glm(family=Gamma)

[Ed Jones <ejones@stat.tamucc.edu>]

This is a bit of a minor question, but can anyone explain why Splus
appears to use 1/mu as the link function for the Gamma distribution
rather than -1/mu ? 

I know this is not a big issue, but it seems inconsistent with the
parameterization presented in popular texts on this subject, such as the
book by McCullagh & Nelder, or the more recent book by Venables and
Ripley. Could it cause some confusion in interpreting the Splus glm
coefficients for the Gamma family?

Thanks - ed