Description
Objectives. We assessed the efficiency of small teams relative to their team members by comparing a team's performance to crowdsourcing algorithms run on the team's members. Crowdsourcing algorithms attempt to find optimal methods for aggregating individual decisions to produce an accurate "crowd" response. They are demonstrably highly effective and vary from a simple consensus to weighted combinations of individual participants' accuracy and confidence.Design. Small teams were tested on problems presented in two phases: first, the Individual Phase, where each team member privately recorded their own answer, and then the Team Phase, where team members came together and agreed on an answer.
Methods. We tested 635 people organised into 211 teams of 2–4 people. Either general knowledge or novel matrices problems were presented to different teams.
Results. Team performance was highly correlated (r > .7) with the output of crowdsourcing models run on individual team members. In every case, 75% or more of real teams outperformed the crowdsourcing algorithms. We also calculated each team's "collective knowledge" as the union of all correct responses made in the Individual Phase. Separate analyses showed that collective knowledge was a hard ceiling on team performance and that teams varied greatly in their ability to tap into this knowledge.
Conclusions. Real teams outperformed even the best crowdsourcing algorithms. The best teams were not necessarily discovering new knowledge but were better at surfacing knowledge they already had.
| Period | 27 Aug 2024 |
|---|---|
| Held at | British Psychological Society |
| Degree of Recognition | National |
Keywords
- team decision making procress