Kelly,
Run an ROC (receiver operating characteristic) analysis and see if your
"continuous predictor" is any better than tossing a coin. See if the area
under your ROC curve is significantly greater than 0.5 by the statistical
tests described on the "Magnificent ROC" web page. Aptly named, in this
case, because it will give you a nice visual assessment of your situation:
http://www.anaesthetist.com/mnm/stats/roc/
For the ROC analysis, go to this web page:
http://www.bio.ri.ccf.org/html/rocanalysis.html
and look for the link to rocPlot.s (written by Hemant Ishwaran, PhD,
Cleveland Clinic).
If your ROC plot has a nice knee in it, the plot can help you choose a
cutoff point. If it is flat as a pancake, you'll have some graphic evidence
to prove to whomever told you to take on this project that their continuous
predictor isn't predicting anything.
Let me know how it works out, and I hope this helps.
Best regards,
Alan
Alan Hochberg
Vice President, Research
ProSanos Corp.
225 Market Street
Suite 502
Harrisburg, PA 17101
Tel. 717-635-2124
Fax 717-635-2575
alan.hochberg@prosanos.com
www.prosanos.com
|