Journal article
We consider the identification of the average treatment effect in models with continuous endogenous variables whose impact is heterogeneous and derive a testable restriction that allows us to assess the degree of unobserved heterogeneity.