A comparison of regression spline smoothing procedures (Q1424601)

The aim of this article is to compare current regression spline smoothing strategies through investigation of their performance in a large simulation study. For simplicity and conciseness the author restricts attention to a univariate smoothing setting with Gaussian noise and truncated polynomial regression spline basis. A short overview of three main types of approaches to fitting nonparametric regression splines, i.e., stepwise selection, Bayesian selection, and penalised shrinkage is presented. Results of simulations are analysed.