In the thesis we aim to analyze the composition of latent trait space of the reading part of HSK (Chinese Proficiency Test), in order to provide some evidence to construct validation of HSK. We choose five compensatory Multidimensional Item Response Theory (MIRT) models in which the number of dimensions is from two to six respectively. The Joint Maximum Likelihood Estimation method is used to estimate the item and trait parameters. After that, the research will check the fit-of-goodness of the models in three aspects: inter-item residual covariance matrix, residuals and chi-square. In the process, we try to use Cluster Analysis Technology to divide the persons into some groups. The research results show that the three dimensions model is the best, and that the traits are of different importance towards different items. No.1 trait has the best impact on differentiating persons in contrast to No.2 trait.
|