A COMPARISON OF SEVERAL ITEM DISCRIMINATION INDICES1

Abstract

The item discrimination indices compared in this study include tetrachoric coefficients of correlation, phi or fourfold point correlation coefficients, biserial coefficients, point biserial coefficients, Davis discrimination indices, and simple differences in proportions of correct response between upper and lower groups given the symbol D. The tetrachoric coefficients, the phi coefficients, and the D indices were secured for test samples both divided at the median total score and into highest and lowest thirds in terms of total scores. These subgroups were also used in obtaining the Davis indices. Use of subgroups comprising highest and lowest 27 per cents of the test sample in terms of total scores was shown by Kelley (1939) to yield the most sensitive item indices. In this study, highest and lowest thirds were used because the test samples were relatively small and it seemed that such subgroups would be more meaningful to instructors to whom item data are reported since they more nearly approximate the proportions of students earning A and B and D and F course marks. In any case, use of highest and lowest thirds does not materially affect the comparisons made in this study. Had highest and lowest 27 per cents been used, the indices concerned would have generally been slightly higher. The sources of data were two samples of 210 answer sheets used with Forms A and B of the 60-item Constitution Test used in the Illinois State High School Equivalency Testing Program. In these tests, series of keylist and multiple-choice exercises assess examinee knowledge of the Illinois Constitution, the United States Constitution, the Declaration of Independence, and flag etiquette. The tests have no time limit and no correction is made for guessing. The total scores of the Form A sample have a mean of 30.10 and a standard deviation of 7.67. The