[R] model selection criteria in "regsubsets"

[Thomas Lumley]

The calculations are in summary.regsubsets.

Sending three copies of questions like this does not increase the chance 
of a response.

 	-thomas


On Thu, 21 Jun 2007, nlei at sfu.ca wrote:

> Hi All,
> 
> I used "regsubsets" in package "leaps" to do the model subset selection.
> I noticed the "bic" and "cp" criterias are both included in this
> function. But seems like they are not calculated by
> 
> "bic=-n*log(RSS/n)-(p+1)*log(n)" 
> and 
> "cp=(RSS/sigma_hat^2)-(n-2*p-2)" 
> 
> Could you please let me know what formula used for these two criterias?
> 
> Thank you !
> 
> Linda
>

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle