A Nonparametric Cognitive Diagnosis Method For Polytomous Attributes And CD-CAT Research

Computer-based Adaptive Test for Cognitive Diagnosis(CD-CAT)is based on traditional CAT and has cognitive diagnostic function.CD-CAT set different test for each subjects.Using intelligent strategy,CD-CAT can quickly and accurately estimate the state of knowledge of subjects.In the actual teaching and quiz situation,the cognitive attributes of the item often include multiple levels,and the polytomous attribute cognitive diagnosis came into being.It can not only indicate the level of the cognitive attribute level of the project measures,but also can further examine which level of the cognitive attribute the subject grasp,so it can provide detailed and additional diagnostic information,more practical value and guiding significance.The current cognitive diagnosis assessment is based on the cognitive diagnosis model,which is a parameter diagnosis model.The model selection needs to meet certain prerequisites.As we know,the parameter estimation method is complex and defective.However,the nonparametric cognitive diagnosis method only needs to define a good Q array in advance.It accords he test response to classify,simply and conveniently,to meet the needs of the actual test situation.Based on the idea of clustering analysis,the present research proposes a clustering diagnosis method(PACDM)based on the ideal response pattern and the real response pattern.And several commonly used CD-CAT item selection strategies are introduced into polytomous attribute CD-CAT.And this research investigates the feasibility of these item selection strategies in the framework of pG-DINA model,while comparing its diagnostic accuracy,item exposure rate and test overlap rate and other aspects.The results indicate that:(1)Under the same condition,the pattern match ration and average attribute match ration of the non-parametric method is higher than the parameter method.(2)When the item quality is low(larger slip and guessing parameters),the nonparametric method has the advantage over the parameter method,and this advantage increases with the improvement of the error level.(3)In terms of pattern match ration and average attribute match ration,PA-SHE and PA-MPWKL are the best,PA-KL strategy is the worst,the performance of PAPWK and PA-HKL are better than PA-KL strategy.(4)In terms of test security,PA-HKL and PA-PWKL are the best and PA-SHE and PA-MPWKL are the worst,the performance of PA-KL is better than PA-SHE and PA-MPWKL strategy.