DETECTING ITEM BIAS IN THE RASCH RATING-SCALE MODEL

This study employs simulated data to assess the appropriateness of using the separate calibration and between fit approaches to detecting item bias in the Rasch rating scale model. The development of these methods in the dichotomous Rasch model is discussed to delineate these methods from those often used in the true-score model or in multi-item parameter latent trait models. The results indicate that the Type I error rates for the null distribution hold even under circumstances in which there are different levels of ability of the reference and focal groups. The power and Type II error rates am estimated using differing levels of bias introduced by varying the numbers of persons in the fowl group, varying the proportion of biased items, and varying the magnitude of the bias introduced. The results indicate that using the traditional +/-2 critical values for these methods will produce unacceptable Type II error rates and that some inspection of the item plots and the use of other information is needed to make a final determination regarding bias.