Diagnostic Classification Models

Diagnostic classification models (DCMs) are psychometric models for classifying respondents or test takers according to their status on a set of discrete latent attributes, inferred from their observed performance or response on assessment items or tasks. DCMs, with their modeling of categorical, often fine-grained latent attributes, are well-suited for making classification-based inferences about attribute mastery. This contrasts with more conventional psychometric models containing continuous latent variables to be more suitable for the measurement of aggregated achievement. DCMs are, minimally, models about the item response process. They can also be understood as restricted latent class models for discrete multivariate data.

This entry begins with a comprehensive overview of the conceptual foundations of DCMs, followed by details of the DCM modeling process, and concludes with discussions about estimation and model evaluation. For consistency, the following terminology is used: items refer to questions or tasks in educational or psychological assessment tools; attributes are the categorical latent variables measured by the items; and attribute profiles include all possible combinations of attribute status. In line with the convention in the DCM literature, mastery is used to mean possession, attainment, or presence of the attribute in question.