Research Of Ghci In Cognitive Diagnostic Assessment

Cognitive models, which describe the cognitive process of students, are the foundation of cognitive diagnostic assessment (CDA). In education practice of CD A, it is crucial to make sure that the cognitive model is accurate and reasonable. The cognitive model can affect the accuracy of cognitive diagnosis directly. Sevral validation indices of cognitive model have been developed for dichotomous data at present. In the polytomous test situation, transforming data into dichotomous data to evaluate accuracy and adequacy of the cognitive model will be inaccurate for the missing information. On the basis of previous research, this paper extends the HCI (Hierarchy Consistency Index) for dichotomous data to GHCI (Generalized Hierarchy Consistency Index) for polytomous data, and discussed the GHCI’s stability and accuracy.This study includes five parts. First, I developed GHCI, illustrated its feasibility in different score situation, illustrated its characteristics. Second, using simulation study, I compared the GHCI and HCI under different conversion ratio, and illustated the HCI along with the change of transfomation ratio in the polytomous test situation. Third, I investigated the GHCI’s power in detecting misspecified attribute hierarchy, include attribute to wrong layer, attribute upside down, reducing hierarchy, increasing the hierarchy. Fourth, I investigated the GHCI’s power in detecting misfitting response. Fifth, I investigated the performance of GHCI in education practice. GHCI was used in analyzing a sample of 888 fifth grade students’ response for arithmetic word problem solving to validate the accuracy of the cognitive model and detect abnormal response. The results are as follow:(1) The extension to GHCI is feasible and reasonable in this paper. GHCI is available in different polytomous test situations and in mixed scoring data. Compared with the HCI, GHCI has certain advantages and is more detailed and accurate in evaluting the students’ response.(2) In the simulation study of four attribute hierarchies in the polytomous test, the mean of GHCI is greater than HCI of different transformation ratio significantly. HCI underestimates the consistency between the students’response and the cognitive model. HCI is affected by transformation ratio. HCI changes differently as the transformation ratio increases in each hierarchy of attributes.(3) GHCI’s can reflect the misspecification of attribute hierarchy. GHCI drops more and more severe as the degree of misspecification increses in each types of model misspecification. GHCI’s sensitivity varies with types of model misspecification. GHCI drops more severely when attribute is hierarchical wrongly, attribute is upside down, and attribute hierarchy increases than when attribute hierarchy decreases.(4) GHCL HCI each has its advantages in detecting abnormal response. GHCI had high power at detecting creative responding, and HCI had high power at detecting had high power at detecting model misspecification. GHCI and HCI’s power tended to increase with the number of items. GHCI and HCI’s power at detecting creative responding tended to increase with the item discrimination. The GHCI and HCI’s power at detecting model misspecification with specific item discrimination varies with the number of items.(5) The mean of GHCI in analyzing students’response for arithmetic word problem solving is.757 greater than the cut-score, which validate the accuracy of the cognitive model specified. The cognitive model correctly described the cognitive processes used by students to solve arithmetic word problem. GHCI’s cut-score waves in a small range which means GHCI is more stable. The abnormal response detected by GHCI is different from by HCI.(6) In empirical research, the mean of HCI decreased as the transformation ratio increased. The mean of GHCI is significantly greater than HCI in different transformation ratio. HCI underestimates most students’ degree to which an observed examinee response is consistent with the attribute hierarchy. HCI also underestimates the accuracy of the attribute hierarchy. GHCI is is more detailed and accurate in evaluting the consistency between student’s response and the cognitive model specified.