Manipulating Prior Causal Beliefs Through Causal Mechanism Information Affects the Outcome-Density Bias

Non-contingent learning data can lead to the (seemingly) illusory perception of causality if the potential cause and effect often co-occur (i.e., the effect prevalence is high) -- an effect known as "outcome-density bias". Bayesian models of causal induction explain this effect as the result of a rational learning process in which the data do not fully override held non-zero causal priors. Convincing evidence of this rational explanation requires the demonstration of an experimentally manipulated effect of causal priors, which has been lacking. We successfully manipulated participants' (N =300) causal priors through a combination of statistical and mechanism information. This manipulation moderated the outcome-density effect in the predicted way and, in one condition, even eliminated it entirely by inducing causal priors close to zero. The results strengthen the rational Bayesian view of causal induction and support computational models that formalize this view.