Are attributes in "true" diagnostic classification modelling studies theoretically distinct from those in other...

Diagnostic classification modelling (DCM) encompasses both a family of latent class models and a specific methodology for using these models with an assessment. Critics argue that research employing the latent models without the methodology should not be labelled DCM. However, most DCM studies fall into this category. This paper evaluates the legitimacy of these studies by testing whether a theoretical boundary exists between the attributes used in cognitive models in "true" and "other" DCM research. This is important because "other DCM" studies often suffer from model fit issues, and efforts to address this rely on the theoretical framework of "true DCM." Theoretical distinctions may arise because "True DCM" typically use fine-grained attributes representing procedural steps, whereas "other DCM" often use broad attributes which refer to curriculum topics. Three intuitive boundaries are evaluated: singular attributes versus composites of finer-grained skills; attributes as real processes versus attributes as constructs; and attributes representing individual-level problem-solving versus group-level behaviour. The paper argues that none of these distinctions hold, suggesting that both "other" and "true DCM" can share the same core theoretical assumptions across these three aspects. However, this should not be interpreted as giving researchers carte blanche; other theoretical distinctions may well exist, and the use of broad attributes can carry notable practical challenges. Instead, DCMs with broad attributes may be best suited for a narrow range of assessment contexts.