--
The number of cutpoints is equal to the number of distinct predictions
from the model, which is in turn equal to the number of distinct
configurations of the predictors. If you have primarily categorical
predictors, smoothing may give a false impression of the
discriminatory power of the model if the plot is not already
smooth-appearing (and then, why smooth?). To get a smoothed plot you
can download -senspec-, written by Roger Newson, from SSC. It will
output variables for plotting. You might need to add artificial
observations at the endpoints of the ROC plot: (0,1) and (1,0). See an
example at: http://www.stata.com/statalist/archive/2009-01/msg00689.html.
Then use -lowess- or -lpoly- to smooth the plot.
-Steve
On Mon, Jul 13, 2009 at 10:32 AM, Stephan
Rudolfer<[email protected]> wrote:
> I am working on a database of about 3,000 patients, and wish to plot the ROC curves of various diagnostic algorithms based on logistic regression, using roctab and roccomp. Perhaps unsurprisingly, the plots are very "fuzzy", presumably since the cutpoints generated depend on the total number of patients in the database and hence are too numerous?
>
> My question: Is it possible to get a "cleaner" ROC curve plot? The roc routines don't seem to allow any control of the number of cutpoints used for the ROCC.
>
> Many thanks in advance for any help you can provide,
> Stephan Rudolfer
>
>
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
--
Steven Samuels
[email protected]
18 Cantine's Island
Saugerties NY 12477
USA
845-246-0774
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
