Causal Information as a Constraint on Similarity

Considerable evidence indicates that causal information provides a vital constraint on conceptual representation and coherence. We investigated the role of causal information as a constraint on similarity, exploiting an asymmetry between predictive causal reasoning (given the cause, predict the effect) and diagnostic causal reasoning (given the effect, diagnose the cause). This asymmetry allowed us to isolate the effects of causal understanding from the effects of sharing non-causal features. We found that judged similarity between two objects that are identical except for one feature was affected by whether that feature was a competing potential cause of an effect or an effect of a common cause. Causation Constrains Similarity Any two objects have indefinitely many features in common. For example, Murphy and Medin (1985) pointed out that the number of features that plums and lawn mowers have in common is, in principle, infinite. Both weigh less than 1000 kg, and both are found on earth, in the solar system. Both cannot hear well, have an odor, are used by people, not by elephants, and so on. But despite these shared features people do not generally consider plums and lawnmowers to be similar. Intuitively, the features plums and lawnmowers have in common are not considered important. But what features are important? Why are some features important and not others? There must be criteria for constraining the sheer number of these features (Goodman, 1972; Medin, Goldstone, & Gentner, 1993). Previous researchers (e.g., Murphy & Medin, 1985) have suggested that features will be considered more important when they are diagnostic of causal function and part of a larger explanatory framework. Whether or not a feature is causal serves as a criterion by which people select important features and separate them from unimportant ones (Sloman, Love, & Ahn, 1998). We propose a new paradigm to investigate the influence of causal knowledge on similarity and categorization. This paradigm provides an optimal way of equating the number of common and distinctive features between two objects across conditions: Equality is ensured because the stimuli were in fact identical across conditions. We use this paradigm to investigate the influence of causal knowledge as a constraint on similarity.