Science 29 October 2010: Vol. 330 no. 6004 pp. 686-688 DOI: 10.1126/science.

1193
147
* REPORT
Evidence for a Collective Intelligence Factor in the Performance of Human Groups
1. Anita Williams Woolley1,*,
2. Christopher F. Chabris2,3,
3. Alex Pentland3,4,
4. Nada Hashmi3,5 and
5. Thomas W. Malone3,5
+Author Affiliations
1. 1Carnegie Mellon University, Tepper School of Business, Pittsburgh, PA 1521
3, USA.
2. 2Union College, Schenectady, NY 12308, USA.
3. 3Massachusetts Institute of Technology (MIT) Center for Collective Intellig
ence, Cambridge, MA 02142, USA.
4. 4MIT Media Lab, Cambridge, MA 02139, USA.
5. 5MIT Sloan School of Management, Cambridge, MA 02142, USA.
1. *To whom correspondence should be addressed. E-mail: awoolley@cmu.edu
ABSTRACTPsychologists have repeatedly shown that a single statistical factor—often
called “general intelligence”—emerges from the correlations among people’s performance
on a wide variety of cognitive tasks. But no one has systematically examined whe
ther a similar kind of “collective intelligence” exists for groups of people. In two
studies with 699 people, working in groups of two to five, we find converging e
vidence of a general collective intelligence factor that explains a group’s perfor
mance on a wide variety of tasks. This “c factor” is not strongly correlated with th
e average or maximum individual intelligence of group members but is correlated
with the average social sensitivity of group members, the equality in distributi
on of conversational turn-taking, and the proportion of females in the group.
As research, management, and many other kinds of tasks are increasingly accompli
shed by groups—working both face-to-face and virtually (1–3)—it is becoming ever more
important to understand the determinants of group performance. Over the past cen
tury, psychologists made considerable progress in defining and systematically me
asuring intelligence in individuals (4). We have used the statistical approach t
hey developed for individual intelligence to systematically measure the intellig
ence of groups. Even though social psychologists and others have studied for dec
ades how well groups perform specific tasks (5, 6), they have not attempted to m
easure group intelligence in the same way individual intelligence is measured—by a
ssessing how well a single group can perform a wide range of different tasks and
using that information to predict how that same group will perform other tasks
in the future. The goal of the research reported here was to test the hypothesis
that groups, like individuals, do have characteristic levels of intelligence, w
hich can be measured and used to predict the groups’ performance on a wide variety
of tasks.
Although controversy has surrounded it, the concept of measurable human intellig
ence is based on a fact that is still as remarkable as it was to Spearman when h
e first documented it in 1904 (7): People who do well on one mental task tend to
do well on most others, despite large variations in the tests’ contents and metho
ds of administration (4). In principle, performance on cognitive tasks could be
largely uncorrelated, as one might expect if each relied on a specific set of ca
pacities that was not used by other tasks (8). It could even be negatively corre
lated, if practicing to improve one task caused neglect of others (9). The empir
ical fact of general cognitive ability as first demonstrated by Spearman is now,
arguably, the most replicated result in all of psychology (4).
Evidence of general intelligence comes from the observation that the average cor
relation among individuals’ performance scores on a relatively diverse set of cogn
itive tasks is positive, the first factor extracted in a factor analysis of thes
e scores generally accounts for 30 to 50% of the variance, and subsequent factor
s extracted account for substantially less variance. This first factor extracted
in an analysis of individual intelligence tests is referred to as general cogni
tive ability, or g, and it is the main factor that intelligence tests measure. W
hat makes intelligence tests of substantial practical (not just theoretical) imp
ortance is that intelligence can be measured in an hour or less, and is a reliab
le predictor of a very wide range of important life outcomes over a long span of
time, including grades in school, success in many occupations, and even life ex
pectancy (4).
By analogy with individual intelligence, we define a group’s collective intelligen
ce (c) as the general ability of the group to perform a wide variety of tasks. E
mpirically, collective intelligence is the inference one draws when the ability
of a group to perform one task is correlated with that group’s ability to perform
a wide range of other tasks. This kind of collective intelligence is a property
of the group itself, not just the individuals in it. Unlike previous work that e
xamined the effect on group performance of the average intelligence of individua
l group members (10), one of our goals is to determine whether the collective in
telligence of the group as a whole has predictive power above and beyond what ca
n be explained by knowing the abilities of the individual group members.
The first question we examined was whether collective intelligence—in this sense—eve
n exists. 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 (11)?
To answer this question, we randomly assigned individuals to groups and asked th
em to perform a variety of different tasks (12). In Study 1, 40 three-person gro
ups worked together for up to 5 hours on a diverse set of simple group tasks plu
s a more complex criterion task. To guide our task sampling, we drew tasks from
all quadrants of the McGrath Task Circumplex (6, 12), a well-established taxonom
y of group tasks based on the coordination processes they require. Tasks include
d solving visual puzzles, brainstorming, making collective moral judgments, and
negotiating over limited resources. At the beginning of each session, we measure
d team members’ individual intelligence. And, as a criterion task at the end of ea
ch session, each group played checkers against a standardized computer opponent.
The results support the hypothesis that a general collective intelligence factor
(c) exists in groups. First, the average inter-item correlation for group score
s on different tasks is positive (r = 0.28) (Table 1). Next, factor analysis of
team scores yielded one factor with an initial eigenvalue accounting for more th
an 43% of the variance (in the middle of the 30 to 50% range typical in individu
al intelligence tests), whereas the next factor accounted for only 18%. Confirma
tory factor analysis supported the fit of a single latent factor model with the
data [χ2 = 1.66, P = 0.89, df = 5; omparative fit index (CFI) =.99, root mean squ
are error of approximation (RMSEA) = 0.01]. Furthermore, when the fa tor loading
s for different tasks on the first general fa tor are used to al ulate a s or
e for ea h group, this s ore strongly predi ts performan e on the riterion task
(r = 0.52, P = 0.01). Finally, the average and maximum intelligen e s ores of i
ndividual group members are not signifi antly orrelated with [r = 0.19, not s
ignifi ant (ns); r = 0.27, ns, respe tively] and not predi tive of riterion tas
k performan e (r = 0.18, ns; r = 0.13, ns, respe tively). In a regression using
both individual intelligen e and to predi t performan

e on the riterion task,
has a signifi ant effe t (β = 0.51, P = 0.001), ut average individual intellig
ence (β = 0.08, ns) and maximum individual intelligence (β =.01, ns) do not (Fig. 1)
.

View this ta le:
* In this window
*
* In a new window


Ta le 1Correlations among group tasks and descriptive statistics for Study 1. n
= 40 groups; *P ≤ 0.05; **P ≤ 0.001.

View larger version:

* In this page
*
* In a new window
* Download PowerPoint Slide for Teaching

Fig. 1Standardized regression

coefficients
for collective intelligence (c) and a
verage individual mem er intelligence when oth are regressed together on criter
ion task performance in Studies

1 and 2 (controlling for group size in Study 2).
Coefficient for maximum mem er
intelligence is also shown for comparison, calcu
lated
in a separate regression ecause
it is too highly correlated with individu
al mem er intelligence to incorporate oth in
a single analysis (r = 0.73 and 0.
62 in Studies 1 and 2, respectively). Error ars, mean ± SE.

In Study 2, we used 152 groups ranging from two to five mem ers.
Our goal was to
replicate these findings in groups of different sizes, using a roader sample o
f tasks and an alternative measure of individual intelligence. As expected, this
study replicated the findings of Study 1, yielding a first factor explaining 44
% of the variance and a second factor explaining only 20%. In addition, a confir
matory factor analysis suggests an excellent fit of the single-factor model with
the data (χ2 = 5.85, P = 0.32, df = 5; CFI = 0.98, NFI = 0.89, RMSEA = 0.03).
In addition, for a subset of the groups in Study 2, we in luded five additional
tasks, for a total of ten. The results from analyses in orporating all ten tasks
were also onsistent with the hypothesis that a general fa tor exists (see Fi
g. 2). The s ree test (13) learly suggests that a one-fa tor model is the best
fit for the data from both studies [Akaike Information Criterion (AIC) = 0.00 fo
r single-fa tor solution]. Furthermore, parallel analysis (13) suggests that onl
y fa tors with an eigenvalue above 1.38 should be retained, and there is only on
e su h fa tor in ea h sample. These on lusions are supported by examining the e
igenvalues both before and after prin ipal axis extra tion, whi h yields a first
fa tor explaining 31% of the varian e in Study 1 and 35% of the varian e in Stu
dy 2. Multiple-group onfirmatory fa tor analysis suggests that the fa tor stru
tures of the two studies are invariant (χ2 = 11.34, P = 0.66, df = 14; CFI = 0.99,
RMSEA = 0.01). Taken together, these results provide strong support for the exi
sten e of a single dominant fa tor underlying group performan e.
View larger version:
* In this page
*
* In a new window
* Download PowerPoint Slide for Tea hing

Fig. 2S ree plot demonstrating the first fa tor from ea h study a ounting for m
ore than twi e as mu h varian e as subsequent fa tors. Fa tor analysis of items
from the Wonderli Personnel Test of individual intelligen e administered to 642
individuals is in luded as a omparison.
 The riterion task used in Study 2 was an ar hite tural design task modeled afte
r a omplex resear h and development problem (14). We had a sample of 63 individ
uals omplete this task working alone, and under these ir umstan es, individual
intelligen e was a signifi ant predi tor of performan e on the task (r = 0.33,
P = 0.009).
When the same task was done by groups, however, the average individual intellige
n e of the group members was not a signifi ant predi tor of group performan e (r
= 0.18, ns). When both individual intelligen e and are used to
predi t group
performan

e, is a signifi ant predi tor (β = 0.36,
P = 0.0001), ut average grou
p mem er intelligence (β = 0.05, ns) and maximum mem er intelligence (β = 0.12, ns)
are not (Fig. 1).

If c exists, what causes it? Com ining the
findings of the two studies, the aver
age intelligence of individual group mem ers was moderately correlated with c (r

= 0.15, P = 0.04), and so was the intelligence

of the highest-scoring
team mem
er (r = 0.19, P = 0.008). However, for oth studies, cwas still a much etter pr
edictor of group performance on the criterion tasks than the average or maximum
individual intelligence
(Fig. 1).

We also examined a num er of group and individual factors that might e good pre
dictors of c. We found that many of the factors one might have expected to predi
ct group performance—such as group cohesion, motivation, and satisfaction—did not.
However, three factors were
significantly correlated with c. First, there was a
significant

correlation

etween c and the average social sensitivity of group me
m ers, as measured y the “Reading the Mind in the Eyes” test (15) (r
= 0.26, P = 0.
002). Second,
c was negatively

correlated
with the variance in
the num er
of spe

aking turns y group mem ers, as measured y the sociometric adges worn y a su
set of the groups (16) (r = –0.41, P = 0.01). In other words, groups where a few
people dominated the onversation were less olle tively intelligent than those
with a more equal distribution of onversational turn-taking.
Finally, was positively and signifi antly orrelated with the proportion of fe
males in the group (r = 0.23, P = 0.007). However, this result appears to be lar
gely mediated by so ial sensitivity (Sobel z = 1.93, P = 0.03), be ause ( onsist
ent with previous resear h) women in our sample s ored better on the so ial sens
itivity measure than men [t(441) = 3.42, P = 0.001]. In a regression analysis wi
th the groups for whi h all three variables (so ial sensitivity, speaking turn v
arian e, and per ent female) were available, all had similar predi tive power fo
r , although only so ial sensitivity rea hed statisti al signifi an e (β = 0.33,
P = 0.05) (12).

These results provide su stantial
evidence for the existence
of c in groups, ana
logous to a well-known similar a ility
in individuals. Nota ly, this collective
intelligence
factor appears to depend oth on the composition of the group (e.g.

average mem er intelligence) and
,
on factors that emerge from the way group mem

ers interact when they are assem led (e.g., their conversational turn-taking e
havior) (17, 18).
These findings raise many additional questions. For example, could a short colle
ctive intelligence test predict a sales team’s or
a top management team’s long-term
effectiveness? More importantly, it would seem to e much easier to raise the in

telligence of
a group than an
individual. Could a group’s
collective intelligence
e increased y, for example, etter electronic colla oration tools?

Many

previous studies have addressed questions like these for specific tasks, u
t y measuring the effects of specific interventions on a group’s c, one can
predi
ct the effects of those interventions on a wide
range of tasks. Thus, the a ilit


y to measure

collective intelligence as a sta le property of groups provides
ot
h a su stantial economy of effort and a range of new questions to explore in ui
lding a science of collective performance.
Supporting Online Materialwww.sciencemag.org/cgi/content/full/science.1193147/DC
1
Materials

and Methods
Ta les S1 to S4
References


* Received for pu
lication 2 June 2010.


* Accepted for pu lication 10 Septem er 2010.
References and Notes
1. ↵
1. S. Wuchty,
2. B. F. Jones,
3. B. Uzzi
, The increasing dominance of teams in production of knowledge. Science 316, 103
6 (2007).


A stract/FREE Full Text

2.
1. T. Gowers,
2. M. Nielsen

, Massively colla orative mathematics. Nature 461, 879 (2009).


CrossRefMedlineWe of Science

3. ↵
J. R. Hackman, Leading Teams: Setting the Stage for Great Performances (Harvard
Business School Press, Boston, 2002).

4. ↵
I. J. Deary, Looking Down on Human Intelligence: From Psychometrics to the Brain
(Oxford Univ. Press, New York, 2000).

5. ↵
J. R. Hackman,

C. G. Morris, in Small Groups and Social Interaction, Volume 1, H
. H. Blum erg, A. P. Hare, V. Kent, M. Davies, Eds. (Wiley, Chichester, UK, 1983
), pp. 331–345.

6. ↵
J. E. M Grath, Groups: Intera tion and Performan e (Prenti e-Hall, Englewood Cli
ffs, NJ, 1984).

7. ↵
1. C. Spearman
, “General intelligen e,” obje tively determined and measured. Am. J. Psy hol.15, 20
1 (1904).
CrossRefWeb of S ien e

8. ↵
C. F. Chabris, in Integrating the Mind: Domain General Versus Domain Spe ifi Pr
o esses in Higher Cognition, M. J. Roberts, Ed. (Psy hology Press, Hove, UK, 200
7), pp. 449–491.

9. ↵
C. Brand, The g Fa tor (Wiley, Chi hester, UK, 1996).

10. ↵
1. D. J. Devine,
2. J. L. Philips
, Do smarter teams do better: A meta-analysis of ognitive ability and team perf
orman e. Small Group Res. 32, 507 (2001).
CrossRef

11. ↵
1. R. R. M Crae,
2. P. T. Costa Jr..
, Validation of the five-fa tor model of personality a ross instruments and obse
rvers. J. Pers. So . Psy hol. 52, 81 (1987).
CrossRefMedlineWeb of S ien e

12. ↵
Materials and methods are available as supporting material on S ien e Online.
13. ↵
1. R. B. Cattell
, The s ree test for the number of fa tors. Multivariate Behav. Res. 1, 245 (196
6).
CrossRefWeb of S ien e

14. ↵
1. A. W. Woolley
, Means vs. ends: Impli ations of pro ess and out ome fo us for team adaptation
and performan e. Organ. S i. 20, 500 (2009).
Abstra t/FREE Full Text

15. ↵
1. S. Baron-Cohen,
2. S. Wheelwright,
3. J. Hill,
4. Y. Raste,
5. I. Plumb
, The “Reading the Mind in the Eyes” test revised version: A study with normal adult
s, and adults with Asperger syndrome or high-fun tioning autism. J. Child Psy ho
l. Psy hiatry 42, 241 (2001).
CrossRefMedlineWeb of S ien e

16. ↵
A. Pentland, Honest Signals: How They Shape Our World (Bradford Books, Cambridge
, MA, 2008).

17. ↵
1. L. K. Mi haelsen,
2. W. E. Watson,
3. R. H. Bla k
, A realisti test of individual versus group onsensus de ision making. J. Appl
. Psy hol. 74, 834 (1989).
CrossRefWeb of S ien e

18. ↵
1. R. S. Tindale,
2. J. R. Larson
, Assembly bonus effe t or typi al group performan e? A omment on Mi haelsen, W
atson, and Bla k (1989). J. Appl. Psy hol. 77, 102 (1992).
CrossRefWeb of S ien e

19. This work was made possible by finan ial support from the National S ien e
Foundation (grant IIS-0963451), the Army Resear h Offi e (grant 56692-MA), the
Berkman Fa ulty Development Fund at Carnegie Mellon University, and Cis o System
s, In ., through their sponsorship of the MIT Center for Colle tive Intelligen e
. We would espe ially like to thank S. Kosslyn for his invaluable help in the in
itial on eptualization and early stages of this work and I. Aggarwal and W. Don
g for substantial help with data olle tion and analysis. We are also grateful f
or omments and resear h assistan e from L. Argote, E. Anderson, J. Chapman, M.
Ding, S. Gaikwad, C. Huang, J. Introne, C. Lee, N. Nath, S. Pandey, N. Peterson,
H. Ra, C. Ritter, F. Sun, E. Sievers, K. Tenabe, and R. Wong. The hardware and
software used in olle ting so iometri data are the subje t of an MIT patent ap
pli ation and will be provided for a ademi resear h via a not-for-profit arrang
ement through A.P. In addition to the affiliations listed above, T.W.M. is also
a member of the Strategi Advisory Board at InnoCentive, In .; a dire tor of Ser
iosity, In .; and hairman of Phios Corporation.

Arti le Views
1. Abstra t
2. Full Text
3. Full Text (PDF)
4. Figures Only
5. Supporting Online Material
6. Pod ast Interview
7. VERSION HISTORY
1. 330/6004/686(most re ent)
 2. s ien e.1193147v2
3. s ien e.1193147v1

Arti le Tools
1. Save to My Folders
2. Download Citation
3. Alert Me When Arti le is Cited
4.
* Post to CiteULike

5. E-mail This Page

6. Submit an E-Letter
7. Commer ial Reprints and Permissions
8. View PubMed Citation
Related ContentSimilar Arti les In:
1. S ien e Magazine
2. PubMed
Sear h Google S holar for:
1. Arti les by Woolley, A. W.
2. Arti les by Malone, T. W.
Sear h PubMed for:
1. Arti les by Woolley, A. W.
2. Arti les by Malone, T. W.
Find Citing Arti les in:
1. Web of S ien e (2)
2. CrossRef
3. S opus (1)

My S ien e
1. My Folders
2. My Alerts
3. My Saved Sear hes
4. Sign In
More InformationMore in Colle tions
* Psy hology
Related Jobs from S ien eCareers
* Psy hology

ADVERTISEMENT

ADVERTISEMENT
To Advertise Find Produ ts
S ien e. ISSN 0036-8075 (print), 1095-9203 (online)
News | S ien e Journals | Careers | Blogs and Communities | Multimedia
| Colle tions | Help | Site Map | RSS
Subs ribe | Feedba k | Priva y / Legal | About Us | Advertise With Us |
Conta t Us
© 2011 Ameri an Asso iation for the Advan ement of S ien e. All Rights Reserved.
AAAS is a partner of HINARI, AGORA, OARE, eIFL, PatientInform, CrossRef, and COU
NTER.