Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

ROC curves comparison

998 views
Skip to first unread message

Diego B

unread,
Nov 30, 2007, 6:16:32 PM11/30/07
to
Hi All,

How can I compare two (or more) ROC curves to see if at least one of
them is significantly different compared to the others ? I know how to
draw a ROC curve in SPSS 15 but I do not know how to compare two
curves to highlight significant (or insignificant) difference, by p-
value.

Thanks

Diego

Bruce Weaver

unread,
Dec 1, 2007, 9:38:25 AM12/1/07
to


The ROCKIT program available at the link below may be your best bet.

http://www-radiology.uchicago.edu/krl/KRL_ROC/software_index6.htm

From the website:

ROCKIT also calculates the statistical significance of differences
between ROC index estimates and parameters. On the basis of a "bivariate
binormal" model, it allows for comparison of 2 paired, partially
paired, or unpaired datasets (which would represent, for example,
different imaging modalities or diagnostic tests) with regard to:

* Differences in the binormal ROC curve parameters a and b (related to
the difference in the mean and of the standard deviations of the two
latest normal distributions used in the fit)
* Difference in the areas (Az) under the two estimated binormal ROC curves
* Difference between the two true-positive fractions (TPF)s on the two
curves at a selected false-positive fraction (FTP)


--
Bruce Weaver
bwe...@lakeheadu.ca
www.angelfire.com/wv/bwhomedir
"When all else fails, RTFM."

0 new messages