The sample size is 20, i'm trying to find the cut-off point between two groups with a prism cover test (10 with symptoms and 10 without)
However, the curve does look odd for this sample. Is it ok to include this in my results?
ROC curve >> http://imgur.com/a/Im37N
You are right, that ROC is odd but it is a plotting issue not an issue with the data. ROC should never show a sloped line, only vertical or horizontal. I don't know what software you are using to plot this but it isn't doing it right. I've heard of some programs fitting functions to smooth the trend, but this is not what this is doing.
AUC is on a constrained scale of 0.5 to 1 (or 0-1 in validation as if you have a really bad model it can invert performance in testing) so the resolution you can obtain from 20 samples is very limited and the error of the estimate will be large. With so few samples you tend to get very steep steps.
It certainly does. However your curve is not so weird.
For publication, it depends on you and the reviewer. I can definitely say that your sample size is too small.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics