Semi-parametric simultaneous confidence bands for an ordinal dominance curve
Abstract
Practitioners in various fields are often confronted with the task of assessing the ability of a diagnostic test to correctly assign one of two labels (healthy/diseased or low/high credit risk) to each observation in a collection of observations. An ordinal dominance curve (ODC) describes the functional relationship between the proportions of correct assignments to the labels. In this paper we derive a semi-parametric confidence band for the ODC and evaluate some of its properties by Monte Carlo simulation. The methodology is illustrated by application to data from a credit risk environment.