Bootstrap Calibrated Empirical Likelihood Confidence Intervals for ROC Curves

Applications of receiver operating characteristic (ROC) curve analysis can be found in almost every scientific field. Under nonparametric settings, this paper proposes to use empirical distributions to estimate the population distributions in ROC curves, and then combine the bootstrap calibrated empirical likelihood method to construct confidence intervals for ROC curves. We demonstrate through simulation studies that intervals based on the bootstrap calibrated normal approximation are less satisfactory for samples of small or moderate size while the bootstrap calibrated empirical likelihood ratio confidence intervals perform well for all samples.