[R] one more piece of info on AIC

[Peter Ehlers]

On 2012-05-31 07:52, Mark Leeds wrote:
> just one other thing about the AIC issue:
>
> there is a line in glm.fit which is the following:
>
> aic = aic(y, n, mu, weights, dev) + 2 * rank
>
> but I couldn't find the function aic so I couldn't investigate further. It
> looks suspicious though because it seems to me like
> it should be
>
> aic =  -2*likelihood + 2 * rank
>
> if anyone could help me find the aic function it's appreciated.

Have a look at ?family to see that the family object used
in glm.fit() is a list most of whose elements are functions.
The code in glm.fit() has a few lines extracting those functions,
one of which is aic(). The 'gf <- Gamma()' example on the help
page for family is informative as is the annotated code for
glm.fit in the sources.

Peter Ehlers
--