Empirical Research Background: Computerized Classification Test (CCT) is efficient in classification of examinees into a few categories. Most studies of CCT were based on unidimensional item response models. In practice, a person might have to be classified into categories on multiple latent traits. For example, a mastery examination may consist of multiple subjects of mathematics, language and science. It is thus desirable to develop a CCT that is based on multidimensional item response models so that the correlation between latent traits can be considered to improve accuracy of classification. Empirical Research Aims: To develop a multidimensional CCT and compare its accuracy and efficiency with a unidimensional CCT in classification of examinees into a few categories when the latent traits are correlated. Empirical Research Sample: 500 simulees were generated from a multivariate normal distribution. Empirical Research Method: Response data were simulated according to the between-item multidimensional Rasch model. A multidimensional CTT and a unidimensional CTT were then performed to classify examinees. Person latent trait estimate was based on maximum a posteriori (MAP) and item selection was based on the Fisher information. The two independent variables were (a) number of dimensions (2 and 5), and (b) correlation between dimensions (.2 and .8). The two dependent variables were (a) correct classification rate (accuracy), and (b) average number of items administered (efficiency). Empirical Research RASCH: The between-item multidimensional Rasch model was used. Empirical Research Results: The whole simulation is not yet finished, however, partial results revealed that the multidimensional CCT was more accurate and efficient than the unidimensional CCT, when the correlation between dimensions was high and the number of dimensions was large. Empirical Research Conclusions: .... multidimensional CCT was more accurate and efficient than the unidimensional CCT, when the correlation between dimensions was high and the number of dimensions was large.
|Published - 2009