Robustness of item parameter estimation programs to assumptions of unidimensionality and normality

The effects of test dimensionality (one- or three-dimensional), theta distr ibution shape (normal, positively skewed, or platykurtic), and estimation p rogram (BILOG, MULTILOG, or XCALIBRE) on the accuracy of item and person pa rameter estimates were assessed. The criterion was the root mean squared er ror of the difference between estimated and true parameter values. There wa s an interaction between program and dimensionality, indicating that the ro bustness of the unidimensionality assumption was a function of the estimati on program. With the sample size and test length used, unidimensional estim ation programs were insensitive to different shapes of the underlying theta distribution. BILOG consistently produced the smallest root mean squared e rror under most conditions. However, MULTILOG and XCALIBRE showed less vari ance in parameter estimation due to the violation of unidimensionality, wit h the exception of estimating the discrimination parameter in MULTILOG. Gui delines for estimating parameters of multidimensional test items using unid imensional item response theory models are suggested.