A Comparison of Several Item Response Theory Calibration Programs With Implications for Equating

• In Event: Factors Impacting Equating Results

Abstract

The purpose of the current study was to compare item parameter estimation results from four different Item Response Theory (IRT) calibration programs (i.e., MULTILOG, PARSCALE, flexMIRT, and IRTPRO) and determine whether differences in results subsequently influence IRT true score and observed score equating relationships. Four IRT model combinations were fit to data from a national placement exam that consisted of both multiple-choice and free-response items. The two- and three-parameter logistic models were used in combination with Samejima’s (1969) graded response model and Muraki’s (1992) generalized partial credit model. Comparisons were made on the basis of (a) parameter estimation results, (b) MC, FR, and test characteristic curves, (c) MC, FR, and test information functions, and (d) estimated equating relationships.

Authors

• Jaime Leigh Peterson, University of Iowa

• Mengyao Zhang, The University of Iowa

• Seohong Pak, The University of Iowa

• Shichao Wang, University of Iowa

• Wei Wang, Educational Testing Service

• Michael J. Kolen, University of Iowa

• Won-Chan Lee, University of Iowa