HOW TO USE PARTAREA.FOR
***********************
PARTAREA.FOR is an interactive FORTRAN program to estimate the area under the 
curve (for either one or two curves) within a certain range of false-positive
rates. In other words, the program calculates the partial ROC area; rather than
looking at the AUC under the entire curve, we look at the curve over a portion
of the X-axis (i.e., false-positive rate). A reference for this methodology 
can be found in: McClish, 1989. Analyzing a portion of the ROC curve. Medical
Decision Making; 9:190-195.

You will first need to copy the program PARTAREA.FOR into your home directory.
Then you will need to issue the following statements in your command tool
after the % prompt:

source /vol/vni/ipt/bin/iptsetup.csh
f77 partarea.for -o partarea
partarea

The program will then prompt you for information which you will then enter 
directly into the command tool window. You will first be asked whether you are
interested in the partial AUC of one curve or two. Then you will be asked for 
the A and B parameter estimates, along with Var(A), Var(B), and Cov(A,B) for
each curve you are interested in (either one curve or two curves, depending on
what you specified earlier). To obtain these parameter estimates, you will need
to first run your data through one of the "parametric" FORTRAN programs 
(ROCFIT, CORROC2, LABROC4, CLABROC...check WHICHPRG.TXT to determine which
program is appropriate for your data). You will then be asked if your two curves
are dependent (i.e., correlated or "paired"), and if they are, you will be asked
for more information (specifically, Cov(A1,A2), Cov(A1,B2), Cov(B1,A2), and
Cov(B1,B2)) from the same parametric FORTRAN program (either CORROC2 or
CLABROC) that you used before. CORROC2 prints out a variance-covariance matrix
that will give you the covariances you need. Keep in mind that Cov(A1,A2), for
example, may be referred to in the program as "Cov(AX,AY)". CLABROC, meanwhile,
will print out the following in a single row, in the following order:
Cov(A,B) for curve 1 (or "X")
Cov(A,B) for curve 2 (or "Y")
Cov(A1,A2)
Cov(A1,B2)
Cov(B1,B2)
Cov(B1,A2)

You will then be asked to type the range of false-positive rates you are
interested in. At this point, the program will compute and display the partial
areas, variances, and (if applicable) the covariance between the partial AUCs of
the two curves. You will then be asked if you wish to rerun the program, and
if so, if you wish to change only the range of FPRs.

The Z-test comparing the two partial AUCs then takes the usual form:

Z= AUC1-AUC2/sqrt(V1+V2-2COV)