Previous studies have repeatedly demonstrated the existence of item position effects in large-scale assessments. Usually, items are answered correctly more often whenadministered at the beginning of a test compared to at the end of a test. In this article, the aspects of item position effects that are investigated are their pattern, whether they remain stable over time, and whether they are affected by changes in the test administration mode. For this purpose, a Bayesian item response model for modeling item position effects is proposed. This model allows for nonlinear position effects on the item side and linear individual differences on the person side. A full Bayesian estimation procedure is proposed as well as its extension to data collected from stratified clustered samples. The model was applied to the reading data collected in the 2009, 2012, and 2015 cycles of the Programme for International StudentAssessment (PISA) for six countries.
Original languageEnglish
JournalPsychological Test and Assessment Modeling
Issue number2
Pages (from-to)241-263
Number of pages23
Publication statusPublished - 2018
No renderer: handleNetPortal,dk.atira.pure.api.shared.model.researchoutput.ContributionToJournal

    Research areas

  • item position effects, Bayesian IRT model, MCMC estimation, stratified clustered sample, PISA

ID: 917956