Abstract
<p>People tend to bias their report of a target item briefly held in visual working memory (VWM) towards novel visual input shown during the delay, especially when the novel non-target input is deemed similar to the target item. We investigate whether the magnitude of this influence, known as the similarity-induced memory bias (SIMB), depends on the certainty with which the target and non-target items are encoded. Across two experiments, participants (N = 64) remembered the direction of a dot motion stimulus (target) for later report. On 80% of trials, a second dot motion stimulus (non-target) appeared during the delay, and participants judged whether its motion direction was similar or dissimilar to that of the target. In Experiment 1, stimulus certainty was manipulated by changing the motion coherence of the target and non-target in tandem. Here, low target and non-target coherence resulted in reduced recall precision, less confidence in memory report, and amplified SIMB. When the target and non-target coherences were manipulated independently (Experiment 2), reduced recall precision and increased SIMB magnitude were most evident when target coherence was low. Furthermore, modeling revealed that the SIMB reflected two dissociable memory errors, namely bias of the memory target toward the non-target, and replacement of the target by the non-target. While both contributed to SIMB in most of the conditions, replacement errors happened more frequently under low target coherence. Overall, these findings suggest that a non-target item can interact with a memory target, evoking two types of memory error that are shaped by stimulus certainty.</p>