The metacognitive factors for biases in visual perception

Serial Dependence (SD), as a sequential perceptual bias, influences current visual percepts by its recent history. While the low-level perceptual origins of SD are well investigated, its modulation by high-level metacognitive factors, e.g., confidence, is not as clear. In the current study, we investigated how local and global confidence judgments influence the magnitude of SD in a delayed orientation estimation task. We also examined whether the relationship between local and global confidence judgments is modulated by SD effects. Participants reproduced orientations of Gabor patches and rated both trial-based (local) and block-level (global) confidence across blocks varying in set size (3, 6, or 9). This design also allowed us to test whether global confidence estimates were directly impacted by local confidence and block set size. We observed a serial dependence effect which was modulated by local confidence judgments: there was a significant attractive bias in following high-confidence trials, and for low-confidence trials, there was no reliable SD. Global confidence, however, did not modulate SD. Linear mixed-effects models showed that both previous confidence and current error independently predicted local confidence. Global confidence was influenced by prior global confidence estimates and block-wise error, but not set size. Taken together, these findings show the dynamic nature of metacognitive computation in relation to recent perceptual history. Local, but not global, confidence estimates shape serial dependence, highlighting a limited time course of the bias.