An investigation of test and sub-tests equating in terms of different conditions

The main purpose of this study is to compare the results obtained from the equating of a multidimensional test with the various application steps at test level and at sub-test level in terms of evaluation criteria. The research was conducted using simulation data. Since the methods and conditions that minimize equating error are tried to be determined in the research, the study is a basic research. In this research, common-item nonequivalent groups (CINEG) design was used to equating the two test forms. The equating process was carried out in six applications according to the execution of the equating process at test level and sub-test level with different parameter estimation paths. In the scope of the research, the performance of the six different applications in which the equating process was conducted was examined according to the relationship level between sub-tests, number of items in sub-tests, common item ratio in sub-tests, sample size, difficulty level between tests and sub-tests and scale conversion methods. Under these conditions, equating processes were realized by the IRT true score equating method. RMSE (equating error), BIAS (equating bias) and SE (standard error) values were calculated for item and ability parameters in order to examine the accuracy of equating results. In the scope of the research, R software was used for data generation, IRTPRO 4.2 was used for estimating item and ability parameters, IRTEQ was used for unidimensional equating and LinkMIRT was used for multidimensional equating. After estimating the parameters of multidimensional test according to unidimensional 3PL MTK model, the highest error values were reached in equating. These error values are followed by the error values obtained from the equating process after estimating the parameters according to the multidimensional 3PL MTK model. In this study, the smallest error values were obtained from the equating after estimating the parameters of each subtest according to unidimensional 3PL MTK model.