So there I was testing ROC for a scale to predict intellectual deficit as the scale tests for some basic skills. So what matters here is that as the scoring goes lower, there should be a higher chance of predicting the clinical group membership. Running that on SPSS is quite easy as I can indicate that lower score increases chance of detection. Still I'm interested in some outputs that SPSS isn't able to provide me.
Then I looked after options in R... stumbled upon many pakcages which weren't THAT user friendly until I found pROC. Well, pROC is simple and incredible. Problem is I miss the option which I can use to indicate the same as I indicate in SPSS (that score decreases and case detection is more likely). Control group is 0, clinical is 1. Everything is set... but here I am missing that one simple information... or let me know if it is not needed at all.
Not sure why, but pROC seems to give me what I want as default. Just found that out while running the same analysis on both, considering the settings I want on SPSS. Funny thing is that one of the curves is strange. That one shows an inversion when compared to SPSS'. Every other curve goes the same way. Strangely that same curve in SPSS is upwards, while on pROC it is inverted.... score isn't inverted and group membership is also the same.....
That should be because SPSS has some option checked (something like "Larger values show disease" etc..). I am not sure if pROC has a similar option like that in SPSS. But an easy workaround is to change the coding of binary variable.
Try this if your controls show higher values than cases:
RocIN
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics