Assessing Partial Knowledge in the Rasch Model With Fixed Random Guessing Parameter: A Modified 1PL-AG Model

• In Event: AERA Poster Session 14
  In Poster Session: Division D Section 1 Poster Session 1

Abstract

A modified ability-based guessing model is developed and applied to the TIMSS 2015 datasets in mathematics and sciences under the Bayesian framework. The modified 1PL-AG model believes the probability of a correct guess depends on the latent ability and distinguishes random guessing from educated guessing based on partial knowledge. With a simulation study, parameter recovery of the modified 1PL-AG model was investigated. Followed by a real data analysis with the TIMSS 2015, this study also examined the relative model fit. The results show that the modified 1PL-AG model can recover the parameters of interest and has real-life implications. That is, takers utilize their incomplete knowledge to guess in multiple-choice tests.

Authors

• Jiaqi Zhang, SKT Education Group

• Paul De Boeck, The Ohio State University

• Jorge González, Pontificia Universidad Católica de Chile