Evaluating the complexity and falsifiability of psychological models

Area of Science:

• Psychological modeling
• Computational neuroscience
• Cognitive science

Background:

• Model complexity is crucial for developing effective psychological models.
• Existing measures of model falsifiability have limitations.
• Falsifiability relates model predictions to empirical evidence.

Purpose of the Study:

• To develop a new, improved measure of model falsifiability.
• To challenge existing intuitions about model complexity and falsifiability.
• To introduce the concept of plausible falsifiability.

Main Methods:

• Developed KL-delta, a novel measure using Kullback-Leibler divergence.
• KL-delta compares model prior predictive distributions with data priors.
• Applied KL-delta to psychophysics, decision-making, and memory recall models.

Main Results:

• Hierarchical models with more parameters can be more falsifiable.
• Models with response determinism can be harder to falsify than simpler cases.
• Informative data priors enhance KL-delta's ability to distinguish models.

Conclusions:

• KL-delta offers a more nuanced understanding of model complexity and falsifiability.
• Adding parameters does not necessarily increase model complexity or decrease falsifiability.
• Plausible falsifiability, considering data likelihoods, is valuable for model evaluation.