Flexible Parsimonious Smoothing and Additive Modeling

Friedman and Silverman describe and demonstrate a new technique for smoothing and fitting additive models. They use regression splines, and select the locations of the knots using subset regression techniques. Two attractive features of their method are 1) By using special piecewise linear basis functions, they can achieve the subset selection computations in in O(n) operations. 2) In the context of additive models, their technique performs variable selection and the amount of smoothing per variable simultaneously.