AREA UNDER THE BINORMAL ROC CURVE USING CONFIDENCE INTERVAL OF MEANS BASING ON QUARTILES

Abstract

Roc curve analysis is a best statistical tool to assess the performance of test accuracy by an area under the curve (AUC). In binormal model, let X and Y be two normal populations with means μx and μy for diseased population (D) and healthy population (H). This paper emphasis, area under the binormal roc curve model and comparisons are made with the help of different AUCs basing on various possible distances (difference between population means) Dj; j=1, 2 ….9. These nine possible distances can be calculated by taking lower and upper limits of confidence interval of means, which can be computed from first and third quartiles (i.e. Q1 and Q3) from their respective normal populations. Estimation of three new method of averages namely i) Simple Average Method ii) Fixed Weights Method (FW-Method) and iii) Proportional Weights Method (PWMethod) are briefly discussed also comparisons made between them and normality is tested by P-P plot.