Nonparametric confidence intervals for the area under the ROC curve

Abstract

Following an idea by Jing et al. (2005), this paper combines the empirical likelihood for the mean functional with jackknife pseudo-values obtained from the Mann-Witney two-sample statistic. This leads to a pseudo-likelihood L(p) for the area p0 under the ROC curve. A Wilks-type theorem is proved, so that L(p) can be used in a standard way to obtain approximate confidence intervals for p0. Some simulation results are given, in order to show the usefulness of the proposed method.