Files and links (1)
Dissertation Main_SeoheePark_Final_Revision3.88 MB
Embargoed Access, Embargo ends: 02/01/2026
Dissertation
Multiple group measurement alignment of the bifactor model within the IRT framework
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Autumn 2021
DOI: 10.17077/etd.006250
Abstract
This dissertation extended the measurement alignment method into the bifactor model. Measurement alignment places all item parameters across multiple groups on the same scale simultaneously. The item parameters aligned by measurement alignment are used to evaluate the presence of biased items across multiple groups. Although measurement alignment is used in various fields, applicable models are somewhat limited. Considering the increasing popularity of bifactor models, this study introduced the procedures of measurement alignment for the bifactor model. A technical description of measurement alignment for the bifactor model was provided. The extended model was evaluated by comparing it to the Wald-2 approach regarding estimated group means, standard deviations, item parameters, and differential item functioning (DIF) analysis through real data analysis. Furthermore, this study evaluated the quality of estimates through a simulation manipulating several conditions: the number of groups, the non-equivalence among groups, the magnitude of DIF, and the patterns of DIF. The empirical data analysis showed that the measurement alignment and Wald-2 approaches produced different estimates of group means, group standard deviations, and group item parameters. Nevertheless, the DIF detection results were similar between the measurement alignment and Wald-2 approaches. Measurement alignment found 80% of DIF items, and Wald-2 found 88% of DIF items. The simulation study highlighted that measurement alignment for the bifactor model recovered group means, group standard deviations, and item parameters well in general. The accuracy of estimation of measurement alignment for the bifactor model was high although the stability of estimation was slightly low. In addition, the simulation study demonstrated that the equivalence of group distributions, the magnitude of DIF, and the patterns of DIF influenced the quality of estimation of measurement alignment for the bifactor model, whereas the number of groups did not affect the quality of estimation. These verified results of measurement alignment for the bifactor model would be utilized for research using the bifactor model. In addition, this study would be beneficial to future studies, which aim to extend measurement alignment into other multidimensional models, because this study provided the details of the procedure as an initial study of measurement alignment for a multidimensional
Details
- Title: Subtitle
- Multiple group measurement alignment of the bifactor model within the IRT framework
- Creators
- Seohee Park
- Contributors
- Ariel M Aloe (Advisor)Michael E Walker (Committee Member)Lesa Hoffman (Committee Member)Won-Chan Lee (Committee Member)Jonathan Templin (Committee Member)Jacob B Priest (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Psychological and Quantitative Foundations
- Date degree season
- Autumn 2021
- DOI
- 10.17077/etd.006250
- Publisher
- University of Iowa
- Number of pages
- xvi, 211 pages
- Copyright
- Copyright 2021 Seohee Park
- Language
- English
- Description illustrations
- color illustrations
- Description bibliographic
- Includes bibliographical references (pages 115-131)
- Public Abstract (ETD)
Measurement alignment places all item parameters across multiple groups on the same scale simultaneously. The item parameters aligned by measurement alignment are used to evaluate the presence of biased items across multiple groups. These features of measurement alignment are necessary to compare test or item properties across multiple groups. Although measurement alignment is used in various fields, applicable models are somewhat limited. Considering the increasing popularity of bifactor models, this study introduced the procedures of measurement alignment for the bifactor model. A technical description of measurement alignment for the bifactor model was provided. The extended model was evaluated through empirical data analysis by comparing it with a commonly used method, Wald-2. In addition, this study evaluated the quality of estimates through a simulation manipulating several conditions.
The empirical data analysis showed that the measurement alignment provided similar results to ones of Wald-2 approach in terms of identifying biased items while the two methods produced different estimates of parameters. The simulation study highlighted that measurement alignment for the bifactor model recovered the parameters well in general. The accuracy of estimation of measurement alignment for the bifactor model was high although the stability of estimation was slightly low. These verified results of measurement alignment for the bifactor model would be utilized for research using the bifactor model. In addition, this study would be beneficial to future studies, which aim to extend measurement alignment into other multidimensional models, because this study provided the details of the procedure as an initial study of measurement alignment for a multidimensional model.
- Academic Unit
- Psychological and Quantitative Foundations
- Record Identifier
- 9984210840702771
Metrics
112 Record Views