[Rd] standard errors from glm (PR#3180)

Peter Dalgaard BSA p.dalgaard at biostat.ku.dk
Wed Jun 4 11:24:27 MEST 2003


Prof Brian Ripley <ripley at stats.ox.ac.uk> writes:

> On Tue, 3 Jun 2003, Thomas Lumley wrote:
> 
> > On Wed, 4 Jun 2003 mcneney at db.math.sfu.ca wrote:
> > 
> > >
> > > The QR decomposition being returned by glm.fit uses the fitted values from
> > > parameter estimates at the (final-1)st iteration. But don't we want
> > > fitted values from parameter estimates at the final iteration?
> > > I tried inserting the following lines upon completion of the iterative
> > > weighted least squares loop:
> > >
> > 
> > Well, yes, perhaps.  OTOH it's a whole lot easier just to tighten the
> > tolerance for convergence, which has been done for 1.7.1
> 
> For 1.8.0, in fact.

Hmm. Would we want that for 1.7.1 too? There is some use in having
output unchanged in the patch series, but not if it is (too)
inaccurate, I suppose?

BTW, are we sure that tightening convergence criteria also works in
the divergent case? Probably not. If the parameter estimate doubles on
each iteration, the SEs could be seriously out of whack if they refer
to the parameters from last time around (Hauck-Donner effect
notwithstanding).

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907



More information about the R-devel mailing list