Evaluating Intersectional DIF via Logistic Regression: A Simulation Study (Poster 2)

• In Event: AERA Poster Session Saturday 3:20 p.m.
  In Poster Session: SIG-Educational Statistician Poster Session

Abstract

This simulation study explores logistic regression (LR) for detecting intersectional Differential Item Functioning (DIF). Previous studies examined intersectional DIF using state assessment data. Analyses used many pairwise comparisons rather than cohesive model-based analyses and lacked attention to error issues. We use LR analyses to explore intersectional DIF in a new way, examining power and error rates via a simulation study. Preliminary results showed that a LR model including interactions detected intersectional DIF. Rejection rates were high when DIF was present, especially for items with high discrimination and difficulty values. Rates were lower for items without DIF, but 12 out of 90 no-DIF rejection rates were greater than 8%, suggesting a tendency to over-detect DIF when none is present.

Authors

• Wenyi Li, Florida State University

• Saeyan Yun, Florida State University

• Yiting Yao, Florida State University

• Niaz Morshed, Florida State University

• Gertrudes Velasquez, HumRRO

• Jielin Ming, Aretum Inc.

• Yachen Luo, HumRRO

• Salih Binici, Florida Department of Education

• Yanyun Yang, Florida State University

• Betsy J. Becker, Aretum, Inc.