When similarity is less disruptive: feature-sharing distractors in visual working memory under diffuse attention.

Visual working memory (VWM) is a capacity-limited system that enables visual information to be maintained, prioritized, and used after perceptual input has disappeared. Although attention is often assumed to protect VWM from distraction, recent work suggests that distraction is not uniformly harmful and that its impact depends on the representational and attentional state of the memorized information. The present study examined how attentional prioritization and distractor similarity jointly shape VWM across encoding and maintenance. In two change-detection experiments, cue load and distractor type were manipulated within participants. Experiment 1 used pre-cues to guide external attention before encoding, whereas Experiment 2 used retro-cues to guide internal attention during maintenance. Trial-level binomial generalized linear mixed-effects models for accuracy and linear mixed-effects models for log-transformed correct-trial reaction time showed that performance was best when one or two items were prioritized and declined when attention was distributed across more items or when no cue was provided. Distractor effects depended on cue condition. Under broad or absent attentional guidance, non-feature-sharing distractors were generally more disruptive than distractors that shared one feature with an unprobed memory item; however, feature-sharing distractors did not improve accuracy relative to the no-distractor baseline. Accuracy and reaction time converged for the cue-load effect but often dissociated for distractor contrasts, indicating that RT-only effects should be interpreted cautiously as changes in retrieval or decision efficiency rather than memory fidelity. These findings support a conditional account in which the effect of feature overlap depends on attentional allocation and the accessibility of the maintained representations.