[R] Likelihood of a ridge regression (lm.ridge)?

[Ravi Varadhan]

Joris,

Ridge regression is a type of regularized estimation approach.  The objective function for least-squares, (Y - Xb)^t (Y - Xb) is modified by adding a quadratic penalty, k b^t b.  Because of this the log-likelihood value (sum of squared residuals), for a fixed k, does not have much meaning, and is not really useful.  However, a key issue in such regularized estimation is how to choose the regularization parameter "k".  You can see that lm.ridge gives two different ways to estimate k (there are other ways).

Ravi.

____________________________________________________________________

Ravi Varadhan, Ph.D.
Assistant Professor,
Division of Geriatric Medicine and Gerontology
School of Medicine
Johns Hopkins University

Ph. (410) 502-2619
email: rvaradhan at jhmi.edu


----- Original Message -----
From: joris meys <jorismeys at gmail.com>
Date: Tuesday, March 17, 2009 7:37 pm
Subject: [R] Likelihood of a ridge regression (lm.ridge)?
To: R-help Mailing List <r-help at r-project.org>


> Dear all,
>  
>  I want to get the likelihood (or AIC or BIC) of a ridge regression model
>  using lm.ridge from the MASS library. Yet, I can't really find it. As
>  lm.ridge does not return a standard fit object, it doesn't work with
>  functions like e.g. BIC (nlme package). Is there a way around it? I would
>  calculate it myself, but I'm not sure how to do that for a ridge regression.
>  
>  Thank you in advance
>  Kind regards
>  Joris
>  
>  	[[alternative HTML version deleted]]
>  
>  ______________________________________________
>  R-help at r-project.org mailing list
>  
>  PLEASE do read the posting guide 
>  and provide commented, minimal, self-contained, reproducible code.