Quantifying effects of inattention on serial dependence in numerosity and orientation judgments using a Bayesian integration approach

Abstract Background Our perceptual judgments are influenced by recent experience. In three experiments, we examined how such serial dependence in numerosity and orientation judgments is influenced by attention. Participants estimated numerosity in dot clouds (Experiment 1) or the orientation of Gabor patches (Experiment 2: high contrast, Experiment 3: low contrast). These tasks were paired with an attention-demanding central task to modulate attention in a dual-task design. Results We found stronger serial dependence under divided attention across all experiments. Furthermore, we employed an ensemble of Bayesian models to disentangle the influence of attention, sensory evidence, and prior responses in numerosity and orientation judgments. Our six models differed in which of the three components of prior experience – average sensory evidence, prior sensory evidence and prior response – were included and how they expressed the influence of history on perceptual judgments. We found that all three factors influence perceptual judgments with attractive effects for prior responses and average sensory evidence, but a repulsive effect for prior sensory evidence. Importantly, the strength of the influence on individual components depended on the nature of the task (numerosity or orientation judgment), the quality of sensory evidence, and availability of attention. Conclusions Our findings suggest that perceptual decisions arise from a context-specific integration of prior experiences. This means that the magnitude and even the sign of serial dependence, whether it is overall attractive or repulsive, reflect specific combinations of factors (task, quality of evidence, attention, etc.) and highlight the importance of computational modeling for disentangling the individual effects.