Abstract
Abstract
Serial dependence, the phenomenon where current perception is biased toward the previous stimulus, has been interpreted as a mechanism for perceptual stability, smoothing variability in sensory input. However, analyses that include both the previous stimulus and previous response unmasked repulsive effects at the sensory level that were otherwise concealed. Moreover, attractive serial biases fail to scale with environmental correlations, challenging their primary role in maintaining stability. We propose the “precision hierarchy hypothesis,” grounded in predictive processing, as a unifying framework for these seemingly contradictory findings. Repulsive effects naturally emerge when priors—shaped by recent trials—are highly precise, making small deviations in incoming input more probable under alternative priors. We further propose that prior precision decreases across hierarchical levels. A higher precision at lower sensory levels leads to repulsive effects, as even small deviations in incoming signal are in conflict with the prior. In contrast, attractive biases emerge at higher, more abstract levels, where deviations are integrated with broader conceptual priors. By dynamically shaping prior precision according to environmental correlations, perception remains adaptive, balancing the need for stability with the sensitivity to detect meaningful changes in the environment. This framework reconciles diverse findings on serial effects and provides testable predictions for future research.