Abstract
A procedure for detecting DIF in polythomous and dichotomous items. In the present paper a modification of a DIF detection procedure is presented. The DIF procedure was proposed by Linn and Harnisch (1981) and used in applied settings by Barbero, Sánchez Bruno, Prieto and San Luis (1995), Prieto and Barbero (1996), Prieto, Barbero and San Luis (1997), Barbero and Prieto (1997), among others. The purpose of this modification was to be able to detect adequately the DIF when some items from the test don't fit to the data in the reference group. Basically, the procedure consists in several steps: first, to check the fit of the model to the subject responses in the focal group. Second, to estimate the item parameters in the reference group and to check the fit of these parameters in the focal group and, lastly, to calculate the difference between both fit indices in each item and to compare with a critical value obtained throught a simulation procedure. This procedure can be used with both polythomous and dichotomous items. In this stimulation study, the procedure was applied to polythomous items and four factors were manipulated: percent of DIF presented in each item, type of DIF, percent of the items with DIF and significance level. The results indicates that the proposed procedure is effective in the detection of DIF, mainly when the DIF is non-uniform and mixed. Likewise, the rates of false alarms was very low, specially if it is compared with the rates from other simulation studies with similar features.
