Help interpreting ROC Curve

[Rachel Jones]

Hello!

I am trying to get ROC Curve results for a logistic regression. When I run a logistic regression and then the lroc command after, I receive a single AUC score of .66.

However, when I run the rocreg command with all of the same variables, I get an output that gives an AUC for each variable. However, each variable had a significantly lower AUC than .66 with some as low as .25!
How does the same logistic regression give AUCs that are so different? Which one would I report? Also, does anyone have any suggestions to learn about interpreting ROC curves? I already read the help file on Stata and I need more help!

Thank you!!

[Josh Wang]

Hi Rachel,
I would report the overall AUROC. An AUROC of 0.5 would refer to the fact that the model has no discrimination whereas 1 is a perfect prediction. In general, your model performance is acceptable when it is between 0.7 and 0.8, no less than 0.8 represents excellent discrimination.
You can find details from this reference: Hosmer, D. W. Jr., Lemeshow, S., & Sturdivant, R. X. (2013). Applied logistic regression (3rd ed.). New Jersy: John Wiley & Sons.

Best
Josh