The Study of Relationships between BW Item-fit Indices and DIFs

103

The purpose of this study was to examine the relationship between BW Item-fit indices and DIF. Based on WinGen3.0, response-matrices were simulated by Monte Carlo Simulation. Response-matrices were analyzed by Winsteps and WBStar, and then the degrees of DIF, w index (hint), b index (disturbance), c index (difficulty), and m index (misfit) could be evaluated. The relationship between DIF and BW Item-fit indices was examined by general linear regression. The research was using ANOVA and t test to analyze whether there was difference of relationship at different conditions. Finally binary logistic regression was used to assess the accuracy in which items was classified into a “DIF” or “normal” group by the BW Item-fit indices. The results were demonstrated as following: 1. As a whole, R2 of general linear regression was .224. The regression coefficient of indices of w, b, c and m were .150, .079, .411 and .256, respectively. The difficulty of item made the most influenceon DIF. 2. Under varied conditions, the R2 between BW Item-fit indices and DIF existed significant differences. 3. The accuracy rate in which items were classified into a “DIF” or “normal” group by BW Item-fit indices reached 94%. Based on the results of this study, the researcher provided some suggestions for practitioners and future research.