A report was published in Science last week titled “Evidence for a collective intelligence factor in the performance of human groups”. They set up the “c factor”, for collective intelligence, somehow a parallel of the g (for general intelligence. I told a bit about this after Prof. Haier’s conference at the EMBO meeting). In brief, what the authors report, is that individual intelligence of people constituting a group (a team) is not correlated with the success of the team to solve a problem. Further, when extending the tests, it came up that having more women in the team increases the so called “social sensitivity” and improves performance. Groovy, eh?

Let’s begin from somewhere. The motivation of this study seem a good start. So, the authors introduce this study in the light of the importance of the group for the way we accomplish tasks. In other words: consider research, management, and many other tasks, completed both by groups who work face-to-face or virtually. How is those teams’ performance determined? Psychologists have set up metrics for individual intelligence, but not for a group’s one. The main question addressed by this study is therefore to assess how well a single group is able to perform different tasks ranging widely; moreover, is it possible to use this information in order to predict the future performance of a group?

I should say: I like the question. Even if it is a bit stressful. However, my concern is the following part (which is the whole study, actually…): this “collective intelligence” metrics. In brief, they use the individual intelligence metric but applied to groups. In order to go deeper into this study, we need to define the general intelligence g. This is first an empirical observation: if you do well on one mental task, you tend to do well on most of the others. Importantly, performance on various cognitive tasks may be uncorrelated, which is because each relied on a given specific set of capacities and abilities not used to complete other tasks. As in the extreme cases pointed out by Prof. Haier in his talk at the EMBO meeting, people who show extraordinary intelligence solving one very particular problem (learning by heart more than 20,000 of the Pi digits is one) perform bad at other tasks. It is even observed that if practicing to improve one task makes you neglect of others, the correlation we told about above is negative.

Coming back to the g definition: statistically speaking, it is defined as the first factor extracted in a factor analysis accounting for 30-50% of the variance (subsequent factors extracted accounting for substantially less variance). The factor analysis is one carried on a dataset of individual’s performance scores on diverse set of cognitive tasks; the average correlation among those scores is positive. Thus, g is the main thing intelligence tests measure. Analogically, the authors of the current study define “a group’s collective intelligence (c) as the general ability of the group to perform a wide variety of tasks”. If we admit a group’s ability to do well on a given task is positively correlated to its ability to perform a wide range of tasks, we therefore consider that this collective intelligence is a property of the group as an entity and not just the individuals in it. In a very logical fashion, here comes the question: does the collective intelligence of a given group as a whole have predictive power “above and beyond what can be explained by knowing the abilities of the individual members”?

As one may imagine it, the first experience the authors performed was aimed at examining whether collective intelligence exists at all:

Is there a single factor for groups, a c factor, that functions in the same way for groups as general intelligence does for individuals? Or does group performance, instead, have some other correlational structure, such as several equally important but independent factors, as is typically found in research on individual personality?

They recruited 120 people and randomly assigned them in 40 teams of 3. These groups worked on a variety of tasks: brainstorming to come up with possible uses for a brick, solving Raven’s Advanced Progressive Matrices, but also more real-world ones such as planning a shopping trip for a group of people sharing the same car. Each participant’s individual intelligence was also measured independently, in the beginning of each session.

What came up supports the hypothesis that a general collective intelligence factor c exists.To go back to the factor stuff I wrote above: factor analysis of team scores yielded one factor accounting for 43% of the variance whereas the next factor accounted for only 18%. Furthermore, the average and maximal intelligence scores of individual group members are not significantly correlated with c.

This being concluding, the authors went more deeply. They recruited around 600 more people and assigned them to groups of 2 to 5. A subset of the people participating in this second experiment completed several additional tasks. This study replicated the factor analysis from the first one.

So, if c exists, what causes it? Even if combining the findings of the 2 studies showed a moderate correlation between the average intelligence of individual group members and c, the latter was still a much better predictor of group performance. When the authors examined a number of group and individual factors that might be good predictors for c, they found that many of the factors one would expect to predict group performance – such as cohesion, motivation, satisfaction – did not. However, 3 factors were found to be significantly correlated with c:

1. a significant correlation was observed between c and the average social sensitivity of group members, as measured by the Reading the mind in the eyes test;
2. c was negatively correlated with the variance in the number of speaking turns by group members (measured by sociometric badges worn by the participants). Humanly speaking, this means that groups where few people dominated the conversation were less collectively intelligent in comparison to those where there was a more equal distribution of conversational turn-taking;
3. c was positively correlated with the proportion of females in the group. However, write the authors, “this result appears to be  largely mediated by social sensitivity since (consistent with previous research) women in our sample scored better on the social sensitivity measure than men”. When performing a regression analysis with all the 3 variables, only social sensitivity reached statistical significance for performance prediction.

In conclusion, a collective intelligence exists and this c factor appears to depend upon both the composition of the group ( = average member intelligence) and factors emerging from the way group members interact (for instance, their conversational turn-taking behaviour). Of course, all this raises many questions: “could a short collective intelligence test predict a sales team’s or a top management team’s long-term effectiveness?”; more importantly, “could a group’s collective intelligence be increased by, for example, better electronic tools?”

Before finishing on this, I wanted just to point out the gender question slightly brought around. This presumably increased social sensitivity in women is to me more a gender stereotype. Indeed, there are studies showing that women can improve teams by virtue of their social acuity; but there are also studies which have found that women tend to remain quiet and let others decide, sometimes in the detriment of their team. Therefore, it is very delicate to associate women to increased social sensitivity in this direct fashion as one could think it is done in the current study. This “female factor” is to be considered with extreme care.

Anita Williams Woolley, Christopher F. Chabris, Alexander Pentland, Nada Hashmi, Thomas W. Malone (2010). Evidence for a Collective Intelligence Factor in the Performance of Human Groups Science : 10.1126/science.1193147

Image (CC-by)