Bias In Confidence: A Critical Test For Discrete-State Models Of Change Detection

Ongoing discussions on the nature of storage in visual working memory have mostly focused on 2 theoretical accounts: On one hand we have a discrete-state account, postulating that information in working memory is supported with high fidelity for a limited number of discrete items by a given number of "slots," with no information being retained beyond these. In contrast with this all-or-nothing view, we have a continuous account arguing that information can be degraded in a continuous manner, reflecting the amount of resources dedicated to each item. It turns out that the core tenets of this discrete-state account constrain the way individuals can express confidence in their judgments, excluding the possibility of biased confidence judgments. Importantly, these biased judgments are expected when assuming a continuous degradation of information. We report 2 studies showing that biased confidence judgments can be reliably observed, a behavioral signature that rejects a large number of discrete-state models. Finally, complementary modeling analyses support the notion of a mixture account, according to which memory-based confidence judgments (in contrast with guesses) are based on a comparison between graded, fallible representations, and response criteria.