Table 1 Findings and previous studies

Global-level implication

An integrative framework, created by merging individual mechanisms, is very effective, even more so than neural networks.

not previously reported

Interactions between items

Items influence each other’s biases and retention rates.

different from them¹⁸

3.9

The interactions between items are based on pre-categorical, not category-based, color information.

not previously reported

3.7

The effect of interaction on biases is governed by a Mexican-hat-like function.

confirms them³⁰

Chunking

Chunking magnitude = reduction in the number of storage units.

Likelihood = between-chunk variability - within-chunk variability.

different from them^19,21,26,33

3.9

Chunking is based on pre-categorical, not category-based, color information.

not previously reported

3.7

Better-chunked patterns are less likely to be swapped, and less attracted toward category centers.

not previously reported

Contrary to previous findings, better-chunked patterns are no more likely to be remembered.

different from them^19,21,26,33

Category-

biased component

Memory of colors are affected by color categories.

confirms them²³

3.4

The category-based encoding is Bayesian-like but does not strictly follow Bayesian rules.

different from them^18,23,43

3.6

Red advantage: reddish colors are represented more precisely than other colors in category-biased component (i.e., red 2 category).

not previously reported

3.5

Unbiased component

Red disadvantage: reddish colors are represented less precisely than other colors in the unbiased component.

not previously reported

3.5

Swap-based component

Spatial binding errors occur at the representation stage, but not at the response stage.

different from them^13,20

3.8

Concentration and Crosstalk

The weights of items affect each other.

not previously reported

3.9

Retention rates/ Random guess

More typical colors, as defined by the categories, are more likely to be remembered.

confirms them²⁷

Consistent with the spirit of the slot model, only the retention rates, not the precision, are affected by interactions between items and spatial attention.

different from them^15,24

3.3

Trade-off

There is a trade-off between the quantity and quality of representations. This is consistent with the spirit of resource model.

confirms them^16,28,41

3.3

Precision of representations

There are low-precision components.

confirms them^17,29,31,32

3.2

Spatial attention

Better-attended items are more likely to be remembered.

confirms them^{15,51,52,53,54}

3.1

Better-attended items’ color categories are narrower and taller, and less effective at attracting the color-category biased component.

not previously reported

3.1

Category/ representation

The truncated normal distribution is superior to the von Mises distribution.

different from them^15,16,38,55

3.10