In ROC curves, the general rule is that our test should do a performance better than Pd=Pfa. What is the meaning of a curve being present below this line? How to analyze ROC?
In my question Pfa being Probability of False Alarm and Pd being Probability of Detection and ROC is obviously Receiver Operating Curve/Characteristics.
Thanks in Advance.
It means you are making a coding error. You need to flip your dependent binary variable.
Being less is equivalent to being better. Thinking about using Logistic regression to construct a ROC, you can define either event=1 or event=0. If using event=1 you get a ROC above diagonal line, then you will get a ROC below if you switch to event=0. Many software automatically flip curve to make it always above diagonal line, but I know SPSS will give you ROC below.
SPSS could flip it if one flips an option named "smaller test results show more positive results".
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Probability