|
ROC Analysis
|
|
Web-based Calculator for ROC Curves
Highlights:
New support for continuously distributed data (Format 5).
To access this page, you may use the link
www.jrocfit.org.
|
John Eng, M.D.
Russell H. Morgan Department of Radiology and Radiological Science
Johns Hopkins University, Baltimore, Maryland, USA
|
|
Instructions: This web page calculates a receiver operating characteristic
curve from data pasted into the input data field below. To analyze your data, use the following steps
to fill out the data form on this page.
- Select the data format. (See explanation of data formats.)
- Paste or enter your data into the "Input Data" field or click the "Paste Example Data" button. Data may be pasted from programs such as
Microsoft Excel or Word (on the Macintosh, there is a bug).
- Enter the number of rating categories. (See explanation of data formats.) This is not needed for the example data or Format 5.
- Click the "Run Program" button.
- To export the ROC plot to Microsoft Word or Excel, see instructions below.
Web browser requirements for this page: For Windows, Microsoft Internet
Explorer 5.0, Netscape Navigator 6.0, Mozilla 1.4, or later versions of these browsers.
For Macintosh, Microsoft Internet Explorer 5.0, Mozilla 1.4, Safari 1.0, or later versions of these browsers.
This page may not work with earlier browser versions. If you have trouble, you can try the
earlier version of this page.
Please send any bugs, questions, comments, or suggestions to
electronic mail will be answered.
|
|
Suggested citation: The citations below conform to the styles used by the
National Library of Medicine and the
American Psychological Association, respectively.
- Eng J. ROC analysis: web-based calculator for ROC curves. Baltimore: Johns Hopkins University
[updated 2006 May 17; cited ____]. Available from: http://www.jrocfit.org.
- Eng, J. (n.d.). ROC analysis: web-based calculator for ROC curves. Retrieved
____, from http://www.jrocfit.org.
Acknowledgment: This web page contains JROCFIT and JLABROC4,
Java programs for calculating receiver operating
characteristic (ROC) curves. JROCFIT performs calculations for discrete ratings data,
and JLABROC4 performs calculations for continuously distributed data. JROCFIT and
JLABROC4 are direct translations of the ROCFIT
and LABROC4 programs, respectively. The latter two were Fortran programs
developed by Charles Metz and colleagues at the University of Chicago.
JROCFIT and JLABROC4 reproduce all output of the original programs.
See below for program translation details.
|
|
|
|
(Not needed for example data or for Format 5.)
|
| Input Data: (paste or enter) |
|
Program Output: (may be copied and pasted into other programs) |
|
|
|
| ROC Curve |
| Summary Statistics: |
| Points for Plotting: (copy & paste to Excel) |
|
 |
|
|
|
| |
 |
Key for the ROC Plot
RED symbols and BLUE line: Fitted ROC curve.
GRAY lines: 95% confidence interval of the fitted ROC curve.
BLACK symbols ± GREEN line: Points making up the empirical ROC curve (does not apply to Format 5).
Exporting the ROC plot to Word or Excel: Due to limitations of web technology,
there is no one-step method for exporting the ROC plot to Microsoft Word or Excel. However,
two methods are available for exporting:
- Copy a snapshot of the browser window by pressing Alt-PrintScreen, switch to the
Microsoft Word window, and paste the image by pressing Control-V. In Word, you need
to use Format...Picture (or the cropping tool from the Picture toolbar) to crop and
scale the image as needed.
- Select all of the text in the "Points for Plotting" field, which is located to
the right of the graph above. Copy and paste this text into Microsoft Excel. In
Excel, create a graph from the data by usual methods. This is a good way to obtain a
publication-quality graph of the ROC curve. Note: this method does not work with
the Microsoft version of Java. You must have the Sun version of Java, which can
be downloaded free from java.com.
|
|
Software bug affecting data pasted from Excel: With the Apple Mac OS X
operating sytem, copying and pasting data from applications such as Excel may not work properly
due to interference from embedded text formatting commands. This is probably due to a bug in Excel
that not only affects JROCFIT, but it also affects RTF text editors such as Apple's TextEdit
application. To
work around this problem, text copied from Excel must first be pasted into a plain text editor
(such as TextEdit with RTF turned off) to remove the embedded formatting. The text can then be
copied and pasted from the plain text editor into the JROCFIT data input field. Another
potential workaround is to use Mozilla
or Firefox. These web browsers may not be affected by the bug.
Program translation details: The colored box at the top of this page contains
JROCFIT and JLABROC4, programs for fitting receiver operating characteristic (ROC) curves using the
maximum likelihood fit of a binormal model. JROCFIT and JLABROC4 are literal Java translations
of the Fortran source code for ROCFIT and LABROC4, programs developed and maintained by
Dr. Charles Metz and colleagues at the University of Chicago. During translation, data input
routines were added to JROCFIT and JLABROC4 in order to
provide a more convenient user interface. The
translation was performed with kind permission from Dr. Metz,
but neither he nor his colleagues were
otherwise involved in the translation, nor do they share any responsibility for the translation's
accuracy. While the translation was performed as carefully as possible, the author and Johns
Hopkins University assume no
responsibility for the accuracy of the programs' operation, regardless of cause.
|
|
|