[R] glm() for log link and Weibull family

[Prof Brian Ripley]

On Fri, 8 Jun 2007, Robert A. LaBudde wrote:

> I need to be able to run a generalized linear model with a log() link
> and a Weibull family, or something similar to deal with an extreme
> value distribution.

The Weibull with log link is not a GLM, but survreg() in package survival
can fit it, as well as other extreme-value distributions.

> I actually have a large dataset where this is apparently necessary.
> It has to do with recovery of forensic samples from surfaces, where
> as much powder as possible is collected. This apparently causes the
> results to conform to some type of extreme value distribution, so
> Weibull is a reasonable starting point for exploration.
>
> I have tried ('surface' and 'team' are factors)
>
> glm(surfcount ~ surface*team, data=powderd, family=Gamma(link='log'))
>
> but this doesn't quite do the trick. The standardized deviance
> residuals are still curved away from normal at the tails.
>
> Thanks for any info you can give on this nonstandard model.

It's perfectly standard, just not a GLM.

-- 
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