Research And Application Of Cat Algorithm For Processing Response Style(CAT-RS)

Computerized Adaptive Testing(CAT),as a main application of item response theory(IRT),is popular and applied in fields such as educational examination,talent selection,and psychological educational measurement.At present,CAT is more widely used.Researchers have developed multidimensional CAT and polytomously-scored CAT for different test purposes,test conditions and data types.Likert rating scale is the main method to measure subjective psychological phenomena(such as attitude and belief).It requires individuals to choose whether the description of the problem conforms to their own and to what extent.However,this form of question and answer will lead individuals to make choices based on their preferences for response categories.Simply,the response behavior of the subject is driven by the format of the question rather than the content of the question.This phenomenon is called response styles(RS).As a stable individual characteristic,reaction style will produce systematic errors in the test process,distort data results and affect the accuracy of the test.Therefore,it is very important to pay attention to them.In recent years,there have been a lot of studies on Response style,and researchers have proposed a variety of methods to detect and deal with response style.Among them,Item Response Tree(IRTree)model is a very powerful tool.This kind of model is a project response model with tree structure,which is flexible and rich in information.It shows the internal decision-making process of an individual macroscopically with tree structure,providing a clear explanation for researchers to analyze how an individual makes a final decision,and can deal with problems that cannot be solved by other models.Nowadays,with the increasing demand for testing,an ideal theoretical model of testing should be able to describe the psychological characteristics it measures effectively and realistically.The previous CAT assumption was that there was no reaction style and the measured characteristics were their real potential characteristics,but this assumption was not always satisfied in practice.Therefore,this study combines IRTree model with the theory and method of CAT to solve this problem.On the basis of response style tree model,CAT-RS was developed that can handle reaction styles.Three studies were carried out in this paper.In study 1,the response style tree model--MPP model was incorporated into CAT,the maximum likelihood function and Fisher information matrix of the model were derived,and the corresponding CAT-RS selection algorithm was developed.In order to verify the feasibility and effectiveness of a developed CAT-RS algorithm,studies 2 and 3 respectively plan to test and demonstrate it through simulation research and empirical research.The results show that: the simulation study shows that both of the two newly developed item selection strategies have ideal estimation accuracy.In contrast,the Doptimality method has higher estimation accuracy,higher question bank security,and faster operation.In the empirical study,the developed CAT-RS algorithm is applied to the real item bank to further verify its efficacy in the actual data.The results show that it also has satisfactory estimation accuracy and high accuracy.In conclusion,the CAT item selection strategy developed in this study that can handle response style is feasible both in theory and in practice.It meets the requirements of computerized adaptive test,and provides reliable methodological support for handling response style in CAT practice and improving the estimation accuracy of personality trait parameters of subjects.