Development And Application Of Polytomously-scored IRT Model With Response Time

With the development of computer testing technology,collecting response time has become a routine work of many large-scale tests.However,most current IRT models with response time are only applicable to dichotomously scored data,which greatly limits the application of IRT model in practice.Based on the traditional dichotomously scored response time IRT model,this paper intends to develop a polytomously scored response time model.Under the framework of hierarchical modeling,the generalized partial scoring model(GPCM)and the log-normal model(JRT-GPCM)were used to construct the graded IRT model with response time(JRT-GPCM),and the parameter estimation of the new model was realized by the bayesian MCMC algorithm.In order to verify the feasibility of the proposed JRT-GPCM and investigate its application in practice,this paper carried out two studies: The Study for simulation experiment research,which used a 2 x 2 two factor experiment design,one factor for the number of participants(1000 and 2000 respectively,the two level),another factor for the test number(20 and 30 respectively two levels);The other study for JRT-GPCM model in the application of the NEO scale with 845 college students,in which male students accounted for 47.9%,rural students accounted for 58.9%,and students who were the only one child of their parents accounted for 28.5%.The results show that:(1)Under the JRT-GPCM model,the MCMC algorithm can accurately estimate both item parameters and person parameters under all simulated conditions,which indicated the MCMC algorithm used here was reasonable and the proposed JRT-GPCM model was acceptable.The study also showed that the estimation precision of item parameters increased with the increase of the number of person and the estimation precision of person parameters increased with the increase of the number of item.(2)The fitting of jrt-gpcm model in real data is good,which indicates that jrt-gpcm model has good applicability and advantages,and improves the accuracy and precision of "potential trait" estimation.(3)the R indexes(Brooks Gelman,1998)of all item and person parameterswere all less than 1.1,which indicated all estimated parameter estimation of MCMC algorithm were convergent.(4)The estimation standard errors of all parameter were small and ranged from 0.01 to 0.2,which indicated the model had robustness in empirical research.The12 items' discrimination parameters of the neurotic subscale were ranged from 0.895 to 1.209,all of which were greater than 0.7(Fliege,2015),indicating that the 12 items were of good quality.(5)There was a positive correlation between the potential traits and the response speed of the subjects.The higher the neurotic tendency of the subjects was,the higher the potential traits and the faster the response speed was.Item step parameters(the location parameter)were correlated to the time intensity of item,and the absolute value between adjacent item location parameters was greater,the time spend on this item would be less.This phenomenon was called "distance-difficult hypothesis" in Ferrando and Lorenzo(2007)which was also validated in our real data study.In conclusion,this study proposed a new method which incorporated the response time into the plolytomously IRT model,which can further expand the application of response time information in psychological and educational measurement.