New study shows social networks reduce group decision accuracy by 20%

A new paper on arXiv (arXiv:2607.14288) by Dan Braha and Marcus de Aguiar explores how social network structure affects majority correctness in binary decisions. Building on the Condorcet Jury Theorem, the authors model voters who interact repeatedly on networks in the presence of committed zealots (leaders pushing a correct answer) and contrarians (who oppose the majority view). They find that while post-deliberation individual correctness can exceed random chance, the collective wisdom of the crowd degrades as social influence strengthens vote correlations.

Key findings show that hub-dominated networks (e.g., scale-free) generate stronger positive correlations among voters, reducing the effective number of independent judgments and lowering majority accuracy by up to 20% compared to well-mixed electorates. Conversely, spatially structured networks (e.g., ring or small-world) preserve weaker correlations and yield higher accuracy. The study also identifies a tipping point: if any persistent contrarian updating is present, both individual and majority correctness collapse to random chance (50%) as the electorate grows, unless social influence is purely conformist. These results have direct implications for designing social platforms and voting systems that aim to aggregate diverse opinions effectively.