Denk Olmayan Gruplarda Ortak Madde Deseni Kullanılarak Madde Tepki Kuramına Dayalı Eşitleme Yöntemlerinin Karşılaştırılması

Abstract

The purpose of this research was to equate the test forms which were constructed in different conditions (sample size, ability distribution, length of test and model type) through scaling methods based on item response theory (mean-mean, mean-sigma and Stocking Lord) and to compare the results obtained from these methods. In addition, it was also investigated that which method gave better results under different conditions. The research was conducted with using dichotomous simulated data which was consistent with two and three parameter logistic model. In order to equate two test forms “the common-item nonequivalent groups” was used in this research. WINGEN3 program was utilized for data generation and 50 replication were done for 36 different condition used in this research. PARSCALE 4.1 was utilized for the prediction of item and ability parameters and IRTEQ was utilized for test equating and scaling in separate calibration. Item and ability parameters were predicted through expected a posteriori (EAP) in this research. The results obtained from this simulation study were evaluated based on equating bias (BIAS) and equating error (RMSE) criterions. The results revealed that, when the conditions evaluated generally, the best equating occurred in 3000-subjects samples, 80-item tests, groups have similar ability distribution, using 2PLM and SL methods. Moreover, the results indicated that methods had less biased and less equating errors when large sample sizes together with long tests were used in groups which had similar ability distributions under the conditions considered in this research.