Abstract
Detection of non-uniform DIF: Mantel-Haenszel and logistic regression methods. The present simulation study compared the performance under different conditions of the Mantel-Haenszel (MH) statistic, the variation of the MH statistic proposed by Mazor, Clauser and Hambleton (1994), and the logistic regression (LR) on detection of nonuniform differential item functioning (DIF). Simulated data were obtained from a real study (Ferreres,1998). Item parameters used in the simulation are from a aplication of Verbal Mental Ability Scale of Middle BADYG (Yuste,1988). Variables manipulated were: sample size, differences in the group trait level averages, type of nonuniform DIF, proportions of DIF items and signification level. These methods were evaluated by examining changes in the power and the Type I error rate over 100 replications. Results obtained ha ve contributed to clarify the utility of these statistical methods in the presence of nonuniform DIF. Conclusions to consider in future studies of the deteccion of nonuniform DIF are given.
