Abstract
We use simulated data to evaluate the adequacy of the Multiple-Group (MG) extension of the Confirmatory Factor Analysis (CFA) with Mean and Covariance Structure (MACS) mo del for the detectio n of differential item functioning (DIF) in po lytomous graded response items. Two factors are varied to generate the DIF conditions: type of DIF (u niform and non-u niform) and size of DIF (low, medium and high ). For each condition, 10 0 samples (N= 800) are generated according to th e Graded Response model. Ten items with 5-response categories are consid ered, and only on e item is man ipulated to show DIF. The data are analyzed by means of LISREL 8. Results show that th e DIF detection procedure analyzed reasonably maintained control of its Type I error under non-uniform DIF conditions, but showed some difficulties under uniform-DIF con dition s. With respect to power, the procedure was satisfactory for detecting med ium and h igh uniform DIF, and lacked power for detecting n on-un iform DIF.
