An investigation on computer-adaptive multistage testing panels for multidimensional assessment

UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
Xinrui Wang (Creator)
The University of North Carolina at Greensboro (UNCG )
Web Site:
Richard Luecht

Abstract: The computer-adaptive multistage testing (ca-MST) has been developed as an alternative to computerized adaptive testing (CAT), and been increasingly adopted in large-scale assessments. Current research and practice only focus on ca-MST panels for credentialing purposes. The ca-MST test mode, therefore, is designed to gauge a single scale. The present study is the first step to investigate ca-MST panels for diagnostic purposes, which involve the assessment of multiple attributes in the same test. This study employed computer simulation to compare multidimensional ca-MST panels and their unidimensional counterparts, and to explore the factors that affect the accuracy and efficiency of multidimensional ca-MST. Nine multidimensional ca-MST panel designs - which differed in configuration and test length - were simulated under varied attribute correlation scenarios. In addition, item pools with different qualities were studied to suggest appropriate item bank design. The comparison between the multidimensional ca-MST and a sequential of unidimensional ca-MST suggested that when attributes correlated moderate to high, employing a multidimensional ca-MST provided more accurate and efficient scoring decisions than several unidimensional ca-MST with IRT scoring. However, a multidimensional ca-MST did not perform better than its unidimensional counterpart with MIRT scoring. Nevertheless, multidimensional panels are still promising for diagnostic purposes given practical considerations. The investigation on multidimensional ca-MST design indicated the following: Higher attribute correlation was associated with better scoring decision because more information carried by a correlation matrix was available for estimation. This held true across all item pool conditions. An optimal item pool would be the one that was discriminative, appropriately located and specifically designed for a configuration. The accuracy and efficiency of a multidimensional ca-MST panel would be diminished if its item pool was too easy, or not informative. According to the results, the 1-2-3 configuration design was most promising. In terms of test length, an appropriate decision would largely depend on the attribute correlation and the item pool characteristics.

Additional Information

Language: English
Date: 2013
Computer-Adaptive Testing, Multidimensional Assessment, Multistage Testing
Educational tests and measurements
Computer adaptive testing

Email this document to