幾種改良式貝氏估計法在多向度電腦化適性測驗與多向度電腦化分級測驗中的效能

Latent trait estimation methods are important technique in item response theory (IRT). The traditional estimation methods were modified by recent researchers in order to improve their disadvantages. For example, Chen(2009) suggested to use the weighted MAP & EAP (EMAP & WEAP) to attenuate the regression bias. Raiche, Blais, & Magis(2007) suggested to use the adaptive MAP & EAP(AMAP & AEAP) as well. The goals of the research are to compare the efficiency of six Bayesian estimation methods on computerized adaptive testing (CAT), multidimensional CAT (MCAT), computerized classification testing (CCT), multidimensional CCT (MCCT). Results indicated that weighted maximum a posterior (WMAP) estimation is the most efficient method which can also reduce the regression error of Bayesian estimation. The results of CCT are correspondent with the results of CAT, which means that the higher the reliability of CAT, the higher the accuracy of CCT. When using MCCT for categorization, compensatory loss function yields higher accuracy of categorization than conjunctive loss function. Besides, the higher the correlation between latent traits, the higher the categorization accuracy. Application and suggestion of choosing the ability estimation methods in CAT, MCAT, CCT and MCCT are addressed in this research.