Bias-Corrected RMSD Item Fit Statistic via SIMEX

Source abstract

This study evaluates the simulation extrapolation (SIMEX) method as a bias-correction approach for the distribution-weighted and difficulty-weighted root mean square deviation (RMSD) item fit statistics. The results indicate that SIMEX reduces the positive bias of the original RMSD statistic and can be applied in the context of differential item functioning (DIF) analysis. Although the SIMEX-based RMSD statistics showed slightly greater bias than previously proposed analytic corrections, they yielded lower RMSE for items with DIF. For items without DIF, the analytic bias-correction methods performed better with respect to both bias and root mean square error (RMSE). An empirical example further showed that the SIMEX-based and analytically bias-corrected RMSD statistics produced very similar estimates.