Date of Award

Spring 2007

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Psychology

Committee Director

Terry L. Dickinson

Committee Member

Valerian J. Derlaga

Committee Member

James M. Henson

Committee Member

Robert M. McIntyre

Abstract

The present Monte Carlo study compared four confirmatory factor analysis (CFA) methods for detecting differential item functioning (DIF). The four methods were the noniterative and iterative mean and covariance structure analysis (MACS) methods, the modification index (MI) method, and the modification index-divided sample (MI-divided) method. Reference and focal groups responded to 12 items with 3 of the 12 items designed to exhibit DIF. Sample sizes of 250 and 500 were examined. In addition, three types of DIF were examined: DIF on loadings, DIF on thresholds, and DIF on both loadings and thresholds. Results indicated that for sample size 250, all methods had good DIF detection rates for DIF on thresholds and for DIF on loadings and thresholds; all methods were not sensitive to DIF on loadings. For sample size 500, all methods had good DIF detection rates for DIF on thresholds and for DIF on loadings and thresholds. With the greater sample size, the noniterative and iterative MACS methods were more sensitive to DIF on loadings. The MI-divided method improved to a lesser degree, but the MI method did not improve at all. For DIF on thresholds for sample size 500, the noniterative MACS, MI, and MI-divided methods had false positive rates that were greater than expected by chance. Only the iterative MACS method maintained the false positive rates at or below that expected by chance.

DOI

10.25777/kdec-gp98

ISBN

9780549069539

Share

COinS