Referat pt. 鈥淩obust estimation and inference for panels with attrition and refreshment鈥� (joint work with Jinyong Hahn, Pierre Hoonhout, Arie Kapteyn, and Geert Ridder) wyg艂osi zaproszony prof. Grigory Franguridi () z Center for Economic and Social Research at University of Southern California.
Abstract: It has long been established that if a panel dataset suffers from attrition, auxiliary (refreshment) sampling may restore full identification under semiparametric assumptions on the attrition process that are significantly weaker than selection on observables. Despite their generality, these identification strategies have not gained wide applicability in empirical research due to the complexity of induced estimation and inference procedures. We show that this is nevertheless possible and suggest a computationally attractive two-step estimator employing the iterative proportional fitting (raking) algorithm. We derive the influence function of this estimator and show that it depends on a nonparametric additive regression, which can be computed efficiently using the well-studied smooth backfitting algorithm. This further leads to a Neyman-orthogonal (debiased) estimator and a procedure for asymptotic inference. Finally, we demonstrate the satisfactory performance of our methodology in simulations and provide an empirical illustration using the Understanding America Study panel.