Re: [R] how 'stepAIC' selects?

From: Ruben Roa <RRoa_at_fisheries.gov.fk>
Date: Sat 18 Jun 2005 - 23:42:27 EST


> -----Original Message-----
> From: r-help-bounces@stat.math.ethz.ch
> [mailto:r-help-bounces@stat.math.ethz.ch]On Behalf Of Bókony Veronika
> Sent: 18 June 2005 14:00
> To: r-help@stat.math.ethz.ch
> Subject: [R] how 'stepAIC' selects?
>
>
> Dear all,
> Could anyone please tell me how 'step' or 'stepAIC' works? Does it
> simply select the model with the smallest AIC from all the possible
> models? Or does it perform any test eg. whether the decrease
> in "information content" between a model with a given predictor and
> another without it is "significant"?
> Thanks for help!
> VB

As a complement to what Uwe said, you may want to consider that Sakamoto et al., in p. 84, Ch. 4, Remark 2, of "Akaike Information

Criterion Statistics", 1986, KTK Scientific Publishers, write "From the relation
between AIC and the entropy, if the differences of AIC's for MODEL(j) and 
MODEL(k) is larger than 1~2, then the difference is considered to be 
significant". Unfortunately I could not find any further ellaboration of this corollary anywhere in the book, but maybe I didn't look hard enough. Additionally, you may want to work with "evidence ratios" and Akaike weights, as recommended in Burnham and Anderson, "Model Selection and Multimodel Inference", 2002, Springer. I am under the impression that to try to interpret the AIC in terms of sampling-distribution inference theory, such as in signficance tests, is to miss the point.
Ruben

R-help@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 Received on Sun Jun 19 01:50:12 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:32:51 EST