4 ms·

Thanks for the link. ROC analysis is mostly useful for actual signal detection paradigms, like radar for which it was invented. It's generally not terribly us

by mhneu 10y ago

Thanks for the link. ROC analysis is mostly useful for actual signal detection paradigms, like radar for which it was invented. It's generally not terribly useful as a statistical test, as tests like Kolmogorov-Smirnoff have much more power while remaining non parametric. And other descriptive statistics are usually more interpretable than ROC.

hprotagonist 10y ago
Interesting. Thanks for pointing me to K-S -- non-parametric tests have their uses and I like to have as many dirty tricks as I can get. I have a signals and controls background, not really a stats one -- so ROC makes some intuitive sense to me. It's gained currency in the biomedical world, probably because so many of us come from EE backgrounds.