Abstract
Traditional theories of visual working memory (VWM) characterize it as a durable short-term memory store. This mainstream view has been challenged by results of Makovski and colleagues (2010), who reported that performance was impaired on 2-alternative-forced-choice (2-AFC) relative to single-probe change detection (CD) tasks. Makovski et al. (2010) proposed that the need to evaluate two stimuli in the 2-AFC testing procedure increases interference costs, which suggests that VWM representations may be more fragile than previously thought. Recent work replicated this difference in performance between 2-AFC and CD procedures (Schurgin, Wixted, & Brady, 2018). However, it is difficult to infer memory costs without first entertaining the range of decision strategies that people might employ under different testing conditions. We consider an alternative explanation for the purported cost in performance found in the 2-AFC task, which is grounded on the recall-to-reject strategy documented in the long-term memory literature (Rotello, Macmillan, & Van Tassel, 2000). According to this view, people treat the CD procedure as a source memory test, which allows them to use a process-of-elimination type strategy, e.g., when presented with a green probe, rather than simply judging how familiar green is, they use their entire set of knowledge: if they are very sure the item was red, they know it was not green. We show that such a strategy can lead to overestimates of sensitivity in CD tasks. Furthermore, this decision strategy would lead to symmetric receiver operating characteristic curves, even if underlying noise and signal distributions are unequal in variance. As such, this view also provides insight into why modeling results reveal consistent evidence for equal variance signal detection models in CD tasks (Schurgin, et al., 2019; Robinson, Irwin, & Benjamin, 2019), even when one might expect effects of item variability on memory.