Measuring Sources of Identification in Nonlinear Econometric Models

We propose a measure of the extent to which a given parameter is identified by a given empirical moment. The measure has simple interpretations in terms of decision theory and sensitivity to model misspecification. It is applicable to a wide class of nonlinear models, and imposes a negligible computational burden even for computationally difficult models. We validate the measure on a linear example and apply it to a recent paper in industrial organization.