Abstract
Logistic regression: analytic strategies in differential item functioning detection. The present study compares several analytic strategies, whose relative efficacy has yet to be evaluated, for detecting Differential Item Functioning (DIF) by means of logistic regression. The strategies are compared by checking different item types in which various DIF conditions are simulated. In all cases 75-item, dichotomous response tests were used. Three types of DIF (uniform, symmetric non-uniform and asymmetric non-uniform) were simulated with three conditions for the amount of DIF (0.0, 0.3 and 1.0 for the difficulty parameter, and 0.0, 0.25 and 1.0 for the discrimination parameter). In general, and according to the criteria of greatest power in detecting DIF and low computational cost, it is recommended that applied psychologists and educators who analyse, translate and adapt tests, use logistic regression analysis with the strategy that checks for the presence of both uniform and non-uniform DIF; this should be complemented by calculating interaction significance in order to distinguish between the two types of DIF.
