Modeling And Application Of Item Response Theory Based On Response Time Process Date

With the development of the modern science and information technology,assessments have moved from traditional paper-and-pencil administration to computer-based administration in the educational and psychological assessments.With the prevalence of computer-based testing,it is very easy to record and present students' interactions with items.The students' response performance on items,i.e.,process data,can be recorded by the log files of computers.Process data include response time(RT)data,behavioral process data,sequential data and so on.The recording of response process data provides an important guarantee for psychometricians and statistical researchers to explore students' response behaviors more deeply and to estimate students' abilities more accurately.The application of response process data includes the following aspects: detect test-taking behaviors(e.g.,rapid guessing behavior and cheating behavior),identify test execution problems(e.g.,identification of students' groups whose real response times are longer than expected response times),identify test speededness and fatigue,study the differences of response time allocations in different countries,test the influence of the settings of item locations,explore the speed differences of information processing among students of different ages,and analyze the factors;Sequential data can provide the students' progress during the test,such as the number of times that items were revisited,the number of times the mice were clicked or dragged,and the paths of the students' responses.A combination of sequential data and response time data can help test analysts to identify the items,units,groups of students and so on.Response times essentially depict the process of information processing.This paper mainly explored response time process data and incorporated response times as auxiliary information into the study of item response theory,devoting to solve several practical application problems that mentioned above.This paper focused on test-taking behaviors and proposed a mixture model for responses and response times with a higher-order ability structure to detect rapid guessing behavior,which incorporates auxiliary information from other subtests and the correlation structure of the abilities,thus it is more powerful to detect rapid guessing behavior.A Markov chain Monte Carlo method is used to estimate the parameters of the model.Simulation studies reveal that all model parameters could be recovered well,and the parameter estimates had smaller absolute bias and mean squared error than the mixture unidimensional item response theory(UIRT)model.Moreover,the true positive rate of detecting rapid guessing behavior is also higher than when using the mixture UIRT model separately for each subscale,while the false detection rate is much lower than for the mixture UIRT model.The deviance information criterion and the logarithm of the pseudo-marginal likelihood are employed to evaluate the model fit.Finally,a real data analysis is presented to demonstrate the practical value of the proposed model.Item nonresponses are prevalent in standardized testing.A main contribution of the paper is to propose a cohesive RT process model to model both types of missingness,i.e.,not-reached items and omitted items.The conventional response time model could very well explain the not-reached items due to time limit.We propose a RT process model to explain the omitted items,which also provides a rational behavior process interpretation.Then,we use Markov chain Monte Carlo method to estimate the parameters of the new model.We use Bayesian information criterion to fit the real data,the real data analysis verified the application value of the RT process model.A simulation study was conducted to further evaluate the performance of the proposed RT process model when there are not-reached items and omitted items.The response time process data is an important part of testing data.The record of this type data provides important auxiliary information for test analysts to explore the examinees' solving/behavioral process.This paper was based on the response time process data,proposed corresponding statistical models for several practical test problems.The models can provide rational interpretations and inferences for the students' behavior process.Therefore,our proposed models can improve the accuracy and precision of the ability estimates,meanwhile,it can provide important reference and guarantee for subsequent decision-making.