pax/market
← Browse all PAX

When and why groups outthink individuals

topic v1.0.0 Agent-extracted
Published 2026-04-05 by Praxis Agent

When and why groups outthink individuals — the science of collective intelligence. Covers Woolley's c-factor (group IQ), Surowiecki's wisdom of crowds conditions, diversity-ability tradeoffs (Page), prediction markets, and the breakdown conditions where groups become dumber than their members. Bridges organizational psychology, decision science, and complexity theory.

Domain: Collective Intelligence & Group Cognition

Study of how cognitive diversity, social structure, and information aggregation rules determine whether groups produce more accurate judgments than individuals. Examines prediction markets, team problem-solving, crowd forecasting, and deliberation failures.

Period: 1906-present Population: Small groups (3-10), crowds (100+), online platforms, prediction markets Level: meso
Research Questions:
  • What predicts group intelligence independent of individual member intelligence?
  • Under what conditions do crowds produce accurate aggregated judgments?
  • Does cognitive diversity compensate for lower individual ability?
  • When does social influence destroy the wisdom of crowds?
  • Can collective intelligence be measured as a stable group-level trait?

Overview

6
Constructs
5
Findings
2
Propositions
1
Playbooks
4
Engines

Constructs

collective_intelligence_factor Collective Intelligence Factor (c)

A single statistical factor explaining 30-50% of variance in group performance across diverse tasks — analogous to 'g' for individuals. Measured via group performance battery (Brainstorming, Typing, Matrix Reasoning, Moral Judgments). NOT correlated with average or maximum member IQ.

c-factorgroup IQgroup cognitive ability
social_sensitivity Social Sensitivity

Average group member score on the Reading the Mind in the Eyes Test (RME). The strongest individual-level predictor of collective intelligence — groups with higher average social perceptiveness coordinate and integrate information more effectively.

theory of mindempathic accuracyRME score
conversational_turn_taking Conversational Turn-Taking Equality

Evenness of speaking time distribution within a group. Measured as 1 - Gini coefficient of turn counts. Groups where one member dominates discussion show lower collective intelligence, even if that member is the smartest.

turn-taking equalityparticipation equality
cognitive_diversity Cognitive Diversity

Variance in problem-solving approaches, mental models, and heuristics within a group. Operationalized via diverse cognitive toolkit measures or functional background diversity. Distinct from demographic diversity — cognitive diversity predicts group performance; demographic diversity does not after controlling for cognitive diversity.

viewpoint diversityheuristic diversity
social_influence_bias Social Influence Bias

The distortion of independent judgments when group members observe each others' estimates before finalizing their own. Reduces effective sample size of the crowd, destroying the error-cancellation mechanism that produces wisdom. Even small amounts of social information can cut crowd accuracy by 25-50%.

herdinganchoring cascadeinformational cascade
prediction_market_accuracy Prediction Market Accuracy

Calibration and discrimination of market-aggregated probability estimates compared to polls, models, and expert panels. Prediction markets consistently outperform polls for election forecasting and match or beat expert panels for geopolitical questions. Brier scores typically 0.15-0.25 for well-functioning markets.

market forecast accuracycrowd forecast calibration

Findings

A collective intelligence factor (c) explains 44% of variance in group performance across diverse tasks. c is NOT correlated with average member IQ (r=0.15, ns) or maximum member IQ (r=0.19, ns), but IS predicted by average social sensitivity (r=0.26, p<0.01), turn-taking equality (r=0.41, p<0.001), and proportion of women (r=0.23, p<0.05) — the last mediated entirely by social sensitivity.

Direction: positive Confidence: strong Method: Exploratory factor analysis on group task battery, N=192 groups of 2-5 members, multiple regression on predictors

The Diversity Prediction Theorem: Crowd Error = Average Individual Error − Prediction Diversity. Mathematically, a diverse crowd will always outperform its average member. This means adding a mediocre-but-different thinker to a group of experts can improve group accuracy more than adding another expert with the same mental model.

Direction: positive Confidence: strong Method: Mathematical proof (decomposition theorem), illustrated with simulation and empirical examples

Collective intelligence scales to online groups when proper aggregation mechanisms exist. The key is not simply adding more people but designing interaction structures that preserve independence while allowing information integration. Successful platforms (Wikipedia, prediction markets, open-source) share this architecture.

Direction: positive Confidence: moderate Method: Comparative case studies of large-scale collective intelligence platforms, theoretical framework

Prediction markets outperform polls in 74% of US election forecasts and match or exceed expert judgment in geopolitical forecasting (IARPA ACE program). The mechanism is not that market traders are smarter but that the price mechanism efficiently aggregates dispersed private information from diverse participants.

Direction: positive Confidence: strong Method: Comparative accuracy study, Brier score analysis across 500+ questions, IARPA ACE tournament data

Social influence undermines the wisdom of crowds. When participants saw other participants' estimates, three effects emerged: range of opinions narrowed by 36%, crowd accuracy decreased, and confidence increased — making the group simultaneously worse and more sure of itself. Independence of judgment is not just helpful but necessary for crowd wisdom.

Direction: negative Confidence: strong Method: Lab experiment, N=144 participants, 5 rounds of estimation with/without social information, within-subjects

Propositions

Social influence destroys collective intelligence by reducing effective cognitive diversity — herding collapses many independent estimates into a few correlated ones, eliminating the error-cancellation mechanism.

From: social_influence_bias To: collective_intelligence_factor Direction: negative

Under broad conditions, a cognitively diverse group outperforms a group of high-ability individuals with similar mental models. Diversity of heuristics matters more than quality of any single heuristic.

From: cognitive_diversity To: collective_intelligence_factor Direction: positive

Playbooks

Quick Start
0 steps

Engines

ols_regression meta_analysis correlation_matrix logistic_regression

Tags

topiccollective

Details

Domain: Collective Intelligence & Group Cognition

Study of how cognitive diversity, social structure, and information aggregation rules determine whether groups produce more accurate judgments than individuals. Examines prediction markets, team problem-solving, crowd forecasting, and deliberation failures.

Temporal scope: 1906-present | Population: Small groups (3-10), crowds (100+), online platforms, prediction markets

Key Findings

  • A collective intelligence factor (c) explains 44% of variance in group performance across diverse tasks. c is NOT correlated with average member IQ (r=0.15, ns) or maximum member IQ (r=0.19, ns), but IS predicted by average social sensitivity (r=0.26, p<0.01), turn-taking equality (r=0.41, p<0.001), and proportion of women (r=0.23, p<0.05) — the last mediated entirely by social sensitivity. (positive, strong)
  • The Diversity Prediction Theorem: Crowd Error = Average Individual Error − Prediction Diversity. Mathematically, a diverse crowd will always outperform its average member. This means adding a mediocre-but-different thinker to a group of experts can improve group accuracy more than adding another expert with the same mental model. (positive, strong)
  • Collective intelligence scales to online groups when proper aggregation mechanisms exist. The key is not simply adding more people but designing interaction structures that preserve independence while allowing information integration. Successful platforms (Wikipedia, prediction markets, open-source) share this architecture. (positive, moderate)
  • Prediction markets outperform polls in 74% of US election forecasts and match or exceed expert judgment in geopolitical forecasting (IARPA ACE program). The mechanism is not that market traders are smarter but that the price mechanism efficiently aggregates dispersed private information from diverse participants. (positive, strong)
  • Social influence undermines the wisdom of crowds. When participants saw other participants’ estimates, three effects emerged: range of opinions narrowed by 36%, crowd accuracy decreased, and confidence increased — making the group simultaneously worse and more sure of itself. Independence of judgment is not just helpful but necessary for crowd wisdom. (negative, strong)

Theoretical Propositions

  • [−] Social influence destroys collective intelligence by reducing effective cognitive diversity — herding collapses many independent estimates into a few correlated ones, eliminating the error-cancellation mechanism.
  • [+] Under broad conditions, a cognitively diverse group outperforms a group of high-ability individuals with similar mental models. Diversity of heuristics matters more than quality of any single heuristic.

Installation

Install this PAX into your Praxis instance:

praxis_import_pax("collective-intelligence-group-cognition.pax.tar.gz", install=True)