A multidimensional IRT model for ability-item-based guessing: the development of a two-parameter logistic extension model

Abstract

Multidimensional three-parameter logistic model (M3PLM) often faces poor recovery of parameters due to overestimation or underestimation of guess parameters. In order to more reasonably consider the participant’s guesses, this study proposed a new multidimensional IRT modeling based on 2PLE, which equal to the guessing parameter with a function that integrates item characteristic according to the ability level of an examinee to quantify guessing behavior. Our new model, named as multidimensional two-parameter logistic model with ability-item-based guessing (M2PL-AIG) model, not only the number of parameters is smaller than the M3PL model, but also considers the participant’s guesses just like M3PL model. Two simulation studies were conducted to fully comparing the performance of the M2PL-AIG model with the M3PL model. In addition, real data from TIMSS 2011 is used as an application illustration and to compare with the M3PL model. The results show that the M2PL-AIG model outperforms the M3PL model in parameter recovery and model fit, which indicates that the M2PL-AIG is a rather interesting proposal, surely worth some further study and the consideration by the IRT community.