Proposing The Three-parameter DIF Detecting Model For Testlets And Its Detecting Efficiency

The classic IRT depends heavily on LID, which blocks its applicationin a lot of testlet-based tests. Accordingly,the DIF methods based theclassic IRT failed to take into account the influence of the LID. Thisstudy, based on three-parameter TRT model and B-MIRT DIF detection modelfor testlets, put forward the three-parameter DIF detection model fortestlets,The proposed model contains the testlet effect parameter and thepseudo-guessing parameter, which fits better to the testlet data,and thesetting of the pseudo-guessing parameter make the proposed model can beused in the multiple-choice items presenting the guessing behavior.At the same time, this paper applied MCMC algorithm under the Bayesianapproach to estimate the parameter of the magnitude of DIF,β,of theproposed DIF detecting model. This study examine the effect of recoveringthe βin the two prior information level. Simulation researches show: Inthe accurate prior information conditions, the random error of recoveringβis0.0182,and the system error is0.1586,and the correlationcoefficient is0.9741;In no information prior conditions, the randomerror of recovering βis0.1347,and the system error is0.1825,and thecorrelation coefficient is0.9705.The results achieve the ideal effect，and in both cases at0.01level no significant differences exists. andthe study of the efficiency of the detecting model found, the model methodexists higher DIF detecting rate, and higher DIF detecting error rate.The latter will waste some items that don’t exist DIF.