On a Reparameterization of the MC-DINA Model

Area of Science:

• Psychometrics
• Educational Measurement
• Cognitive Psychology

Background:

• Cognitive Diagnosis Models (CDMs) are essential for understanding student mastery of specific skills.
• Traditional CDMs often use dichotomous responses and do not account for distractor selection.
• The Multiple-Choice Diagnostic Model (MC-DINA) extends CDMs by allowing nominal responses and modeling distractor effects.

Purpose of the Study:

• To re-express the MC-DINA model as a multinomial logit model with latent discrete predictors.
• To clarify the model's structure, assumptions, and parameter restrictions.
• To explore implications for psychological interpretations and statistical estimation, particularly for small sample sizes.

Main Methods:

• Re-parameterization of the MC-DINA model.
• Utilizing a multinomial mixture model framework.
• Applying a signal detection-like parameterization.

Main Results:

• Demonstrated that MC-DINA can be represented as a multinomial mixture model.
• Identified parameter restrictions inherent in the model structure.
• Showcased the applicability of the reparameterization using data from the TIMSS 2007 fourth-grade exam.

Conclusions:

• The reparameterization provides a clearer understanding of MC-DINA, especially regarding distractor effects.
• Identified parameter restrictions have significant implications for psychological interpretations and statistical estimation.
• The proposed approach facilitates parsimonious models suitable for practical applications, including those with limited sample sizes.