This paper studies the identification of partial differences of nonseparable structural functions. The paper considers triangular structures with no more stochastic unobservables than observable outcomes, that exhibit a degree of monotonicity with respect to variation in certain stochastic unobservables. It is shown that, the existence of a set of instrumental values of covariates, over which the stochastic unobservables exhibit local quantile invariance and over which a local order condition holds, defines a model which identifies certain partial differences of structural functions. This result is useful when covariates exhibit discrete variation. The paper also considers the identification of partial derivatives in smooth structures when covariates exhibit continuous variation.
Centre for microdata methods and practice