Comparing Local Fitting to Other Automatic Smoothers

In a public service enterprise by Breiman and Peters (1991) various automatic smoothers, such as the supersmoother (SSMU), cross-validated smoothing splines (BART), delete-knot regression splines (DKS) and the cross-validated kernel smooth (KERNEL) were compared by simulation on a variety of sample sizes, noise levels and functions. The intention was to give practitioners guidelines when to use which type of smoother. The given work completes those simulations by including the increasingly popular local fitting approach, that was introduced to the statistical literature by Cleveland (1979). Fedorov et al. (1993) have modified the technique in order to take account possible misspecification bias, termed 'optimized moving local regression', and here we use an automated version (by crossvalidation) of it as given in Fedorov et al. (1994). (author's abstract)