Marcus Fischer wrote:
areas under multiple ROC curves can be compared using "roccomp". How can I
compare areas under ROC curves if I want to adjust the model for additional
covariates? I guess, "rocfit" does not allow the inclusion of covariates;
and after multiple logistic regression you will get of course only 1 ROC
curve.
-------------------------------------------------------------------------------
The general method is illustrated below, using -roccomp-. The example is to
compare the fit of a multiple logistic regression against one of the predictors
alone, so the dataset is configured wide. If the samples are independent in
your case, then, as the help file indicates, configure the dataset long and use
the -by()- option to indicate grouping.
If you're going to be involved in evaluations of medical diagnostic tests, then
take a look at -somersd-, too, and related papers and presentations (many
downloadable) from its author, Roger Newson. (-findit newson-)
Joseph Coveney
sysuse auto, clear
// multiple logistic regression results
logit foreign price trunk
lroc, nograph
// AUC of the ROC is 75%
predict xb, xb
// compare ROC of multiple regression against, say, price, alone
roccomp foreign xb price
// That for price alone is 58%
// If, on the other hand, your'e interested in hypothesis testing
// for the improvement in prediction, then
logit foreign price trunk
estimates store A
logit foreign trunk
lrtest A .
exit
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
