Impact of multidimensionality on unidimensional IRT linking and equating methods
- UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
- Uk Hyun Cho (Creator)
- Institution
- The University of North Carolina at Greensboro (UNCG )
- Web Site: http://library.uncg.edu/
- Advisor
- Kyung Yong Kim
Abstract: The present study investigates the influence of multidimensionality on linking and equating in a unidimensional IRT. Two hypothetical multidimensional scenarios are explored under a nonequivalent group common-item equating design. The first scenario examines test forms designed to measure multiple constructs, while the second scenario examines a test aimed to measure a primary latent trait but contaminated with a nuisance factor. Classification measures and equating equity properties are used to compare the baseline multidimensional IRT and unidimensional IRT under these scenarios. The findings suggest that multidimensionality is not the primary factor influencing the behavior of linking constants A and B. However, interacting factors such as mean shift, covariance structure, and linking method do have an impact. Test structure alignment is crucial for achieving quality equating results, as equating bias constitutes a substantial proportion of the total error. Classification indices demonstrate that unidimensional IRT generally outperforms the baseline MIRT, with semi-equivalent test structures showing higher performance. Equating equity properties indicate that test structure alignment and choice of linking methods significantly influence equating quality and predictability. The study highlights the importance of considering factors in achieving accurate and precise equating results. Further, Approximate Multidimensional IRT True Score (AMT) equating is proposed as a possible solution to assess the impact of multidimensionality to address the limitations of conventional equating methods in capturing dimension-specific changes in scores between test forms.
Impact of multidimensionality on unidimensional IRT linking and equating methods
PDF (Portable Document Format)
4923 KB
Created on 5/1/2024
Views: 202
Additional Information
- Publication
- Dissertation
- Language: English
- Date: 2024
- Keywords
- Classification, Equating equity, MIRT TSE
- Subjects
- Educational tests and measurements $x Evaluation
- Psychological tests $x Evaluation
- Multidimensional scaling
- Item response theory