Toward a principled Bayesian workflow in cognitive science

Area of Science:

• Cognitive Science
• Computational Linguistics
• Psychology

Background:

• Bayesian methods are increasingly used in cognitive science research.
• Probabilistic programming languages (e.g., Stan) and packages (e.g., brms) facilitate Bayesian analysis.
• Model relevance and accurate data structure capture are crucial for Bayesian utility.

Purpose of the Study:

• Introduce a principled Bayesian workflow for cognitive science.
• Demonstrate model verification techniques for robust data analysis.
• Guide the proper employment of Bayesian methods in scientific inquiry.

Main Methods:

• Applied a principled Bayesian workflow to cognitive science.
• Used a working example of reading times data with linguistic manipulation (object vs. subject relative clauses).
• Described key model verification questions: prior predictive checks, computational faithfulness, model sensitivity, and posterior predictive checks.

Main Results:

• The workflow demonstrates using domain knowledge to inform prior distributions.
• Provided guidelines and checks to avoid overfitting and capture relevant data structure.
• Showcased how to obtain robust answers to scientific questions using Bayesian methods.

Conclusions:

• The principled Bayesian workflow enhances the reliability of cognitive science research.
• Analysts must verify model utility, even with advanced software.
• Proper application of Bayesian methods ensures valid data analysis and meaningful scientific conclusions.