Parameter Estimation And Simulation Study Of Two-Parameter Testlet Response Model With Structural Parameters

With the increasing development of the mordern information technology and item response theory (IRT), computerized adaptive testing becomes more and more popular, for example, LSAT(Law School Admission Test) in the USA. Its items in the Analytical Reasoning and Reading Comprehension sections consist of testlets. Testlets refer to a common stimulus with a set of test questions (i.e.,items). The use of the testlets greatly improve test efficiency and save the items setting cost, so it has great application value to study testlet response model.This paper studies two-parameter extended testlet response model with the structural parameters by identifying and modeling the relationship between testlet features and test-taker response behavior on testlets. We utilize Markov chain Monte Carlo method to estimate the structural parameters, and use MATLAB to program and make a simulation study. The absolute values of the bias between simulation results and true values are within the range of 0.01 to 0.2, and the estimation results are better. The structural parameters in this paper include the following aspects:item parameters, the regression coefficients on testlet features, and the variances of respondents’ ability on testlet and testlet features.