Foundations of human causal reasoning
Fugelsang, Jonathan Albert
When evaluating the efficacy of causal candidates, peoples' judgments may be influenced by both the observed empirical evidence (e.g., Cheng, 1997), and their belief in a causal mechanism (e.g., White, 1989). Models of causal reasoning, however, have typically addressed each source of information in isolation. An integrative model for how people combine belief- and covariation-based cues when assessing causal hypotheses is proposed, which is based on the findings of several experiments. Fugelsang and Thompson (2000) presented a series of experiments that demonstrated that people weigh empirical evidence in light of their pre-existing beliefs, such that empirical evidence is given more weight for believable than for unbelievable candidates. The goal of the current series of experiments was twofold: (1) to determine how one's knowledge about cause and effect is represented, and (2) to determine how the specific nature of this knowledge representation affects one's evaluation of empirical evidence. Results from five experiments provided strong support for the conclusion that causal beliefs can be represented by two non redundant properties: mechanism-, 'and' covariation-based. Furthermore, by independently manipulating believability in terms of mechanism- and covariation-based information, the results of Experiments 4 and 5 demonstrated that it is the belief in a causal mechanism, rather than the belief that the cause and effect covary, that determines the use of empirical evidence. In addition, by correlating participants' subjective use of causal cues (i.e., belief- and covariation-based) with their actual use of causal cues, Experiment 5 revealed that the application of causal beliefs was largely unconscious and thus beyond the realm of metacognitive awareness. In contrast, participants' subjective use of covariation-based cues was highly correlated with their actual use of such cues, suggesting that participants were metacognitively aware of their use of covariation-based cues. These data are formalized in a descriptive model that provides an account of belief/evidence interactions using multiple belief representations. Implications of the proposed model are discussed with reference to current theories of causal reasoning in particular (e.g., Cheng, 1997; White, 1989) and human decision making in general (e.g., Evans & Over, in press).