Comprasion of multidimensional item response theory test equating methods

MIRT observed score equating (BF-MIRT), the full MIRT observed score equating (MOSE), and the unidimensional approximation observed score equating (AOSE) methods under common item non-equivalent groups design in terms of standard error of equating (SEE), BIAS, and root mean square error (RMSE) of these equated scores across various conditions. The results aim to provide guidelines on which MIRT test equating methods yield more accurate results under specific conditions. In the study, simulation data were used. Sample size, ratio of common items, level of relationship between dimensions, calibration methods, and test equating methods are the factors that were manipulated. 162 conditions were examined. The data generation and equating studies were conducted using the R programming language. It was observed that the equated scores obtained using concurrent calibration were similar. For both calibration methods, it was found that as the sample size increased, the error values decreased. A sample size of at least 3000 is recommended. In test equating applications performed using concurrent calibration, the smallest bias values were observed when the proportion of common items was 20%. In test equating applications performed using separate calibration, the lowest error values were observed when the proportion of common items was 50%. The proportion of common items should be at most 20% in concurrent calibration and at least 50% in separate calibration. Smaller error values were observed in concurrent calibration compared to separate calibration.