A Comparison of the Accuracy of Multidimensional IRT equating methods for Mixed-Format Tests

Abstract

This research aims to compare the accuracy of multidimensional IRT equating between concurrent calibrated MOSE procedure (CMOSE) and separated calibrated MOSE with Test characteristic function procedure (SMOSE) under the Monte Carlo simulation for mixed-format tests with approximate simple structure under the nonequivalent group with anchor test design (NEAT) with different common item score proportions. The consideration of accuracy of score equating is based on coefficient of variance of standard error of equating (CVSE). When comparing the CVSE value between the CMOSE and the SMOSE, the finding shows that there is interaction between MOSE procedures and common item proportions affecting CVSE value. The result of descriptive statistics shows that the CMOSE has lower CVSE value than that of the SMOSE. The result of simple-effect analysis shows that the CMOSE has lower CVSE value than that of the SMOSE when the proportions are 30 percent and 10 percent (p = .004 and p = .000, respectively). When comparing the CVSE value between the common item proportions for the CMOSE, the finding shows that the CVSE values among the proportions are not different. While the result of simple-effect analysis for the SMOSE shows that the 30% proportion has lower CVSE value than the 20% proportion (p = .000), the 20% proportion has lower CVSE value than the 10% proportion (p = .000) and the CVSE value of the proportion of 20 percent and 10 percent are not different. However, both types of the MOSE have the lowest CVSE when the proportions are 20 percent, 30 percent and 10 percent, respectively.