Development And Calibration Of A Multidimensional Item Response Model Incorporating Response Times

The test based on computer have two advantages: one is in the computerized adaptive test(CAT),it can according to the examinees' latent trait level to choose the items which phase match their level from the item bank;and the another one is it can record the response data of each item,meanwhile record the response time of each item.item selection methods of the traditional ignore response time CAT that considered the information of the items' score,but considered little the information of response time in it.Incorporate response time information in CAT well combines the advantages of the test based on computer,and it also overcome the disadvantages of traditional ignore response time CAT,that mostly to use the response time information will making the CAT has more efficient.In this paper,4PL-RT model was applied to the computerized adaptive test for the first time and came up with a CAT algorithm that based on 4PL-RT model.It is mainly from the perspective of mathematical statistics,the one is based on the Fisher information method,the response data and response time data were used in select the appropriate item for each candidate;the another one is derived EAP method that to estimate the candidates' latent trait.The 4 new item selection methods of incorporate response time information were developed and evaluated in this paper,verified the performance of those methods by the latent trait's estimation accuracy and the test safety,and comprehensive comparison with the traditional ignore response time CAT.This article used the monte carlo simulation experiment and empirical application research,that has three studies were carried out.Study 1 developed the CAT algorithm based on 4PL-RT model,that is the information function based on 4PL-RT model and the 4 new item selection methods.Study 2 verified CAT algorithms' effectiveness which use Monte Carlo simulation.Study 3 applied the incorporate response time information in CAT to Raven's standard reasoning test.The results show that:(1)the CAT successfully achieved incorporate response time data and response data of the candidates.The 4 new item selection methods were proposed that has highly effects of test accuracy,item exposure rate and test overlap rate.(2)the performance of incorporate response time information in CAT was better than the traditional CAT,mainly manifested in higher parameter estimation accuracy and test security.Among them,the test length was proportional to the test accuracy,and inversely proportional to the test accuracy.As the result of when the test length was 20,the comprehensive performance of various item select methods were the best.(3)Applied the incorporate response time information in CAT to Raven's standard reasoning test,found that,the new methods not only verified the response time information was an important factor affected the estimation of subjects' ability,and has higher parameter estimation accuracy,that compared with the traditional ignore response time CAT method.This provides a new CAT method and technical support to measure the intelligence use the raven's reasoning test(or even other psychological tests).