Skip to content

ROC curves using R

2 messages · Cheng, Yiling, Frank E Harrell Jr

#
ROC (receiver operating characteristic) curve is used to determine of
cutpoints for screening or diagnostic tests. It have been widely used in
Epidemiology area. Usually we use logistic regression to get this kind
curve. If anyone knows how to get this curve and do statistic test for two
curve, do please let me know too. Thanks in advance.
Yiling Cheng

-----Original Message-----
From: baron at cattell.psych.upenn.edu
[mailto:baron at cattell.psych.upenn.edu]
Sent: Friday, February 01, 2002 11:17 AM
To: jg_liao at yahoo.com; r-help at stat.math.ethz.ch
Subject: Re: [R] ROC curves using R


What is it that you want to compute?  I have a set of
scripts - written by my colleaged David Brainard in
Matlab and translated to R by me - which we used to teach
signal-detection theory in a grad "methods" course.
But they are very straightforward and not worth putting
into a package.  (I'm sure there is something much more
sophisticated one can do, but you don't say what.)

Some useful R functions (illustrated, and probably
self-explanatory):

zHit <- qnorm(Ph)
zFa <- qnorm(Pfa)
dPrime <- zHit-zFa

zCr <- qnorm(1-Pfa)
Beta <- dnorm(zCr,dPrime,1)/dnorm(zCr,0,1)

And here is the optimal "beta" (criterion):

logBetaOpt <- log((pn/ps)*(Vcr + Vfa)/(Vh + Vm))

Jon Baron
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.
-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._.
_._
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
#
For the purposes for which you mention, using ROC curves results in a number of serious problems.  Full post-test probability models are generally recommended, without dichotomization of post-test probabilities until the point at which individual-patient utilities can be estimated.

Frank Harrell

On Fri, 1 Feb 2002 14:27:07 -0600
"Cheng, Yiling" <ycheng at cooperinst.org> wrote: