bug in glm (PR#397)

[ripley@stats.ox.ac.uk]

On Mon, 10 Jan 2000, Peter Holzer wrote:

> Prof Brian Ripley writes:
>  > > Date: Mon, 10 Jan 2000 11:53:50 +0100 (MET)
>  > > From: holzer@stat.math.ethz.ch
>  > > 
>  > > Dear R-team
>  > > 
>  > > As I didn't get any answer to my bug-report last week I have taken the
>  > > effort and extracted a minimal data set from my data (see below) where the
>  > > following bug occurs:
>  > > 
>  > > > glm(SKR.ein.aus ~ ., family = binomial, data = bugdata, na.action = na.omit)
>  > > Error in names<-.default(*tmp*, value = ynames) : names attribute must be the 
>  > same length as the vector
> > Indeed, thank you (we needed an example). Alter
>  > 
>  >     names(w) <- ynames
>  >     
>  > to
>  >     names(w) <- ynames[good]
>  > 
>  > and
>  > 
>  >     names(fit$effects) <-
>  > 	c(xxnames[seq(fit$rank)], rep("", nobs - fit$rank))
>  > 
>  > to 
>  >     names(fit$effects) <-
>  > 	c(xxnames[seq(fit$rank)], rep("", sum(good) - fit$rank))
>  > 
>  > as the vector w is dropping observations with fitted probs 0 or 1.
> 
> This is actually a solution I thought of as well. However it has the
> disadvantage that plot (and other functions?) gets problems:
> 
> > fit.tst <- glm(SKR.ein.aus ~ ., family = binomial, data = bugdata)
> Warning messages: 
> 1: fitted probabilities of 0 or 1 occurred in: (if (is.empty.model(mt))
> glm.fit.null else glm.fit)(x = X, y = Y, 
> .
> .
> .
> 7: Algorithm did not converge in: (if (is.empty.model(mt)) glm.fit.null else glm.fit)(x = X, y = Y,  
> > plot(fit.tst)

We already know that plot.lm does not work correctly with glm objects with
zero weights.  I am not at all sure it should be used for glm objects
(except Gaussian ones).  However, I think you missed Martin M's point.
This behaviour indicates an inappropriate model, one with separation
of the groups, and all the linear approximations needed for a local lm
model are inappropriate.

We also know that the whole glm area needs a long hard look.

-- 
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 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._