I show how to identify and estimate the average partial effect of explanatory variables in a model where unobserved heterogeneity interacts with the explanatory variables and may be unconditionally correlated with the explanatory variables. To identify the populationaveragedeffects, I use extensions of ignorability assumptions that are used for estimating linear models with additive heterogeneity and forestimating average treatment effects. New stimators are obtained for estimating the unconditional average partial effect as well as the average partial effect conditional on functions of observed covariates.
Estimating average partial effects under conditional moment independence assumptions
1 March 2004
Working Paper (CWP03/04)