Abstract
Humans can extract statistical representations quickly over a set of elements that vary over space and time. Previous studies have investigated the computational processes underlying ensemble perception. Here, we investigated how observers make judgments about their own performance amid spatiotemporal uncertainty in an ensemble-perception task. We adapted the stimulus from Yashiro et al. (2020) and compared two types of ensembles with different spatiotemporal uncertainty: Temporally-smooth ensembles had orientations varied across locations, but with the same average orientation across frames (i.e., temporal SD = 0), while spatially-smooth ensembles had orientations varied across frames, but with the same orientation across locations in every frame (i.e., spatial SD = 0). To obtain a performance-matched confidence comparison between the two ensemble types, we adopted the confidence forced-choice paradigm (Barthelmé & Mamassian, 2010; Knotts et al., 2018). On each trial, participants performed a clockwise-vs-counterclockwise orientation-discrimination task on the average orientation of the two ensemble types, one after the other in randomized order, and then chose the response in which they were more confident in being correct. We manipulated task difficulty by varying the ensemble’s spatiotemporal average orientation relative to the vertical reference. In Experiment 1, when locations of elements varied from frame to frame for both ensemble types, participants chose spatially-smooth ensembles more frequently than temporally-smooth ensembles even when performance was matched. This systematic metacognitive bias was eliminated in Experiment 2 when element locations were kept constant both across frames and between the two ensemble types. Furthermore, we replicated the “recency effect” (Yashiro et al., 2020) on confidence judgments in that later frames in an ensemble had greater influence on confidence choices than earlier ones. Our findings suggest that the structure of spatiotemporal uncertainty of a stimulus ensemble could influence not only first-order performance, but also second-order metacognitive judgments.