Evaluating the Multiple-Group Mean and Covariance Structure Analysis model for the detection of Differential Item Functioning in polytomous ordered items

reat to the validity of a test. Both the APA (American Psychological Association) and the ITC (International Test Commission) standards emphasize the necessity of checking for DIF in order to guarantee the fair use of a test (AERA, APA, NCME, 1999; COPITC, 2000). With these recommendations in mind, and with the intention of spreading the application of DIF analyses, researchers should develop and assess DIF detection methods that would be easily implemented by practitioners lacking a highly technical or statistical background. Focusing on polytomous graded response items, a number of methods have been proposed to detect both uniform and non-uniform DIF (see Potenza & Dorans, 1995; Hidalgo & Gómez, 1999). One of these methods is based on a popular technique among psychologists: Factor Analysis. In this paper, we evaluate a FactorAnalysis-based DIF detection method: the Multiple Group Confirmatory Factor Analysis with Mean and Covariance Structure (MG-CFA-MACS). In the last few years, the MG-CFA-MACS has attracted the attention of DIF researchers and has been frequently applied for the evaluation of both uniform and non-uniform DIF on polytomous items with ordered response alternatives (Everson, Millsap & Rodriguez, 1991; Byrne, 1998; Tomás, González-Romá, & Benito, 2000; Chan, 2000; Wasti, Bergman, Glomb, & Drasgow, 2000). Taking into account that all these studies have used empirical data, the extent to which using a continuous response model such as the MG-CFA-MACS is adequate for the correct detection of DIF on polytomous ordered response items must be clarified. The present study addresses this issue using simulated data.