The Wisdom of the Crowds effect - a form of collective intelligence - refers to the ability of crowds or groups to solve problems or make predictions better than the best of its individuals.
Although this is not an entirely new concept, the way new technologies have enabled the access to data analysis is fairly recent: crowdsourcing has made possible for complex real-world problems to benefit from collaborative knowledge production and decision-making processes otherwise impossible to collect or to coordinate.
Social influence has been largely suggested to have a negative impact on the wisdom of the crowds effect. Also referred as imitation, social influence diminishes the diversity of groups by introducing pressure towards conformity, limiting the independency of individual perspectives. However, non-socially influenced behaviour is extremely rare and difficult to quantify in natural environments. In fact, many social and biological systems rely on the observation of others and of the environment to adapt and revise their behaviour as a process of learning and innovating.
© Scott Page
— Experiment flow
The key for a more efficient use of social influence resides in the form social information is presented: less aggregation produces equally good results.
Departing from this idea, I explored the hypothesis that sharing the opinions of other individuals within a group might have a positive contribution to the wisdom of the crowds effect.
By displaying full information we enable multiple comparisons between options: several iterations between individual heuristics and the collective information, where the risk of copying cascades of poor estimates is unlikely because it would take several individuals to come to the same conclusion independently to reach the same quorum threshold.
We chose a variation of the classic jelly bean experiment and compared the performance of 4 groups regarding their exposure to social influence (the estimates of others). In each group, we asked approximately 90 subjects to guess the number of jelly beans in a jar.
On group 0 we showed no information about what others have guessed. On group 3, we displayed a hint as a dispersion map where all previously made estimates represented as dots, so that subjects had access to the values of previous estimates in that group.
The other two groups showed hints containing variations of aggregated information: a modified mode and a qualitative hint.
The results in our study suggest that social influence might not inhibit the WoC as suggested in previous literature. Quite the opposite, we verified that providing unaggregated social information equally produces an accurate estimation of the right number of jellybeans.
The key for a more efficient use of social influence resides in the form social information is presented: the less aggregation produces equally good results.
Our results contribute to the improvement of the collection and quality of crowd sourced data.