Examining the Discrimination of Binary Scored Test Items with ROC Analysis

In this study, it was claimed that ROC analysis, which is used to determine to what extent medical diagnosis tests can be differentiated between patients and non-patients, can also be used to examine the discrimination of binary scored items in cognitive tests. In order to obtain various evidence for this claim, the 2x2 contingency table used in the ROC analysis was adapted in accordance with the logic of item discrimination. It was suggested in the article that the areas under the ROC curves (AUC) obtained by using the sensitivity and specificity values calculated with the adapted contingency table can be considered as a measure of item discrimination. The results of the statistical analyses made on the simulation data showed that the AUC values were positively and highly correlated with the D, r(bis) and a parameter values of the items, and the AUC values from different sized samples were consistent. Additionally, ROC analysis was more stable against range narrowing than other methods. In this respect, it was concluded that very large groups were not needed to examine item discrimination with the proposed method.