Consistency of nonparametric classification in cognitive diagnosis

Area of Science:

• Psychometrics
• Educational Measurement
• Cognitive Science

Background:

• Latent class models are standard for cognitive diagnosis, classifying individuals based on skill profiles.
• These models account for deviations from deterministic response patterns.
• An alternative approach is needed that bypasses parameter estimation.

Purpose of the Study:

• To present a non-parametric method for cognitive diagnosis.
• To demonstrate the consistency of this distance-minimization approach.
• To analyze its performance across common latent class models.

Main Methods:

• Minimizing the distance between observed and ideal response patterns.
• Utilizing a non-parametric approach without stochastic terms for deviations.
• Presenting theorems to prove classification consistency.

Main Results:

• The proposed method achieves classification consistency.
• Consistency is independent of sample size as no parameters are estimated.
• Simultaneous consistency for large groups is demonstrated under specific conditions.

Conclusions:

• A novel, non-parametric cognitive diagnosis method is validated.
• This approach offers consistent examinee classification without parameter estimation.
• It provides a robust alternative to traditional latent class models.