The Effect of the New Variance-Covariance Estimation Method on Differential Functioning of Items and Tests: BILOG-MG3 Versus IRTPRO

• In Event: Roundtable Session 6
  In Roundtable Session: Differential Item Functioning: Methods and Applications

Abstract

A new calibration program for item response theory (IRT), IRTPRO (Cai, Thissen, & du Toit, 2011), claims that estimation of the covariance matrix is improved by using the supplemented expectation maximization (SEM) algorithm (Cai, 2008). The purpose of this paper was to investigate the effect of the new variance-covariance estimation method on the performance of NCDIF, a DIF technique in the DFIT framework (Raju, van der Linden, & Fleer, 1995). The comparison was made between BILOG-MG3, the calibration program used in the past, and IRTPRO in terms of DIF detection using NCDIF. Results showed that IRTPRO in fact produced better variance-covariance estimates than BILOG-MG3, and the performance of NCDIF was enhanced by using IRTPRO.

Authors

• T. C. Oshima, Georgia State University

• Keith Darnell Wright, Secondary School Admission Test Board

• David R J Fikis, Georgia State University

• Scott B Morris, Illinois Institute of Technology