A Comparative Analysis of Accuracy in Ability Estimation Methods in Multistage Adaptive Testing(MST)

The purpose of this study is to provide direction for establishing ability estimation methods in the design process of the multistage adaptive testing(MST) by reviewing methods with high accuracy and consistency when each subtest(module) concludes in MST. Accordingly, a simulation experiment was conducted to compare the accuracy of ability estimation methods based on item response theory in a 1-3-3 MST design. Using MSTGen in the simulation experiment, item and ability parameters were generated, and ability estimates and correlations among parameters were calculated under different test conditions. The results indicated that accuracy was lowest when ability was estimated using the Maximum Likelihood Estimation(MLE) method at all stages, while Bayesian methods(EAP, MAP) showed the highest accuracy. Furthermore, consistently applying Bayesian methods for ability estimation at all stages resulted in higher accuracy and smaller errors compared to applying different estimation methods at each stage.