テストの信頼性が等化に与える影響

概要

論文の詳細を見る
This study examines the effect of test reliability on equating as a test result by using a simulation. Four types of tests having different difficulty grades were used in the simulation. These tests were assigned to five hundred subjects whose ability corresponded to the test grade, and correct/error data was created by assuming the item response theory. Each test consisted of forty items and adjacent tests, arranged according to the difficulty grades, included four common items. The item parameters were estimated from this data using Bock's EM algorithm and the average and variance of the subjects' ability estimated using Mislevy's method. This means that equating is used in the process of such alternate estimations. The simulation results show that the equating results were not affected by a decrease in the reliability of non-common items between tests. On the contrary, a decrease in the reliability of common items（that is a decrease of the discrimination of common items ）had an adverse effect on the equating of test results. However, it was found that the equating result is improved by increasing the number of subjects, even if the discrimination of common items is low.
文教大学の論文