[R] cv.glm error function

[Prof Brian Ripley]

On Fri, 20 Jul 2007, Rachel Davidson wrote:

> I have a couple quick questions about the use of cv.glm for
> cross-validation.
>
> 1. If we have a Poisson GLM with counts Y~Poisson(mu) and
> ln(mu)=beta0+beta1*x1+..., is the prediction error (delta) that is output
> from cv.glm provided in terms of the counts (y) or the (mu)?

It is of prediction error as measured by the cost function.  Since y and 
mu are on the same scale I wonder if you meant was it on log scale?  If 
so, it is on the response scale.

The default cost function is probably not appropriate for a log-linear 
model.

> 2. Can cv.glm be used for negative binomial models fit using glm.nb? It
> appears to work, but since NB models aren't strictly GLM's I wanted to
> check.

NB models are strictly glms, but glm.nb estimates parameters that glm does 
not, so no.

>
> Thanks!
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>

-- 
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