Overall Similarity Overrides Element Similarity when Evaluating the Quality of Analogies

Abstract

Dominant computational models of analogical reasoning (e.g., SME, ACME and LISA) consider that two facts or situations are more analogous as the similarity between corresponding propositional elements increases. We report the results of two experiments demonstrating that when people judge the quality of an analogy, the similarity between matched elements is overridden by another type of similarity that implies comparing the meaning of whole propositions. In Experiment 1, participants received a base fact followed by two structurally identical target facts. Whereas in one of them propositional elements resembled their counterparts in the base, in the other they did not, but the meaning of the whole proposition resembled that of the base. Participants chose as more analogous the targets maintaining this second type of similarity. In Experiment 2, participants received a base cause followed by an effect, and were told that such effect reoccurred later as a consequence of an analogous cause. Participants had to decide which of two structurally identical facts was the cause of the target effect. Again, participants based their choices on global similarities, passing over similarities between propositional elements, but in a more ecologically valid task that involves comparing systems of relations. We conclude with some intuitions about the mechanisms underlying how people assess the quality on an analogy, and discuss their implications for future theories of analogical thinking.