A Cat Comparative Study Among Different Item Selection Strategies:Item Exposure And Estimation Precision

Computerized Adaptive Testing(CAT)is a combination of modern measurement theory and computer technology.The basic idea is "due to human test","tailored",according to the item response theory,the computerized automatically selected for the level of subjects to test subjects to answer items.And when the difficulty of the subject and the ability to test the level of matching,the test the highest efficiency.Compared with the traditional pen and paper test,CAT has the advantages of high efficiency,flexible measurement and high safety,so domestic and foreign scholars on CAT theory research and application practice more and more.The computerized adaptive testing mainly includes the following six parts: the item response model,the construction of the item bank,the selection of the initial item,the selection strategy,the ability estimation method and the condition of the termination of the test.The item selection strategy is the most critical and the core part.In the actual test process,CAT in the common item strategy will lead to the use of low item bank,the item exposure is not uniform and other issues(Chen Ping,Li Zhen,Xin Tao,2011).Mao's article(2013)pointed out that the accuracy of measurement and item exposure are often mutually constrained,the relationship between the shift.Therefore,a good item selection strategy should be able to balance the accuracy of the measurement and item exposure uniformity.There are many researches on computerized adaptive testings at home and abroad,but there is no discussion on many factors that affect the item exposure.This study will explore the factors that may affect the accuracy of item exposure and ability estimation in order to find a balance between item exposure and ability estimation accuracy under various conditions,and provide a reference for later CAT applications,especially in high interest test,For decision-makers in the item bank design,selection of items such as strategy to provide reference.In order to solve the above problems,this paper uses Monte Carlo simulation and 3PLM model as an example to start a series of studies,in order to find the balance of item exposure and ability estimation accuracy under various conditions.We use both simulation and comparative analysis to carry out these two studies.Finally,the influence of test length,item bank distribution and item bank number on the ability estimation accuracy and the item exposure control index are also discussed in this paper.This study found that:(1)Under different experimental conditions,MFI's ability to estimate the accuracy is higher;if the comprehensive ability to estimate the accuracy and item exposure indicators,a-STR,followed by OID method.(2)In general,the accuracy of comprehensive test and the uniformity of test items,test efficiency and other indicators,a-STR in the first echelon,the rate of moderate,the item exposure control is better;OID method and a-STR of comparable,Slightly worse than a-STR.K-L for the second echelon,the ability to estimate the ability of the item exposure control in general;MFI worst,the highest accuracy of the estimated ability,but the item exposure is very serious.(3)Test length and ability estimation accuracy is proportional to,and item exposure is inversely proportional.When the length of the test is 30 items,the ability to estimate the accuracy of each item selection strategy and the item exposure to achieve the best balance of the state.The length of the test has played a regulatory role between the ability to estimate the accuracy and the item exposure.(4)Under the distribution of four kinds of item bank,the accuracy of the selection strategy and the item exposure index are not unified trend.In the item bank(a ~ U(0.2,2.5),b ~ N(0,1),c ~ Beta(5,17),the best ability of various strategies is estimated and the item exposure is the lowest.(5)The more the number of item bank,for the subjects to choose more richer,the ability to estimate the more accurate,the item exposure uniformity is also better.When the item bank is 2000,the accuracy of the selection strategy and the item exposure are the best.