The Crowd Classification Problem
Decades of research has argued that social information exchange can improve judgement accuracy as measured at both the group level and the individual level. However, we show both theoretically and empirically that this effect is limited to numeric judgements. In discrete choice estimates, also known as classification tasks—such as yes/no decisions, or selecting the better of two options—social influence simply amplifies the majority opinion, regardless of the accuracy of that opinion. As a result, initially inaccurate groups become less accurate after social information exchange but display stronger consensus. This effect is not due only to the type of information exchanged, but applies more generally to any case where group members are polled on a discrete choice, as in a voting process. These results point to the need for a contingency theory of collective intelligence identifying the types of decisions for which social information processing can improve outcomes. In the case of estimation accuracy, these results also point to a simple but effective strategy: organizations should focus on aggregating beliefs about numeric quantities, and avoid framing problems as a discrete choice until as late as possible in the decision process.
Joshua Aaron Becker, Douglas Guilbeault, Edward (Ned) Smith
Becker, Aaron Joshua, Douglas Guilbeault, and Edward (Ned) Smith. 2019. The Crowd Classification Problem.