Many achievement tests, particularly in large-scale assements, deal with measuring abilities that are themselves assumed to be composed of other more specific abilities. Many models were used for analyzing these data. First, using the unidimensional model, it neglects the correlations between items and causes an overestimate in examinee’s ability. Second, the correlations between items are taken into account by using the multidimensional model, but it neglects the impact of local item dependence (LID). Third, the LID is accounted by the testlet model, and assuming the testlect effects are zero. This paper was accounts for the situation of testlect effects are not zero and proposed the subdimension model. Introducing the relationships between the sundimension model with multidimensional model and testlet model, and it is also a pesical case of the multidimensional random coefficients multinomial logit model. Demonstrating how to use the ConQuest for analysis and give an empirical example in finally.
Comment
1. In the “ESTIMATION USING CONQUEST” part, it is very clear to demonstrate and describe commends for how to use ConQuest. It is useful for us to learning and implementing our studies to ConQuest.
2. No simulation study for evaluate the performance of the subdimension model, just using an empirical data for analysis, why can says the subdimension model is outperformed other models?