[R] AIC in bestglm, glm, and lm - why do they differ?

[Bill.Venables at csiro.au]

AIC is only defined up to an additive constant (as is log-likelihood).

It should not surprise you that the values for AIC differ between packages.

The real question is whether the change in AIC when going form one model to anoth is the same.  If not, one is wrong (at least). 

-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On Behalf Of Darren M Gillis
Sent: Friday, 15 October 2010 1:37 PM
To: r-help at r-project.org
Subject: [R] AIC in bestglm, glm, and lm - why do they differ?

I recently found the "bestglm" package while trolling CRAN for a function to
save some keystrokes with simple (k<10) nested models.  I am interested in
using the best of several nested linear models which I have explored using
lm(), glm() and now bestglm().  When I compare the AIC values for the 'best'
candidate model I get the same AIC values with lm() and glm() but a
different AIC (and likelihood) value from bestglm().  Is this the result of
some difference in likelihood calculation that I am missing in reviewing the
documentation and help files? I can provide code if there is interest in
looking into this, otherwise I will continue to assemble my tables the long
way with glm() and lm(), though the options and potential of the bestglm()
package has me very interested.

 

Cheers, Darren Gillis

 

Biological Sciences

University of Manitoba

Winnipeg, MB

 

 


	[[alternative HTML version deleted]]

______________________________________________
R-help at r-project.org 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.