[Rd] Bug in glm.fit? (PR#1331)

[ripley@stats.ox.ac.uk]

On Fri, 5 Apr 2002 ripley@stats.ox.ac.uk wrote:

> Berwin,
>
> Thank you for this.
>
> Unfortunately, although the analysis is mainly correct, but the solution
> has almost as many problems as the current state.  Suppose pivoting
> occurs. Then for the internal code, coef needs to contain zeroes at the
> omitted cols. But you have
>
> coef <- coefold <- start  # initialise coef here in order to
>                           # be able to subset it below
>
> so it will contain whatever start had for those cols.  Further, the test

Sorry, that's wrong as the innards of dqrls (which I have now read) will
do the zeroing.

[...]

However, there's more:

> >             while (!(valideta(eta) && validmu(mu))) {
> >                 if (ii > control$maxit)
> >                   stop("inner loop 2; can't correct step size")
> >                 ii <- ii + 1
> >                 coef <- (coef + coefold)/2
> >                 mu <- linkinv(eta <- eta + offset)
> >             }

How can that loop work?  eta is just incremented by the offset.

-- 
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
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._