Some New Classification Methods In Attribution Hierarchy Model

In comparison with the Standard Test Theory, a New Generation of Test can provide the detailed information of the examinees. As the focus of a New Generation of Test Theory, Cognitive Diagnosis is widely concerned by domestic and foreign researchers. Cognitive Diagnostic Models(CDMs) help to assess the test, get individual details. Attribute Hierarchy Method (AHM) proposed by Leighton et al.(2004) is a Cognitive Diagnostic Model based on statistical pattern recognition and classification. According to the identified hierarchical ordering of attributes, AHM will be assigned the Observed examinee Response Patterns(ORP) on items to the Expected examinee Response Patterns(ERP), to be estimate examinees'attribute mastery patterns associated with different cognitive skills and further help design effective and targeted remedy for individual examinees.There are two classification methods with AHM ,named Method A and Method B. Some investigation, discussion and research on these two kinds of method are carried out; some defects of the two methods are analyzed; several kinds of new classification methods based on AHM are proposed in this paper. The new classification methods proposed is based on establishing a series of indices of the similarity between the expected response pattern (ERP) and the observed response pattern (ORP). In order to promote the correct rate of the classification, two factors are considered carefully: the probability of ERP (or ORP), i.e. the likelihood of the ERP (or ORP), and the probability of coinciding part of ERP and ORP. Then a series of new classification methods are developed by combining the similarity index and the two kinds of the probabilities mentioned above. The new classifications are partitioned into two parts. One is the analog to Method A (new A methods), another is the analog to Method B (new B methods). The performance of all the methods mentioned above has been judged in the simulation research. According to the simulation method brought forward by Ying Cui et al.(2006) carries out the condition of simulating an experiment , under the condition that guessing and slipping(Leighton et al. call them slip) are drawn from the uniform distribution on (0 , 1) and the number of the expected response pattern drawn from the normal distribution. The results indicate that all kinds of new A method have wide rise in attribute pattern classification rightness rate than Leighton's method A, and with the slips increasing the advantage is especially obvious; all kinds of new B method have obvious advantage on pattern classification rightness rate than Leighton's method B; for the marginal attribute rightness rate, all the new methods are also a bit more excellent than method A and method B brought forward by Leighton et al.