Structure and strength in causal judgments

Several recent theories have attempted to account for the judgments people make about causal relationships. We argue that questions about causal structure, such as whether or not a causal relationship actually exists, make an important contribution to these judgments. We use graphical models, a formal tool for describing causality developed in computer science, to illustrate that two leading rational accounts of causal judgment assume that a causal relationship exists and estimate the strength of that relationship. The important question of whether or not a causal relationship actually exists can be modeled as a Bayesian inference, and we present a measure of the evidence in favor of this conclusion that we term “causal support”. We show that causal support is consistent with previous results, and test its predictions with three experiments. The first and second experiments explore a novel effect predicted by this structural account, and use this effect to demonstrate that structure learning and parameter estimation can be dissociated in human judgments. The third experiment shows that learning from the rates of different events also reflects causal structure. Together, these results illustrate the importance of structural considerations in causal induction.