NotesFAQContact Us
Search Tips
ERIC Number: ED527291
Record Type: Non-Journal
Publication Date: 2008
Pages: 194
Abstractor: As Provided
Reference Count: N/A
ISBN: ISBN-978-1-1242-9562-6
A Simulation Study on the Performance of Four Multidimensional IRT Scale Linking Methods
Wei, Youhua
ProQuest LLC, Ph.D. Dissertation, University of Florida
Scale linking is the process of developing the connection between scales of two or more sets of parameter estimates obtained from separate test calibrations. It is the prerequisite for many applications of IRT, such as test equating and differential item functioning analysis. Unidimensional scale linking methods have been studied and applied frequently over the past two decades. The development of multidimensional linking methods is at the infancy stage and more research is needed to obtain definitive results. As an extension of previous research, the purpose of this study was to use simulated data to evaluate the performance of four multidimensional IRT scale linking methods, the direct method, equated function method, test characteristic function method, and item characteristic function method, under various testing conditions, which include different test structures, test lengths, sample sizes, and ability distributions. There were one hundred and ninety-two experimental conditions in this study and five hundred replications were conducted for each of the conditions. The linking performance evaluation was based on the differences between the item parameter estimates for base group and the transformed item parameter estimates for the equated group across the test items. The mean and standard deviation of the differences across the 500 replications were computed to examine the accuracy and stability of the four linking methods. Our results indicate that for approximate simple test structure, each of the four linking methods worked approximately equally well under all testing conditions. The results also suggest that for complex test structure: (a) The equated function method did not work well under any testing conditions, (b) the performance of other three linking methods depended on other testing conditions including sample size, test length, and ability distribution difference between groups, and (c) the direct method was the best linking procedure for most testing conditions. In addition, the study shows that the item parameter values influenced the linking performance. Under most of the testing conditions, the linking results for the discrimination parameter tended to be less accurate and less stable when the item parameter had extreme values. The linking accuracy for the difficulty parameter was not dependent on the item parameter values. The linking stability for the difficulty parameter depended on the item parameter values only when the sample size was large. Then, the linking results were less stable when the item parameter had extreme values. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page:]
ProQuest LLC. 789 East Eisenhower Parkway, P.O. Box 1346, Ann Arbor, MI 48106. Tel: 800-521-0600; Web site:
Publication Type: Dissertations/Theses - Doctoral Dissertations
Education Level: N/A
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: N/A