Cognitive Diagnosis Models (DINA / G-DINA)
Also known as: Diagnostic Classification Model, Skills Assessment Model, Attribute Mastery Model, Bilişsel Tanı Modeli
Cognitive Diagnosis Models (CDMs) are a family of latent variable models designed to classify examinees according to their mastery of a set of discrete cognitive attributes or skills. The Generalized DINA (G-DINA) framework, introduced by Jimmy de la Torre in 2011, provides a unifying structure that encompasses many specific CDMs — including the DINA, DINO, ACDM, and LLM models — as special cases, enabling fine-grained diagnostic feedback beyond a single total score.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use CDMs when the goal is fine-grained diagnostic feedback rather than rank ordering — for example, formative assessments, learning analytics, or clinical skills inventories. Key assumptions include: attributes are binary (mastered vs. not), a valid Q-matrix can be specified by domain experts, and the sample size is sufficient (commonly N > 500 per K attributes). CDMs are not appropriate when a continuous ability scale is the primary goal; in that case IRT models such as the 2PL or Rasch model are preferable.
Strengths & limitations
- Provides actionable, attribute-level diagnostic profiles rather than a single ability score
- The G-DINA framework unifies many specific CDMs, allowing model selection and comparison within one framework
- Well-suited to criterion-referenced and mastery-based educational contexts
- Model fit can be evaluated at the item level, facilitating iterative Q-matrix refinement
- Requires a correctly specified Q-matrix; misspecification leads to biased classifications
- The number of latent classes grows exponentially with the number of attributes (2^K), demanding large samples
- Binary attribute assumption may oversimplify continuous or partially developed skills
- Software options are fewer than for IRT, and estimation can be computationally intensive for large K
Frequently asked
What is the difference between DINA and G-DINA?
DINA (Deterministic Input, Noisy And-gate) assumes a conjunctive rule — an examinee must master all required attributes to have a high probability of correct response, and the only free parameters are slip and guessing. G-DINA relaxes this by allowing main effects and interactions for each required attribute combination, making it a general framework that reduces to DINA and other specific models as constrained special cases.
How large a sample do I need to apply CDMs?
Required sample size depends on the number of attributes K and the specific CDM. Simulation studies suggest a minimum of roughly 500 examinees for models with K = 3–4 attributes and the G-DINA framework. Models with more attributes or sparser Q-matrices may require several thousand observations for stable estimation and reliable classification.
Can CDMs be applied to polytomous or continuous response data?
Standard CDMs are designed for binary (correct/incorrect) item responses. Extensions for polytomous data exist (e.g., the sequential G-DINA model for ordered responses), but these are less commonly implemented. Continuous response data generally do not fit the CDM framework and should be analyzed with factor-analytic or IRT approaches instead.
Sources
- de la Torre, J. (2011). The generalized DINA model framework. Psychometrika, 76(2), 179–199. DOI: 10.1007/s11336-011-9207-7 ↗
How to cite this page
ScholarGate. (2026, June 2). Cognitive Diagnosis Models (DINA / G-DINA). ScholarGate. https://scholargate.app/en/psychometrics/cognitive-diagnosis-model
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Latent Class AnalysisStatistics↔ compare
- Rasch ModelPsychometrics↔ compare