Abstract
When presented with temporal synchronous events, like a finger snap followed by a light going out, we are likely to interpret them as having a causal relationship. Contrary to other forms of causal inference, this impression of causality is immediate, irresistible and intrinsically perceptual: it depends on low-level spatiotemporal factors rather than elaborated reasoning. However, causality does not require exact spatial and temporal matching as long as the two events occur within a spatial and temporal binding window. When the spatial or temporal offset is near the limit, causality judgments vary from trial to trial. To better understanding the mechanisms underlying perceptual judgments, we investigated the influence of recent perceptual history on visual causality: does performance on the previous trial influence the current trial? To investigate this question, participants were shown canonical Michotte-like launching sequences where a moving circle collides with another circle, ‘causing’ it to move. We manipulated the duration of a temporal lag at the time of collision. Trials were divided post hoc into two bins based on whether they were following a causal or non-causal response. For each group and participant, we fitted a psychometric function to the percentage of causal responses for each temporal lag to estimate the perceptual threshold. Nonparametric permutation testing revealed a significant difference between the thresholds of the bins: stimuli following a causal impression were more likely to be judged as causal and vice versa. Moreover, this retrospective bias could be traced up to five trials back. We interpret this result in the context of attractive serial dependence effects. Serial dependence has been suggested to benefit stable representations against perceptual noise and low environmental variability. This series dependence may offer a tool to study the mechanisms underlying visual causality.