Factorial comparison of working memory models.

Area of Science:

• Cognitive Psychology
• Neuroscience
• Computational Modeling

Background:

• Visual working memory (VWM) research faces key questions on mnemonic precision, capacity, and binding errors.
• Previous models often tested single hypotheses, ignoring interactions between these factors.
• A comprehensive factorial approach is needed to understand VWM limitations.

Purpose of the Study:

• To systematically evaluate all combinations of proposed answers to fundamental VWM questions.
• To determine the nature of mnemonic precision, item capacity, and the role of binding errors.
• To identify the best-fitting model of VWM limitations across diverse experimental data.

Main Methods:

• A factorial design was employed, creating 32 unique models by combining answers to precision, capacity, and binding error questions.
• All models were compared against data from 10 delayed-estimation experiments (164 participants, 131,452 trials).
• Model performance was assessed based on consistency across laboratories and trials.

Main Results:

• Mnemonic precision is continuous and variable across items/trials, not quantized.
• Item recall capacity is variable, with a best-model average of 6.4 items.
• Spatial binding errors are present but account for only a small proportion of responses (16.5% at set size 8).

Conclusions:

• The best-fitting model reveals continuous, variable mnemonic precision, variable item capacity, and limited impact of binding errors.
• Existing documented models of VWM limitations are strongly contradicted by the data.
• Factorial model comparison is crucial for advancing the understanding of visual working memory.