Incorrect AUC value and CI [bug] about proc HOT 5 CLOSED

cf. over there for more such examples
https://gist.github.com/UnixJunkie/d38de911f7fac31aca48651c8684d1de

from proc.

The value of AUC is obviously correct.

With the dataset you provided in score_labels.txt:

> roc_curve <- roc(df, "labels", "scores")
Setting levels: control = 0, case = 1
Setting direction: controls > cases
> auc(roc_curve)
Area under the curve: 0.6789
> ci.auc(roc_curve)
95% CI: 0.6382-0.7196 (DeLong)

If you have prior knowledge of which group has higher values of the predictor, you should change the direction argument to match.

> roc_curve <- roc(df, "labels", "scores", direction = "<")
Setting levels: control = 0, case = 1
> auc(roc_curve)
Area under the curve: 0.3211
> ci.auc(roc_curve)
95% CI: 0.2804-0.3618 (DeLong)

See ?roc for more details.

from proc.

That's very dangerous this direction parameter.
The default behavior is not right all the time.

from proc.

There's no default that will be right all the time - otherwise there would be no need for a parameter in the first place.

This behavior is documented in ?roc, in FAQs and extensively in forums/stackoverflow. Output is created to communicate about the decision transparently.

from proc.

Funnily, the python CROC package never gets the AUC wrong...

from proc.