Abstract: Without correction, kernel density estimates suffer from boundary effects. Many boundary corrections now exist, but almost all those with good theoretical performance allow the corrected estimator to become negative. An exception is provided by some recently proposed sophisticated transformation methodology. In this paper, we propose much simpler nonnegative boundary corrected estimators which are analogues of the wide class of simple, but possibly negative, boundary corrections based on generalized jacknifing.
Key words and phrases: Boundary kernels, generalized jackknifing, positivity, smoothing.