Differential Item Functioning Analysis Models in Large-Scale Assessments

• In Event: Roundtable Session 28
  In Roundtable Session: Topics in Differential Item Functioning

Abstract

The current research introduced two approaches for the incorporation of examinee characteristics into multilevel DIF analyses in large-scale assessments: multilevel Mantel-Haenszel (MMH) procedure with propensity score model and the MMH procedure with regression-based covariate control. We examined whether two proposed approaches would perform equally to detect DIF, when applied to large-scale assessments. PISA 2009 reading literacy test and survey data were used. Reading strategies and activities in which significant gender differences were found were employed as examinee characteristics. Pooled booklet method was applied to MMH procedures for the balanced incomplete booklet design that large-scale assessments often adopt. The results indicated that PSM and RBM did not perform equally in the detection of DIF. The PSM approach tended to detect more items as DIF, compared with PSM approach.

Authors

• HyeSun Lee, University of Nebraska - Lincoln

• Kurt F. Geisinger, University of Nebraska