Dynamics of variability and bias in working memory

Visual working memory (VWM) performance declines with the quantity of material stored and the duration of its maintenance, but it is disputed whether these constraints are independent or interact. Specifically, different studies have reached contradictory conclusions on whether greater memory load accelerates deterioration of the stored information during retention, a question with implications for storage mechanisms. Meanwhile, prominent models of VWM based on stable attractors predict that recall should become increasingly biased towards a fixed set of canonical features with longer retention, but empirical support is mixed. Uncertainty is further exacerbated by limited data, methodological variation, and inconsistent analytical approaches across studies. Moreover, previous studies have not always distinguished response variability from stimulus-specific biases, nor considered the possibility of swap errors (intrusions). To address these issues, we conducted six new VWM experiments and reanalysed data from seven published experiments using consistent analytical methods to examine the effects of set size and delay on variability and bias. Our results show that recall variability increased with longer delays and higher set sizes, consistent with prior findings, and importantly, that these effects did not operate independently: the effect of delay was amplified at higher set sizes. We also observed consistent evidence for systematic biases across stimuli, but unlike variability, bias amplitude did not change with either set size or retention interval. These results support diffusion accounts of VWM maintenance, in which variability grows through the accumulation of random error, while challenging stable attractor accounts that predict systematic drift towards canonical values during maintenance.