Close Does Count: Evidence of a Proximity Effect in Inference from Causal Knowledge

Two studies are reported in which participants drew inferences about variables in systems of causal relationships. Previous work has shown that such inferences are influenced by information about variables that, on a normative account of causal reasoning, should be irrelevant. The present studies tested two hypotheses about how relevance is assigned to these normatively irrelevant variables. Though results were mixed, they suggest that greater relevance is assigned to variables that are closer in known causal structure to the variable about which an inference is being made. Though the learning of causal relationships from correlational evidence has received a fair amount of attention from psychologists, the use of causal knowledge to make inferences and predictions has not. This may be due to the fact that until recently most studies have focused on how learners detect the existence or strength of a causal relationship between just two dichotomous variables: a single cause and a single effect, each either present or absent. Drawing inferences or predictions from knowledge of a single causal relationship is presumably straightforward in the sense that each variable is predictive of the other: An effect is more likely present when its cause is present, and vice versa. This paper deals with inferences about variables in complex causal systems, that is, systems of causal relationships among three or more variables. Such inferences are often less straightforward than inferences from single causal relationships, in that it is not always so clear whether or under what conditions variables are relevant to one another. Suppose, for example, that a doctor knows that virus X causes a certain enzyme deficiency, which in turn causes liver damage. That is, the doctor knows this three-variable causal chain: virus X