Many thanks to those who responded to my question about a bug in
predict.glm.
David ("MathSoft") Smith said "it looks like a bug..[and] it appears to be
fixed in S-PLUS 6." Thanks!
Kunio Takezawa had two very helpful replies. First he noted that
predict.gam() called explicitly produces the correct answers without the
warning. This is the approach I'm going to take when those warning messages
prove to be too much of an annoyance. Second, he pointed me to
http://www.stat.auckland.ac.nz/PEOPLE/yee/smartpred/index.shtml ,
a page maintained by Thomas Yee which contains Yee and Hastie's "smart
prediction" code. This set of functions (with both S-PLUS and R versions)
takes care of predicting with data-dependent transformations, like x -
min(x), or with splines and other stuff. Thanks for the tip!
Finally, Gerard ("All Others Bring Data") Jean noted an additional problem
with predict.glm(); it chokes when a data frame has had a new level added to
some factor variable, even if no observation takes on that level. Again
predict.gam() can handle this.
Thanks to all,
Sam Buttrey
buttrey@nps.navy.mil
|