Causal Models Interact with Structure Mapping to Guide Analogical Inference

We recently proposed a theoretical integration of analogical transfer with causal learning and inference (Lee & Holyoak, 2008). A Bayesian theory of learning and inference based on causal models (Lee, Holyoak & Lu, 2009) accounts for the fact that judgments of confidence in analogical inferences are partially dissociable from measures of the quality of the mapping between source and target analogs. The integrated theory postulates a dual role for causal relations, which can guide both analogical mapping and also subsequent inferences about the target. It follows that depending on whether or not a mapping is structurally ambiguous, dropping a preventive cause from the target can either decrease or increase confidence in the same analogical inference. We report an experiment that yielded data in close agreement with predictions of the Bayesian theory. These results provide further support for the importance of integrating analogical transfer with the broader framework of causal models.