On question 3, for calibration plot c, why are there 4 points instead of 5? Can we assume that two of the points have been collapsed in to one point? If so, which two points have been collapsed together? It seems like the answer to the question depends on this... for example, if the two lowest risk groups were collapsed then discrimination would be worse among low risk... is that thinking correct?
Hi Sarah,
The citation for this article is included in the problem set, and after some sleuthing it looks like one of the categories in the CRB65 had zero patients (and thus zero deaths) so I can't imagine two points were collapsed.
Be that as it may, this question is asking about the fundamental property of constructing an ROC curve, and whether re-calibration or not changes the curve. I would think, if you applied our methodology of making ROC curves, that you could answer this question without even seeing the plots! Hope that is helpful.
Best,
Kirk