Restricted Latent Class Models for Nominal Response Data: Identifiability and Estimation

Area of Science:

• Psychometrics
• Social Sciences
• Statistics

Background:

• Restricted latent class models (RLCMs) are crucial for analyzing multivariate binary responses.
• Existing research has advanced identifiability conditions for binary and polytomous data.
• Multiclass, nominal response data are common in social sciences and psychometrics.

Purpose of the Study:

• To establish new identifiability conditions for RLCMs with multiclass data.
• To discuss the implications of these conditions for real-world applications.
• To propose a Bayesian framework for parameter inference.

Main Methods:

• Derivation of novel identifiability conditions for multiclass RLCMs.
• Development of a Bayesian framework for parameter estimation.
• Monte Carlo simulation study for parameter recovery assessment.

Main Results:

• New identifiability conditions for RLCMs with multiclass data were successfully established.
• The proposed Bayesian framework demonstrated effective parameter recovery.
• The methodology was validated through application to a real dataset.

Conclusions:

• The new identifiability conditions enhance the applicability of RLCMs to polytomous and nominal data.
• The Bayesian approach provides a robust method for analyzing complex response data.
• This research offers valuable tools for diagnosis and classification in social science and psychometric research.