Evidence for and against a simple interpretation of the less-is-more effect

The less-is-more effect predicts that people can be more accurate making paired-comparison decisions when they have less knowledge, in the sense that they do not recognize all of the items in the decision domain. The traditional theoretical explanation is that decisions based on recognizing one alternative but not the other can be more accurate than decisions based on partial knowledge of both alternatives. I present new data that directly test for the less-is-more effect, coming from a task in which participants judge which of two cities is larger and indicate whether they recognize each city. A group-level analysis of these data provides evidence in favor of the less-is-more effect: there is strong evidence people make decisions consistent with recognition, and that these decisions are more accurate than those based on knowledge. An individual-level analysis of the same data, however, provides evidence inconsistent with a simple interpretation of the less-is-more effect: there is no evidence for an inverse-U-shaped relationship between accuracy and recognition, and especially no evidence that individuals who recognize a moderate number of cities outperform individuals who recognize many cities. I suggest a reconciliation of these contrasting findings, based on the systematic change of the accuracy of recognition-based decisions with the underlying recognition rate. In particular, the data show that people who recognize almost none or almost all cities make more accurate decisions by applying the recognition heuristic, when compared to the accuracy achieved by people with intermediate recognition rates. The implications of these findings for precisely defining and understanding the less-is-more effect are discussed, as are the constraints our data potentially place on models of the learning and decision-making processes involved.