People use analogies for many cognitive purposes such asbuilding mental models, making inspired guesses, andextracting relational structure. Here we examine whether andhow analogies may have more direct influence on knowledge:Do people treat analogies as probabilistically trueexplanations for uncertain propositions?We report an experiment that explores how a suggestedanalogy can influence people’s confidence in inferences.Participants made predictions while simultaneouslyevaluating a suggested analogy and observed evidence. In twoconditions, the evidence is either consistent with or in conflictwith propositions based on the suggested analogy. Weanalyze the responses statistically and in a psychologicallyplausible Bayesian network model. We find that analogies areused for more than just generating candidate inferences. Theyact as probabilistic truths that affect the integration ofevidence and confidence in both the target and sourcedomains. People readily treat analogies not as a one-wayprojection from source to target, but as a mutually informativeconnection.

What does it mean for analogy to be rational? John StuartMill described a probabilistic underpinning for analogicalinference based on the the odds of observing systemicpairwise correspondence across otherwise independentsystems by mere chance. Although proponents and criticshave debated its validity, Mill’s approach has yet to beimplemented computationally or studied psychologically. Inthis paper we examine Mill’s approach and show how it canbe instantiated using Bayes theorem. Then we describe twoexperiments that present subjects with partially-revealed,aligned binary strings with varying degrees of intra- and inter-string regularity. Experimental results are compared to aformal rational analysis of the stimuli revealing conditionswhereby participants exhibit confidence patterns consistentand inconsistent with Mill’s rational basis of analogy.