ABSTRACT
We start by considering the situation that a predictor has been published, that is, we are able to compute predictions Yˆ for any new subject in dependence on its covariate values X∗1 ,X∗2 , . . . , X∗p . Typically this means that we have access to the regression parameters of the fitted model and-if necessary-to some cutpoint. Now, the aim is to describe the performance of this predictor in our current sample. Of course, we can just apply the predictor to all subjects in our sample such that we obtain a prediction Yˆi for each subject. It just remains to describe how close Yˆi is to Yi on average.
