Collective problem decomposition drives the wisdom of deliberative crowds

Abstract

Understanding when and why social interaction improves human judgment is a central question in the behavioural sciences. We examine whether accuracy increases when groups break problems into parts and generate approximate solutions, a process we call Collective Fermi Estimation. We tested this idea across three studies by analysing over 1,000 online group deliberations in text chatrooms. Study 1 (N=500) shows that greater use of problem decomposition in group discussions predicts lower error. Study 2 (N=240) provides causal evidence: instructing groups to apply problem decomposition leads to higher accuracy than combining initial estimates. Study 3 (N=160) shows the advantage arises when applied collectively rather than by individuals working alone. We also introduce a scalable natural language method to detect problem decomposition in deliberation text and predict collective accuracy. These findings identify problem decomposition as a key mechanism behind the wisdom of deliberative crowds and provide tools to detect and promote it.