Explaining Four Psychological Asymmetries in Causal Reasoning : Implications of Causal Assumptions for Coherence

In this chapter, we describe four psychological phenomena that offer clues to how untutored people infer causal relations. We contrast the predictions for these phenomena, all involving asymmetries in causal inferences, according to two psychological approaches—an associationist approach analysis reveals that each phenomenon is inexplicable by associationist models but follows coherently from a causal theory. What distinguishes these approaches is that the causal theory has the goal of explaining the occurrence of a target event by the potentially independent influences of candidate causes and other (background) causes. This goal has no analogue in associationist models. To arrive at a coherent explanation, the causal account creates a theoretical construct of causal power (Cartwright 1989)— the probability with which a cause influences an effect. According to this account, reasoners search for, or define, candidate causes with the goal of arriving at causes that influence a target effect independently of the background causes. In other words, they seek causes whose powers are (ideally) invariant regardless of how frequently the background causes occur (see Woodward 2003, for a discussion of the degree of invariance and depth of explanation; also see Haavelmo 1944, for a discussion of causes varying on degree of autonomy). By ''cause,'' we mean both simple causes that consist of a single element and conjunctive causes that consist of a combination of two or more elements acting in concert; we also mean a direct cause in the sense that, for the purpose of analysis, intermediate causes that lie on the path between the candidate cause and the effect are ignored or treated as part of the candidate. In our view, causal explanation occurs within a hypothesis-testing framework in which predictions based on various sets of assumptions are evaluated to reach the goal of a satisfactory explanation. This testing begins with simpler hypotheses unless there is evidence refuting them. Any processing system that cannot simultaneously evaluate all possible hypotheses needs an ordering bias; two reasons supporting a simplicity bias are: (1) simple causes are an inherent part of the definition of conjunc-tive causes (Novick and Cheng 2004), and (2) they are the elements in more complex networks. We restrict our discussion to causes and effects that are represented by binary variables with a present value and an absent value; this type of cause and effect, compared with the type represented by continuous variables, more clearly reveals the function of causal constructs. As will …