Enhancing experimental design through Bayes factor design analysis: insights from multi-armed bandit tasks.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Bayesian Statistics

Background:

• Bayesian statistics is ideal for iterative hypothesis updating and incorporating prior information.
• Prospective Bayesian analysis combined with computational modeling of behavioral data is less developed.
• Optimizing experimental designs is crucial for robust prospective cognitive neuroscience research.

Purpose of the Study:

• To provide a tutorial on combining Bayes factor design analysis (BFDA) with latent variable modeling.
• To evaluate exploration-exploitation trade-offs in the binary multi-armed bandit task (MAB).
• To assess how design parameters influence the detection of group differences in latent variables.

Main Methods:

• Utilized Bayes factor design analysis (BFDA) and latent variable modeling.
• Simulated data from a binary multi-armed bandit task (MAB).
• Examined the impact of sample size, number of games, and effect size on evidence strength and error metrics.

Main Results:

• BFDA effectively evaluates and optimizes parameter estimation for exploration in MAB tasks.
• Demonstrated how design parameters influence the strength of evidence for group differences.
• Identified scenarios where large samples/effects may still yield high error and low detection probability.

Conclusions:

• BFDA combined with latent variable modeling prospectively informs the design and statistical power of human behavioral tasks.
• This integrated approach enhances the evaluation of experimental designs for cognitive neuroscience.
• BFDA offers a framework for optimizing research in areas like decision-making and exploration-exploitation.