Using the iterative latent-class analysis approach to improve attribute accuracy in diagnostic classification models

The focus of diagnostic classification models is on investigating a respondent's mastery status of the attributes required for completing tasks and/or solving problems. Recent advances in model development have produced saturated model variants such as the log-linear cognitive diagnostic model (LCDM), but works focusing on improving the accuracy of their attribute estimates have not been accomplished commensurably. This article proposes an iterative latent class analysis (ILCA) approach to estimating attributes, such that the accuracy can be higher than that of traditional approaches. Particularly, the needs for the ILCA approach are illustrated within a literature review, the detailed procedures of the ILCA are presented via both pseudo-codes and verbal explanations, a simulation study is conducted to demonstrate the estimation accuracy, and finally, a discussion containing limitations and future research directions is provided. The results of this article show that ILCA outperforms its competitors in many conditions. Thus, it can be used to produce score reports.