R devel - Jul 2004 - Guidelines for writing model term functions (?)

Are there any general guidelines for writing functions that appear in 
model formulas for functions like lm()/glm()?  The ones I can think of 
are ns()/bs() in `splines' and s() in `mgcv'.  It seems that ns()/bs() 
basically create matrix with a number of attributes and a class, while 
s() returns a list of parameters.  It seems that gam() in `mgcv' has 
its own way of dealing with formulas so maybe that's not a great 
example, but are there any suggestions for writing a function that 
might appear in the formula for, say, lm() or glm()?

Thanks,

-roger

-- 
Roger D. Peng
http://www.biostat.jhsph.edu/~rpeng/

Well, there are examples like log, too, and poly() is a good one.

poly, ns and bs are special because they participate in the safe 
prediction scheme (?SafePrediction). If you don't need that, just return a 
vector or matrix (whose columns will be treated as a set of vectors).


On Fri, 2 Jul 2004, Roger D. Peng wrote:
> Are there any general guidelines for writing functions that appear in 
> model formulas for functions like lm()/glm()?  The ones I can think of 
> are ns()/bs() in `splines' and s() in `mgcv'.  It seems that
ns()/bs()
> basically create matrix with a number of attributes and a class, while 
> s() returns a list of parameters.  It seems that gam() in `mgcv' has 
> its own way of dealing with formulas so maybe that's not a great 
> example, but are there any suggestions for writing a function that 
> might appear in the formula for, say, lm() or glm()?
-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595