Abstract
Erroneous detection of Differential Item Functioning. A comparison of methods. Some methods to analyze Differential Item Functioning (DIF) are compared in terms of their false positive rates, namely the normal test (NT) for the difference of difficulty parameters, the Mantel-Haenszel chi-square test and the item classification in 'Type C' or 'Not Type C' according to the only magnitude of the MH DDIF statistic (empirical classification) -the last two on both a single-stage and two-stage proceduresand according to ETS criterion. They were applied to data without DIF which were simulated under different design conditions with respect to sample size, impact and item parameter configurations. The results obtained from two-stage and single-stage procedures were similar when impact is present. The method that generally produces lower false positive rates is the NT one, followed by both types of classification which, holding the same computational cost as Mantel-Haenszel test, have a much lower failure rate in the erroneous detection of DIF.
