Detecting Local Dependence: A Threshold-Autoregressive Item Response Theory (TAR-IRT) Approach for Polytomous Items

Xiaodan Tang*, George Karabatsos, Haiqin Chen

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

In applications of item response theory (IRT) models, it is known that empirical violations of the local independence (LI) assumption can significantly bias parameter estimates. To address this issue, we propose a threshold-autoregressive item response theory (TAR-IRT) model that additionally accounts for order dependence among the item responses of each examinee. The TAR-IRT approach also defines a new family of IRT models for polytomous item responses under both unidimensional and multidimensional frameworks, with order-dependent effects between item responses and relevant dimensions. The feasibility of the proposed model was demonstrated by an empirical study using a polytomous response data. A simulation study for polytomous item responses with order effects of different magnitude in an education context shows that the TAR modeling framework could provide more accurate ability estimation than the partial credit model when order effect exists.

Original languageEnglish (US)
Pages (from-to)280-292
Number of pages13
JournalApplied Measurement in Education
DOIs
StatePublished - 2020
Externally publishedYes

ASJC Scopus subject areas

  • Education
  • Developmental and Educational Psychology

Fingerprint Dive into the research topics of 'Detecting Local Dependence: A Threshold-Autoregressive Item Response Theory (TAR-IRT) Approach for Polytomous Items'. Together they form a unique fingerprint.

Cite this