A hierarchical Bayesian state trace analysis for assessing monotonicity while factoring out subject, item, and trial...

Area of Science:

• Cognitive psychology
• Psychometrics
• Computational neuroscience

Background:

• State trace analysis assesses cognitive process dimensionality by examining monotonic relationships between dependent variables.
• Existing methods often assume independence between measures, potentially biasing inference with unacknowledged dependencies.
• Hierarchical modeling offers a framework to explicitly model and account for these dependencies.

Purpose of the Study:

• To develop and validate a novel hierarchical Bayesian state trace analysis technique.
• To address limitations of existing methods by incorporating subject, item, and trial-level dependencies.
• To provide a more robust assessment of latent dimensionality in cognitive processes.

Main Methods:

• Developed a hierarchical Bayesian model to explicitly model dependencies between two measures.
• Assessed monotonicity by comparing models with and without non-monotonic relations between condition effects.
• Utilized Widely Applicable Information Criterion (WAIC) and Pseudo Bayesian Model Averaging for model comparison.
• Validated the technique through model recovery simulations with varying dependency structures.

Main Results:

• The new method accurately recovers ground truth in simulations, demonstrating robustness to dependencies.
• Model comparison metrics (WAIC, Pseudo-BMA) effectively distinguish between monotonic and non-monotonic structures.
• The hierarchical model successfully accounts for subject and trial-level dependencies, improving inferential accuracy.

Conclusions:

• The proposed hierarchical Bayesian state trace analysis provides a more accurate and reliable method for assessing cognitive dimensionality.
• Explicitly modeling dependencies enhances the validity of state trace analyses, particularly with limited data.
• This technique offers a valuable tool for researchers investigating the structure of cognitive processes.