The Influence of Item Discrimination on Misclassification of Test Takers

Abstract

It has been suggested that low discriminating items can be included in a test with a criterion-referenced score interpretation as long as they measure a highly relevant content. However, low item discrimination increases the standard error of measurement, which might increase the expected proportion of misclassified test takers. In order to test it, responses from 2000 test takers to 100 items were simulated, varying item discrimination values and number and location of cut scores, and classification inaccuracy was estimated. Results show that the expected proportion of misclassified test takers increased as item discrimination decreased, and as the cut scores were closer to the mean of the distribution of test takers. Therefore, a test should include as few items with low discrimination values as possible —or even none— in order to reduce the expected proportion of test takers classified into a wrong performance level.