Examining the Dimensionality of IELTS Academic Reading: A Rasch Testlet Model Approach to Local Item Dependence

This study investigates the presence and impact of testlet effect in the IELTS Academic Reading test using the Rasch Testlet Model (RTM), with particular attention to local item dependence (LID) effects. While the IELTS test is widely used for academic and immigration purposes, its undesirable testlet variance has received limited scrutiny. Given the passage-based structure of the reading test, multiple items are grouped into testlets, potentially introducing local item dependence (LID) and violating the unidimensionality assumption of IRT models. To address this, the study compares the performance of the standard dichotomous Rasch model with the Rasch Testlet Model, which accounts for within-testlet dependencies. A dataset of ۴۰۷ test takers was analyzed using the TAM package in R. Findings show that the RTM provided superior model fit across all statistical indices (−۲LL, AIC, BIC, CAIC), and that ignoring LID resulted in an inflated reliability estimate (.۸۷) compared to a more accurate value (.۷۷) when LID was modeled. Notably, Testlet ۲ exhibited variance exceeding that of the general reading dimension, indicating substantial local dependence. These results provide empirical evidence that the IELTS Academic Reading test is not strictly unidimensional and suggest that testlet modeling is essential for fair and valid score interpretation. The findings have implications for test development, validation, and policy makers, underscoring the importance of test score interpretation high-stakes language assessments.

کلیدواژه ها:

IELTS ، testlet effect ، Rasch Testlet Model ، Local item dependence ، IRT