[R] Inf in regressions

[Prof Brian Ripley]

Actually, R does not accept NA.  First, the action is from lm and not from 
the R base, and second, NAs are in fact deleted and not accepted.

Function lm() has a na.action argument which is by default na.omit.  You 
could easily make na_and_inf.omit and use it. You could also use
subset=(w > 0). The help pages are your friends here.

On Tue, 25 Oct 2005, Dimitri Szerman wrote:

> Suppose I I wish to run
>
> lm( y ~ x + z + log(w) )
>
> where w assumes non-negative values. A problem arises when w=0, as log(0)
> = -Inf, and R doesn't accept that (as it "accepts" NA). Is there a way to
> tell R to do with -Inf the same it does with NA, i.e, to ignore it? (
> Otherwise I have to do something like
>
> w[w==0] <- NA
>
> which doesn't hurt, but might be a bit incovenient sometimes.)

-- 
Brian D. Ripley,                  ripley at 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