Working-Memory Performance 1 Running head: WORKING-MEMORY PERFORMANCE The Evidence for A Guessing State in Working-Memory Judgments

We present the results of four experiments designed to assess whether there is a guessing component to visual working memory, where guessing refers to responses that do not reflect the presented stimuli. The presence of guessing is important because such components are intimately consistent with discrete-slot view of WM and inconsistent with many competing views. We use the production paradigm from Zhang and Luck (2008, Nature, 453, 233-235), except our stimuli, which differ in angular displacement, are free to vary only within a restricted range rather than across all angles. As a result, guesses take on a distinctive pattern: when people guess they tend to select either a stereotypical clockwise or counter-clockwise position, and indeed this pattern is observed across all experiments. We formally fit a large number of discrete-slot and distributed-resource models and find that modified discrete-slot models outperform distributed-resource models in all experiments. One novel phenomenon we find is coarse encoding. Either items are encoded with fine-grain metric information, or they are encoded coarsely up to a categorical value (e.g., a stereotypical clockwise or counter-clockwise orientation). Implications for future theoretical developments are discussed.