[Diagnosismed-list] Fwd: ROC()

[Pedro Emmanuel Alvarenga Americano do Brasil]

---------- Forwarded message ----------
From: Alexander Hapfelmeier <Alexander.Hapfelmeier at tum.de>
Date: 2009/2/19
Subject: ROC()
To: emmanuel.brasil at gmail.com


Hallo Sir,

Today I searched some functionalities for ROC-Curves and found your function
ROC(). It is a very good function to analyse ROC-Curves but its
documentation brings up some questions. How are the CI computed? So I
checked the code itself. CI are computed by Wilsons method. But there seems
to be a problem with the CI auf the AUC. It is not the 95%-KI as you compute
AUC +- SE(AUC). For me it is not clear if you did this by purpose or if it
is a mistake.

So I have two suggestions:
- Maybe you could improve your documentation by including some
specifications about the computational background e.g. mentioning that CI
are computed by Wilsons method and so on...
- A singificance level to the CI of the AUC should be added.

Thank you very much for your efforts.
Sincerely,
Alexander Hapfelmeier



-- 
Abraço forte e que a força esteja com você,
Pedro Emmanuel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.r-forge.r-project.org/pipermail/diagnosismed-list/attachments/20090219/da50490c/attachment.htm