Computing Resources for Nonparametric Regression

Most general statistical computer packages (e.g., Stata, SPSS) now incorporate some facilities for scatterplot smoothing, usually some version of lowess -- that is, local polynomial regression with nearest-neighbor windows.

• The most recent version of SAS (Version 8) has new procedures for local polynomial multiple regression (proc loess) and for smoothing splines (proc tpspline); see <http://www.sas.com/rnd/app/da/new/danpm.html>.
• The statistical programming language S has extensive facilities for nonparametric regression. S-Plus is the standard commericial implementation of S, described at <http://www.splus.mathsoft.com/>.
• R is a free (open-source) implementation of the S language; it contains most of the nonparametric-regression capabilities of S-Plus, with the notable exception of Hastie and Tibshirani's implementation of gam (generalized additive models -- but see below). R (along with add-on packages)  is available through the Comprehensive R Archive Network, at <http://cran.r-project.org/> and several mirror sites. Also see the R home page, at <http://www.R-project.org/>. R provides an attractive means of trying (and using) nonparametric-regression methods. The price is right, and R is available for Unix/Linux, Windows, and Macintosh computers. The software is also of very high quality, and the nonparametric-regression procedures available in R are extensive:
• Some nonparametric-regression facilities are in the base R package.
• The loess function is in the modreg ("modern regression") package, which comes with the standard R distribution.
• Other notable nonparametric regression facilities are available in the add-on packages locfit (associated with C. Loader, Local Regression and Likelihood, New York: Springer, 1999); sm (A.W. Bowman and A. Azzalini, Applied Smoothing Techniques for Data Analysis: The Kernel Approach with S-Plus Illustrations, Oxford, Oxford University Press, 1997); and tree (classification and regression trees -- also see the maptree and rpart packages) . Smoothing-spline implementation of generalized additive models (and more) are available in the mgcv and gss packages.
• The XploRe program, associated with Wolfgang Haerdle (an important contributer to nonparametric-regression methods), also has extensive nonparametric-regression capabilities. Information is available at <http://www.xplore-stat.de/>.