ROC analysis using Ordinal Data

[Ryan]

Hi,

I'm trying to figure out how to create cut points through ROC analyses
based for four ordered categories (1, 2, 3, 4).

I ran ordinal regression and have predicted probabilities for each
category. I assume I'd run three ROC analyses to determine the cut
between 1 and 2, 2 and 3, and 3 and 4?...

* First ROC analysis would use the predicted probability for category
2 as the test variable
* Second ROC analysis would use the predicted probability for category
3 as the test variable
* Third ROC analysis would use the predicted probability for category
4 as the test varaible

Questions:

Would the actual state variable (values 1, 2, 3, and 4) remain the
same for all three ROC analyses?
Would I need to use the same level of specificity and sensitivity for
all three cut points?

Absolutely any thoughts would be greatly appreciated!

[Bruce Weaver]

Ryan, see if this gives you any ideas:

http://www.ncbi.nlm.nih.gov/pubmed/16099683

I found it by searching on <ROC curve ordinal gold standard>.

--
Bruce Weaver
bwe...@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/
"When all else fails, RTFM."

[Ryan]

> bwea...@lakeheadu.cahttp://sites.google.com/a/lakeheadu.ca/bweaver/
> "When all else fails, RTFM."- Hide quoted text -
>
> - Show quoted text -

Thanks, Bruce. I'll review it today.