Q-Matrix Completeness for Multiple-Choice Items in Cognitive Diagnosis

Abstract

Research in the development of cognitive diagnosis (CD) methods for multiple-choice (MC) items has advanced. However, some key statistical properties of the Q-matrix for MC items remain unexplored. One such critical property is completeness, which, when satisfied, ensures that all possible attribute profiles in the data can be identified. An incomplete Q-matrix may result in the misclassification of examinees, leading to distorted interpretations of results and misguided pedagogical interventions. Moreover, a complete Q-matrix is a fundamental assumption in several important areas of CD research, such as model identifiability. In light of these undesirable consequences, this study aims to establish sufficient and necessary conditions for determining the completeness of the Q-matrix for MC items. However, this problem is not a simple extension of the corresponding one for binary items, because an MC item may contain additional diagnostic information through its coded distractors. Consequently, conditions that are sufficient and necessary for completeness in the binary setting may no longer be necessary in the MC setting. To support the proposed theorem, a lemma is developed, followed by corollaries that enhance its applicability. All theoretical advancements are rigorously justified through formal mathematical proofs. Subsequently, a simulation study is conducted to illustrate the implication of the proposed theorem.