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

With the rapid development of information technology,computerized testing is becoming more and more popular in large-scale tests.Of all the secondary data collected during the test,Response time(RT)is probably one of the most important and commonly used sources of information.Using response time information in a test will help the to separate the speed and ability(or trait),better for design experiments,identify abnormal behaviors in the test,improve the accuracy of parameter estimation between examinees and items,improve the item selection efficiency of Computerized Adaptive Test(CAT),and compare individual response strategies.Therefore,how to use the information of the response time data to provide more accurate measurement results has become a research topic that researchers are increasingly concerned about.However,in order to make full use of the RT information in the test,the first problem is to establish an appropriate RT measurement model.In the item response theory(IRT),there is a long history of research on RT modeling.Such as,the response time is modeled separately by referring to the parameter settings of the traditional item response model built for response accuracy(RA).The item response parameters were added to the response time model to obtain more accurate estimation of the subjects’ response time parameters.In order to obtain more accurate estimation of the person parameters,the response time parameters were constructed in the traditional item response model.However,the development and application of these RT related models are mainly focused on the one-dimensional test,which has many limitations on the multidimensional measurement considering the response time.Moreover,the traditional multidimensional item response model(such as M3PL)which can be applied to the multidimensional test ignores the response time information and cannot simultaneously process the response data and response time data of the subjects.Therefore,it is of great significance to develop a multidimensional item response model which is suitable for multidimensional test and can analyze both response data and response time information.The purpose of this study is to propose a multidimensional item response model that incorporating individual response time.The Bayesian Markov Chain Monte Carlo(MCMC)method based on Hamiltonian Monte Carlo(HMC)sampling algorithm was used to estimate the parameters of the proposed model.The applicability of the proposed model and the robustness of parameter estimation are verified and evaluated by using real data and two types of simulation data.The result of this research shows that:(1)M4PL-RT with response time information is theoretically more in line with the real multidimensional measurement situation.(2)Compared with other traditional IRT models,the proposed model shows better fitting performance with real data in this research.(3)Compared with the traditional multidimensional item response model M3 PL,the M4PL-RT with response time information can provide more measurement information.In other words,it can provide higher measurement accuracy for the subjects.(4)In the simulation study based on real data,both the person parameters and item parameters of the M4PL-RT model show good recovery performance.(5)In the simulation data research using 2×2×3 completely randomized experimental design,M4PL-RT shows the better performance of parameter estimation recovery under 12 conditions.The multidimensional item response model proposed in this study has important theoretical and practical significance for estimating the potential parameters of individuals.In addition,this paper further discusses several potential application directions of the proposed model and finally introduces the prospect of further improvement and improvement in the future.