Extensions and applications of higher-order item response theory models

Chi-Ming SU, Wen Chung WANG

Research output: Contribution to conferencePaper

Abstract

The study generalized unidimensional second-order IRT models to multidimensional higher-order IRT models and developed CAT algorithms. Due to high dimensionality of the new models, we proposed to use Bayesian MCMC methods for parameter estimation. The results of simulations show that the parameters can be recovered fairly well using the computer WinBUGS. Furthermore, both the PsBF and the DIC were sensitive in model comparison across a variety of conditions. The CAT algorithms of the new models were successfully developed. The simulation results indicate that the progressive method and the alpha-stratified method, although increasing bank usage, did not always maintain item exposure rates at a prespecified level, whereas the Sympson and Hetter online freeze method did. In practice, it is recommended that both the progressive method and the Sympson and Hetter online freeze method are implemented to maintain both item exposure and bank usage.
Original languageEnglish
Publication statusPublished - 2011

Fingerprint

Parameter estimation

Citation

SU, C.-M., & Wang, W. C. (2011, July). Extensions and applications of higher-order item response theory models. Paper presented at the 76th Annual and the 17th International Meeting of the Psychometric Society, The Hong Kong Institute of Education, China.