Evidence for a collective intelligence factor in the performance of human groups
Citation: Anita Williams Woolley, Christopher F. Chabris, Alex Pentland, Nada Hashmi, Thomas W. Malone (2010) Evidence for a collective intelligence factor in the performance of human groups. Science (RSS)
DOI (original publisher): 10.1126/science.1193147
Semantic Scholar (metadata): 10.1126/science.1193147
Sci-Hub (fulltext): 10.1126/science.1193147
Internet Archive Scholar (search for fulltext): Evidence for a collective intelligence factor in the performance of human groups
Tagged: Psychology (RSS) collective intelligence (RSS), general intelligence (RSS), groups (RSS), turn-taking (RSS), gender (RSS)
In Psychology, there is an central (although very controversial) concept of "general intelligence" or g. Empirical support for the concept of individual intelligence using the g factor comes from the fact that when individuals' performance is measured on a large number of cognitive tests which encompass a wide variety of different tasks, factor analysis of all of these tests shows a single factor that explains between 30-50 percent of the variance in the outcome across tests within a person. Intelligence tests are designed to be highly correlated with this factor with the idea that they will then be useful predictors of a variety of different types of cognitive tasks.
This paper presents initial empirical evidence that a similar type of factor may exist for groups. The authors call this construct "collective intelligence" or c. The authors pull different types of tasks from the taxonomy of group tasks displayed on the the McGrath's circumplex Typology of tasks. These include:
- A brainstorming tasks to find uses for a brick measured by how many things they created.
- A set of matrix reasoning questions completed as a group.
- A moral reasoning issue using a fictitious disciplinary action case.
- A task to plan a shopping trip.
- A group typing task where they had to type up a document as a group.
- The groups played checkers as a group against a standardized opponent.
The authors found that there was a single factor, similar in magnitude to g explaining between 30-50% of the variance in the outcomes of the varied tasks. Scree plots show that the addition of subsequent factors do relatively little to improve the predictive power.
The authors show that this c factor was not strongly correlated either with the average intelligence (i.e., g factor as measured through an intelligence task) or with the maximum intelligence score in the group. However, c was correlated with the average social sensitivity of the group and in the equality of term turn-taking as measured through socio-metric badges in the group. It was also correlated with the proportion of females in the group although OLS regressions implied that this was because females were more likely to take turns and to be more socially sensitive.
Theoretical and Practical Relevance
In psychology and in policy circles, the importance and the role that g should play in analysis is highly controversial. That said, the existence of the factor in the context of the types of texts included in cognitive intelligence batteries is very established empirically over nearly a century of research. By relying so closely on g in the model of c, this article has already been subject to criticism.
Although early, it seems to have important implications for issues of team design in certain contexts.
Press and other descriptions
Defend Your Research: What Makes a Team Smarter? More Women by Anita Woolley and Thomas Malone in Harvard Business Review