The kernel estimator fits a local mean at each point x and thus cannot even estimate a line without bias (Cleveland, Cleveland, Devlin and Grosse 1988). An estimator based on locally-weighted ...