Deep Structures of Collaboration: Physiological Correlates of Collective Intelligence and Group Satisfaction

Abstract

Collective intelligence (CI), a group's capacity to perform a wide variety of tasks, is a key factor in successful collaboration. Group composition, particularly diversity and member social perceptiveness, are consistent predictors of CI, but we have limited knowledge about the mechanisms underlying their effects. To address this gap, we examine how physiological synchrony, as an indicator of coordination and rapport, relates to CI in computer-mediated teams, and if synchrony might serve as a mechanism explaining the effect of group composition on CI. We present results from a laboratory experiment where 60 dyads completed the Test of Collective Intelligence (TCI) together online and rated their group satisfaction, while wearing physiological sensors. We find that synchrony in facial expressions (indicative of shared experience) was associated with CI and synchrony in electrodermal activity (indicative of shared arousal) with group satisfaction. Furthermore, various forms of synchrony mediated the effect of member diversity and social perceptiveness on CI and group satisfaction. Our results have important implications for online collaborations and distributed teams.

Full text

Deep Structures of Collaboration: Physiological Correlates
of Collective Intelligence and Group Satisfaction
Prerna Chikersal
Carnegie Mellon University
Pittsburgh, PA 15213
prerna@cmu.edu
Maria Tomprou
Carnegie Mellon University
Pittsburgh, PA 15213
mtomprou@andrew.cmu.edu
Young Ji Kim
Massachusetts Institute of
Technology
Cambridge, MA 02142
youngji@mit.edu
Anita Williams Woolley
Carnegie Mellon University
Pittsburgh, PA 15213
awoolley@cmu.edu
Laura Dabbish
Carnegie Mellon University
Pittsburgh, PA 15213
dabbish@andrew.cmu.edu
ABSTRACT
Collective intelligence (CI), a group’s capacity to perform a
wide variety of tasks, is a key factor in successful
collaboration. Group composition, particularly diversity and
member social perceptiveness, are consistent predictors of
CI, but we have limited knowledge about the mechanisms
underlying their effects. To address this gap, we examine
how physiological synchrony, as an indicator of
coordination and rapport, relates to CI in computer-
mediated teams, and if synchrony might serve as a
mechanism explaining the effect of group composition on
CI. We present results from a laboratory experiment where
60 dyads completed the Test of Collective Intelligence
(TCI) together online and rated their group satisfaction,
while wearing physiological sensors. We find that
synchrony in facial expressions (indicative of shared
experience) was associated with CI and synchrony in
electrodermal activity (indicative of shared arousal) with
group satisfaction. Furthermore, various forms of
synchrony mediated the effect of member diversity and
social perceptiveness on CI and group satisfaction. Our
results have important implications for online
collaborations and distributed teams.
Author Keywords
Physiological synchrony; Behavioral similarity; Collective
Intelligence; Distributed/ virtual teams
ACM Classification Keywords
H.5.3 Information interfaces and presentation (e.g., HCI):
Group and Organization Interfaces – collaborative
computing, computer-supported cooperative work.
INTRODUCTION
Recent research has demonstrated that groups exhibit
“collective intelligence” (CI) [95] defined as a group’s
capacity to perform a wide variety of tasks, and that CI is
consistently predictive of future performance [27, 28, 44,
94]. Further, CI has been shown to be heavily influenced by
team composition and team structure [94], particularly by
team diversity (in terms of sex composition and cognitive
diversity; [1]) and inclusion of members with higher
average social perceptiveness [28, 65, 95]. These results
have been replicated with groups working online [28] and
in groups in multiple cultures [27]. In addition to
predicting team performance, CI is also associated with
teams’ ability to engage in tacit coordination, or
coordination without communication [1].
Despite advances in our understanding of CI and its
relationship with team performance, we lack understanding
of its so-called deep structure, that is how CI develops, and
how details of physiological responses and behavior are
related to CI and collaboration outcomes. The previous
results on CI lead us to ask whether a basic mechanism via
which group diversity or composition affects CI may reside
in the sensing, and possibly synchronization, of subtle
nonverbal physiological signals. Along with CI, we will
also explore whether members’ satisfaction with the team,
as a measure of how team members “feel” about the
interaction, is associated with similar physiological
mechanisms.
Our study uses new sensing instrumentation to explore the
connections between diversity, physiological synchrony, CI
and group satisfaction. Specifically we investigate how
synchrony in physiological responses such as in
electrodermal activity, heart rate, and in facial expressions
is a mechanism via which diversity affects collective
intelligence and group satisfaction in computer-mediated
interaction.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. Copyrights for
components of this work owned by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, or republish, to
post on servers or to redistribute to lists, requires prior specific permission
and/or a fee. Request permissions from Permissions@acm.org.
CSCW '17, February 25-March 01, 2017, Portland, OR, USA
© 2017 ACM. ISBN 978-1-4503-4335-0/17/03
…$15.00
DOI: http://dx.doi.org/10.1145/2998181.2998250

Our results contribute to the field of CSCW by extending
our understanding of the mechanisms underlying collective
intelligence and group satisfaction. These findings enhance
our ability to evaluate the quality of interaction in ongoing
computer-mediated teams, advance our ability to model and
detect collective intelligence in computer-mediated teams,
and inform interventions that could build collective
intelligence during early stages of collaboration.
BACKGROUND
Physiological Sensing and Computer Supported
Collaborative Work
The previous research in CSCW on sensing group
interaction falls primarily in two categories, intelligent
meeting systems and group feedback systems. Intelligent
meeting rooms use sensors to analyze ongoing group
behavior for capture, annotation and review or intelligent
intervention [97]. For example, computer vision or spatial
microphones are used for speaker identification and
automatic camera control rather than assessing
interpersonal dynamics (e.g. [66]). More recently sensing is
being used to support smart rooms that are aware of
participant activity and location for augmented reality
interface placement and behavior (e.g. the GravitySpace
system tracks users and poses in a smart room to support
AR projections that do not intersect with users bodies [16]).
Research on group feedback systems has also used sensors
to measure behavior of individuals during group
interactions in order to improve interaction quality and
collaborative outcomes through real-time or posthoc
visualization or feedback displays [10, 23, 43, 53, 82].
Much of this work has focused on equalizing participation
in face-to-face meetings. The earliest of these systems
SecondMessenger, was a visualization system for reviewing
speaker participation patterns in a face-to-face discussion
[23]. The goal in group feedback systems research is to
capture patterns of interaction dynamics and reflect them to
participants in real time or after the fact to alter the
dynamics in some way.
Other work beyond these two categories has used
physiological sensors to detect engagement or stress levels
during collaboration. These sensors have been used to
evaluate the quality of collaborative experiences. For
example, Mandryk & Inkpen [57] pioneered the use of
physiological indicators for indicating engagement levels
during collaborative gameplay interactions. In the remote
collaboration setting, recent work by Tan et al [84] used
physiological sensors to give remote collaborators
awareness of their partner’s workload. Their results
suggested that collaboration supported by physiological
feedback provides unobtrusive awareness of confusion or
difficulty during a remote assembly task. These results
suggest the potential for physiological sensing to provide a
more fine-grained understanding of participant experience
during collaborative interaction in addition to facilitating
awareness among partners. Ultimately this sensing may be
able to drive feedback systems and displays that improve
collaboration quality.
Our study is distinct from the previous CSCW research on
sensing in collaboration in that we are using physiological
sensing to examine the underlying dynamics of
collaboration, specifically synchrony, and how it is
connected to collaborative outcomes. The previous work
did not associate synchrony, or physiological sensing at the
group level, with collective intelligence or group
satisfaction as collaboration outcomes. Thus our results will
inform whether and how sensing can provide early
indicators of a group's future potential for collaboration.
Collective Intelligence and Group Satisfaction
Psychologists have repeatedly shown that a single statistical
factor called “general intelligence” or “g” emerges from the
correlations among individuals’ performance on a wide
variety of cognitive tasks, and that it predicts an
individual’s future performance. This general factor is in
addition to more task-specific intelligences [39] with the
majority of empirical analyses of individual ability
supporting the notion of a general intelligence factor [21]
Similarly, researchers recently explored whether such a
general intelligence factor exists for groups, by adopting the
same approach that psychologists have used in examining
general intelligence in individuals [95]. They gave a sample
of groups a wide range of different types of tasks, and
found that teams that did well on one type of tasks tended to
also do well on all of the other tasks. A factor analysis of
the groups' scores revealed a single, dominant, general
factor explaining a large proportion of the variance in all of
the groups' scores. In individuals, this factor is called
“general intelligence” or “IQ;” for groups, they call this
first factor “collective intelligence.” Collective intelligence
(CI) was then shown to predict team’s future performance
on a more complex task [95]. Recent work has replicated
these findings with groups working for just one hour on an
online battery of tasks [28], in student teams [92], and in
groups in multiple cultures [28, 31]. In many of these
settings, collective intelligence has been shown to predict
future performance, consistent with the prior research done
in face-to-face groups.
A consistently puzzling observation in the work on CI is its
lack of relationship with various measures gauging the
quality of member interpersonal relationships [95, 28, 92].
Variables such as group satisfaction or cohesion are
generally treated as reliable indicators of the level of
rapport in a team, even in online collaborations [60]. Since
physiological synchrony has been shown to be an indicator
of interpersonal rapport and relationship quality [71], it may
be the case that some forms of synchrony will relate more
to group satisfaction than CI.
Despite the empirical evidence of collective intelligence
and its utility for predicting performance, research on
collective intelligence is still in its infancy, leaving many
questions unresolved. Prior to the recent studies in human

teams, work on collective intelligence originated in other
species, where it manifests as large scale coordination with
a physiological basis, such as the following of pheromone
scent trails by ants [32] or the reaction to visual signals in
fish shoals [9] raising plausible research questions as to
whether similar effects exist in human interactions. As such
here we examine the physiological signals shown to govern
various aspects of human social interaction in prior studies
to investigate their role in human collective intelligence.
We focus on the synchrony of facial myography or what we
will refer to facial expressions, electrodermal activity and
heart rate as previous experiments have shown synchrony in
these signals relate to the quality of social interactions and
level of cooperation both online and face to face (e.g., [36,
62, 69, 71]).
Physiological Synchrony
Across different social environments people often engage in
group activities that lead the members to act in synchrony
with each other [89]. As a social phenomenon, synchrony
promotes affiliation and closeness among members across
different teams and groups of individuals (e.g., from close
relationships and newly formed teams to armies and group
dancers; [38, 89]). Studies consistently find that behavioral
synchrony promotes affiliation, establishes rapport and
cooperation and supports the pursuit of joint goals [12, 67
87]. Scholars reason that behavioral synchrony functions as
social glue, which is powerful to promote coordinated
action and joint outcomes. More recently, researchers have
begun to test whether physiological synchrony (manifested
in the synchronization of less consciously controllable
physiological processes) reveals similar effects.
Recent studies build on advances in sensing technology that
provide better capabilities to examine physiological
synchrony and how it influences group performance and
collaboration. Mitkidis et al. [69] found that trust has a
positive effect on heart rate synchrony, and that the degree
of heart rate synchrony was predictive of participants’
expectations of their partners in a behavioral economics
game. Mønster et al. [71] showed that synchronous
activation of the zygomaticus major (the smile muscle) was
related to team cohesion and members’ decisions to adopt a
new routine, whereas synchrony in electrodermal activity
was related to negative affect and group tension.
Interestingly, they found no relationship between
physiological synchrony and task performance, but did find
a strong relationship between physiological synchrony and
the emotional aspects of cooperation (i.e., team cohesion
and team tension). During adult-child interactions,
synchrony in electrodermal activity is related with child’s
engagement levels [36] and better emotional attunement
[4]. Synchrony in facial expressions also promotes
emotional contagion among dyads [62].
In this study, we focus on the physiological synchrony of
heart rate, electrodermal activity and facial expressions,
which have been shown in prior research to influence
cooperation among teams [12, 18]. Electrodermal activity
(EDA) or skin conductance is an indicator of emotional
arousal and reactivity, both at conscious and unconscious
levels [2, 73]. Heart rate is a measure of cardiac activity and
also an indicator of arousal [2, 76] that has been used in
different studies related to emotional episodes [37]. Both
measures capture unconscious physiological processes and
variations that link to certain emotional states such as
positive affect, anxiety, and boredom as well as cognitive
states such as level of engagement, arousal and attention
[64, 36, 30, 74, 19, 75, 49, 20]. We interpret HR and EDA
to be indicative of generalized arousal [24, 35, 56], however
there is some empirical evidence that shows that EDA and
HR can alter in opposite ways in certain conditions [49, 50],
referred to as directional fractionation1. However we do not
anticipate those conditions to be relevant to the current
study and thus expect the signals of arousal via HR and
EDA to be consistent. Further, facial myography can also
complement the physiological picture of felt experience
[90]. Previous research has shown that facial expressions
can reliably detect conscious and unconscious experiences
of affect [79] or mimicry [62].
Hypothesis
Based on the existing findings, we now walk through our
predictions for the current study.
Physiological Synchrony and CI
CI has been associated with teams’ ability to engage in tacit
coordination, that is mutual adjustment without explicit
communication [1, 91]. This relationship was observed in
teams of strangers participating in a laboratory study
together over the space of a few hours, versus among
individuals with a long-standing relationship, suggesting
that the level of coordination undergirding CI must exist at
a fairly basic, sensory level. Additional evidence
demonstrates that CI levels established in a team’s first
interaction remain relatively stable over time, even
following a period of months of regular interaction [93].
This reinforces the notion that rather than being based on
relational elements that are developed over time, CI may be
rooted in much more basic, instantaneous, and perhaps
sensory-based mechanisms, which communicate and
perceive all kinds of interpersonal information and
1 For example, in some studies, attentiveness caused EDA
to increase [30, 74] and HR to decrease [19, 50]. These
findings were not consistent though [17], and did not hold
when the subject was moving or performing tasks requiring
cognitive effort or when two or more cognitive constructs
occurred simultaneously (eg: cognitive effort while paying
attention) [50, 42] such as TCI). Additionally, this
phenomenon is usually related to cognitive constructs like
attention or anxiety, and measuring these algorithmically
using physiological signals will require validation against
participant reported values. Hence, we do not analyze the
effects of directional fractionation on CI in this study.

expectancies, similar to those that drive thin-slice
judgments [3].
However, exactly which forms of physiological synchrony
will support CI is unclear, particularly given the lack of
relationship demonstrated in prior research between CI and
group satisfaction, cohesion, and psychological safety [27,
28, 31, 95]. Many of the existing studies of physiological
synchrony support its association with indicators of group
member relationships and rapport. For instance, research
shows that team members’ synchrony in EDA is related to
tension and negative affect [71], and that synchrony in HR
is related to cooperation [69] Some scholars have also
found that spontaneous synchrony in facial expressions are
related to cohesion [71] and resulted in increased levels of
emotional interaction and liking among dyads (e.g., [62]).
Our study will extend these findings in online
collaborations and examine whether the association
between group satisfaction and physiological synchrony
holds for computed-mediated environments. Furthermore,
given the differentiation between CI and group satisfaction
previously noted, we may find that different patterns of
physiological synchrony correspond with CI, deepening our
understanding of these two different building blocks of
collaboration. Therefore, we propose:
Hypothesis 1. Physiological synchrony is positively related
with (a) collective intelligence and (b) group satisfaction.
The Effect of Social Perceptiveness
We also hypothesize that the effect of group members’
social perceptiveness on CI [95, 28] will be explained in
part by physiological synchrony. Social perceptiveness is a
measure of an individual’s ability to infer what others are
thinking or feeling based on subtle, nonverbal cues. It is
correlated with other aspects of emotional intelligence [59]
and consistently related to higher CI [28, 65], and more
effective group functioning [26] both in online and face-to-
face collaborations [28, 94]. We argue that the underlying
reason for the strong relationship of team-level social
perceptiveness with CI and potentially with affect-laden
group satisfaction is physiological synchrony. People who
are high in social perceptiveness are better at
communicating as well as coordinating physical movements
with others, even in the absence of visual access to their
interaction partner [45]. This occurs because highly socially
perceptive team members are more likely to pick up on the
subtle nonverbal cues, and we expect that will also enable
them to physiologically synchronize with others in a
manner that facilitates rapport and coordination. This leads
to our second hypothesis:
Hypothesis 2. Group average social perceptiveness will
affect (a) collective intelligence and (b) group satisfaction
via effects on physiological synchrony.
The Role of Group Composition
As described previously, collective intelligence has
consistently been related to features of group composition
and structure [94]. Specifically, linear and curvilinear
relationships have been reported between CI and both
gender and cognitive diversity [1, 28, 95] and with
members’ level of social perceptiveness [28, 65, 95]. In
addition, some preliminary findings suggest that age
diversity serves to disrupt CI [96]. Based on these findings
and the existing literature on team diversity and social
intelligence, we anticipate we will observe relationships
between these various forms of diversity and CI, but
furthermore that physiological synchrony may serve as a
mechanism.
Typically diversity refers to any attribute that may lead one
person to perceive another one as different from self [86].
In practice this may mean any aspect of differentiation with
research typically focusing on these aspects that relate to
background and social categorization (e.g., gender,
education, age, ethnicity and so on). Diversity in work
teams has been found to relate to both functional outcomes
such as increased information sharing and creativity [63] as
well as dysfunctional outcomes such as increased conflict
[22, 41]. In our study, we investigate the accessible social
categories that people may use to make conclusions, i.e.,
gender, age, and ethnicity. Ethnicity and age are important
variables in team composition research because they are
visible characteristics that may be used for social
categorization [83] which are typically found to disrupt
group relationships and productivity due to stereotyping
and associated conflicts [46].
In addition to exploring whether the previously observed
relationships between CI and diversity are replicated in
dyads working online, we are also interested in exploring
the role of physiological synchrony as a mechanism. There
is some empirical evidence about the relationship of gender
and age with physiological synchrony [54]. In married
couples, when male spouses experience negative affect,
they are also more likely to demonstrate increased
electrodermal activity but the relationship between affect
and arousal is absent in wives. In the same study, the
researchers found that older couples reported higher
positive affect and lower arousal than middle-aged couples
[54]. Regarding ethnic composition, Blascovich and
colleagues [14] showed that White participants interacting
with Black confederates exhibited increased cardiovascular
response and performed poorly on a cooperative task
compared to participants interacting with White
confederates. However, in another experimental setting,
Blascovich and colleagues [14] found no effects in heart
rate activity of the interaction of White participants with
Black confederates. In our experiment, rather than focus on
the main effect of ethnicity, sex or age, we focus on the
effects of group composition on physiological synchrony.
Specifically, we focus on whether dissimilarity in
observable variables such as age, ethnicity, and gender
disrupts physiological synchrony, and the degree to which
that disruption helps to explain their role in collective
intelligence and group satisfaction.

Hypothesis 3. Group composition (sex diversity, age
diversity (or distance), and ethnic diversity) will affect (a)
collective intelligence and (b) group satisfaction via effects
on physiological synchrony.
STUDY DESIGN
We investigated our hypotheses in the context of a
laboratory study. In the experiment, teams of two
completed the Test of Collective Intelligence (TCI). We
start with examining our questions in dyads, as construct
and measures of CI have been demonstrated to apply to
dyads as well as larger groups [95], and focusing on dyads
enables us to look at synchrony without the additional
statistical and phenomenological complexity of subgroups
that may form in larger groups (e.g., [51]). We collected
individual measures of social perceptiveness and
demographics before the TCI and group satisfaction after its
completion. Throughout the TCI, we recorded physiological
measures of electrodermal activity and heart rate. All
sessions were also video recorded to obtain facial
expression data.
METHOD
Participants
We recruited 116 (60 male, 56 female) participants from the
participation pool of a large Northeastern university in the
United States with the age range of 18 to 61 years old (M =
26.4, SD = 8.45). All participants were compensated 15 US
dollars. We ran the study using both same- and mixed-
gender teams (18 male-only dyads, 20 female only dyads
and 20 mixed gender). We failed to capture physiological
signals and video for six dyads due to technical equipment
issues.
Procedure
Each session lasted approximately 30 minutes. Members of
each dyad were seated in different rooms. None of the
participant pairs knew each other before the experiment.
After completing a pretest survey, they were instructed to
wear the E42 wristbands on their non-dominant hand and
relax for two minutes to obtain a baseline in EDA and heart
rate. After that, participants initiated the video conference
call with their partner. Participants then logged onto the
Platform for Online Group Studies (POGS), a web browser-
based platform that supports synchronous multiplayer
interaction, to complete the Test of Collective Intelligence
(TCI) with the other research participant [27, 28, 95]. The
TCI contained six tasks ranging from 2 to 6 minutes each,
and instructions were displayed before each task for 15
seconds to 1.5 minutes. At the end of the test, they were
instructed to sign off the videoconference and proceed to
the post-test survey, which was completed independently.
Participants were then compensated and debriefed.
2 https://www.empatica.com/e4-wristband
Measures
Participants provided demographic information and
completed the test for social perceptiveness individually
prior to working on the TCI with their group. Physiological
synchrony was measured during the group work on the TCI.
After the group work period was over, group satisfaction
was measured at the individual level.
Group composition. We examined group composition in
terms of three surface-level attributes: sex, age, and
ethnicity. For sex, we calculated the number of females in
each dyad (male only = 0, mixed = 1, female only = 2). Age
diversity was operationalized as the distance between two
members’ ages in years. Age distance ranged from 0 to 40
years (M = 9.24 years, SD = 10.04). Ethnic diversity was
dummy-coded; if participants reported identifying with
different ethnic groups, they were considered dissimilar,
coded as 1, otherwise, 0. Thirty-four dyads (56.7%) were
ethnically dissimilar.
Social perceptiveness. To measure social perceptiveness,
we used the Reading the Mind in the Eyes test (RME)
developed by Baron-Cohen and colleagues [6]. The test
consists of 36 images of the eye region of individual faces.
Participants were asked to choose among possible mental
states to describe what the person pictured was feeling or
thinking. The options were complex mental states (e.g.,
guilt) rather than simple emotions (e.g., anger). Individual
participants’ scores were averaged for each dyad.
Collective intelligence. Collective intelligence was
measured using the Test of Collective Intelligence (TCI),
which was completed by dyads working together. The TCI
is an online version of the collective intelligence battery of
tests used by Woolley et al. [95], which contains a wide
range of group tasks [27, 28]. The TCI was adapted into an
online tool to allow researchers to administer the test in a
standardized way, even when participants are not co-
located. There were a total of six tasks in the version of the
TCI used in this study which measured the dyads’ ability to
collaborate in a variety of ways by having them generate
creative ideas, solve word and number puzzles, collectively
remember detailed information, and execute detail-oriented
tasks quickly and accurately. To obtain collective
intelligence scores for all dyads, we first scored each of the
six tasks and then standardized the raw task scores. We then
computed an unweighted mean of the six standardized
scores, a method adapted from prior research on collective
intelligence [95].
Group satisfaction. To measure group satisfaction, we
used six items that reflect the quality of group collaboration
and relationship, adapted from the Team Diagnostic Survey
(TDS, [88]) (e.g., “I am very satisfied working with this
team”). Participants reported their ratings on a five point
Likert-type scale (α = .72, M= 4.12, SD = .42). Since group
satisfaction is conceptualized as a group-level construct, we
aggregated individual dyad members’ mean group
satisfaction score to the dyadic level by computing the

mean of two ratings (ICC(1) = .64, ICC(2) = .75). The
median r*wg(J) [55] was .97. The values ranged from .78 to
1, demonstrating acceptable level of within-group
agreement.
Physiological synchrony. We assessed physiological
synchrony by recording facial expressions, electrodermal
activity (EDA), and heart rate (HR) of each individual in
the dyad throughout their interaction during the TCI.
Synchrony in facial expressions can capture shared
experience or mimicry [79] over time. Synchrony in HR
and EDA captures shared arousal during periods of stress,
excitement, or high levels of engagement [2, 73, 76]. We
processed the individual responses into physiological
response signals, sequences of response scores over time in
the study, and then calculated distances between the series
of scores (or signals) of each individual in a dyad using
Dynamic Time Warping. In the rest of this section we
describe how we translated sensor data into physiological
response signals and calculated interpersonal distances.
Physiological response signals. In order to develop
physiological response signals comparable across
individuals, we first translated each person’s physiological
responses into scores over time for each measure. We then
reduced noise in the scores and normalized the signals. We
did this to account for individual differences that can
change the scale of response for signals like EDA and HR
that are usually influenced by factors such as age and
cardiovascular fitness. For all our physiological signals, we
restricted analyses to the task portions of the experiment,
trimming out data where participants were not collaborating
i.e., while reading instructions. Below, we describe how we
transform raw signals from each sensor to physiological
response signals, and illustrate the process in figure 1.
Facial Expressions
We derived facial expression signals from web conference
videos of participants’ faces recoded by Evaer3. For each CI
task, we manually extracted the respective video and used
OpenFace [5] to detect Facial Action Units (AUs) [25] in
each frame. We coded for two types of expressions in the
video, positive (AU12 with or without AU6), and negative
(AU15 and AU1 and/or AU4).
We coded smiles of different types as positive expressions.
A smile involves pulling the lip corners up (AU12) with or
without raising the cheeks (AU6). Early research [25]
argued that smiles in which AU12 and AU6 co-occur
3 http://www.evaer.com/
Figure 1. Data transformation from raw data to physiological response signals for each measure.

indicate felt positive emotion, while smiles containing only
AU12 are polite, social, or “masking” smiles, and do not
indicate felt emotion. However, recent findings [81, 47]
show that smiles containing only AU12 can also indicate
felt emotions, while smiles containing both AU12 and AU6
can also be feigned. Hence, whenever the system detects
the lip corners being pulled upwards (AU12), it labels the
expression as a positive one, irrespective of whether the
cheeks are raised or not.
We coded activation of a different set of key facial action
units as negative expressions. Negative expressions such as
worry or displeasure are often conveyed by depression of
the lip corners (AU15) and changes to the positioning of
eyebrows forming something close to a frown (AU1 and/or
AU4). Hence, when the system detects depression of the lip
corners, and either raising of inner brows or lowering of
brows or both (AU15 and AU1 and/or AU4), it labels the
expression as a negative one.
We converted the facial expressions identified to scores for
each frame in the video. We assigned a value of “1” to
frames containing a positive expression, “-1” to frames
containing a negative expression, and “0” for frames
containing neither (neutral). The signal obtained is noisy
potentially due to jittery facial motion and the use of an
automatic tool that can often inaccurately detect
expressions. To reduce noise, we smoothed the signal over
29 frames (average frame rate/ approximately 1 second) by
applying a Simple Moving Average filter (SMA). SMA
allows us to calculate a moving average by adding signal
samples over a number of time periods (i.e. 29 samples),
and then dividing this total by the number of time periods.
Finally, individual scores during each task make up the
individual’s physiological response signal for that task.
Since we have six tasks, we get six physiological response
signals for each person. We then calculate the synchrony
of these scores between the two interaction partners as
described below.
Electrodermal Activity (EDA)
We measured EDA to assess each participant’s level of
electrodermal arousal during the session. We recorded EDA
using the E4 wristband with a sample rate of 4 samples per
second (4Hz). The resulting signals have two components -
tonic (skin conductance level) and phasic (skin conductance
response). The tonic component changes gradually over
time, approximates a person’s baseline and is not the result
of stimuli. The phasic component contains quickly
changing peaks that typically occur in response to short-
term events or environmental stimuli. We separate phasic
EDA from tonic EDA using Continuous Decomposition
Analysis [8], and use only phasic EDA (pEDA) henceforth,
since we’re only interested in participant task responses and
not their baselines. We get six task pEDA signals, for each
participant. These signals are subjected to the steps
described below.
We normalized pEDA signals using z-score of each sample
to enable inter-participant comparison. The signals obtained
were very noisy and so we applied a Simple Moving
Average filter (window size = 5 seconds = 5*4 samples,
empirically determined) to smooth the signals, and name
these physiological response signals. As in facial
expressions, we end up with six physiological response
signals (one for each of the six tasks). We then calculate the
synchrony of these scores between the two interaction
partners as described below.
Heart Rate (HR)
We measured HR to assess each participant’s level of
cardiovascular arousal during the session. Since different
people have different resting heart rates, to enable inter-
participant comparison, we normalized the HR data by
dividing it by its mean (assuming mean to be an estimate of
the baseline) and multiplying it by 100. No smoothing was
required for HR signals, since HR is the number of
heartbeats averaged or smoothed over a moving window of
1 minute. We obtained six task HR signals that are
physiological response signals representing a participant’s
percentage change in HR over time, from his/her mean HR.
We then calculate the synchrony of these scores between
the two interaction partners as described below.
Dynamic Time Warping (DTW)
We used Dynamic Time Warping (DTW) [11] to calculate
synchrony between facial expressions, heart rate, and EDA
of partners in a dyad. DTW is an algorithm for measuring
similarity between two temporal sequences that vary in time
and speed. Physiological response signals of each
participant are also temporal sequences that are computed
using the method described above. DTW provides the
distance between the partners’ physiological response
signals for each task, which are then summed across tasks
to give the total distance. We operationalize synchrony as
similarity of a physiological measure between the partners’
response signals of that measure. This similarity is
calculated by subtracting the total distance from the total
distance of the most different (largest total distance) dyad.
Figure 2 shows an example of two signals - Signal A and
Signal B that are different in length, time, and speed. DTW
warps the time axis of these signals to find corresponding
points between the two signals that optimally match. Figure
3 shows some of the matched corresponding points in
signals A and B. A locality constraint in DTW is the
maximum distance allowed between the matched
corresponding points in the two sequences. The algorithm
used to match the signals illustrated in figure D does not
specify a locality constraint, which may or may not be
always favorable. Since, our physiological response signals
are several minutes long, and we only want to match
physiological responses expressed by participants within a
few seconds of their partners as synchrony, having a
locality constraint is necessary for this analysis. The

locality constraint we chose for each measure is
approximately 5 seconds, that is 145 samples for facial
expressions (since average frame rate is 29Hz), 20 samples
for electrodermal activity (frame rate is 4Hz), and 5
samples for heart rate (frame rate is 5Hz). This locality
constraint is arbitrarily chosen since it appears to cover
roughly one cycle of the signals (see [61]). We use
Euclidean distance to calculate distance between any two
corresponding points in DTW.
Unlike other methods like sample-wise Euclidean distance
and cross-correlation, DTW is able to overcome time lag
and flexibility issues in our data. For example, if a
participant smiles 4 seconds after his/her partner (time lag
issue), or if a person smiles 2 seconds longer than his/her
partner (time flexibility issue), DTW matches both these
events. Next, we present our findings.
RESULTS
Table 1 presents zero-order correlations among all
variables: group compositional variables (number of
females in the dyad, age diversity, ethnic diversity, and
social perceptiveness), collective intelligence (CI), group
satisfaction, and physiological synchrony variables (facial
expressions, electrodermal activity (pEDA), heart rate).
Physiological Synchrony and Collective Intelligence
In order to test hypothesis 1, we examined the relationship
between physiological synchrony and CI, and physiological
synchrony and group satisfaction. We found a significant,
positive relationship between synchrony in facial
expressions and CI (r = .30, p = .01). By contrast, CI was
neither significantly correlated with synchrony in pEDA
nor with synchrony in heart rate.
Interestingly, we found a different pattern with respect to
group satisfaction. Group satisfaction was positively
associated with high levels of pEDA synchrony (r = .33, p =
.04). In other words, when both members of a dyad
exhibited similar levels of electrodermal activity, they later
reported a higher level of satisfaction with the interaction
with their partner. However, group satisfaction had no
significant relationship with synchrony in facial expressions
and heart rate. Finally, pEDA and heart rate were
negatively related to one another (r = -.32, p = .02). Dyads
that had higher synchrony in their heart rate also had lower
synchrony in electrodermal activity. Only synchrony in
electrodermal activity was significantly associated with
group satisfaction.
There was no relationship between CI and group
satisfaction, consistent with prior studies [95, 27, 28, 96].
Interestingly, synchrony in facial expressions (correlated
with CI) was not significantly correlated with synchrony in
pEDA (correlated with group satisfaction), reinforcing the
speculation that there might be two separate paths along
which collaborative relationships in groups develop.
Social Perceptiveness
In Hypothesis 2, we predicted that a group’s average social
perceptiveness, measured by the RME test would positively
affect (a) CI and (b) group satisfaction, via effects on
physiological synchrony. To test this hypothesis, we ran a
series of mediation models using PROCESS macro via
SPSS [34]. We tested mediation by first using social
perceptiveness to predict CI and then looked at the change
in effect of social perceptiveness when each form of
synchrony (facial expression vs. pEDA or heart rate) was
also included in the model. We ran a similar series of
models to test whether both forms of physiological
synchrony mediated the relationship between social
perceptiveness and group satisfaction (facial expression vs.
pEDA). For all tests, we used kappa squared as an index of
indirect effect size [78]. Since kappa squared is a ratio, the
direction of each relationship is characterized by the
coefficient of the indirect effect.
Time
0 2 4 6 8 10 12 14 16 18 20
Value
-1
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
4
11
22
3
3
4
4
55
66
99
1212
1515
1616
18
18
19
19
Example of Matched Points
Figure 3. Signals A and B matched using DTW. Only
some of the corresponding points matched between the
two signals are shown for readability purposes.
Time
0 2 4 6 8 10 12 14 16 18 20
Value
-1
0
1
2
3
4Signal A
Time
0 2 4 6 8 10 12 14 16 18 20
Value
-1
0
1
2
3
4Signal B
Figure 2. Examples of signals A and B of different
lengths and varying time and speed

The results showed that the average social perceptiveness
had a positive indirect effect on CI via synchrony in facial
expressions (kappa squared = .06; 95% bias-corrected
10000 bootstrap confidence intervals ranged from .0038 to
.1976). Average social perceptiveness had a significant,
positive direct effect on CI, as well (p = .04). This suggests
that the effect of average social perceptiveness on CI, which
had been repeatedly shown in previous research [27, 95], is
in part explained by a physiological mechanism,
specifically synchrony in facial expressions. Average social
perceptiveness had a small but positive indirect effect on
group satisfaction via synchrony in pEDA (kappa squared =
.03; 95% bias-corrected 10000 bootstrap confidence
intervals ranged from .0006 to .1130). On the other hand,
average social perceptiveness did not have a direct effect on
group satisfaction after controlling for pEDA.
Group Composition
Finally, in Hypothesis 3, we predicted that (a) CI and (b)
group satisfaction would be influenced by group
composition in terms of sex, age, and ethnicity, and that the
effects would be mediated by physiological synchrony.
Results revealed that the number of females in the dyad had
no indirect effect on CI via synchrony in facial expressions.
The number of females in the dyad did not have a direct
effect on CI controlling for facial expressions synchrony.
On group satisfaction, the number of females did not have a
direct effect on group satisfaction, but had a positive
indirect effect on group satisfaction via synchrony in pEDA
(kappa squared = .03; 95% bias-corrected 10000 bootstrap
confidence intervals ranged from .0006 to .1294).
Age diversity, measured by distance, had a negative indirect
effect on CI through reduced synchrony in facial
expressions (kappa squared = .07; 95% bias-corrected
10000 bootstrap confidence intervals ranged from .0065 to
.1726). That is, dyads whose members greatly differ in age
were less likely to synchronize in facial expressions, thus
having lower CI. Age distance did not have a significant
direct effect on CI, however. With respect to group
satisfaction, age diversity had a positive indirect effect via
synchrony in pEDA (kappa squared = .04; 95% bias-
corrected 10000 bootstrap confidence intervals ranged from
.0027 to .1457). The direct effect of age diversity on group
satisfaction controlling for synchrony in pEDA was not
significant.
Ethnic diversity had a positive indirect effect on CI through
increased synchrony in facial expressions (kappa squared =
.03; 95% bias-corrected 10000 bootstrap confidence
intervals ranged from .0003 to .0940). Ethnic diversity had
a positive direct effect on CI as well, controlling for
synchrony in facial expressions. Thus, having ethnically
dissimilar members in the dyad seems to increase CI with
and without synchrony in facial expressions. For group
satisfaction, ethnic diversity had a negative direct effect,
controlling for synchrony in pEDA (p = .01). However, it
had a positive indirect effect on group satisfaction via
synchrony in pEDA (kappa squared = .03; 95% bias-
corrected 10000 bootstrap confidence intervals ranged from
.0008 to .1358). It suggests that ethnically dissimilar dyads
are less likely to report satisfaction with their partner;
however, ethnically dissimilar dyads who synchronized in
electrodermal activity reported higher levels of group
satisfaction.
DISCUSSION
In this study, we examined whether collective intelligence
of human groups is associated with the deep structures of
collaboration manifested by synchronization of
Mean
S.D.
1.
2.#
3.#
4.#
5.#
6.#
7.#
8.#
9.#
10.#
1. Age distance
9.240
10.044
1
!
!
!
!
!
!
!
!
!
2. Number of females
1.010
0.813
.214
1
!
!
!
!
!
!
!
!
3. Ethnicity similarity
1.42
0.498
-.019
-.271*
1
!
!
!
!
!
!
!
4. Education distance
1.51
1.135
.342**
.083
.131
1
!
!
!
!
!
!
5. RME
26.350
2.943
-.198
-.240
-.085
.050
1
!
!
!
!
!
6. CI
-0.016
0.630
-.223
.101
-.295*
-.041
.390**
1
!
!
!
!
7. Group satisfaction
4.120
0.425
.115
-.050
.266*
-.006
.020
-.140
1
!
!
!
8. DTW_EDA
3674.8
411.553
.114
.211
-.085
.071
.077
-.064
.325*
1
!
!
9. DTW_HR
15571.0
5935.2
.207
.047
-.225
.072
-.053
-.140
-.002
-.340*
1
!
10. DTW_FACE
7772.1
3931.1
-.255
.168
-.068
.152
.222
.304*
-.093
-.057
-.200
1
** p <.01, * p <.05, N = 60 (in subjective measures), N = 52 ( in facial expressions) , N = 53 (in EDA and HR)
Number of females is coded as 0 (male only), 1 (mixed sex), 2 (female only), ethnicity similarity coded as 1 (dissimilar), 2
(similar). CI = Collect intelligence (z-score), RME= Read the Mind through the Eyes, DTW = Dynamic Time Warping
(measure of similarity or synchrony in dyads), EDA= electrodermal activity, HR= heart rate, FACE = facial expressions.
Table 1. Correlation coefficient for synchrony strength in facial expressions, electrodermal activity and heart rate with
collective intelligence, group satisfaction and RME.

physiological responses. In addition, building upon the
established teams and organizational literatures on group
interpersonal processes, specifically group satisfaction, we
further tested whether the interpersonal aspect of group
processes is similarly governed by synchrony in
physiological signals. Finally, we hypothesized that
synchrony in physiological signals is one key mechanism
for previously studied effects of group composition on
collective intelligence and group satisfaction. All of this
was tested in a computer-mediated communication
environment, one in which physiological synchrony has not
been extensively examined.
To test these hypotheses, we conducted an experiment
where 60 dyads interacted in virtual collaborative
environments while being measured for physiological
signals such as electrodermal activity (EDA), heart rate, and
facial expressions. We found that collective intelligence
was positively correlated with synchrony in facial
expression, but not with EDA nor heart rate. On the other
hand, group satisfaction, which captures the quality of
group interaction and relationships, was positively
correlated with synchrony in EDA, but not with facial
expression synchrony. These findings suggest that the
physiological structures of group collaboration are not
monolithic, but perhaps comprised of different building
blocks. Specifically, similarity in group members’ facial
expressions is a symptom of a higher level of attentiveness
to other members which may facilitate coordination and
collective effort. In contrast, similarity in EDA indicates
shared arousal, capturing how the group members feel
during the interaction, and thus has effects on members’
level of satisfaction. This finding is also consistent with
Mønster et al.’s, [71] work that showed physiological
synchrony in EDA is related to emotional aspects of the
group dynamics but not to task performance.
Social Perceptiveness
We observed that groups with high social perceptiveness on
average were more collectively intelligent, consistent with
previous research [95, 27] and interestingly, this effect was
mediated by synchrony in facial expressions. That is, a
group’s average social perceptiveness increases collective
intelligence because members in such group synchronize
their facial expressions more, facilitating coordination. It is
important to note that the direct effect of social
perceptiveness on collective intelligence, controlling for the
mediator, was also significant. A question worth further
exploring is what other mechanism explains the positive
effect of social perceptiveness on collective intelligence. It
is likely that other communication and coordination
behaviors manifest in groups with higher levels of social
perceptiveness [45, 95], which further enhance collective
intelligence.
Group Composition
Physiological synchrony also appeared to be an underlying
mechanism for the effect of group composition on
collective intelligence and group satisfaction, but to varying
degrees. All of the effects on CI operated via their impact
on facial expression synchrony, which can be viewed as a
gauge of shared attention and concentration [29]. Ethnic
diversity indirectly increased collective intelligence via
increased synchrony in facial expressions. Ethnicity is a
surface-level characteristic, and diversity in such
characteristic has been shown to prime heightened levels of
mutual attentiveness among group members [77],
reinforcing our interpretation of facial expression
synchrony as an index of shared attention. Age diversity, on
the other hand, negatively affected collective intelligence
because the more dissimilar members are in terms of age,
the less synchronous in facial expressions, reinforcing the
negative effect of age heterogeneity on group performance
[85]. Taken together, these findings suggest that ethnic
diversity perhaps contributed to a heightened level of
attention among members, a favorable condition for
collective intelligence; however, age diversity can evoke a
sense of hierarchy between members, obstructing the
development of collective intelligence [33, 96].
Group satisfaction was also indirectly affected by the group
composition variables we examined. Similar to collective
intelligence, sex and ethnic diversity had positive indirect
effects on group satisfaction, but via synchrony in EDA, a
signal of shared arousal in the group. However, age
diversity, measured by distance, also had a positive indirect
effect on group satisfaction via synchrony in EDA. That is,
groups composed of members with a greater age gap
demonstrated a high level of synchrony in EDA, which was
positively associated with group satisfaction. It is possible
that a greater age gap between members created a non-
competitive, caring environment for group members;
however, without the greater attentiveness engendered via
facial expression synchrony, this did not translate into the
group’s collective intelligence. Finally, social
perceptiveness indirectly increased group satisfaction via
synchrony in EDA, albeit to a very small degree.
Implications
There are a number of important implications of this study
and interesting opportunities for future work on CI and
CSCW associated with the additional insight that the
physiological mechanisms provide.
Different Processes May Drive Cohesion vs Performance
First, it appears that the dissociation between CI and group
satisfaction repeatedly observed in prior studies [27, 96, 95]
has a parallel in physiological signals. Here facial
expression synchrony and EDA synchrony, were also
dissociated from each other but related to CI and group
satisfaction, respectively. This dissociation harkens back to
debates of a few decades ago regarding the cohesion-
performance connection or lack thereof [58, 72] and
suggests additional mechanisms to gain further insight into
that relationship, namely group composition variables and
their differential effects of group member physiological

response and synchrony. This distinction is important if we
want to develop more fine-grained, sensory-driven
predictive models of group performance to drive intelligent
environments or feedback systems.
Diversity, Social Perceptiveness and Collective Intelligence
While scholars have documented the benefits of diversity
for cognition [68, 70] the implications for physiological
measures and their independent impact on collective
intelligence and group member relationships has only
begun to be explored [2, 76]. Our study suggests that
diverse groups may engage in fundamentally different
interpersonal processes as a function of heightened social
perceptiveness. Our study provides only an initial glimpse
at answers to questions about diversity and group process,
but suggests a host of other relationships to explore. Future
work should examine the effects of interventions to regulate
physiology as a means of improving working relationships
in diverse groups. In addition, it may be possible to improve
physiological synchrony and ultimately performance in
diverse and homogeneous groups through social
perceptiveness training interventions. It would also be
interesting and fruitful to look at the relationship of facial
synchrony and more detailed process behaviors in diverse
groups to unpack exactly how and why it may support CI.
Shared attention, facial synchrony and turn-taking
The positive relationship between collective intelligence
and synchrony in facial expressions in our study confirms
the importance of visual, nonverbal cues about team
members in facilitating collective intelligence,
complementing previous work in CSCW. Theory and
research in CSCW has long noted the value of nonverbal
cues for supporting language understanding and
coordinating turn-taking during remote collaboration (e.g.
[7]). Our results suggest facial expression synchrony may
be a critical aspect of collective intelligence, and that
systems may be able to enhance team performance by
making faces more visible and salient. Designers may
consider, for example, screen layouts for video
conferencing systems that could help group members attend
to and synchronize other members’ facial expressions
easily. However, future work is needed to examine what
additional mechanisms or collaboration tools might
facilitate facial expression synchrony. This can be done by
for example, repeating the experiment without
videoconferencing (audio only) to see how the effect of
synchrony in facial expressions on CI changes. Similar
results could indicate visual processes do not primarily
drive that facial expression synchrony, but by more
generalized sensing processes.
Enhancing collective intelligence
One potential application for our results is improving
collective intelligence via technological or behavioral
intervention. This application raises a host of additional
research questions about the relationship between
synchrony and collective intelligence and intervention
design.
Is synchrony controllable?
Our work raises the question of whether synchrony is a
controllable or unconscious process. Can individuals
consciously increase facial or physiological synchrony?
And by doing so increase collective intelligence or group
satisfaction? Synchrony may depend on individual
differences in perspective taking or empathic abilities [40].
We found potential evidence for this in group level
variations in synchrony.
Intervention design
Next there is a question of how to design interventions for
increasing synchrony. Should such interventions be direct
or unobtrusive?
Indirect environmental interventions
Unobtrusive interventions could increase the level of
physiological synchrony among members through shared
activities or other manipulations outside of participants’
conscious awareness. Intelligent meeting rooms could
attempt to unobtrusively intervene or modify the
collaboration environment to enhance group performance in
response to sensed levels of synchrony early in a group’s
life cycle to improve the quality of interaction. For
example, an intelligent system that sensed pEDA
asynchrony across members could increase room
temperature or ambient noise in one location until that
member was ‘in synch’ in terms of their arousal level. It
remains to be seen, however, whether synchrony evoked via
this kind of unobtrusive background intervention would
have a similar positive effect on group satisfaction.
Visual feedback
Alternatively, technological interventions could be more
obtrusive or directive, and presented at the individual
versus the group level. For example, video conferencing
systems could integrate a facial synchronometer showing
the level of similarity across participants’ facial expression
or provide commands to individual group members for
increasing synchrony (e.g. “smile more to match your
partner”). This kind of on-screen instruction introduces
other tradeoffs like mental and visual attention demand.
However, previous work in CSCW has successfully applied
in-situ visual feedback on group processes such as floor
sharing behavior to increase equality of participation and
ultimately team performance [23, 43, etc.]. Future work
should explore the use of directive feedback versus real
time or post-hoc visualization of individual and group
responses to see whether it is possible to enhance
synchrony.
Training

Training is another more direct way to potentially enhance
teams synchrony. Social perceptiveness training has been
used in other settings effectively to increase individual
levels of attention to social cues [80]. If facial synchrony is
controllable it may be possible to train team members to
better attend to facial expressions and synchronize their
own in response. Alternatively, participants can be primed
with a pro-social task [52] before they begin collaborating.
Limitations
As with any study that represents an initial attempt at
applying a new methodology to a novel context, our study
has some limitations that readers must bear in mind. First,
heart rate synchrony turned out to be a less sensitive
measure of synchrony of arousal for the present study. A
potential reason for this is that the tasks in TCI did not
induce strong enough arousal, compared to other studies
that show such an effect [66]. On the other hand, existing
studies in married couples and infant-mother pairings show
heart rate responses to rather subtle changes in activity [29]
and thus it is hard to discern the reasons for a lack of
relationship observed here. Alternatively, wearable
wristbands (like E4) capture HR as averaged over a minute
instead of instantaneous HR. ECG sensors are more
accurate but also more intrusive for use in lab experiments
as these require straps placed close to heart. Future
research could explore additional techniques for measuring
heart rate, and/or tasks to induce larger changes during the
interaction to see if greater effects associated with heart rate
synchrony might be observed.
A second potential limitation here is associated with our use
of dyads versus larger groups. CI has been examined in
dyads in the past [95] and shown to be predicted by the
same variables associated with it in larger groups. However,
it is unknown in this context whether the more complex
form of synchrony that would need to develop in a larger
group would (a) develop in larger groups collaborating
online, and (b) show the same effects on CI and group
satisfaction. Thus future research will need to explore
replication of these effects in larger groups.
CONCLUSION
In conclusion, this study represents an initial foray into
exploring the role of deep collaboration mechanisms,
represented by physiological synchrony, in the development
of CI and group satisfaction in computer-mediated teams.
We find fairly strong evidence of the role of facial
expression synchrony and EDA synchrony in explaining the
effects of group diversity and social perceptiveness.
We believe this study is only the beginning of a series of
experiments that use sensors to capture the deep structures
of collective intelligence. Future research can expand our
findings by adopting other sensors such as facial
electromyography (EMG), cortisol levels, eye-tracking,
motion sensors (for body language), audio (for voice
quality, jitter, turn-taking, etc), and EEG. Results will
further our understanding of the mechanisms underlying
group performance and cohesion.
As collaboration becomes ever more dispersed and
technology-mediated, we see that these fairly basic,
primitive human responses to one another remain. The real
challenge will be to develop new and innovative
interventions that harness this newly acquired knowledge to
enable teams to reach new heights in collective intelligence
and satisfaction with their relationships.
ACKNOWLEDGEMENT
We would like to thank Thomas Rasmussen, Mikahla
Vicino, Zhong Yu Bing, and Sean Tao for helping us with
data collection. This project was supported by the National
Science Foundation, under awards CNS-1205539, VOSS-
1322278, and VOSS-1322241.
REFERENCES
1. Ishani Aggarwal, Anita Williams Woolley, C.F. Chabris,
and T.W. Malone. 2015. Cognitive diversity, collective
intelligence, and learning. In Proceedings of Collective
Intelligence 2015.
2. Modupe Akinola. 2010. Measuring the pulse of an
organization: Integrating physiological measures into the
organizational scholar’s toolbox. Research in
Organizational Behavior 30: 203–223.
3. Nalini Ambady and Robert Rosenthal. 1992. Thin slices
of expressive behavior as predictors of interpersonal
consequences: A meta-analysis. Psychological bulletin 111,
2: 256.
4. Jason K Baker, Rachel M Fenning, Mariann A Howland,
Brian R Baucom, Jacquelyn Moffitt, and Stephen A Erath.
2015. Brief report: A pilot study of parent–child
biobehavioral synchrony in autism spectrum disorder.
Journal of autism and developmental disorders 45, 12:
4140–4146.
5. Tadas Baltrusaitis, Marwa Mahmoud, and Peter
Robinson. 2015. Cross-dataset learning and person-specific
normalisation for automatic Action Unit detection. In
Automatic Face and Gesture Recognition (FG), 2015 11th
IEEE International Conference and Workshops on, 1–6.
6. S. Baron-Cohen, S. Wheelwright, J. Hill, Y. Raste, and I.
Plumb. 2001. The “Reading the Mind in the Eyes” Test
revised version: a study with normal adults, and adults with
Asperger syndrome or high-functioning autism. The
Journal of Child Psychology and Psychiatry and Allied
Disciplines 42, 02: 241–251.
7. Mathilde M Bekker, Judith S Olson, and Gary M Olson.
1995. Analysis of gestures in face-to-face design teams
provides guidance for how to use groupware in design. In
Proceedings of the 1st conference on Designing interactive
systems: processes, practices, methods, & techniques, 157–
166.

8. Mathias Benedek and Christian Kaernbach. 2010. A
continuous measure of phasic electrodermal activity.
Journal of neuroscience methods 190, 1: 80–91.
9. A. Berdahl, C. J. Torney, C. C. Ioannou, J. J. Faria, and I.
D. Couzin. 2013. Emergent sensing of complex
environments by mobile animal groups. Science 339, 6119:
574–576.
10. Tony Bergstrom and Karrie Karahalios. 2007.
Conversation Clock: Visualizing audio patterns in co-
located groups. In System Sciences, HICSS 2007. 40th
Annual Hawaii International Conference on, pp. 78-78.
IEEE, 2007
11. Donald J. Berndt and James Clifford. 1994. Using
dynamic time warping to find patterns in time series. In
KDD workshop, 359–370.
12. Frank J Bernieri. 1988. Coordinated movement and
rapport in teacher-student interactions. Journal of
Nonverbal behavior 12, 2: 120–138.
13. Frank J. Bernieri, J. Steven Reznick, and Robert
Rosenthal. 1988. Synchrony, pseudosynchrony, and
dissynchrony: Measuring the entrainment process in
mother-infant interactions. Journal of personality and
social psychology 54, 2: 243.
14. Jim Blascovich, Wendy Berry Mendes, and M. Seery.
2002. Intergroup Encounters and Threat. From Prejudice to
Intergroup Emotions, Differentiated Reactions to Social
Groups, edited by Diane M. Mackie and Eliot R. Smith: 89–
109.
15. Jim Blascovich, Steven J. Spencer, Diane Quinn, and
Claude Steele. 2001. African Americans and high blood
pressure: The role of stereotype threat. Psychological
science 12, 3: 225–229.
16. Alan Bränzel, Christian Holz, Daniel Hoffmann, et al.
2013. GravitySpace: tracking users and their poses in a
smart room using a pressure-sensing floor. In Proceedings
of the SIGCHI Conference on Human Factors in
Computing Systems, 725–734.
17. Joseph J Campos and Harold J Johnson. 1967. Affect,
verbalization, and directional fractionation of autonomic
responses. Psychophysiology 3, 3: 285–290.
18. Jonas Chatel-Goldman, Marco Congedo, Christian
Jutten, and Jean-Luc Schwartz. 2014. Touch increases
autonomic coupling between romantic partners. Frontiers
in behavioral neuroscience 8.
19. Michael G Coles. 1972. Cardiac and respiratory activity
during visual search. Journal of Experimental Psychology
96, 2: 371.
20. George E Deane. 1961. Human heart rate responses
during experimentally induced anxiety. Journal of
Experimental Psychology 61, 6: 489.
21. Ian J. Deary. 2012. Intelligence. Annual Review of
Psychology 43, 1: 453–482.
22. Carsten K. W. De Dreu and Laurie R. Weingart. 2003.
Task versus relationship conflict, team performance, and
team member satisfaction: A meta-analysis. Journal of
Applied Psychology 88, 4: 741–749.
23. Joan Morris DiMicco, Katherine J Hollenbach, Anna
Pandolfo, and Walter Bender. 2007. The impact of
increased awareness while face-to-face. Human–Computer
Interaction 22, 1-2: 47–96.
24. Elizabeth Duffy. 1957. The psychological significance
of the concept of“ arousal” or“ activation.” Psychological
review 64, 5: 265.
25. Paul Ekman and Wallace V. Friesen. 1978. Manual for
the facial action coding system. Consulting Psychologists
Press.
26. Hillary Anger Elfenbein. 2006. Team emotional
intelligence: What it can mean and how it can affect
performance. In Linking emotional intelligence and
performance at work, Druskat V, F. Sala and G. Mount
(eds.). Lawrence Erlbaum Associates, Publishers, 165–184.
27. D. Engel, Anita Williams Woolley, Ishani Aggarwal, et
al. 2015. Collective intelligence in online collaboration
emerges in different contexts and cultures. CHI ’15
Proceedings of the SIGCHI Conference on Human Factors
in Computing Systems.
28. D. Engel, Anita Williams Woolley, Lisa X. Jing,
Christopher F. Chabris, and Thomas W. Malone. 2014.
Reading the mind in the eyes or reading between the lines?
Theory of mind predicts collective intelligence equally well
online and face-to-face. PLoS ONE 9, 12: e115212.
29. Ruth Feldman. 2007. Parent–infant synchrony
biological foundations and developmental outcomes.
Current Directions in Psychological Science 16, 6: 340–
345.
30. Christopher D Frith and Heidelinde A Allen. 1983. The
skin conductance orienting response as an index of
attention. Biological psychology 17, 1: 27–39.
31. Ella Glikson, Raveh Harush, Young Ji Kim, Anita
Williams Woolley, and M. Erez. 2016. Psychological safety
and collective intelligence in multicultural globally
dispersed teams. In the 2016 INGRoup Conference.
Helsinki, Finland.
32. Deborah M. Gordon. 2016. Collective Wisdom of Ants.
Scientific American 314, 2: 44–47.
33. David A. Harrison and Katherine J. Klein. 2007. What’s
the difference? diversity constructs as separation, variety, or
disparity in organizations. Academy of Management Review
32, 4: 1199–1228.

34. Andrew F. Hayes. 2013. Introduction to mediation,
moderation, and conditional process analysis: A
regression-based approach. Guilford Press.
35. Donald Olding Hebb. 1955. Drives and the CNS
(conceptual nervous system). Psychological review 62, 4:
243.
36. Javier Hernandez, Ivan Riobo, Agata Rozga, Gregory D
Abowd, and Rosalind W Picard. 2014. Using electrodermal
activity to recognize ease of engagement in children during
social interactions. In Proceedings of the 2014 ACM
International Joint Conference on Pervasive and
Ubiquitous Computing, 307–317.
37. Mary M. Herrald and Joe Tomaka. 2002. Patterns of
emotion-specific appraisal, coping, and cardiovascular
reactivity during an ongoing emotional episode. Journal of
personality and social psychology 83, 2: 434.
38. Michael J. Hove and Jane L. Risen. 2009. It’s all in the
timing: Interpersonal synchrony increases affiliation. Social
Cognition 27, 6: 949.
39. Gardner Howard. 1999. Intelligence reframed: Multiple
Intelligences for the 21st century. Howard Gardner.
40. Zac E Imel, Jacqueline S Barco, Halley J Brown, et al.
2014. The association of therapist empathy and synchrony
in vocally encoded arousal. Journal of counseling
psychology 61, 1: 146.
41. K. A. Jehn, G. B. Northcraft, and M. A. Neale. 1999.
Why differences make a difference: A field study of
diversity, conflict, and performance in workgroups.
Administrative Science Quarterly 44, 4: 741–763.
42. Daniel Kahneman, Bernard Tursky, David Shapiro, and
Andrew Crider. 1969. Pupillary, heart rate, and skin
resistance changes during a mental task. Journal of
experimental psychology 79, 1: 164-167.
43. Taemie Kim, Agnes Chang, Lindsey Holland, and Alex
Sandy Pentland. 2008. Meeting mediator: enhancing group
collaboration with sociometric feedback. In CHI’08
Extended Abstracts on Human Factors in Computing
Systems, 3183–3188.
44. Young Ji Kim, D. Engel, Anita Williams Woolley,
Jeffrey Lin, Naomi McArthur, and Thomas W. Malone.
2015. Work together, play smart: collective intelligence in
League of Legends teams. In Proceedings of Collective
Intelligence 2015.
45. Bradley L. Kirkman, Benson Rosen, Paul E. Tesluk,
and Cristina B. Gibson. 2004. The impact of team
empowerment on virtual team performance: The
moderating role of face-to-face interaction. Academy of
Management Journal 47, 2: 175–192.
46. Daan van Knippenberg and Michaéla C. Schippers.
2007. Work Group Diversity. Annual Review of Psychology
58, 1: 515–541.
47. Eva G. Krumhuber and Antony SR Manstead. 2009.
Can Duchenne smiles be feigned? New evidence on felt and
false smiles. Emotion 9, 6: 807.
48. Meredyth Krych-Appelbaum, Julie Banzon Law, Dayna
Jones, Allyson Barnacz, Amanda Johnson, and Julian Paul
Keenan. 2007. “I think I know what you mean”: The role of
theory of mind in collaborative communication. Interaction
Studies 8, 2: 267–280.
49. John I Lacey. 1959. Psychophysiological approaches to
the evaluation of psychotherapeutic process and outcome.
In Research in Psychotherapy, Apr, 1958, Washington, DC.
50. John I Lacey and Beatrice C Lacey. 1970. Some
autonomic-central nervous system interrelationships. In P.
Black (Ed.), Physiological correlates of emotion (pp. 205-
227). New York: Academic Press.
51. Dora C Lau and J Keith Murnighan. 1998.
Demographic diversity and faultlines: The compositional
dynamics of organizational groups. Academy of
Management Review 23, 2: 325–340.
52. Jane Leighton, Geoffrey Bird, Caitlin Orsini, and
Cecilia Heyes. 2010. Social attitudes modulate automatic
imitation. Journal of Experimental Social Psychology 46, 6:
905–910.
53. Gilly Leshed, Diego Perez, Jeffrey T Hancock, et al.
2009. Visualizing real-time language-based feedback on
teamwork behavior in computer-mediated groups. In
Proceedings of the SIGCHI Conference on Human Factors
in Computing Systems, 537–546.
54. Robert W. Levenson, Laura L. Carstensen, and John M.
Gottman. 1994. Influence of age and gender on affect,
physiology, and their interrelations: A study of long-term
marriages. Journal of personality and social psychology 67,
1: 56.
55. Michael K. Lindell, Christina J. Brandt, and David J.
Whitney. 1999. A revised index of interrater agreement for
multi-item ratings of a single target. Applied Psychological
Measurement 23, 2: 127–135.
56. Robert B Malmo. 1959. Activation: A
neuropsychological dimension. Psychological review 66, 6:
367.
57. Regan L Mandryk and Kori M Inkpen. 2004.
Physiological indicators for the evaluation of co-located
collaborative play. In Proceedings of the 2004 ACM
conference on Computer supported cooperative work, 102–
111.
58. John E. Mathieu, Michael R. Kukenberger, Lauren
D’Innocenzo, and Greg Reilly. 2015. Modeling reciprocal
team cohesion–performance relationships, as impacted by
shared leadership and members’ competence. Journal of
Applied Psychology 100, no. 3: 713-734.

59. John D. Mayer, Richard D. Roberts, and Sigal G.
Barsade. 2008. Human abilities: Emotional intelligence.
Annual Review of Psychology 59: 507–536.
60. Martha L. Maznevski and Katherine M. Chudoba. 2000.
Bridging space over time: Global virtual team dynamics
and effectiveness. Organization science 11, 5: 473–492.
61. Michael P. McAssey, Jonathan Helm, Fushing Hsieh,
David A. Sbarra, and Emilio Ferrer. 2013. Methodological
advances for detecting physiological synchrony during
dyadic interactions. Methodology 9, 2: 41–53.
62. Daniel N McIntosh. 2006. Spontaneous facial mimicry,
liking and emotional contagion. Polish Psychological
Bulletin 37, 1: 31.
63. Poppy Lauretta McLeod, Sharon Alisa Lobel, and
Taylor H. Cox. 1996. Ethnic diversity and creativity in
small groups. Small group research 27, 2: 248–264.
64. Colleen Merrifield and James Danckert. 2014.
Characterizing the psychophysiological signature of
boredom. Experimental brain research 232, 2: 481–491.
65. Nicoleta Meslec, Ishani Aggarwal, and P. L Curşeu.
2016. The insensitive ruins it all: Compositional and
compilational influences of social sensitivity on collective
intelligence in groups. Frontiers in Psychology.
66. Ivana Mikic, Kohsia Huang, and Mohan Trivedi. 2000.
Activity monitoring and summarization for an intelligent
meeting room. In Human Motion, 2000. Proceedings.
Workshop on, 107–112.
67. Lynden K Miles, Louise K Nind, and C Neil Macrae.
2009. The rhythm of rapport: Interpersonal synchrony and
social perception. Journal of experimental social
psychology 45, 3: 585–589.
68. F. J Milliken, C. A Bartel, and T. R Kurtzberg. 2003.
Diversity and creativity in work groups: A dynamic
perspective on the affective and cognitive processes that
link diversity and performance. Group creativity:
Innovation through collaboration: 32–62.
69. Panagiotis Mitkidis, John J. McGraw, Andreas
Roepstorff, and Sebastian Wallot. 2015. Building trust:
Heart rate synchrony and arousal during joint action
increased by public goods game. Physiology & behavior
149: 101–106.
70. S. Mohammed and E. Ringseis. 2001. Cognitive
diversity and consensus in group decision making: The role
of inputs, processes, and outcomes. Organizational
Behavior and Human Decision Processes 85, 2: 310–335.
71. Dan Mønster, Dorthe Døjbak Håkonsson, Jacob Kjær
Eskildsen, and Sebastian Wallot. 2016. Physiological
evidence of interpersonal dynamics in a cooperative
production task. Physiology & Behavior 156: 24–34.
72. Brian Mullen and Carolyn Copper. 1994. The relation
between group cohesiveness and performance: An
integration. Psychological bulletin 115, 2: 210.
73. Reiner Nikula. 1991. Psychological correlates of
nonspecific skin conductance responses. Psychophysiology
28, 1: 86–90.
74. Redmond G O’Connell, Mark A Bellgrove, Paul M
Dockree, Adam Lau, Michael Fitzgerald, and Ian H
Robertson. 2008. Self-alert training: Volitional modulation
of autonomic arousal improves sustained attention.
Neuropsychologia 46, 5: 1379–1390.
75. JF Papillo and D Shapiro. 1990. The cardiovascular
system In JT Cacioppo and LG Tassinary (Eds.) Principles
of psychophysiology: Physical, social, and inferential
elements (pp. 456-512). Cambridge: Cambridge University
Press.
76. Suzanne J. Peterson, Christopher S. Reina, David A.
Waldman, and William J. Becker. 2015. Using
physiological methods to study emotions in organizations.
In New Ways of Studying Emotions in Organizations.
Emerald Group Publishing Limited, 1–27.
77. Katherine W. Phillips. 2003. The effects of
categorically based expectations on minority influence: The
importance of congruence. Personality and Social
Psychology Bulletin 29: 3–13.
78. Kristopher J. Preacher and Ken Kelley. 2011. Effect
size measures for mediation models: Quantitative strategies
for communicating indirect effects. Psychological methods
16, 2: 93.
79. Rainer Reisenzein, Markus Studtmann, and Gernot
Horstmann. 2013. Coherence between emotion and facial
expression: Evidence from laboratory experiments. Emotion
Review 5, 1: 16–23.
80. Idalmis Santiesteban, Sarah White, Jennifer Cook, Sam
J Gilbert, Cecilia Heyes, and Geoffrey Bird. 2012. Training
social cognition: from imitation to theory of mind.
Cognition 122, 2: 228–235.
81. Karen L. Schmidt, Zara Ambadar, Jeffrey F. Cohn, and
L. Ian Reed. 2006. Movement differences between
deliberate and spontaneous facial expressions: Zygomaticus
major action in smiling. Journal of Nonverbal Behavior 30,
1: 37–52.
82. Janienke Sturm, Olga Houben-van Herwijnen, Anke
Eyck, and Jacques Terken. 2007. Influencing social
dynamics in meetings through a peripheral display. In
Proceedings of the 9th international conference on
Multimodal interfaces, 263–270.
83. Henri Tajfel and John C. Turner. 1986. The social
identity theory of intergroup behavior. In (2nd ed.), Stephen
Worchel (ed.). Nelson-Hall Publishers, Chicago, 7–24.

84. Chiew Seng Sean Tan, Johannes Schöning, Kris Luyten,
and Karin Coninx. 2014. Investigating the effects of using
biofeedback as visual stress indicator during video-
mediated collaboration. In Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems, 71–
80.
85. Thomas A. Timmerman. 2000. Racial diversity, age
diversity, interdependence, and team performance. Small
Group Research 31, 5: 592–606.
86. H. Triandis, L. Kurowski, and M. Gelfand. 1994.
Workplace diversity. In H. Triandis, M. Dunnette and L.
Hough (eds.). Consulting Psychologists Press., Palo Alto,
CA, 769–827.
87. Piercarlo Valdesolo, Jennifer Ouyang, and David
DeSteno. 2010. The rhythm of joint action: Synchrony
promotes cooperative ability. Journal of Experimental
Social Psychology 46, 4: 693–695.
88. Ruth Wageman, J. Richard Hackman, and Erin Lehman.
2005. Team Diagnostic Survey Development of an
Instrument. The Journal of Applied Behavioral Science 41,
4: 373–398.
89. Scott S. Wiltermuth and Chip Heath. 2009. Synchrony
and cooperation. Psychological science 20, 1: 1–5.
90. Piotr Winkielman and John T. Cacioppo. 2001. Mind at
ease puts a smile on the face: Psychophysiological evidence
that processing facilitation elicits positive affect. Journal of
personality and social psychology 81, 6: 989.
91. Gwen M. Wittenbaum, Garold Stasser, and Carol J.
Merry. 1996. Tacit coordination in anticipation of small
group task completion. Journal of Experimental Social
Psychology 32, 2: 129–152.
92. Anita Williams Woolley and Ishani Aggarwal.
Collective intelligence and group learning. In Handbook of
Group and Organizational Learning, Linda Argote and J.
M Levine (eds.). Oxford University Press, London, UK.
93. Anita Williams Woolley and Ishani Aggarwal. The
mind and the heart of the group: Collective intelligence and
relationship quality in task performing teams. Under
review.
94. Anita Williams Woolley, Ishani Aggarwal, and Thomas
W. Malone. 2015. Collective intelligence and group
performance. Current Directions in Psychological Science
24, 6: 420–424.
95. Anita Williams Woolley, Christopher F. Chabris, Alex
Pentland, Nada Hashmi, and Thomas W. Malone. 2010.
Evidence for a collective intelligence factor in the
performance of human groups. Science 330, 6004: 686–
688.
96. Anita Williams Woolley, Rosalind M. Chow, Anna T.
Mayo, Jin Wook Chang, and Christoph Riedl. 2016.
Competition and collective intelligence: Do women always
make groups smarter? Under review.
97. Zhiwen Yu and Yuichi Nakamura. 2010. Smart meeting
systems: A survey of state-of-the-art and open issues. ACM
Computing Surveys (CSUR) 42, 2: 8.