Abstract
Items maintained in visual working memory (VWM) can be biased toward visual input shown during the delay, as indexed by systematic shifts in people’s recall responses. Such biases are exaggerated when the item in VWM is actively compared with input shown during the delay, especially when that input is judged as similar to the VWM contents. In this study, we tested the hypothesis that the magnitude of this similarity-induced memory bias is determined by the precision of the VWM representation, such that lower VWM precision invites larger memory biases for items judged as similar. Across two experiments, participants (N = 55) remembered the direction of a dot motion target stimulus over a 2500 ms delay, with subsequent recall via continuous report. On 80% of trials, a second dot motion stimulus (probe) appeared during the delay, and participants compared its direction (offset from the target by 22.5º, 45º, or 67.5º) to that of the target (indicating “similar” or “dissimilar”) prior to reporting the target direction at the end of the trial. Importantly, behavioral precision was directly manipulated by changing the coherence of the dot motion (from 25% to 100%). Indeed, the precision of the reported target direction dropped with lower coherence. Critically, the magnitude of the similarity-induced bias increased as coherence decreased, thus confirming our hypothesis. The effect of coherence on the similarity-induced memory bias magnitude was primarily driven by an increase in swap errors (i.e., more reports centered around the direction of the probe). Taken together, our results demonstrate that low-precision VWM is more vulnerable to apparent similarity-induced biases due to an increased risk of replacement with incoming visual input.