Comparison of some detection methods on unidimensional IRT calibration
Abstract
Details
- Title: Subtitle
- Comparison of some detection methods on unidimensional IRT calibration
- Creators
- Lu Wang
- Contributors
- Robert Ankenmann (Advisor)Catherine Welch (Committee Member)Deborah Harris (Committee Member)Stephen Dunbar (Committee Member)Kate Cowles (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Psychological and Quantitative Foundations
- Date degree season
- Autumn 2019
- DOI
- 10.17077/etd.005181
- Publisher
- University of Iowa
- Number of pages
- xiii, 147 pages
- Copyright
- Copyright 2019 Lu Wang
- Language
- English
- Description illustrations
- color illustrations
- Description bibliographic
- Includes bibliographical references (pages 90-95).
- Public Abstract (ETD)
Currently, unidimensional item response theory (UIRT) models are widely employed in calibration and scoring processes in large-scale tests. However, successful applications of UIRT models rely heavily on whether the unidimensional IRT assumptions are satisfied. When a test is intended to assess examinees’ ability levels, but time constraints are imposed, examinees’ test performance is affected not only by their ability levels, but also by how quickly they answer the test questions. That is, a speed component plays a nonnegligible role in affecting examinees’ test performance. Under such circumstances, using UIRT models could lead to a violation of both unidimensionality and local independence assumptions of the UIRT models (Hambleton & Swaminathan, 1985; Yen, 1993).
To avoid speededness effects on the application of UIRT models, several speededness detection methods were proposed to identify either speeded responses or examinees. This study compared the performance of various speededness detection methods, combined with various calibration treatments in reducing speededness effects under both 2PL and 3PL IRT models. The study found that item parameter estimates were more affected than person parameter estimates by speededness effects. In addition, treating the identified speeded responses as “not-presented” in calibration provided more accurate parameter estimates compared to using the other calibration treatments across various speededness levels under both the 2PL and 3PL IRT models. Moreover, the detection methods that were based on response time distribution to identify speeded responses showed better performance than the other detection methods in dealing with speededness effects.
- Academic Unit
- Psychological and Quantitative Foundations
- Record Identifier
- 9983779398002771