Research On Factors Influencing Diagnostic Accuracy In AHM And DINA

Cognitive diagnosis are often applying statistical pattern recognition methods to classify the knowledge state of examinees. And the attribute hierarchy method (AHM; Leighton, Gierl, & Hunka, 2004; Gierl, Cui, & Hunka, 2007) more emphasis on attribute hierarchy; the DINA mode(lDeterministic Inputs, Noisy"And"Gate Model)adds the slipping parameter(Slipping) and guessing parameter(Guessing)of item itself to the parameters assigned to each item. DINA and AHM are two models relatively widely used in cognitive diagnosis. They are both conjunctive model, that is, in order to correctly answer the item, the examinee must master all the attributes the item requires; otherwise he cannot make a correct response.For the two models, when using different classification methods, classification accuracy will vary accordingly. We can ask the following questions: "Using the same test of different models, what will happen to their classification accuracy? When measuring the same group of examinees the test making are of different types, what will happen? How much dose the optimal test length to set, regarding to the number of attributes need to be examined?"In this paper, Monte Carlo simulation is used to determine the various factors that affect the classification accuracy. This paper mainly analyzes test construction (TC) and test-length(TL), type of attribute hierarchy(TOAH),cognitive diagnosis model (CDM), quality of item (QOI), and other factors on the impact of classification diagnosis.Those factors will affect the accuracy of diagnosis to some extent. The paper also talked the classification accuracy under the influence of different factors, based on comparing the two diagnosis models AHM and DINA. The results show different factors having varying degree of influence on the accuracy of diagnosis. TL is not the longer the better when TOAH is linear. Different TC has varying degree of influence on the classification accuracy rate (CAR) of models. A cognitive test that contains reachability matrix has a high CAR than that of did not contain. The higher the slip, the lower the pattern and the marginal CAR. And attribute construction with higher degree of loose often has the lower CAR. AHM is more sensitive to attribute construction and sometimes performs much reasonablely than DINA . But, generally speaking, the CAR based on DINA is higher than that based on AHM .