Stephen, thanks for that pointer to the Delong, Delong & Clarke-Pearson article.  The reference I knew for this topic was Hanley & McNeil (1983), which Delong et al. cite.  But here's what they say about it:

"Hanley and McNeil (1983) use some properties of this nonparametric statistic to compare areas under ROC curves arising from two measures applied to the same individuals. Their approach involves calculating for both the normal and the abnormal sample the correlation between the values of the original measures. The average of the two correlations is used along with the average of the areas under the two curves to arrive at an estimated correlation between the two areas. A table that applies when the average area is at least .70 is given. However, for measures that are not continuous or nearly so, their method relies on Gaussian modeling assumptions for estimating the variances of the two areas. In Section 2 we present an alternative methodology using a more completely nonparametric approach which exploits the properties of the Mann-Whitney statistic." (p. 838)

Thank you very much Steve and Bruce!

So there is no ready to use program established yet? Then I'll try to find a way to implement it myself. Thank you both very much for the references!

Dear Katrin Skerl , 

The best way to determine the optimal cut value is apllying ODA (Optimal data analysis)

Helena

Dear  Steve, Dear Helena,

thank you very much for your replies.

Helena, can you explain me how to do it? Can you recommend a program for it?

Thank you very much in advance!

kind regards,

Katrin

Dear Katrin, 

In terms of ODA you should ask Paul Yarnold. if you know satitistics , than his book is very easy to read and because has lots of examples you would find it easy to apply to your problem. 

And Paul Yarnold is in the RG.

let me know about your decision.

Kind regards, 

Helena

PS: I have exmples about ROC in Spss. To compare the 2 results you could have used a dicotomic variable  and use it as the respondent variable in logistic regression . with save command you obtain the variable to introduce in ROC.

But the best way I think is for you  to learn ODA.

You can compare AUC values from ROC analysis using Medcalc.

https://www.medcalc.org/features/roccurves.php

Thank you very much Helena, I'll study ODA accuratly because I found the results of many statistical analyzes be illogical with applying classical models .

I have found a large number of valuable articles in the eJournal of Optimal Data Analysis (ODA) and I want to buy the book titled (Optimal Data Analysis: A Guidebook with Software for Windows) to start from the basics. 

Dear Helena, is there an another web for buying this book instead of this web site.

http://www.amazon.com/Optimal-Data-Analysis-Guidebook-Software/dp/15579898181

Thanks in advanced

Dear Huda,

I bought  the book via amazon. It was very easy and rapid...

But you could by it via link from ODA journal...could be less expensive.. If i was you I would ask him...

I saw this for you:

Contact Us

As of January 13, 2015, and until further notice, only standard mail communications will initially be available to new individuals who wish to communicate with ODA LLC. This is due to the accelerating use of the internet for spam pollution. For all matters, please mail a letter to: Optimal Data Analysis, LLC; 6348 N. Milwaukee Ave., #163; Chicago, IL 60646.Please include your eMail address in the letter. Once a legitimate letter has been received, an eMail address that may be used for all subsequent communications with ODA LLC will be eMailed to you.

The Journal

All the best for you

Helena

Thank you my dear Helena, I think the book is very cheap compared with its scientific value, but the problem lies in the lack of recognition of the Iraqi credit card which belong to all Iraqis researchers. In any case. Thay send me this massage:

 (Sorry, this item can't be shipped to your selected address. Learn more. You may either change the shipping address or delete the item from your order. You can also see if this item is available to ship to your address from another seller. )

I will repeat the attempt with this site and I do not have a problem because a lot of my family members lived outside of Iraq.

Many thanks for your kindness

Best Regards

Hi all,

As this question seems to be interesting for more people, I just figured out that a function comparing depending ROCs is available for R using the pROC package.

Best wishes,

Katrin

Hi all,

As this question seems to be interesting for more people, I just figured out that a function comparing depending ROCs is available for R using the pROC package.

Best wishes,

Katrin

Hi all,

As this question seems to be interesting for more people, I just figured out that a function comparing depending ROCs is available for R using the pROC package.

Best wishes,

Katrin

A function for comparing ROC curves is also available in Stata: roccomp and can be applied to both ROC curves derives from the same sample and from different samples.

http://www.stata.com/manuals13/rroccomp.pdf

Kind regards,

Daniela

I have just seen the topic. Yes, after five years it might have lost actuality, however, it is my duty to report some updates for such basic objectives like AUC comparison in order to enable access to various softwares. JMPPro - licensed versions are easily provided by many universities/companies - can also do that in two clicks for a quick and simple modeling and models comparison report.

Just take a look into a case enclosed where eight models for > 6000 stratified data are compared. Although for the validation set is well described by one of the Neural networks, almost all the models' AUC > 0.8. Thus, under the ROC curves the comparison analysis is provided, as the idea in this case is to propose the simpliest model with an acceptable AUC.

As seen, only 3 cases have a significant difference related to the weakest neural network model (highlighted in bold*). Thus, such a case can be easily described by simple logistic regression model whose AUC (0.8205) is not different significantly from the strongest one (AUC = 0.8211) assigned to Neural Network with a single hidden layer of triple hyperbolic tangent function.

Michael Nazarkovsky thank you for your reply. It was very helpful. DO you know which mmethod is used here to compare the two AUCs. Is it De-Long?

If you are used to SPSS and just want to analyse the ROC curves for statistical difference, which is not possible with SPSS, from my point of view MedCalc is the most easiest options. It is a free ware software and easy to use. JMPPro might be a little bit confusing - if you are not used to it.

I did the same analysis using Medcalc. I downloaded the free trait version. It is easy to use, no need for complicated syntax. I recommend using it.

https://analyse-it.com/ can be used.

You can do so by using compare ROC function offered in Medcal software.

R can be used to compare ROC curves. pROC package (maybe others too) enables this.