Investigation of IRT-Based Equating Methods in the Presence of Outlier Common Items

• Published in: Applied psychological measurement Vol. 32; no. 4; pp. 311 - 333
• Main Authors: Huiqin Hu, Rogers, W. Todd, Vukmirovic, Zarko
• Format: Journal Article
• Language: English
• Published: Los Angeles, CA SAGE Publications 01.06.2008
• Subjects: Achievement Tests; Comparative Analysis; Computer Simulation; Computer Software; Equated Scores; Evaluation Methods; Item Analysis; Item Response Theory; Test Items; item response theory; outliers; transformation; equating; calibration
• ISSN: 0146-6216; 1552-3497
• DOI: 10.1177/0146621606292215

Common items with inconsistent b-parameter estimates may have a serious impact on item response theory (IRT)—based equating results. To find a better way to deal with the outlier common items with inconsistent b-parameters, the current study investigated the comparability of 10 variations of four IRT-based equating methods (i.e., concurrent calibration, separate calibration with test characteristic curve [TCC] and mean/sigma [M/S] transformations, and calibration with fixed common item parameters [FCIP]) when outliers were either ignored or considered. Simulated data were generated for the common-item nonequivalent groups matrix design to reflect the manipulated factors: group ability differences and nonequivalent groups, number/score points of outliers, and types of outliers. When no outliers were present, the TCC and M/S transformations performed the best. When there were outliers, overall, the methods that considered them (except the M/S transformation with outliers weighted) resulted in a vast improvement compared to the methods that ignored them.