition, which states that the potential propensities to be treated with or without the instrument should have the same distribution, conditional on potential outcomes. This holds for instance if the slippages between these potential propensities and the average propensity are independent of potential outcomes. In this framework, the Wald statistic identifies a LATE on a population which comprises both compliers and always takers.

Auteur(s)

Clément de Chaisemartin, Xavier d’Haultfoeuille

Date de publication
  • 2012
Mots-clés
  • Instrumental Variables
  • Monotonicity
  • Defiers
  • Rank Invariance
  • Exchangeable random variables
Référence interne
  • PSE Working Papers n°2012-23
Version
  • 1