Scale Transformation for Multidimensional Tests According to Unidimensional and Multidimensional Methods

In this study, it was aimed to examine the performance of scale transformation methods using MIRT and UIRT. For this purpose, two-dimensional and simple structured test data were generated. The conditions addressed in the study are sample size, interdimensional correlation, common item ratio, difference in ability distribution and parameter estimation model. Mean-mean, mean-sigma and Stocking-Lord methods were used as scale transformation methods and RMSE was taken as an evaluation criterion. RMSE values were estimated as a result of the scale transformations using NEAT design and compared with the independent samples t-test. Lower RMSE values were estimated in unidimensional methods for parameter a, multidimensional methods for parameter b and ability parameter, and statistically significant differences were found in favour of unidimensional methods for parameter a and multidimensional methods for parameter b and ability parameter. Lower RMSE values were obtained in conditions with high sample size and common item ratio. Multidimensional methods performed better when the interdimensional correlation was low, and unidimensional methods performed better when the interdimensional correlation was high. Lower RMSE values were estimated when the difference in ability distribution was low.