A rational process model of reasoning causally with continuous variables

Abstract

People have been shown to be effective causal reasoners. Yet, they also commit systematic errors. A model referred to as the mutation sampler explains this dual pattern of results by positing that causal inferences arise from a rational process, that is, an algorithm that is able to compute normatively correct answers but sometimes falls short due to cognitive resource limitations. To date, tests of this account has been limited to binary variables and usually generative causal relations, relations in which a cause makes its effect more probable. This study conducts new empirical tests of how people draw causal inferences with continuous variables that form a common cause network that are related by a mixture of generative and inhibitory relations (a cause lowers the value of its effects). The results showed that people commit the same qualitative errors with continuous variables that they do with binary ones and, moreover, that these effects are explained by a new version of the mutation sampler developed for continuous variables. Additional analyses indicate that the sampling process that the mutation sampler posits had a quantitative effect on all of participants’ causal inferences, not just those that exhibit qualitative violations of normative reasoning.