Hi all
I have generated logistic regression models, GAMs and
classification tree models for presence absence data.
I want to compare their predictive performance using
the statistic of the area under the curve of a ROC
plot. I want to estimate the Area under the curve of a
ROC plot by dividing the Wilcoxon two-sample test
statistic by the product of the absence and presence
data (Hanley and McNeil, 1982).
The problem is that S-Plus gives the normal Z
statistic without specifying the Wilcoxon two-sample
test statistic. How can I tell S-Plus to give me the
Wilcoxon two-sample test statistic?
Any other ways of estimating the area under the curve
of a ROC plot using S-Plus would be most appreciated.
Hanley, J.A. and McNeil, B.J. (1982), "The Meaning and
Use of the Area under a Receiver Operating
Characteristic (ROC) Curve," Radiology, 143 29 -36.
thanks and cheers
James
__________________________________________________
Do You Yahoo!?
Yahoo! Greetings - Send FREE e-cards for every occasion!
http://greetings.yahoo.com
|