R help - Sep 2009 - Design Package - Penalized Logistic Reg. - Query

Dear R experts,

The lrm function in the Design package can perform penalized (Ridge)
logistic regression. It is my understanding that the ridge solutions are not
equivalent under scaling of the inputs, so one normally standardizes the
inputs. Do you know if input standardization is done internally in lrm or I
would have to do it prior to applying this function.

Also, as I'm new in R (coming from SAS) I don't know how well R will
handle
relatively large data sets (e.g. 1/2 million observations on 40 variables).

I'll appreciate your comments.

Many thanks in advance.

Lars/

	[[alternative HTML version deleted]]

On Sep 25, 2009, at 8:33 PM, Lars Bishop wrote:
> Dear R experts,
>
> The lrm function in the Design package can perform penalized (Ridge)
> logistic regression. It is my understanding that the ridge solutions  
> are not
> equivalent under scaling of the inputs, so one normally standardizes  
> the
> inputs. Do you know if input standardization is done internally in  
> lrm or I
> would have to do it prior to applying this function.
>
> Also, as I'm new in R (coming from SAS) I don't know how well R
will
> handle
> relatively large data sets (e.g. 1/2 million observations on 40  
> variables).
I don't have the answer to your first question but I routinely work  
with a dataset that is several times that large using the Design (and  
now) the rms packages. (You do need to have sufficient physical  
memory, but it is not R that is the limiting factor.)

-- 

David Winsemius, MD
Heritage Laboratories
West Hartford, CT

Lars Bishop wrote:
> Dear R experts,
> 
> The lrm function in the Design package can perform penalized (Ridge)
> logistic regression. It is my understanding that the ridge solutions are
not
> equivalent under scaling of the inputs, so one normally standardizes the
> inputs. Do you know if input standardization is done internally in lrm or I
> would have to do it prior to applying this function.
It's done internally, as buried in the documentation somewhere. 
Actually lrm puts the scaling factors (standard deviations for 
continuous variables) into the penalty matrix.

Frank
> 
> Also, as I'm new in R (coming from SAS) I don't know how well R
will handle
> relatively large data sets (e.g. 1/2 million observations on 40 variables).
> 
> I'll appreciate your comments.
> 
> Many thanks in advance.
> 
> Lars/
> 
> 	[[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.
> 

-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University