We consider the identification of the average treatment effect in models with continuous endogenous variables whose impact is heterogeneous. We derive a testable restriction that allows us to assess the degree of unobserved heterogeneity. Our analysis uses assumptions relating to the Local Instrumental Variables (LIV) approach and the control function approach.
Instrumental variables, local instrumental variables and control functions
Jean-Pierre Florens, James Heckman, Costas Meghir, Edward Vytlacil
5 July 2002
Working Paper (CWP15/02)