[S] summary: extrapolating natural splines

Pierre Joyet (pierre.joyet@bluewin.ch)
Sun, 15 Feb 1998 19:50:58 +0100


Some days ago, I asked how to extrapolate "naturally" (i.e., with straight
lines) natural splines. The problem is that predict and predict.gam use
extreme knots that depend on the new data instead of taking the extreme
knots used in the fitting procedure. I am using S-Plus 3.2 for Windows.

I wish to thank Tim Hesterberg, Frank E. Harrell Jr and Trevor Hastie who
answered and solved my problem.

Tim Hesterberg mentioned code put by Trevor Hastie and Doug Bates on the S
archive on statlib,
in the safe.predict and splines archives respectively, 'that offers
improvement'. The functions of Trevor Hastie allow to specify the extreme
knots explicitly, which is exactly what I needed. Tim thinks that 'the key
functionality of Bates code has been incorporated into recent versions of
S-PLUS, as has some of Hastie's code'.
With the Design library of Frank Harrell (to be found on statlib), 'you can
use the rcs
(restricted cubic spline) function inside a formula, and all predictions
with Design are "safe" as knots are "remembered" with the fit'.

Pierre Joyet
Basle, Switzerland

-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news