I have posted an update to the GAM package. Note that this package 
implements gam() as described
in the "White" S book (Statistical models in S). In particular, you can 
fit models with lo() terms (local regression)
and/or s() terms (smoothing splines), mixed in, of course, with any 
terms appropriate for glms.

A number of bugs in version 0.92 have been fixed; notably

1) some problems with predict and newdata
2) plot.gam now works with any model for which predict( ..., 
type="terms") is appropriate
   (well, at least several). Examples are lm, glm, gam and coxph models.

  So for example, if you have fit a Cox model

	cox1 <- coxph( Surv(Survival, death) ~ Grade + ns(Age,4) + ns(Size,4))

  Then plot.gam(cox1, se=T) will produce three plots, one for each term 
in the model, with standard error bands.

3) I have implemented the fast versions of backfitting for models 
consisting of all local regression
     terms (lo.wam) or all smoothing spline terms (s.wam).

Please let me know of any problems with the gam package

Trevor Hastie

-------------------------------------------------------------------
  Trevor Hastie                                   hastie@stanford.edu 
  Professor, Department of Statistics, Stanford University
  Phone: (650) 725-2231 (Statistics)          Fax: (650) 725-8977 
  (650) 498-5233 (Biostatistics)   Fax: (650) 725-6951
  URL: http://www-stat.stanford.edu/~hastie 
    address: room 104, Department of Statistics, Sequoia Hall
           390 Serra Mall, Stanford University, CA 94305-4065 
  --------------------------------------------------------------------

	[[alternative text/enriched version deleted]]