Group Judgments: Statistical Means, Deliberation, and Information Markets
Cass R. Sunstein
How can groups elicit and aggregate the information held by their individual members? There are three possibilities. Groups might use the statistical mean of individual judgments; they might encourage deliberation; or they might use information markets. In both private and public institutions, deliberation is the standard way of proceeding; but for two reasons, deliberating groups often fail to make good decisions. First, the statements and acts of some group members convey relevant information, and that information often leads other people not to disclose what they know. Second, social pressures, imposed by some group members, often lead other group members to silence themselves because of fear of disapproval and associated harms. As a result, deliberation often produces a series of unfortunate results: the amplification of errors, hidden profiles, cascade effects, and group polarization. A variety of steps can be taken to ensure that deliberating groups obtain the information held by their members; restructuring private incentives, in a way that increases disclosure, is the place to start. Information markets have substantial advantages over group deliberation; such markets count among the most intriguing institutional innovations of the last quarter-century and should be used far more frequently than they now are. One advantage of information markets is that they tend to correct, rather than to amplify, the effects of individual errors. Another advantage is that they create powerful incentives to disclose, rather than to conceal, privately held information. Information markets thus provide the basis for a Hayekian critique of many current celebrations of political deliberation. They also provide a valuable heuristic for understanding how to make deliberation work better. These points bear on the discussion of normative issues, in which deliberation might also fail to improve group thinking, and in which identifiable reforms could produce better outcomes. Applications include the behavior of juries, multimember judicial panels, administrative agencies, and congressional committees; analogies, also involving information aggregation, include open source software, Internet “wikis,” and weblogs.