Testing for Item Level Disparities Using Zero-Inflation Models in NAFS

• In Event: AERA Roundtable Session Sunday 1:45 pm - Gold 3
  In Roundtable Session: RT1: Measurement and Related Issues in Quantitative Research Designs, Analysis, and Simulation

Abstract

The present study proposes an enhanced Zero-Inflated (ZI) Mixture IRT framework to validly estimate person ability scores in the presence of an excessive number of zeros. Using a sample of 6th grade and the measurement of mathematical operations results highlighted the superiority of the proposed model over traditional IRT approaches. A two-class solution significantly improved model fit, capturing a distinct zero-inflated subgroup and was associated with significant reductions in the bias of estimating trait scores. Additionally, inclusion of covariates revealed significant effects of gender and language of testing on mathematics achievement, with females and exhibiting lower performance compared to males and international students, respectively. The study suggests the value of the ZI model in enhancing psychometric accuracy and fairness.

Authors

• Fathima Mohrrag Jaffari, National Center for Assessment

• Georgios D. Sideridis, Harvard University