The comparison of kernel equating methods: TIMSS 2019 science test case

Test equating studies are generally performed using common items. However, in practice, there is no common substance in the tests. In this case, equating through common variables can contribute to equating applications in practise. For this purpose, this study, which used binary scoring items, carried out in nonequivalent groups with common items design (NEAT) and nonequivalent groups with covariates design (NEC). When using methods Kernel poststratification equipercentile equating, Kernel poststratification linear equating, Kernel chain equipercentile and Kernel chain linear equating methods under NEAT design, were used Kernel eqipercentile and Kernel linear equating methods in NEC design. On the other hand, gender and computer/tablet variables are used as common variables in the NEC design. This study aimed to examine the amount of errors that will be given by whether or not you have a computer or tablet considering the introduction of eTIMSS in 2019. Test forms are equated via the kequate package in program R and the standard error of equating (SEE) was calculated. When the results of the research are evaluated, it was determined that the equated scores obtained from the NEC design were closer to the raw scores. On the other hand, the error values obtained from the NEC design were found to be higher than the NEAT design. According to the results, the least error was obtained from poststratification equating in NEAT design. The lowest error in the NEC design was achieved in the Kernel linear equating method with the computer/tablet variable.