How bad is incoherence

Rationality is often defined in terms of coherence, assuming that a single syntactical rule such as consistency, transitivity, or Bayes’ rule suffices to evaluate behavior. Many normative claims made in psychological research follow this assumption. We argue that coherence-based norms are of limited value for evaluating behavior as rational. Specifically, we maintain that (a) there is little evidence that coherence violations are costly, or if they are, that people would fail to learn to avoid them; (b) adaptive rules of behavior can in fact imply incoherence; (c) computational intractability and conflicting goals can make coherence unattainable; and (d) coherence plays a key role in situations where it is instrumental in achieving functional goals. These observations lead to the conclusion that coherence cannot be a universal benchmark of rationality. We argue that smart choices need to be defined in terms of ecological rationality, which requires an analysis of the environmental structure and its match with cognitive strategies.