Redundancy Repartitions Capacity in Visual Working Memory: The Sample-Size Model of Neural Activation

Visual working memory (VWM) has a limited capacity of around four items, yet recall is also improved by structure in the memory array. We reconcile these views by testing how color redundancy — a low-level structural regularity — alters recall precision and response time in a continuous color-reproduction task using a between-subjects sample (21 participants) and a small-N within-subjects design (5 participants). To explain precision, we combine the attention-weighted sample-size model of Smith et al. (2018) with Oberauer’s (2023) signal discrimination model of neural population coding. This attention-weighted sample-size model of neural activation predicts that tuning amplitude will follow a power law (β) such that β = -0.5 indexes the canonical 1/√N capacity decline observed as the number of to-be-recalled elements increases. Constrained amplitude fits matched or exceeded free baseline models and captured redundancy gains (β > sample-size) and non-redundant costs (β < sample-size). For four of five individuals, distinct color count — not item count — defined the unit N; one participant was best described by item count. Small-N model fits revealed that redundancy gains occurred through attention reallocation away from non-redundant elements in three participants. Redundant and non-redundant item precision improved in one participant, suggesting a shared capacity improvement — a ‘spill-over’ benefit. These results motivate individual-level model fitting to reveal nuanced redundancy gains and potential spill-over effects. We conclude that VWM capacity is defined over feature-based units that reflect array regularities, which individuals use strategically, producing redundancy gains and spill-over effects that may be masked by group statistics.