Hastie and Tibshirani's (1990) monograph on GAMs proposed a method for estimating the matrices R and [R.sub.i], but this algorithm was considered too computationally expensive to be implemented in the S-Plus gam function.
(2003) suggest that in GAMs for time-series data, the standard error bias of GAMs may be avoided by using fully parametric smoothers in place of nonparametric smoothers.