An assessment of fixed-capacity models of visual working memory

Cowan, Nelson · 2008 · OpenAlex

DOI: 10.1073/pnas.0711295105

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the debate regarding the capacity limits of visual working memory (WM), specifically testing the hypothesis that WM consists of a fixed number of discrete slots for storing items. While previous research suggested a capacity of approximately three to four items, those demonstrations relied on aggregate data and asymptotic trends that could be mimicked by alternative models. The authors sought to provide a rigorous, formal test of the fixed-capacity model using a change-detection task, contrasting it with signal-detection theories that propose variable mnemonic strength rather than discrete storage. The study employed a within-participant factorial design with 23 participants. Participants performed a change-detection task where they viewed an array of colored squares (set sizes of 2, 5, or 8) and later determined if a single test square matched the corresponding studied square. Crucially, the probability of a change occurring was manipulated across blocks (0.3, 0.5, or 0.7) to assess response biases. The authors derived predictions for receiver operating characteristic (ROC) curves based on a fixed-capacity ideal observer model, which predicts linear ROCs with a slope of 1.0. To account for occasional lapses in attention, they extended this into a five-parameter model incorporating an attention parameter. This model was compared against a variable-capacity discrete model and signal-detection models using maximum-likelihood estimation and model-selection statistics (Akaike and Bayesian information criteria). The results provided strong support for the fixed-capacity model. Observed ROC functions exhibited slopes near 1.0, consistent with all-or-none mnemonic representations. The five-parameter fixed-capacity model fit the data well for 20 of the 23 participants, significantly outperforming the vacuous binomial model. Model-selection statistics indicated that the fixed-capacity model offered the best fit compared to variable-capacity and signal-detection alternatives. The estimated average capacity was 3.35 items, with participants attending to the display on 88% of trials. While the fixed-capacity assumption held for the majority, four participants showed variable capacity, likely due to idiosyncratic factors such as intimidation by larger arrays or encoding strategies. The significance of these findings lies in the robust empirical validation of discrete, fixed-capacity models of visual working memory. The study demonstrates that the fixed-capacity model is more parsimonious and accurate than signal-detection alternatives for this task. By isolating WM capacity through specific procedural controls—such as unique colors and single-item testing—the authors confirm that visual WM operates via a limited number of discrete slots. This supports the theoretical framework that human visual WM capacity is fundamentally constrained to approximately four simple objects, providing a solid foundation for further research into mechanisms like chunking and individual differences in memory performance.

Key finding

Receiver operating characteristic slopes of 1.0 and superior model fit statistics confirm that visual working memory capacity is fixed at approximately 3.35 items rather than varying with array size.

Methodology

lab_experiment

Sample size: 23

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via openalex_abstract on 2026-05-08 (7 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• working memory