True Q-Matrix Estimation for the Examination for the Certificate of Proficiency in English Test Using Regularization Techniques

• In Event: Cognitive Diagnosis Model: Innovative Methods and Applications

Abstract

Cognitive assessment has recently gained prominence in psychological and educational measurement for assessing students’ mastery of skills. One critical issue is to correctly identify the item-attribute relationship, which is provided by the so-called Q-matrix. Previous studies with these cognitive diagnostic models assume a known Q-matrix designed by domain experts, which is not empirical but subjective. The misspecification of the Q-matrix will lead to incorrect attribute profiles so that we cannot identify students' mastery of skills. We intend to estimate the true Q-matrix based on item response data by using a probabilistic model. Furthermore, the logarithmic barrier function was chosen as the penalty function to regularize the maximum likelihood so that the results will be more accurate and robust.

Authors

• Jihang Chen, Boston College

• Zhushan Mandy Li, Boston College