On Thu, 28 Sep 2000, Laffont, Jean-Louis wrote:
>
> Dear S-Plus users,
>
> In the S-Plus 2000 version of predict.glm, the scale parameter defaults to 1
> for binomial and Poisson fits when type="link" or type="response" but not
> for type="terms".
> What is the reason for this?
Here is my guess. (S-PLUS 2000 incorporates a number if fixes formerly
in library MASS.)
The scale parameter is only used if se.fit=T. Standard errors of the
predictions for terms do not make much sense (as the matrix of predictions
depends on some pretty arbitrary assumptions, for one). So althogh Bill
Venables and I supplied a fix for predictions on link or response scale, we
did not consider fixing the standard errors for the terms as we had never
had cause to use it. The fix would be much easier if predict.lm took
a scale parameter, but it did not at the time we revised predict.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 272860 (secr)
Oxford OX1 3TG, UK Fax: +44 1865 272595
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
|