On Bayesian estimation of a latent trait model defined by a rank-based likelihood

Area of Science:

• Statistics
• Psychometrics

Background:

• Maximum likelihood estimation (frequentist) and Bayesian estimation are standard parameter estimation techniques.
• Maximum likelihood estimation has limitations including sensitivity to outliers, computational demands, and challenges with ordinal categorical data, potentially causing biased estimates and inaccurate coverage.

Purpose of the Study:

• To introduce and evaluate a novel parameter estimation method addressing the limitations of classical approaches.
• To enhance the accuracy and reliability of parameter estimation, especially for ordinal categorical data.

Main Methods:

• Employed a latent trait model incorporating Bayesian marginal likelihood and rank-based estimation.
• Utilized simulation studies to assess the performance of the proposed Bayesian method.
• Analyzed convergence using trace plots and potential scale reduction factors.
• Performed posterior predictive checks to evaluate model fit.

Main Results:

• The proposed Bayesian method demonstrated favorable performance in simulations.
• Trace plots and potential scale reduction factors indicated good convergence without issues.
• Posterior predictive checks confirmed the model effectively captures data variations.
• The Bayesian method yielded superior performance metrics (MAE, RMSE, coverage) compared to classical methods.

Conclusions:

• A latent trait model with Bayesian marginal likelihood and rank-based estimation is a superior parameter estimation technique.
• This Bayesian approach effectively handles ordinal categorical data and overcomes limitations of classical methods.