Garry
I have had a similar situation arise - i.e. when there is perfect separation in a small sample, there is no direct way to get a sampling error estimate of the ROC area. In this case, the reported confidence interval is bogus. As an alternative, I fit a model to the data and used simulation, drawing samples of size (109 in your case), did the classification on each sample and looked at the empirical distribution of the ROC area.
AL Feiveson
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Garry Anderson
Sent: Friday, October 30, 2009 3:42 AM
To: [email protected]
Subject: st: Roctab and binomial exact confidence interval
Dear Statalist,
webuse hanley
gen ratingm5 = rating
replace ratingm5 = rating - 5 if disease==0
roctab disease ratingm5,binomial
ROC -- Binomial Exact --
Obs Area Std. Err. [95% Conf. Interval]
--------------------------------------------------------
109 1.0000 0.0000 0.00023 0.05006
How does one interpret the 95% CI of 0.00023 to 0.05006 when the ROC
area is 1.00?
I have seen a dataset (n=47, 15 +ve) where the ROC area was 1.00 and I
wish to determine the lower 95%CI.
Cheers, Garry
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
