When is a causal illusion an illusion? Separating discriminability and bias in human contingency judgements.

Abstract

Humans often behave as if unrelated events are causally related. As the name suggests, such causal illusions imply failures to detect the absence of a causal relation. Taking a signal detection approach, we asked whether causal illusions indeed reflect failures of discriminability, or whether they reflect a general bias to behave as if events are causally related. Participants responded in a discrete trial procedure in which point gains, point losses, or no change in points occurred dependently on or independently of responding. Participants reported whether each event was response-dependent or response-independent by choosing between two stimuli, one corresponding to reporting "I did it" and the other to "I didn't do it." Overall, participants responded accurately in about 80% of trials and were biased to report that events depended on responding. This bias was strongest after point gains and for higher-performing participants. Such differences in event-specific biases were not related to response rates; instead, they appear to reflect more fundamental differences in the effects of appetitive and aversive events. These findings demonstrate that people can judge causality relatively well, but are biased to attribute events to their own behaviour, particularly when those events are desirable. This highlights discriminability and bias as separable aspects of causal learning, and suggests that some causal illusions may not really be "illusions" at all-they may simply reflect a bias to report causal relations.