The Comparison Analysis On Parameter Estimation Accuracy Of TRM And Traditional IRM

Item response theory (IRT) has become the representative of themodern psychological and educational measurement theory. Based onthe presuppositions of developed basic model, a large number ofmodels and theories for the actual test application problems havebeen derived and extended. This thesis mainly analyzed the Testletresponse model (TRM), which was developed by local independencen(LI) under the situation of dissatisfaction. On the other hand, thestability and accuracy of traditional IRT and TRT model parameterswere discussed under the existence of the title of projects.This thesis used the structure of English test as basis, appliedtraditional IRT and TRT （2Parameter Logistic） forparameter estimation, respectively. Furthermore, R languageprogram was used to simulate the test structure, subjects reactionmatrix of five problem sets effect level was generated repeatedly.Several conclusions were obtained based on the change of levelparameters:(1) Compared to TRT model, the traditional IRT model are moreeffective on overestimated item parameters and accuracy of thesubjects ability based on obtained data.(2) Based on simulation results, TRT model are more stable and accurate on parameter estimation compared to traditional IRT modelunder the condition of individual group level. However, there wereless differences for IRT and untitled group effect level parameterswhen.(3) This research also indicated that the stability and accuracyof TRT model on question group effect parameter estimation shouldbe improved under frequent simulations.