Estimation of Partially Defined Q-Matrix
Qianru Liang · Proceedings of the 2019 AERA Annual Meeting · 2019
Q-matrix is an important component of most cognitive diagnosis models (CDMs), and is typically created by domain experts, where its specification and validation process are time-consuming.Due to the subjectivity of expert judgments, the Q-matrix may be inaccurate and the misspecification may affect the model-data fit.The aim of this research is to propose an efficient Q-matrix estimation method based on model fit indices, namely, Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) under the generalized deterministic input, noisy "and" gate (G-DINA) model framework.The performance of the method is examined using both simulated and real data in terms of the element-wise recovery rate, and compared to the method using G-DINA model discrimination index.The results suggest that this method can serve as a viable method in specifying the Q-matrix under some conditions.It can also provide supplemental information when the Q-matrix of additional items or a subsequent test form need to be constructed.