[R] Predictions from a logistic regression model with validation for ROCR

[David Winsemius]

On Aug 26, 2011, at 2:09 PM, Jon Toledo wrote:

>
> Dear experts,
> I am looking for a package that does logistic regression with  
> corssvalidation and gives me the probabilites of all the  
> corssvalidations so that I can plot them in ROCR.
> Would also like to know if the corssvalidate model would give me a  
> summary coefficient for the intercept and my to predictors.

I see you have the same left-right typing dyspraxia that I suffer form.

The rms package will do much of what you request. Please note that  
it's author is fairly dismissive of the value of using ROC measures  
(including the C-statistic often called the "Harrell c-statistic") for  
the purposes you appear to be considering. Harrell himself neither  
takes credit for nor recommends the c-index for measuring predictive  
abilit. He also points out literature by others that shows that cross- 
validation can often be misleading when estimating error rates. See  
pages 247-249 of the RMS book.


> Thank you.
> Best regards,
>
> Jon Toledo, MD
>
> Postdoctoral fellow



David Winsemius, MD
West Hartford, CT