Content uploaded by Martin Lang
Author content
All content in this area was uploaded by Martin Lang on Jul 22, 2021
Content may be subject to copyright.
https://doi.org/10.1177/13684302211016961
Group Processes & Intergroup Relations
1 –21
© The Author(s) 2021
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/13684302211016961
journals.sagepub.com/home/gpi
G
P
I
R
Group Processes &
Intergroup Relations
Outgroup threat and the emergence
of cohesive groups: A cross-cultural
examination
Martin Lang,1 Dimitris Xygalatas,2 Christopher M. Kavanagh,3,4
Natalia Boccardi,5 Jamin Halberstadt,6 Chris Jackson,6
Mercedes Martínez,7 Paul Reddish,8 Eddie M. W. Tong,8
Alexandra Vázquez,7 Harvey Whitehouse,4
Maria Emilia Yamamoto,5 Masaki Yuki9 and Angel Gomez7,10
Abstract
Evolutionary models and empirical evidence suggest that outgroup threat is one of the strongest
factors inducing group cohesion; however, little is known about the process of forming such cohesive
groups. We investigated how outgroup threat galvanizes individuals to affiliate with others to form
engaged units that are willing to act on behalf of their in-group. A total of 864 participants from six
countries were randomly assigned to an outgroup threat, environmental threat, or no-threat condition.
We measured the process of group formation through physical proximity and movement mirroring
along with activity toward threat resolution, and found that outgroup threat induced activity and
heightened mirroring in males. We also observed higher mirroring and proximity in participants who
perceived the outgroup threat as a real danger, albeit the latter results were imprecisely estimated.
Together, these findings help understand how sharing subtle behavioral cues influences collaborative
aggregation of people under threat.
Keywords
activity, cohesion, mirroring, outgroup threat, proximity, willingness to fight
Paper received 26 November 2019; revised version accepted 27 March 2021.
1 Masaryk University, Czech Republic
2
University of Connecticut, USA
3
Rikkyo University, Japan
4
University of Oxford, UK
5
Federal University of Rio Grande do Norte, Brazil
6
University of Otago, New Zealand
7
Universidad Nacional de Educación a Distancia (UNED),
Spain
8National University of Singapore, Singapore
9Hokkaido University, Japan
10ARTIS International, Spain
Corresponding author:
Martin Lang, LEVYNA: Laboratory for the Experimental
Research of Religion, Masaryk University, Arna Nováka 1,
Brno 602 00, Czech Republic.
Email: martinlang@mail.muni.cz
1016961GPI0010.1177/13684302211016961Group Processes & Intergroup RelationsLang et al.
research-article2021
Article
2 Group Processes & Intergroup Relations 00(0)
Introduction
Intergroup conflict is a pervasive problem in
human societies and affects the lives of millions
around the world. Despite international diplo-
macy handling conflicts at the nation-state level,
grassroots movements of active individuals often
rise up to protect parochial sacred values in the
face of outgroup threat, as exhibited in various
protest movements, public militias, suicide terror-
ists, and other embattled communities (Atran &
Ginges, 2012; Newson et al., 2018). Through self-
organized assortment into cohesive groups, indi-
viduals strongly committed to their brothers in
arms, nation, ethnicity, and/or religion are willing
to take costly actions against outgroups (Glowacki
et al., 2016).
Evolutionary models suggest that human psy-
chology has been molded by a long history of fierce
intergroup conflict (Bowles, 2008), and predict that
increased outgroup threat will produce higher rates
of parochial altruism (Bowles, 2009; Whitehouse
et al., 2017), that is, prosocial behaviors directed
only to in-group members. Experimental studies
conducted in areas with recent histories of inter-
group conflict indicate that such violent conflict
translates into heightened progroup behavior and
increased fairness during within-group interactions
in children and early adolescents (Bauer et al., 2014;
Voors et al., 2012). More broadly, intergroup com-
petition is associated with increased public-good
contributions to the in-group (Francois et al., 2018;
Majolo & Maréchal, 2017). Furthermore, a combi-
nation of ethnographic and historical evidence indi-
cates that during intergroup competition, groups
endorse tighter norms (Gelfand, 2019; Gelfand
et al., 2011) and costlier forms of ritual behavior
that signal norm adherence (Sosis et al., 2007).
These and other commitment signals (e.g., increas-
ing similarity and proximity) may serve as mecha-
nisms to maintain or reinforce coalitional safety in
the face of an outgroup threat (Boyer et al., 2015).
At the psychological level, threatening a
group’s values incentivizes individuals to express
willingness to fight for their group (Atran, 2016)
and to protect their values at all costs (Ginges &
Atran, 2011). Such behaviors often take the form
of proactive harm to outgroups, as illustrated by
observational studies of sports fans (Newson
et al., 2018; Wann et al., 1999), experiments using
economic games (De Dreu et al., 2015, 2016),
and experiments manipulating closeness to the
victims of the 9/11 terrorist attacks in the US
(Dumont et al., 2003). The effect of outgroup
threat on willingness to act is especially strong in
individuals whose identity is “fused” with their
group identity (Gómez et al., 2011, 2017); in a
series of laboratory studies, participants scoring
higher on identity fusion (a visceral feeling of
oneness with the group) expressed increased will-
ingness to fight or die for their country (Gómez
et al., 2011; Swann et al., 2009). Furthermore, a
field study of Libyan soldiers showed that, on a
forced-choice question, almost half of frontline
combatants chose fellow fighters rather than
family as their primary fusion target (Whitehouse
et al., 2014). On the basis of such results and a
decade of research on identity fusion (e.g.,
Gómez et al., 2020), Whitehouse (2018) has pro-
posed a “general theory of extreme self-sacri-
fice,” which posits a relationship between identity
fusion and willingness to engage in extreme self-
sacrifice for a group that is moderated by percep-
tions of threat. As briefly described, there is
preliminary evidence in support of this theoreti-
cal model, but much stronger empirical tests are
necessary to examine the proposed relationships.
Together, evolutionary models and empirical
evidence suggest that intergroup conflict is posi-
tively correlated with group cohesion and pro-
group behavior, and that such behavior increases
a group’s survival and success in intergroup com-
petition. However, little is known about the pro-
cess of forming such cohesive groups under
threat from antagonistic groups. In other words,
what are the low-level dynamic processes that
guide interpersonal interactions between anony-
mous individuals to come together to defend a
common identity against outgroups? While
group support through verbal commitment is
often necessary, behavioral nonverbal cues are
generally more reliable signals of group commit-
ment and willingness to fight (Fessler &
Holbrook, 2014; Sosis et al., 2007; Tracy et al.,
Lang et al. 3
2015), and reflect the dynamics of group forma-
tion. In the present study, we aimed to develop
novel measurements of these behavioral cues
(for a review of previous approaches, see Salas
et al., 2015) and examine their dynamics under
external threat. To this end, we followed Carron
and Brawley (2000) and identified two key behav-
ioral cues that reflect the dynamic process of
group formation: increasing willingness of group
members to affiliate with each other in the face
of danger and pursuing the group’s defensive
goals through instrumental action.
Regarding nonverbal behavioral cues of affiliative
tendencies, previous research has long recognized
two crucial mechanisms: movement mimicry and
physical proximity. Movement mimicry is defined as
adopting behavioral patterns, postures, and manner-
ism of interaction patterns, often automatically and
without conscious processing (Lakin et al., 2003).
Importantly, heightened mirroring increases liking,
rapport, empathy, and prosociality among anony-
mous individuals (for a review, see Duffy &
Chartrand, 2015a), and signals romantic interest in
other people (Farley, 2014; Karremans &
Verwijmeren, 2008). In teams, movement mimicry
reflects group alignment from the motor to the
intentional level (Hasson & Frith, 2016), whereby
mimicry facilitates effective team communication
and collaboration (Zhang et al., 2018). Testing the
relationship between movement mimicry and team
cooperation, postural mimicry has been shown to
correlate with student engagement in college semi-
nars (Lafrance & Broadbent, 1976), and artificially
manipulating mimicry between teachers and stu-
dents affected the rating of their rapport (Bernieri,
1988). Moreover, a study of six crew members
deployed on a 4-month simulation of space explora-
tion mission revealed that movement mimicry was
positively correlated with reported group cohesion
in their daily tasks (Zhang et al., 2018).
Proximity, on the other hand, is defined as
physical interpersonal closeness (Allen, 1970;
Cook, 1970). While a related concept to move-
ment mimicry, proximity captures different
aspects of social rapport, specifically, the prefer-
ence to engage in trust-based interaction and face-
to-face communication (with one extreme end
being dyadic intimate relationships). By allowing
others to frequently share the same physical loca-
tion, individuals may exchange key information
and share emotionally charged experiences (Hoegl
& Proserpio, 2004), which are crucial for people
to bond together (Whitehouse & Lanman, 2014).
A study of 145 software developer teams in
Germany revealed that physical proximity was
correlated with self-reported teamwork quality,
including the cohesion of each team (Hoegl &
Proserpio, 2004), and similar results were obtained
in the study of 67 nurses in a Boston area hospital
(Olguin Olguin, 2011). Crucially, physical proxim-
ity was shown to be a reliable indicator of within-
group cohesion and liking (Jackson et al., 2018),
and priming with interdependent/social con-
structs of self, similarly produced higher interper-
sonal proximity (Holland et al., 2004). Note that
we do not claim that being in close proximity is
always beneficial for a group; rather, we under-
stand it as an opposition to individual dispersal
and an indicator of interpersonal liking. In our
conceptualization, both mimicry and proximity
are automated behavioral patterns that reflect the
dynamical process of group emergence through
affiliative tendencies.
Apart from building social rapport through
mimicry and proximity, Carron and Brawley (2000)
identified a group’s ability to self-organize and act
in the face of danger as another important compo-
nent contributing to group functioning. Focusing
on instrumental activity related to an outgroup
threat, previous research showed that threat from
antagonistic groups positively affects self-declared
willingness to act and even sacrifice for the group
(Gómez et al., 2017; Swann et al., 2009), and an
ability to effectively organize for group defense
(Böhm et al., 2016; De Dreu et al., 2016). However,
while relevant to our current question, these stud-
ies do not capture the automated dynamical pro-
cess of group building but reflect a one-shot
conscious decision. In everyday situations, the
group-building enterprise often requires specific
physical action (such as helping with labor, coordi-
nating in defense) to benefit others rather than
direct financial costs (Lockwood et al., 2017); in
other words, physical action and related energy
4 Group Processes & Intergroup Relations 00(0)
expenditure dedicated to solving a group problem
might better reflect the dynamic process of build-
ing functioning groups. Indeed, average daily
movement energy was shown to correlate with
self-reported creativity of research teams tracked
over a 2-week period (Tripathi & Burleson, 2012),
and movement energy was another crucial predic-
tor of perceived team cohesion in the studies of
Boston nurses (Olguin Olguin, 2011) and simu-
lated space exploration mission (Zhang et al.,
2018) described before.
Based on the review of these automated group-
building processes, we should expect that the pres-
ence of an outgroup threat should trigger affiliative
behaviors (proximity and mirroring) and behaviors
directed toward conflict resolution (physical
effort). To investigate this hypothesis, we sampled
864 participants in six different countries. See
Table 1 for the list of countries and the supple-
mental material for detailed description of field
sites. To provide cross-cultural robustness for our
results, these countries were selected to represent
diverse and geographically distant cultures from
five continents and six dominant languages.
In each of our field sites, we randomly assigned
participants to either an outgroup threat, environ-
mental threat, or nonthreatening condition
(approximately 45 participants per condition in
each country) and used innovative and unobtrusive
methods to quantify the effects of these condi-
tions on emergent behavioral properties that indi-
cate group formation. Specifically, groups of four
same-sex participants read an article about an
upcoming international conference addressing
either the threat posed by the Islamic State of Iraq
and the Levant (ISIS) terrorist group (outgroup
threat experimental condition), the threat posed by
an earthquake (environmental threat control con-
dition), or vaguely specified “international politics”
(no-threat baseline condition). Following this
manipulation, participants were instructed to make
a group decision, during a 20-minute discussion
period, on which three delegates (from among six
potential candidates differing on their endorse-
ment of parochial and national politics) should
represent their country at the conference. To quan-
tify the individual progroup behavioral patterns
during the 20-minute discussion period, we
employed sociometric badges (Kim et al., 2012;
Waber et al., 2011). These devices collected indi-
vidual-level data on physical activity, the extent of
mirroring the movements of the other three par-
ticipants in the experimental session, and physical
proximity to those participants. These unobtrusive
and continuously collected measures of mirroring,
proximity, and activity revealed the dynamics of
spontaneous behavior that is not necessarily con-
sciously reflected and may be too subtle to afford
video-coding, thus inaccessible to typical psycho-
metric and social psychology methods. On top of
these main outcome variables, we also asked par-
ticipants about their willingness to fight for their
country (see Figure 1 for the raw distribution and
cross-cultural variation of the outcome variables).
Building on evidence that sociopolitical and
environmental threats promote coordinated
Table 1. Averages with SD of demographic variables.
Site NFemales Age Terrorism World risk
Brazil 144 82 23.3 (4.2) 1.572 4.23%
Japan 136 72 19.9 (1.2) 3.595 13.47%
Mauritius 172 88 21.3 (1.9) 0 15.11%
New Zealand 144 92 21.1 (6.3) 0.611 4.42%
Singapore 116 92 20.9 (1.4) 0 2.36%
Spain 152 84 32.3 (8.3) 1.701 3.23%
Total/Grand M864 510 23.3 (6.4) 1.247 7.14%
Note. Terrorism = Global Terrorism Index (GTI) in 2016, ranging from 0 to 10; World risk = World Risk Index (WRI)
showing 2016 data; it is a score computing exposure and institutional vulnerability to natural disasters (see Section S1.2 in the
supplemental material for site details).
Lang et al. 5
collective action (Gelfand, 2019; Gelfand et al.,
2011), we hypothesized that both threat conditions
(outgroup and environmental) would increase par-
ticipants’ affiliative behaviors and activity compared
to the no-threat baseline condition. Furthermore,
we predicted that there would be a greater willing-
ness to fight for one’s country in the outgroup
threat condition compared to the no-threat base-
line condition. Besides these main models, we
tested three additional hypotheses related specifi-
cally to our manipulation of outgroup threat. First,
since intergroup conflict, coalitional aggression,
and warfare have been historically and cross-cultur-
ally dominated by males (McDonald et al., 2012;
Yuki & Yokota, 2009), we hypothesized that males
will display more affiliative behaviors and activity in
the outgroup threat condition compared to the no-
threat baseline condition. We did not expect such a
difference between the environmental and baseline
conditions. Second, since individuals differed on
their appraisal of the threat posed by international
terrorism, we hypothesized that higher sensitivity
to such threats should moderate the effects of our
treatment such that mirroring, proximity, and activ-
ity would be stronger in the outgroup treatment
compared to the baseline condition. Finally, based
on identity fusion theory, we hypothesized that the
effects of identity fusion with one’s country on
increasing participants’ affiliative and instrumental
behaviors would be stronger in the outgroup threat
condition compared to the baseline condition, with
no difference between the environmental and base-
line conditions (see supplemental material for addi-
tional predictions).
Methods
Participants
Data collection took place in six countries (see
Figure 1) over a period of 2 years (2015–2016).
We recruited university students in groups of
four, 864 participants in total (510 females; Mage
= 23.3, SD = 6.4). We excluded 33 participants
whose native language did not correspond to the
study site; 12 participants who did not fill out
questionnaires; and 77 participants from the anal-
ysis of sociometric data due to malfunctions of
the sociometric badge. All procedures were
approved by the Institutional Review Board at the
University of Connecticut, and additional
approval was obtained in all countries where an
ethics committee was locally available (see sup-
plemental material, Section S2 for additional
information about specific sites).
Procedure and Materials
Five participants of the same sex were invited to a
laboratory, with one participant serving as a sur-
rogate. When all five participants arrived, the sur-
rogate participant was paid a show-up fee (except
Figure 1. Density plots of our four dependent variables.
6 Group Processes & Intergroup Relations 00(0)
for Brazil, where rules did not allow payment for
research participation) and did not take part in the
experiment. We standardized the laboratory rooms
across our sites to include four cubicles with com-
puters, a desk in the middle of the room for group
discussion, two desks by opposite walls with vari-
ous tools, and a white board (see Figure S1).
Research assistants were blind to our hypotheses.
First, participants were fitted with sociometric
badges and then filled out questionnaires assess-
ing identity fusion with their country and demo-
graphic variables. Subsequently, each group was
randomly assigned to one of three conditions: an
outgroup threat condition, an environmental
threat condition, and a no-threat baseline condi-
tion. For 6 minutes, participants read individually
an article detailing an upcoming conference on
one of the three topics: the threat posed by ISIS
(outgroup threat condition), an unspecified earth-
quake disaster (environmental threat condition),
and generic international cooperation (baseline
condition). The texts in all three conditions were
identical except for one paragraph detailing the
potential threat, and each text was anchored by a
relevant picture: an ISIS soldier with a knife and a
hostage kneeling in front of him (outgroup threat
condition); a girl amid debris following an envi-
ronmental disaster (environmental threat condi-
tion); and a generic conference picture (baseline
condition; see supplemental material, Section S5).
The content of all primes was identical across our
field sites to assure between-site comparability of
the obtained results. While the risk for outgroup
and disaster threats naturally varies across field
sites, we conducted additional robustness analyses
to account for this variation (see following lines).
After reading the priming texts, participants
answered questions about the content of the arti-
cle to ensure they paid sufficient attention (we
controlled for inattention in our statistical mod-
els) as well as to remind them of the main topic
of the conference (they were asked to describe
the image and the main topic of the conference).
Next, participants were collectively (i.e., in their
groups of four) introduced to a modified version
of the hidden profile task (Stasser & Titus, 2003),
which is a form of group-decision task where
some information is shared collectively, while
some is accessible only to certain members of the
group. In the present experiment, each partici-
pant received an information sheet at their cubi-
cle with information about six candidates that
might negotiate on behalf of the participant’s
country at the upcoming conference introduced
in the priming material. Each sheet comprised six
characters with three statements for each one:
two statements were shared among all partici-
pants, and one statement was unique for each
participant. Participants were then given 3 min-
utes to study the information. The candidates
were defined based on two variables: (a) their
degree of parochialism (based on statements of
in-group devotion/outgroup hostility) and (b)
whether they had a military or civilian back-
ground (see supplemental material, Section S6).
After reading the materials, participants were
instructed to get together to discuss and decide,
in 20 minutes, on which candidate should repre-
sent their country at the upcoming conference
with the condition-specific topic (ISIS threat,
earthquake, or control topic). Participants had the
following tasks: (a) attach printed symbols of
three selected candidates to the whiteboard and
(b) create a “poster” providing at least one reason
from each participant, which indicated the ration-
ale for selecting the specific candidates. The hid-
den profile interaction task was the primary
measurement period for our sociometric data; the
task was designed to encourage dynamic move-
ment and regrouping, providing raw material for
our behavioral measures.
Following the hidden profile task, participants
were asked to leave the experimental room and
wait in a hallway until the experimenter called
them back. This 5-minute period allowed us to
assess our main measures of interest during a free
interaction period, rather than during a structured
task. Finally, the experimenter called participants
back into the room, asking them to fill out final
questionnaires that assessed their willingness to
fight, die, and make other costly sacrifices on
behalf of their country. At the end of the experi-
ment, participants received either class credit or a
show-up fee paid at standard rates for an
Lang et al. 7
hour-long experiment in each location (except for
Brazil, where participants were not allowed to
receive money as per national policy).
Measures
Our measures were divided into survey items and
behavioral measures obtained through the socio-
metric badges. While the behavioral measures
comprised our outcome variables, the survey
measures comprised both outcome variables (will-
ingness to fight) as well as predictor and control
variables (e.g., identity fusion, conflict salience).
Surveys. All materials were translated and then
back-translated into local languages to ensure
comprehension. Questionnaires were presented
through the computer program Qualtrics, except
in Mauritius, where we used pen and paper. Before
creating latent variables out of individual scale
items, we assessed measurement invariance of the
theoretical constructs across our sites (Boer et al.,
2018). Specifically, we used multigroup confirma-
tory factor analysis (MG CFA; Muthén, 1989) to
test for configural, metric, and scalar invariance
(using R code developed by Fischer and Karl
[2019]). For each invariance test, we obtained basic
fit indices and assessed the model fit (well-fitting
models indicated by CFI and TLI > .95; RMSEA
< .06; SRMR < .08; Vandenberg & Lance, 2000),
as well as the difference between fits of the invari-
ance models (ΔCFI ⩽ .02; Rutkowski & Svetina,
2014).
First, we created a latent variable pertaining to
willingness to fight for one’s country (Swann et al.,
2010), measured with a five-item scale. Since the
model revealed metric variance, we removed one
item, which improved the overall fit of the configu-
ral model as well as metric invariance (CFI = .99,
TLI = 0.97, RMSEA = .05, SRMR = .02; see
Table S3 for loadings and intercepts by country).
However, we also detected scalar variance
(ΔCFIscalar-metric = −.81), which was driven by the
Japanese site. We provide two remedies for the
detected scalar variance: first, we let the intercepts
for willingness to fight vary between sites (see fol-
lowing lines), effectively focusing on the within-site
variance in this measure. Second, in the supplemen-
tal material, we provide the same analyses excluding
the Japanese site.
Similarly, we analyzed the invariance of preex-
isting levels of participants’ fusion with their
country, which served as a moderating factor in
our models. Fusion with country was measured
using the seven-item Fusion Scale (Swann et al.,
2009), which was previously tested in various
countries (Swann et al., 2014) and whose visual
analogue predicted cooperation in small-scale
societies (Purzycki & Lang, 2019). After eliminat-
ing two items that increased metric variance, the
configural invariance model revealed a sufficient
fit to the data (CFI = .95, TLI = 0.89, RMSEA
= .15, SRMR = .04; see Table S3 for loadings
and intercepts by country). To account for
detected scalar variance, we z-scored the identity
fusion measure by site (see supplemental mate-
rial, Section S1.2 for further discussion of the
MG CFA analysis).
We obtained participants’ assessments of the
threat posed to their country by international
conflict to account for the fact that our sites dif-
fered on their potential exposure to conflict
(answered on a 7-point Likert scale; 1 = strongly
disagree, 7 = strongly agree). As control variables, we
collected data on the Ten Item Personality
Measure (TIPI; Gosling et al., 2003), from which
we used two items (“I see myself as anxious, eas-
ily upset” and “I see myself as extraverted, enthu-
siastic”) to assess individual levels of neuroticism
and extraversion, because both may affect behav-
ioral measures (Olguin Olguin, 2011).
Furthermore, we asked participants to place
themselves on a liberal–conservative political
spectrum using a 7-point Likert scale (1 = very
liberal, 7 = very conservative), and how much they
perceived earthquakes to threaten their country
(1 = strongly disagree, 7 = strongly agree). We also
asked participants whether they had met any of
the other participants from their session before
the experiment, and we controlled for this poten-
tial familiarity in our statistical models (in 18% of
the sessions, at least two people knew each other;
however, removing these sessions from the
regression models did not change our results; see
8 Group Processes & Intergroup Relations 00(0)
supplemental R code). To ensure that participants
paid attention to our manipulation, we asked
them three control questions (year and name of
the conference, and participating countries), and
we used the number of mistakes to adjust the
coefficients in our statistical models. Finally, we
asked participants to rate the credibility of the
provided article, to control for interindividual
variability in the prime’s effectiveness (see supple-
mental material, Section S4 for the full
questionnaire).
Behavioral measures. To obtain continuous behav-
ioral measures during the hidden profile task and
the free interaction period, we employed the soci-
ometric badge (Kim et al., 2012; Waber et al.,
2011). This badge is of similar size to that of a
common smartphone (although much lighter)
and is placed on the chest, hanging on a lanyard
around the neck. Each badge records activity
through an accelerometer, computed as the abso-
lute value of the first derivative of energy (see
Figure S2 for an illustration). For easier interpre-
tation, we multiplied the activity values by 9.8 to
get acceleration in m/s2. Thus, difference in
activity means both difference in the vigor and
the frequency of activity. Note that to make sure
that this activity did not reflect coping with a
common threat or reluctance to associate with
others (e.g., walking far from the group), we con-
trolled for neuroticism and introversion in our
supplemental analyses.
Furthermore, each badge sends Bluetooth sig-
nals to other badges with a frequency of 1 Hz and
measures the strength of the signal reciprocated by
other badges that are within interaction proximity
(received signal strength indicator [RSSI]). The
strength of the returned signal is thus a measure of
relative distance between two badges (independent
of whether participants faced each other or not).
Since the RSSI measure is indicated in negative
numbers of decibel (dB), with −90 dB being the
detection threshold (around 1.5 meters) and 0 dB
the maximal proximity, we transformed the RSSI
such that zero was the detection threshold and 90
the maximal signal strength. This transformation
affords intuitive reading of proximity, with
increasing positive numbers indicating increasing
proximity (see Figure S3 for an illustration). The
proximity score for each participant is computed as
the sum of RSSI values of all detected interactions
with the other three participants in the same session,
divided by the number of minutes for each task, to
arrive at average proximity per minute. Similar to
activity, the sum of RSSI values subsumes both the
temporal and spatial dimensions to account for the
fact that some participants may have had less fre-
quent but very close encounters. Since one proximal
encounter is counted for both individuals in prox-
imity (irrespective of who was the approaching indi-
vidual), the number of these interactions is, to some
extent, session-specific. That is, while each partici-
pant has a unique proximity score reflecting their
interaction with the other three participants, these
scores are, to some extent, correlated between par-
ticipants in one session. We adjusted our regression
models for this overlap in encounters by letting the
intercepts for individual sessions vary; that is, by fit-
ting a session-specific intercept and analyzing only
the variance not explained by belonging to a specific
session (see the Analysis section and supplemental
material, Section S3 for the amount of variance
explained by varying intercepts by session; see also
supplemental R code).
Finally, by combining the accelerometer meas-
urements with proximity values and detection of
face-to-face interactions (via infrared sensors), the
sociometric badge provides measurements of
movement mirroring. When two participants are
in close encounter (defined by the proximity
measurement) and face each other, the badge
compares their activity levels—utilizing a 5-sec-
ond sliding window, the percentage of mirrored
movements is computed for each second. Thus,
the value of movement mirroring corresponds to
an average percentage of movements mirrored
between participants in a group during our tasks.
Similar to proximity, the mirroring measure
reflects a dyadic encounter and is, therefore,
counted for both individuals (imitator and imita-
tee). The overlap of the movement mirroring
measure between participants in one session is
again absorbed by letting the intercepts of our
regression models vary by session. All sociometric
Lang et al. 9
data were extracted using the Sociometric DataLab
software, Version 3.1.2468 (Waber et al., 2011).
Analysis
The data were analyzed in R, Version 3.4.1 (R Core
Team, 2020). Each model is a hierarchical linear
model with three levels: participants nested within
sessions that are nested within sites. First, we
nested participants within sessions to account for
the fact that individual-level measures are corre-
lated between participants from the same session.
Second, since our data set comprised data from six
different countries, we varied intercepts by coun-
try, effectively adjusting the models for between-
site differences in our outcome variables and other
potential unmeasured between-site variability.
Together, this hierarchical structure allowed us to
investigate individual-level predictors of mirror-
ing, proximity, activity, and willingness to fight
while accounting for the interdependencies within
our data. We set the no-threat baseline condition as
the reference category.
As a starting point, we built linear mixed mod-
els (LMMs) examining the main effects of our
manipulation (outgroup threat, environmental
threat, and baseline conditions). In the second
step, we added individual characteristics, adjust-
ing our models for the potentially confounding
effects of sex, identity fusion, salience of interna-
tional conflict, salience of natural disaster, extra-
version, neuroticism, and conservatism. In the
third step, we adjusted our models for variables
assessing the quality of our manipulation, namely
participants’ rating of the credibility of our prim-
ing material, the number of wrong answers dur-
ing the attention check, and whether participants
knew someone else in their session. The general
model structure was as follows:
Y=
+ u+ u + T
+ T+ X
+ Z+
ijk
0i 0j 0k 1i 1
2i 2ix
iz i
ββ
ββ
βε
()
()
~ N , 2
µσ (1).
where Yijk is our behavioral measure of individual
i within session j and site k. β0i is a fixed intercept,
u0j is a varying intercept for session, and u0k is a
varying intercept for site. T1iβ1 is the individual-
level parameter for the fixed effect of outgroup
threat treatment (no-threat baseline vs. outgroup
threat), and T2iβ2 is the individual-level parameter
for the fixed effect of environmental threat treat-
ment (no-threat baseline vs. environmental
threat). Xiβxi is the group of individual-level
parameters for the effects of sex, identity fusion,
salience of international conflict, salience of nat-
ural disaster, extraversion, neuroticism, and con-
servatism. Ziβzi is the group of individual-level
parameters for the effects of participants’ rating
of the credibility of our priming material, the
number of wrong answers during the attention
check, and whether participants knew someone
else from their session. εi represents the error
term for the assumed normal distribution.
After examining these models, we interacted
the treatment factor variable with three types of
theoretically important moderators: sex, salience
of international conflict, and identity fusion.
Next, we performed four robustness checks: (a)
we fitted generalized linear mixed models
(GLMMs) that allowed us to more rigorously
account for the specific data-generation process
of our outcome variables; (b) we built the same
LMMs as in our main analyses but excluded the
New Zealand site that had multiple nationalities
in the sample; (c) we analyzed group formation
also during the free interaction task that followed
immediately after the hidden profile task; and (d)
we let the slopes of credibility of our primes, sali-
ence of international conflict, salience of natural
disaster, and individual conservatism vary by site,
to account for the variables’ potentially different
effects across sites (see supplemental material,
Section S1.3 and supplemental R code).
Results
Manipulation Check
As a manipulation check, both the outgroup
threat (β-estimate = 1.49, 95% CI [1.21, 1.78])
and environmental threat (β-estimate = 1.52,
95% CI [1.23, 1.81]) conditions increased the
10 Group Processes & Intergroup Relations 00(0)
feeling of being threatened to a similar level,
compared to the no-threat baseline condition.
However, only in the outgroup threat condition
(β-estimate = 0.31, 95% CI [0.05, 0.58]) and not
in the environmental threat condition (β-estimate
= −0.02, 95% CI [−0.28, 0.24]) were participants
more likely to elect more parochial candidates,
compared to the baseline condition (for further
checks, see supplemental material, Section S3.1).
These results suggest that our manipulation
successfully conditioned specific feelings being
threatened.
Main Models
First, when participants were in face-to-face
contact, we observed that, on average, about
28% of their movements within the 5-second
moving window were mirrored in the no-threat
baseline condition. While the amount of mir-
roring in the outgroup threat condition did not
differ from the baseline condition (β-estimate
= −0.06, 95% CI [−0.69, 0.58]), we observed
higher movement mirroring in the environmen-
tal threat condition (β-estimate = 0.71, 95% CI
[0.07, 1.35]).
Next, the differences between the baseline and
the outgroup and environmental conditions in
the measure of proximity varied around zero (see
Table 2). Regressing participants’ activity (indi-
cated as a mean acceleration per second) on the
experimental conditions, the LMM revealed that
participants were more active in both the out-
group threat (β = 0.74, 95% CI [−0.08, 1.55])
and environmental threat conditions (β = 0.82,
95% CI [0.005, 1.64]), compared to the no-threat
baseline condition, albeit the 95% CI for the out-
group threat condition crossed zero.
Finally, we observed the highest willingness to
fight for one’s country in the outgroup threat
condition (β = 0.13, 95% CI [−0.05, 0.30]), albeit
the 95% CI again crossed zero. The difference
between the baseline no-threat condition and the
environmental threat condition revealed larger
variability (β = 0.07, 95% CI [−0.11, 0.25]).
Excluding the Japanese sample from the analysis
of willingness to fight (see the Methods and
Procedure and Materials sections) strengthened
these results (βoutgroup = 0.20, 95% CI [0.00, 0.41];
βenvironmental = 0.08, 95% CI [−0.12, 0.29]).
These results were robust toward adding a set
of theoretically important predictors as well as
adding a set of variables related to the experi-
mental manipulation. Furthermore, we also let
the slopes of these control variables vary by site
and conducted analyses with different distribu-
tional assumptions for the residuals. None of
these robustness checks suggested that the
reported relationships are unstable (see Table 2
and supplemental material, Section S3 for detailed
results and site-specific analyses).
Interaction Models
Sex as a moderator. Investigating the moderating
role of sex on mirroring, we observed that the
difference between sexes in the outgroup threat
condition was larger compared to the no-threat
baseline condition; specifically, for each 5-second
window, males mirrored each other’s movements
1% more than females in the outgroup threat
condition (βdifference = 1.38, 95% CI [0.06, 2.71]).
There was no interaction effect between sex and
environmental threat condition (βdifference =
−0.02, 95% CI [−1.32, 1.28]). A post hoc analysis
of simple effects from the interaction model
revealed that males in the outgroup condition dis-
played higher mirroring rates than males in the
baseline condition, although confidence intervals
of this effect showed high uncertainty (βoutgroup, males
= 0.70, 95% CI [−0.23, 1.62]). Furthermore, the
absence of an interaction effect in the environ-
mental threat condition was likely caused by a
lower intercept for mirroring in the outgroup
threat condition (i.e., lower mirroring rates
observed in females in the outgroup condition
compared to females in the baseline condition;
see intercept differences in Table 3).
For proximity, we did not observe a moderating
sex effect: males were generally lower in proximity,
but this difference did not vary across our manipu-
lations (see Table 3 and Figure 2). Looking at the
moderating role of sex on activity, we observed, on
average, 1.47 times faster accelerations for male
Lang et al. 11
activity compared to females when participants
faced the outgroup threat, and this sex difference
was higher compared to the sex effect in the no-
threat baseline condition (βdifference = 2.04, 95% CI
[0.37, 3.71]). As predicted, the sex effect did not
differ between the environmental and baseline con-
ditions (βdifference = 0.56, 95% CI [−1.08, 2.21]). A
post hoc analysis revealed that compared to males
in the baseline condition, males in the outgroup
condition were more active (βoutgroup, males = 1.81,
95% CI [0.42, 3.20]; see Figure 2 and Table 3).
Conflict salience as a moderator. Focusing on the mod-
erating effects of conflict salience, we observed
that the slope of conflict salience predicting the
amount of mirroring was more positive in the out-
group condition compared to the no-threat base-
line condition (βdifference = 0.20, 95% CI [−0.01,
0.40]), while the slopes of the environmental and
baseline conditions were indistinguishable (βdifference
= 0.09, 95% CI [−0.12, 0.29]). A post hoc analysis
of these interaction effects showed that partici-
pants scoring highest on the conflict salience scale
displayed higher mirroring rates compared to the
similarly scoring participants in the baseline condi-
tion (βoutgroup, conflict salience > 6 = 1.67, 95% CI
[−0.02, 3.37]). However, analogically to the moder-
ating effects of sex on mirroring, this interaction
effect was driven mostly by low mirroring rates of
participants with low conflict salience scores in the
outgroup condition (see intercept differences in
Table 3).
Conflict salience also showed important mod-
erating effects for our measures of proximity. In
the no-threat baseline condition, increasing con-
flict salience was associated with decreasing prox-
imity, but the opposite trend was observed in the
outgroup threat condition (βdifference = 9.67, 95%
CI [0.45, 18.88]). No such slope difference was
observed between the no-threat baseline and
environmental threat conditions (βdifference = 1.81,
95% CI [−7.26, 10.88]). However, while post hoc
analysis of the interaction effect for the outgroup
threat condition revealed that the effects for par-
ticipants scoring highest on this latent variable
indeed showed the highest mirroring rates from
all of the comparison groups, the difference from
the baseline condition was quite variable and does
not allow us to draw confident inferences
(βoutgroup, conflict salience > 6 = 17.71, 95% CI [−37.10,
72.53]). Possibly, the significant interaction for
the outgroup threat condition was partially driven
by the fact that participants who were not wor-
ried about international conflict were lower in
proximity in this condition compared to the no-
threat baseline (see intercepts of this model in
Table 3). There was no moderating effect of con-
flict salience for the effects of treatment on
activity.
Identity fusion as a mediator. For our measure of
identity fusion with one’s country, the 95% CI of
the moderating effects on mirroring, proximity,
and activity always crossed zero. While the moder-
ating effects for the outgroup condition were posi-
tive (as predicted), the moderating effects of
identity fusion were too variable to allow unequiv-
ocal interpretation (see Table 3 and Figure 2). The
Table 2. Beta estimates with 95% CI for main effects of treatment.
Mirroring Proximity Activity Fight
Treatment: Outgroup −0.06
[−0.69, 0.58]
−4.72
[−30.01, 20.58]
0.74†
[−0.08, 1.55]
0.13
[−0.05, 0.30]
Treatment:
Environmental
0.71*
[0.07, 1.35]
−4.39
[−29.76, 20.99]
0.82†
[0.005, 1.64]
0.07
[−0.11, 0.25]
Intercept 28.82***
[26.87, 30.77]
267.93***
[221.34, 314.51]
9.53***
[8.07, 10.99]
2.96***
[2.38, 3.54]
N participants 761 762 761 824
Note. The no-threat baseline condition is the reference category for the treatment variable.
†p < .1. *p < .05. **p < .01. ***p < .001.
12 Group Processes & Intergroup Relations 00(0)
Table 3. Beta estimates with 95% CI for the interaction of treatment with sex, international conflict salience, and fusion.
Moderator: Sex Moderator: Conflict salience Moderator: Identity fusion
Mirroring Proximity Activity Mirroring Proximity Activity Mirroring Proximity Activity
Treatment: Outgroup −0.73†3.88 −0.18 −0.97†−43.51†0.60 −0.15 −3.00 0.68
[−1.58, 0.12] [−29.64, 37.39] [−1.25, 0.90] [−2.07, 0.13] [−89.81, 2.80] [−0.99, 2.19] [−0.81, 0.51] [−28.59, 22.59] [−0.16, 1.51]
Treatment: Enviro. 0.67 1.31 0.66 0.26 −12.42 1.10 0.63†−5.10 0.85***
[−0.18, 1.51] [−31.93, 34.54] [−0.41, 1.72] [−0.84, 1.36] [−58.69, 33.85] [−0.49, 2.69] [−0.02, 1.28] [−30.39, 20.20] [0.03, 1.68]
Moderator −0.34 −22.00 −0.57 −0.09 −9.02* −0.12 −0.20 −6.91 −0.10
[−1.32, 0.64] [−60.48, 16.49] [−1.80, 0.66] [−0.25, 0.06] [−15.91, −2.12] [−0.36, 0.13] [−0.47, 0.07] [−18.92, 5.09] [−0.52, 0.32]
Outgroup * Mod 1.38* −17.05 2.04* 0.20†9.67* 0.02 0.15 5.96 0.05
[0.06, 2.71] [−69.25, 35.16] [0.37, 3.71] [−0.01, 0.40] [0.45, 18.88] [−0.31, 0.34] [−0.23, 0.52] [−10.62, 22.55] [−0.53, 0.63]
Enviro. * Mod −0.02 −15.99 0.56 0.09 1.81 −0.06 0.19 8.72 −0.04
[−1.32, 1.28] [−67.25, 35.27] [−1.08, 2.21] [−0.12, 0.29] [−7.26, 10.88] [−0.38, 0.26] [−0.18, 0.55] [−7.49, 24.94] [−0.61, 0.52]
Intercept 27.96*** 309.32*** 9.03*** 28.12*** 327.08*** 8.57*** 27.77*** 314.23*** 8.64***
[25.69, 30.23] [239.52, 379.11] [6.78, 11.29] [25.83, 30.41] [254.97, 399.19] [6.24, 10.90] [25.52, 30.01] [245.62, 382.84] [6.44, 10.84]
N participants 691 692 691 691 692 691 691 692 691
Note. Abbreviated results from our full moderator models. Outgroup denotes the outgroup threat condtion; Enviro denotes the envioronmental threat condition; Mod stands for Moderators. The full models
adjust for sex, fusion, conflict salience, natural disaster salience, extraversion, neuroticism, conservatism, prime credibility, mistakes, and acquaintance (see Tables S5, S11, and S16 in the supplemental material).
The no-threat baseline condition is the reference category for the treatment variable. Interactions compare the slopes of moderating variables across the outgroup and environmental threat conditions with the
no-threat baseline condition.
†p < .1. *p < .05. **p < .01. ***p < .001.
Lang et al. 13
supplemental material, Section S3 offers further
details on the interaction analyses, including the set
of robustness checks as for our main models.
Discussion
Across six societies, we assessed how outgroup
and environmental threats affect group dynamics
by utilizing unobtrusive devices that quantify
between-subject movement mirroring, proximity,
and individual activity. We found that both types
of threat increased activity aimed at threat resolu-
tion, however, this effect was the strongest for
males in the outgroup threat condition. The
measures of affiliative behavior revealed more
complex patterns: we observed higher mirroring
rates in the environmental condition compared to
the baseline condition, but outgroup threat
increased movement mirroring only for males,
suggesting that females were galvanized only by
the environmental threat (albeit this sex differ-
ence was imprecisely estimated). Similarly, the
Figure 2. A 3 x 3 mesh of the interaction models showing regression estimates with 95% CI for the no-threat
baseline condition.
Note. Plots show simple interaction effects adjusted only for participants’ nesting in sessions and sites. Mirroring was, on
average, higher in both threat conditions. The only stable effect on proximity was observed for participants in the outgroup
condition who worried about international conflict. The activity measure displayed in the last row was, on average, higher in
the outgroup and environmental threat conditions compared to the no-threat baseline condition, but the effect in the outgroup
threat condition was moderated by sex.
14 Group Processes & Intergroup Relations 00(0)
treatment with outgroup threat increased mirror-
ing for people who perceived armed conflict with
another country as a real threat. There were no
main effects of either threat on proximity,
although, similar to mirroring, participants who
considered international conflict as a real threat
to their country tended to spend more time in
proximity in the outgroup threat condition.
However, this result was substantially variable
and does not allow confident inferences.
Per our conceptualization of cohesive groups,
we focused on two distinct facets of affiliative
behavior, namely movement mirroring and prox-
imity, and hypothesized that they will increase with
outgroup threat. Both are well-established meas-
ures of interpersonal rapport (Grahe & Bernieri,
1999; Lakin et al., 2003; Tickle-Degnen &
Rosenthal, 1990), and may be recruited as behavio-
ral strategies to increase interpersonal liking
(Bernieri et al., 1996; Lakin & Chartrand, 2003)
and facilitate team cohesion during instrumental
tasks (Hoegl & Proserpio, 2004; Olguin Olguin,
2011; Zhang et al., 2018). While proximity reflects
mutual attraction and willingness to interact,
movement mirroring is a more dynamic measure
that results from actual ongoing interaction
(Tickle-Degnen & Rosenthal, 1990), and is well
correlated with instrumental and helping behavior
(Chartrand & Lakin, 2013; van Baaren et al., 2004).
We did not observe any main effects of the
threat conditions on proximity. Possibly, the hid-
den profile task constrained participants to interact
within a relatively narrow distance (group discus-
sion was concentrated around a table in the center
of the room), masking the expected effects of our
threat manipulation (Zhang et al., 2018). A stronger
signal was detected in the measure of movement
mirroring, which does not rely on variance in spa-
tial distance and is more indicative of interpersonal
rapport during instrumental tasks (Tickle-Degnen
& Rosenthal, 1990; Zhang et al., 2018). Specifically,
participants in the environmental threat group
mirrored each other more compared the no-threat
baseline condition. However, no such effect was
observed for the outgroup threat condition, and
we interpret this absence of effect as masked by
the theoretically important moderators.
First, interacting the rating of conflict sali-
ence with our manipulation revealed that par-
ticipants who were more worried about the
possibility of their country engaging in an
international conflict displayed higher levels of
movement mirroring in the outgroup threat
condition compared to the baseline condition.
Since the interaction effect between condition
and salience of international conflict on pre-
dicting affiliative behaviors held also after con-
trolling for neuroticism (see Tables S5 and
S11), it is most likely not driven by affiliative
tendencies of anxious individuals (Schachter,
1959). On the contrary, we suggest that the
affiliative tendencies observed in the mirroring
measure reflect processes related to building a
cohesive team by aligning group intentions and
coordinating communication as well as increas-
ing rapport, that is, factors needed for success-
ful teamwork (Hasson & Frith, 2016). The fact
that these processes were observed only for
people worried about conflict supports the
notion that perceived intergroup competition
increases the need for within-group coopera-
tion and increased affiliation (Francois et al.,
2018; Majolo & Maréchal, 2017). We also
observed similar results for the proximity
measure, where conflict salience positively pre-
dicted proximity only in the outgroup condi-
tion. However, due to reasons discussed in the
previous paragraph, these results are uncertain
and do not allow confident interpretation.
We also observed that participants who were
not worried about international conflict were
lower in proximity and movement mirroring in
the outgroup threat condition compared to the
no-threat baseline (see intercepts of this model in
Table 3). This finding suggests that outgroup
threats may lead to the dispersal of less worried
participants, possibly due to the increased impor-
tance of avoidant strategies (Ein-Dor et al., 2011).
In other words, displaying an outgroup threat
prime to participants who are not usually worried
about such threats may have decreased their need
for sociality. However, this interpretation is spec-
ulative, and future research should investigate this
relationship in more detail.
Lang et al. 15
Second, we found that males were more sensi-
tive to our outgroup threat manipulation. While
environmental threat increased movement mirror-
ing for both sexes, outgroup threat increased mir-
roring only for males. This result is consistent with
the male warrior hypothesis, which states that men
are more sensitive to cues of intergroup conflict
than women (van Vugt et al., 2007) and, upon the
detection of such cues, display higher rates of
parochial cooperation (McDonald et al., 2012;
Yuki & Yokota, 2009). In light of this theory, our
results might be interpreted as supporting the
notion that threats increase the need for affiliation
and cooperative behavior (Gelfand, 2019; Gelfand
et al., 2011), with the caveat that women are less
sensitive to outgroup threats and/or are unwilling
to form cohesive groups under such threats. The
latter notion is supported by the fact that women
in the outgroup threat condition displayed even
lower mirroring rates than women in the baseline
condition. Nevertheless, due to the imprecise esti-
mate of this effect (95% CI crossed zero), this
result should be interpreted with caution.
Analogically to the mirroring movement results,
the analysis of movement energy revealed that
both threatening conditions increased participants’
activity compared to the baseline condition; how-
ever, outgroup threat increased activity only for
males. In contrast to the mirroring results, women
in the outgroup threat condition showed similar
activity levels as women in the baseline condition,
supporting the notion that the decrease in female
mirroring described in the previous paragraph may
have been a chance result. Furthermore, males in
the outgroup threat condition showed even higher
rates of activity than participants in the environ-
mental threat condition (see Figure 2). This result
suggests that males may be sensitive to cues of
intergroup conflict, which is reflected in their
physical action and related energy expenditure
dedicated to solving a group problem, as in the
hidden profile task. Since movement energy is cor-
related with perception of increased effectiveness
of teamwork (Olguin Olguin, 2011), males in the
outgroup threat condition were mobilized to take
action, which also contributes to building a cohe-
sive group (Zhang et al., 2018).
The extent to which threat manipulation
affected self-reported willingness to fight for
one’s country was lower than expected. While the
outgroup threat condition revealed higher ratings
of willingness to fight compared to the baseline
condition, the 95% CI for this difference crossed
zero. Excluding the Japanese sample from our
analysis increased this effect (see the Methods
section for rationale), yet the detected effect
remained relatively small. Nevertheless, this effect
shows that our manipulations affected both self-
reported and behavioral measures aimed at con-
flict resolution. In our supplemental analyses, we
also observed a substantial sex effect, with males
more willing to fight (consistent with the male
warrior hypothesis), an effect also predicted by
salience of international conflict. However, these
variables did not moderate the effect of the out-
group condition (see Table S22).
Likewise, we did not detect any interaction
between sex and outgroup threat affecting the
proximity measure. Males were, on average, fur-
ther apart from each other, but this effect was
constant across all conditions. While mirroring
and activity do not require physical proximity to
establish rapport and facilitate collective action in
males, it can be speculated that proximity is a type
of affiliative behavior favored by females rather
than males. The absence of a moderating effect
of conflict salience on our measure of activity is
likely due to the fact that activity in the outgroup
and environmental threat conditions was already
higher compared to the no-threat baseline condi-
tion. Finally, our measure of identity fusion with
one’s country did not moderate any of the treat-
ment effects on our behavioral variables, despite
the fact that previous studies showed a positive
correlation between identity fusion and coopera-
tive behavior (Purzycki & Lang, 2019), and that
theoretical work predicted that this relationship
should strengthen during intergroup conflict
(Whitehouse, 2018).
Together, these results indicate that outgroup
threat prompts active group behavior particularly
for men and, to some extent, also increases affili-
ative behaviors for men and for individuals who
experience the threat as a real danger to their
16 Group Processes & Intergroup Relations 00(0)
country. To bolster these results, we performed
several robustness checks. Specifically, we
adjusted the models for sensitivity to threats
unrelated to an outgroup (natural disaster) as well
as for neuroticism, to eliminate the variance
explained by general sensitivity to threats that
may provoke affiliative behaviors (Schachter,
1959; Taylor, 2006). We also adjusted the models
for (a) the effects of individual conservatism and
associated norm tightness (Gelfand et al., 2011);
(b) individual extraversion, which may positively
affect affiliative behaviors (Duffy & Chartrand,
2015b); (c) the strength of our manipulation;
and (d) the familiarity of participants within the
group.
Furthermore, by letting the intercepts vary
across sites in our models, we adjusted the model
estimates for the fact that the measured outcome
variables were differentially distributed across sites
(see Figure 1). To illustrate the extent of between-
site variation in the tested relationships, we also
provide site-specific results for the main models in
the supplemental material. Site-specific results sug-
gest that our manipulations usually affected the
outcome variables in the same direction across the
six sites, although the magnitude of those effects
was variable. The only exception was the measure
of proximity, where the direction of coefficients
related to our manipulation varied between sites,
suggesting that proximity may be a measure most
sensitive to the specific cultural milieu and associ-
ated conventions regarding interpersonal space
(Talhelm et al., 2018). Finally, since our sites may
have potentially differed on the extent of individu-
als’ sensitivity to various threats, we also built
LMMs where we let the slopes of conflict salience,
natural disaster salience, conservatism, and credibil-
ity of our manipulation on the outcome variables
vary by site. Nevertheless, the variation explained
by these varying slopes was usually negligible and
did not affect the interpretation of the fixed factors
in our models.
Despite these various control measures, our
results have important limitations. While the
strength of our study lies in the use of a large
cross-cultural sample, this sample comprised only
student populations. Furthermore, our New
Zealand sample was multinational, which may
have impacted both our behavioral and self-
reported measures directed at creating cohesive
groups in order to defend one’s country. While
excluding these international participants from our
analyses and supplemental analyses without the
New Zealand site did not result in any qualitative
differences, future studies manipulating nation-
level perception of intergroup conflict should
carefully select nationally homogenous popula-
tions. Likewise, our composite measures of iden-
tity fusion and willingness to fight revealed
measurement variance across sites. We improved
the invariance of these latent variables by exclud-
ing items that had variable factor loadings across
sites, but such a practice runs the risk of under-
representing the concept (Fischer & Karl, 2019).
Furthermore, whereas using the same priming
materials across different field-sites confers mul-
tiple advantages (cf. Lang et al., 2016; Nichols
et al., 2020), their strength and effectiveness
might have been too low at sites that have never
experienced a terrorist attack. Combining locally
salient materials that would be comparable across
sites should overcome these issues, although
finding such materials would be extremely chal-
lenging. Finally, while providing unique unobtru-
sive measurements, the sociometric badges often
malfunctioned (we lost around 10% of data) and
require increased stability for reliable data collec-
tion, especially if they are to be deployed in
demanding real-life social situations (Xygalatas
et al., 2019). Despite these limitations, our find-
ings offer preliminary evidence of dynamic pro-
cesses of bottom-up group formation for people
who feel under threat by other groups.
Acknowledgements
We thank Joey Chua, Nethy Chunwan, Laura Meunier,
Mehreen Mey, Avishkar Soonea, and Maneesha Soonea
for their help with data collection. We also thank
Gabriela Baranowski Pinto, Radek Kundt, Radim
Chvaja, Vitor Profeta, and members of the Culture,
Cognition, and Coevolution Lab at Harvard University
for their feedback on earlier versions of this manu-
script. We are also grateful to Ronald Fischer for pro-
viding statistical advice.
Lang et al. 17
Author contributions
C. K., M. L., D. X., H. W., and A. G. conceived the
study, prepared protocols, and managed data collection.
N. B., J. H., C. J., C. K., M. L., M. M. D., P. R., E. T., A.
V., H. W., M. E. Y., and M. Y. collected data. M. L. con-
ducted all analyses and created the graphs and tables. C.
K. prepared all illustrations and materials. M. L. drafted
the manuscript and supplemental material. All authors
participated in refining the protocols, experimental
designs, and in manuscript preparation.
Data availability
All data, materials, and R code are available at https://
osf.io/fwztr/
Funding
The author(s) disclosed receipt of the following finan-
cial support for the research, authorship, and/or publi-
cation of this article: This study was funded by a
subgrant from Oxford University as part of the Ritual’s
Impact on the Contemporary World Project, awarded
by the Templeton Foundation. A. G. acknowledges
funding from the Spanish government’s Ministerio de
Ciencia, Innovación y Universidades (RTI2018-
093550-B-100). C. K. and H. W. were supported by an
Advanced Grant from the European Research Council
under the European Union’s Horizon 2020 Research
and Innovation Programme (Grant Agreement
694986).
ORCID iDs
Martin Lang https://orcid.org/0000-0002-2231
-1059
Chris Jackson https://orcid.org/0000-0003-2340
-8116
Alexandra Vázquez https://orcid.org/0000-0002
-6040-9102
Supplemental material
Supplemental material for this article is available online.
References
Allen, T. J. (1970). Communication networks in R&D
laboratories. R&D Management, 1, 14–21. https://
doi.org/10.1017/CBO9781107415324.004
Atran, S. (2016). The devoted actor: Unconditional
commitment and intractable conflict across cul-
tures. Current Anthropology, 57, S192–S203. https://
doi.org/10.1086/685495
Atran, S., & Ginges, J. (2012). Religious and sacred
imperatives in human conflict. Science, 336, 855–
857. https://doi.org/10.1126/science.1216902
Bauer, M., Cassar, A., Chytilová, J., & Henrich, J.
(2014). War’s enduring effects on the develop-
ment of egalitarian motivations and in-group
biases. Psychological Science, 25, 47–57. https://doi.
org/10.1177/0956797613493444
Bernieri, F. J. (1988). Coordinated movement and
rapport in teacher–student interactions. Journal
of Nonverbal Behavior, 12, 120–138. https://doi.
org/10.1007/BF00986930
Bernieri, F. J., Gillis, J. S., Davis, J. M., & Grahe, J. E.
(1996). Dyad rapport and the accuracy of its judg-
ment across situations: A lens model analysis. Jour-
nal of Personality and Social Psychology, 71, 110–129.
https://doi.org/10.1037/0022-3514.71.1.110
Boer, D., Hanke, K., & He, J. (2018). On detecting
systematic measurement error in cross-cultural
research: A review and critical reflection on
equivalence and invariance tests. Journal of Cross-
Cultural Psychology, 49, 713–734. https://doi.
org/10.1177/0022022117749042
Böhm, R., Rusch, H., & Gürerk, Ö. (2016). What makes
people go to war? Defensive intentions motivate
retaliatory and preemptive intergroup aggression.
Evolution and Human Behavior, 37, 29–34. https://
doi.org/10.1016/j.evolhumbehav.2015.06.005
Bowles, S. (2008). Conflict: Altruism’s midwife. Nature,
456, 326–327. https://doi.org/10.1038/456326a
Bowles, S. (2009). Did warfare among ancestral hunter-
gatherers affect the evolution of human social
behaviors? Science, 324, 1293–1298. https://doi.
org/10.1126/science.1168112
Boyer, P., Firat, R., & van Leeuwen, F. (2015).
Safety, threat, and stress in intergroup rela-
tions: A coalitional index model. Perspectives on
Psychological Science, 10, 434–450. https://doi.
org/10.1177/1745691615583133
Carron, A. V., & Brawley, L. R. (2000). Cohe-
sion: Conceptual and measurement issues.
Small Group Research, 31, 89–106. https://doi.
org/10.1177/104649640003100105
Chartrand, T. L., & Lakin, J. L. (2013). The antecedents
and consequences of human behavioral mimicry.
Annual Review of Psychology, 64, 285–308. https://
doi.org/10.1146/annurev-psych-113011-143754
Cook, M. (1970). Experiments on orientation and
proxemics. Human Relations, 23, 61–76. https://
doi.org/10.1177/001872677002300107
De Dreu, C. K. W., Dussel, D. B., & Ten Velden, F.
S. (2015). In intergroup conflict, self-sacrifice is
18 Group Processes & Intergroup Relations 00(0)
stronger among pro-social individuals, and paro-
chial altruism emerges especially among cognitively
taxed individuals. Frontiers in Psychology, 6, Article
572. https://doi.org/10.3389/fpsyg.2015.00572
De Dreu, C. K. W., Gross, J., Méder, Z., Giffin, M.,
Prochazkova, E., Krikeb, J., & Columbus, S.
(2016). In-group defense, out-group aggres-
sion, and coordination failures in intergroup
conflict. Proceedings of the National Academy of Sci-
ences of the USA, 113, 10524–10529. https://doi.
org/10.1073/pnas.1605115113
Duffy, K. A., & Chartrand, T. L. (2015a). Mim-
icry: Causes and consequences. Current Opinion
in Behavioral Sciences, 3, 112–116. https://doi.
org/10.1016/j.cobeha.2015.03.002
Duffy, K. A., & Chartrand, T. L. (2015b). The
extravert advantage: How and when extra-
verts build rapport with other people. Psy-
chological Science, 26, 1795–1802. https://doi.
org/10.1177/0956797615600890
Dumont, M., Yzerbyt, V., Wigboldus, D., & Gordijn,
E. H. (2003). Social categorization and fear reac-
tions to the September 11th terrorist attacks. Per-
sonality and Social Psychology Bulletin, 29, 1509–1520.
https://doi.org/10.1177/0146167203256923
Ein-Dor, T., Mikulincer, M., & Shaver, P. R. (2011).
Effective reaction to danger: Attachment insecu-
rities predict behavioral reactions to an experi-
mentally induced threat above and beyond
general personality traits. Social Psychological
and Personality Science, 2, 467–473. https://doi.
org/10.1177/1948550610397843
Farley, S. D. (2014). Nonverbal reactions to an attrac-
tive stranger: The role of mimicry in communicat-
ing preferred social distance. Journal of Nonverbal
Behavior, 38, 195–208. https://doi.org/10.1007/
s10919-014-0174-4
Fessler, D. M. T., & Holbrook, C. (2014). Marching
into battle: Synchronized walking diminishes the
conceptualized formidability of an antagonist in
men. Biology Letters, 10, Article 592. https://doi.
org/10.1098/rsbl.2014.0592
Fischer, R., & Karl, J. A. (2019). A primer to (cross-
cultural) multi-group invariance testing possibili-
ties in R. Frontiers in Psychology, 10, Article 1507.
https://doi.org/10.3389/fpsyg.2019.01507
Francois, P., Fujiwara, T., & van Ypersele, T. (2018).
The origins of human prosociality: Cultural group
selection in the workplace and the laboratory.
Science Advances, 4, Article eaat2201. https://doi.
org/10.1126/sciadv.aat2201
Gelfand, M. J. (2019). Rule makers, rule breakers: How
tight and loose cultures wire our world. Scribner.
Gelfand, M. J., Raver, J. L., Nishii, L., Leslie, L. M., Lun,
J., Lim, B. C., Duan, L., Almaliach, A., Ang, S.,
Arnadottir, J., Aycan, Z., Boehnke, K., Boski, P.,
Cabecinhas, R., Chan, D., Chhokar, J., D’Amato,
A., Ferrer, M., Fischlmayr, I. C., . . .Yamaguchi, S.
(2011). Differences between tight and loose cul-
tures: A 33-nation study. Science, 332, 1100–1104.
https://doi.org/10.1126/science.1197754
Ginges, J., & Atran, S. (2011). War as a moral imperative
(not just practical politics by other means). Proceed-
ings of the Royal Society B: Biological Sciences, 278, 2930–
2938. https://doi.org/10.1098/rspb.2010.2384
Glowacki, L., Isakov, A., Wrangham, R. W., McDer-
mott, R., Fowler, J. H., & Christakis, N. A. (2016).
Formation of raiding parties for intergroup vio-
lence is mediated by social network structure. Pro-
ceedings of the National Academy of Sciences of the USA,
113, 12114–12119. https://doi.org/10.1073/
pnas.1610961113
Gómez, Á., Brooks, M. L., Buhrmester, M. D.,
Vázquez, A., Jetten, J., & Swann, W. B., Jr. (2011).
On the nature of identity fusion: Insights into the
construct and a new measure. Journal of Personal-
ity and Social Psychology, 100, 918–933. https://doi.
org/10.1037/a0022642
Gómez, Á., Chinchilla, J., Vázquez, A., López-Rod-
ríguez, L., Paredes, B., & Martínez, M. (2020).
Recent advances, misconceptions, untested
assumptions, and future research agenda for
identity fusion theory. Social and Personality Psy-
chology Compass, 14, Article e12531. https://doi.
org/10.1111/spc3.12531
Gómez, Á., López-Rodríguez, L., Sheikh, H., Ginges,
J., Wilson, L., Waziri, H., Vázquez, A., Davis, R.,
& Atran, S. (2017). The devoted actor’s will to
fight and the spiritual dimension of human con-
flict. Nature Human Behaviour, 1, 673–679. https://
doi.org/10.1038/s41562-017-0193-3
Gosling, S. D., Rentfrow, P. J., & Swann, W. B. (2003).
A very brief measure of the Big-Five person-
ality domains. Journal of Research in Personality,
37, 504–528. https://doi.org/10.1016/S0092-
6566(03)00046-1
Grahe, J. E., & Bernieri, F. J. (1999). The importance
of nonverbal cues in judging rapport. Journal
of Nonverbal Behavior, 23, 253–268. https://doi.
org/10.1023/A:1021698725361
Hasson, U., & Frith, C. D. (2016). Mirroring and
beyond: Coupled dynamics as a generalized
Lang et al. 19
framework for modelling social interactions.
Philosophical Transactions of the Royal Society B: Bio-
logical Sciences, 371, Article 20150366. https://doi.
org/10.1098/rstb.2015.0366
Hoegl, M., & Proserpio, L. (2004). Team member
proximity and teamwork in innovative pro-
jects. Research Policy, 33, 1153–1165. https://doi.
org/10.1016/j.respol.2004.06.005
Holland, R. W., Roeder, U. R., van Baaren, R. B.,
Brandt, A. C., & Hannover, B. (2004). Don’t
stand so close to me: The effects of self-construal
on interpersonal closeness. Psychological Science,
15, 237–242. https://doi.org/10.1111/j.0956-
7976.2004.00658.x
Jackson, J. C., Jong, J., Bilkey, D., Whitehouse, H.,
Zollmann, S., McNaughton, C., & Halberstadt,
J. (2018). Synchrony and physiological arousal
increase cohesion and cooperation in large natu-
ralistic groups. Scientific Reports, 8, Article 127.
https://doi.org/10.1038/s41598-017-18023-4
Karremans, J. C., & Verwijmeren, T. (2008). Mim-
icking attractive opposite-sex others: The role
of romantic relationship status. Personality and
Social Psychology Bulletin, 34, 939–950. https://doi.
org/10.1177/0146167208316693
Kim, T., Mcfee, E., Olguin, D. O., Waber, B., & Pent-
land, A. S. (2012). Sociometric badges: Using
sensor technology to capture new forms of col-
laboration. Journal of Organizational Behavior, 33,
412–427. https://doi.org/10.1002/job.1776
Lafrance, M., & Broadbent, M. (1976). Group rapport:
Posture sharing as a nonverbal indicator. Group
and Organization Studies, 1, 328–333. https://doi.
org/10.1177/105960117600100307
Lakin, J. L., & Chartrand, T. L. (2003). Using noncon-
scious behavioral mimicry to create affiliation and
rapport. Psychological Science, 14, 334–339. https://
doi.org/10.1111/1467-9280.14481
Lakin, J. L., Jefferis, V. E., Cheng, C. M., & Char-
trand, T. L. (2003). The chameleon effect as
social glue: Evidence for the evolutionary sig-
nificance of nonconscious mimicry. Journal of
Nonverbal Behavior, 27, 145–162. https://doi.
org/10.1023/A:1025389814290
Lang, M., Mitkidis, P., Kundt, R., Nichols, A.,
Krajčíková, L., & Xygalatas, D. (2016). Music as
a sacred cue? Effects of religious music on moral
behavior. Frontiers in Psychology, 7, Article 814.
https://doi.org/10.3389/fpsyg.2016.00814
Lockwood, P. L., Hamonet, M., Zhang, S. H., Ratnavel,
A., Salmony, F. U., Husain, M., & Apps, M. A. J.
(2017). Prosocial apathy for helping others when
effort is required. Nature Human Behaviour, 1, Article
0131. https://doi.org/10.1038/s41562-017-0131
Majolo, B., & Maréchal, L. (2017). Between-group
competition elicits within-group cooperation
in children. Scientific Reports, 7, Article 43277.
https://doi.org/10.1038/srep43277
McDonald, M. M., Navarrete, C. D., & van Vugt, M.
(2012). Evolution and the psychology of inter-
group conflict: The male warrior hypothesis. Philo-
sophical Transactions of the Royal Society B: Biological
Sciences, 367, 670–679. https://doi.org/10.1098/
rstb.2011.0301
Muthén, B. O. (1989). Latent variable modeling in het-
erogeneous populations. Psychometrika, 54, 557–
585. https://doi.org/10.1007/BF02296397
Newson, M., Bortolini, T., Buhrmester, M., da Silva,
S. R., da Aquino, J. N. Q., & Whitehouse, H.
(2018). Brazil’s football warriors: Social bonding
and inter-group violence. Evolution and Human
Behavior, 39, 675–683. https://doi.org/10.1016/j.
evolhumbehav.2018.06.010
Nichols, A. D., Lang, M., Kavanagh, C., Kundt, R.,
Yamada, J., Ariely, D., & Mitkidis, P. (2020).
Replicating and extending the effects of auditory
religious cues on dishonest behavior. PLoS ONE,
15, Article e0237007. https://doi.org/10.1371/
journal.pone.0237007
Olguin Olguin, D. (2011). Sensor-based organizational
design and engineering. [Doctoral thesis, Massachu-
setts Institute of Technology]. MIT Libraries.
http://hdl.handle.net/1721.1/67756
Purzycki, B. G., & Lang, M. (2019). Identity fusion,
outgroup relations, and sacrifice: A cross-
cultural test. Cognition, 186, 1–6. https://doi.
org/10.1016/j.cognition.2019.01.015
R Core Team. (2020). R: A language and environment for
statistical computing (Version 3.4.1) [Computer soft-
ware]. R Foundation for Statistical Computing.
https://cran.r-project.org/mirrors.html
Rutkowski, L., & Svetina, D. (2014). Assessing the
hypothesis of measurement invariance in the
context of large-scale international surveys. Edu-
cational and Psychological Measurement, 74, 31–57.
https://doi.org/10.1177/0013164413498257
Salas, E., Grossman, R., Hughes, A. M., & Coultas, C.
W. (2015). Measuring team cohesion: Observa-
tions from the science. Human Factors, 57, 365–374.
https://doi.org/10.1177/0018720815578267
Schachter, S. (1959). The psychology of affiliation: Experi-
mental studies of the sources of gregariousness. The
20 Group Processes & Intergroup Relations 00(0)
University of Chicago Press. https://doi.
org/10.2307/1419397
Sosis, R., Kress, H. C., & Boster, J. S. (2007). Scars for
war: Evaluating alternative signaling explanations
for cross-cultural variance in ritual costs. Evolu-
tion and Human Behavior, 28, 234–247. https://doi.
org/10.1016/j.evolhumbehav.2007.02.007
Stasser, G., & Titus, W. (2003). Hidden profiles: A brief
history. Psychological Inquiry, 14, 304–313. https://
doi.org/10.1080/1047840X.2003.9682897
Swann, W. B., Buhrmester, M. D., Gómez, A., Jetten,
J., Bastian, B., Vázquez, A., Ariyanto, A., Besta,
T., Christ, O., Cui, L., Finchilescu, G., González,
R., Goto, N., Hornsey, M., Sharma, S., Susianto,
H., & Zhang, A. (2014). What makes a group
worth dying for? Identity fusion fosters percep-
tion of familial ties, promoting self-sacrifice. Jour-
nal of Personality and Social Psychology, 106, 912–926.
https://doi.org/10.1037/a0036089
Swann, W. B., Gómez, A., Huici, C., Morales, J. F., &
Hixon, J. G. (2010). Identity fusion and self-sac-
rifice: Arousal as a catalyst of pro-group fighting,
dying, and helping behavior. Journal of Personality
and Social Psychology, 99, 824–841. https://doi.
org/10.1037/a0020014
Swann, W. B., Gómez, A., Seyle, D. C., Morales, J. F.,
& Huici, C. (2009). Identity fusion: The interplay
of personal and social identities in extreme group
behavior. Journal of Personality and Social Psychology,
96, 995–1011. https://doi.org/10.1037/a0013668
Talhelm, T., Zhang, X., & Oishi, S. (2018). Mov-
ing chairs in Starbucks: Observational studies
find rice–wheat cultural differences in daily life
in China. Science Advances, 4, Article eaap8469.
https://doi.org/10.1126/sciadv.aap8469
Taylor, S. E. (2006). Tend and befriend: Biobehavioral
bases of affiliation under stress. Current Directions
in Psychological Science, 15, 273–277. https://doi.
org/10.1111/j.1467-8721.2006.00451.x
Tickle-Degnen, L., & Rosenthal, R. (1990). The nature
of rapport and its nonverbal correlates. Psychologi-
cal Inquiry, 1, 285–293. https://doi.org/10.1207/
s15327965pli0104_1
Tracy, J. L., Randles, D., & Steckler, C. M. (2015). The
nonverbal communication of emotions. Current
Opinion in Behavioral Sciences, 3, 25–30. https://doi.
org/10.1016/j.cobeha.2015.01.001
Tripathi, P., & Burleson, W. (2012). Predicting crea-
tivity in the wild: Experience sample and socio-
metric modeling of teams. In Proceedings of the
ACM Conference on Computer Supported Cooperative
Work CSCW (pp. 1203–1212). ACM. https://doi.
org/10.1145/2145204.2145386
Van Baaren, R. B., Holland, R. W., Kawakami, K., & van
Knippenberg, A. (2004). Mimicry and prosocial
behavior. Psychological Science, 15, 71–74. https://
doi.org/10.1111/j.0963-7214.2004.01501012.x
Vandenberg, R. J., & Lance, C. E. (2000). A review
and synthesis of the measurement invariance
literature: Suggestions, practices, and recom-
mendations for organizational research. Organi-
zational Research Methods, 3, 4–69. https://doi.
org/10.1177/109442810031002
Van Vugt, M., De Cremer, D., & Janssen, D. P. (2007).
Gender differences in cooperation and competi-
tion. Psychological Science, 18, 19–23. https://doi.
org/10.1111/j.1467-9280.2007.01842.x
Voors, M., Nillesen, E., Verwimp, P., Bulte, E., Len-
sink, R., & van Soest, D. (2012). Violent conflict
and behavior: A field experiment in Burundi.
American Economic Review, 102, 941–964. https://
doi.org/10.1257/aer.102.2.941
Waber, B. N., Aral, S., Olguin Olguin, D., Wu, L.,
Brynjolfsson, E., & Pentland, A. (2011). Socio-
metric badges: A new tool for I.S. research.
SSRN Electronic Journal. https://doi.org/10.2139/
ssrn.1789103
Wann, D. L., Carlson, J. D., & Schrader, M. P. (1999).
The impact of team identification on the hos-
tile and instrumental verbal aggression of sport
spectators. Journal of Social Behavior and Personal-
ity, 14, 279–286. http://web.csulb.edu/~jmiles/
psy100/wann.pdf
Whitehouse, H. (2018). Dying for the group: Towards
a general theory of extreme self-sacrifice. Behavio-
ral and Brain Sciences, 41, Article E192. https://doi.
org/10.1017/S0140525X18000249
Whitehouse, H., Jong, J., Buhrmester, M. D., Gómez,
Á., Bastian, B., Kavanagh, C. M., Newson, M.,
Matthews, M., Lanman, J. A., McKay, R., &
Gavrilets, S. (2017). The evolution of extreme
cooperation via shared dysphoric experiences.
Scientific Reports, 7, Article 44292. https://doi.
org/10.1038/srep44292
Whitehouse, H., & Lanman, J. A. (2014). The ties that
bind us. Current Anthropology, 55, 1–23. https://
doi.org/10.1086/678698
Whitehouse, H., McQuinn, B., Buhrmester, M., &
Swann, B. W. (2014). Brothers in arms: Libyan revo-
lutionaries bond like family. Proceedings of the National
Academy of Sciences of the USA. Advance online publi-
cation. https://doi.org/10.1073/pnas.1416284111
Lang et al. 21
Xygalatas, D., Khan, S., Lang, M., Kundt, R., Kund-
tová Klocová, E., Krátký, J., & Shaver, J. H.
(2019). Effects of extreme ritual practices on
psychophysiological well-being. Current Anthro-
pology, 60, 699–707. https://doi.org/10.1086
/705665
Yuki, M., & Yokota, K. (2009). The primal warrior:
Outgroup threat priming enhances intergroup
discrimination in men but not women. Journal of
Experimental Social Psychology, 45, 271–274. https://
doi.org/10.1016/j.jesp.2008.08.018
Zhang, Y., Olenick, J., Chang, C.-H., Kozlowski, S., &
Hung, H. (2018). TeamSense: Assessing personal
affect and group cohesion in small teams through
dyadic interaction and behavior analysis with
wearable sensors. Proceedings of the ACM on Interac-
tive, Mobile, Wearable and Ubiquitous Technologies, 2,
Article 150. https://doi.org/10.1145/3264960
1
Supplementary Material
Outgroup Threat and the Emergence of Cohesive Groups: A Cross-Cultural Examination
Lang, M.1*, Xygalatas, D.2, Kavanagh, C.M,3,4, Boccardi, N.5, Halberstadt, J.6, Jackson, Ch. 6,
Martínez, M.7, Reddish, P.8, Tong, E. M. W. 8, Vazquez, A. 7, Whitehouse, H. 4, Yamamoto, M. E.
5, Yuki, M.9, & Gomez, A. 7,10
Affiliations:
1 LEVYNA: Laboratory for the Experimental Research of Religion, Masaryk University, Czech
Republic
2 Department of Anthropology, University of Connecticut, USA
3 College of Contemporary Psychology, Rikkyo University, Japan
4 Centre for the Study of Social Cohesion, Oxford University, United Kingdom
5 Department of Psychobiology, Federal University of Rio Grande do Norte, Brazil
6 Department of Psychology, University of Otago, New Zealand
7 Department of Psychology, Universidad Nacional de Educación a Distancia, UNED
8 Department of Psychology, National University of Singapore, Singapore
9 Department of Behavioral Science, Hokkaido University, Japan
10ARTIS International, Spain
*Correspondence to: martinlang@mail.muni.cz
2
Contents
1. Supplementary Methods ....................................................................................................................... 3
1.1. Behavioral Measures ..................................................................................................................... 3
1.2. Survey measures ........................................................................................................................... 6
1.3. Analysis......................................................................................................................................... 9
2. Site Descriptions ................................................................................................................................. 11
2.1 Brazil ........................................................................................................................................... 11
2.2 Japan ........................................................................................................................................... 12
2.3 Mauritius ..................................................................................................................................... 12
2.4 New Zealand ............................................................................................................................... 13
2.5 Singapore .................................................................................................................................... 14
2.6 Spain ........................................................................................................................................... 15
3. Supplementary Results ........................................................................................................................ 16
3.1 Manipulation Checks .................................................................................................................. 16
3.2 Mirroring ..................................................................................................................................... 18
3.3 Proximity..................................................................................................................................... 26
3.4 Activity ....................................................................................................................................... 33
3.5 Fight ............................................................................................................................................ 41
3.6 Costly Sacrifice ........................................................................................................................... 49
4. Questionnaire ...................................................................................................................................... 55
5. Primes ................................................................................................................................................. 65
5.1 No-threat prime ........................................................................................................................... 65
5.2 Outgroup threat prime ................................................................................................................. 67
5.3 Environmental threat prime ........................................................................................................ 69
6. Candidate Profiles ............................................................................................................................... 71
7. References ........................................................................................................................................... 76
3
1. Supplementary Methods
1.1. Behavioral Measures
Fig. S1. Laboratory set-up for the Hidden Profiles task. The participants’ task was to choose
three out of six candidates to represent their country at an upcoming conference. They were
instructed to write down reasons for supporting each candidate on a whiteboard, creating a poster
explaining their choice. Exp. = Experimenter presence in the room.
4
Fig. S2. An Illustration of Participants’ Activity during the Hidden Profiles Task. The first half of
the task is characterized by stationary discussion with bursts of synchronized activity, while the second
half is characterized by generally heightened levels of less-coupled activity.
Tab. S1. Means (SD) of Main Dependent Variables
Site
Mirroring
Proximity
Activity
Fight
Brazil
30.93 (1.59)
314.88 (113)
10.28 (2.63)
2.86 (1.15)
Japan
30.65 (0.88)
258.9 (100.55)
9.7 (3.29)
2.61 (0.83)
Mauritius
29.82 (3.05)
225.29 (103.88)
11.53 (4.48)
3.97 (1.01)
New Zealand
30.55 (1.75)
211.62 (75.74)
11.64 (2.63)
2.62 (0.95)
Singapore
24.9 (2.38)
236.5 (86.36)
6.91 (3.27)
3.85 (0.91)
Spain
27.32 (3.38)
350.1 (105.24)
10.32 (3.42)
2.24 (1.05)
Grand M/Total
29.15 (3.16)
269.17 (111.77)
10.22 (3.7)
3.03 (1.19)
5
Fig. S3. An Illustration of the Between-Subject Proximity Measurement during the Hidden Profiles
Task. A. Low-proximity group. B. High-proximity group. The proximity of an individual participant was
calculated by averaging their proximity to the other three session members (e.g., compare average
proximity of Participant 1 in Figure A and B).
6
1.2. Survey measures
The variable pertaining to willingness to fight for one’s country (Swann, Gómez, Huici, Morales,
& Hixon, 2010) was measured with a five-item scale. The configural invariance model revealed a
sufficient fit, although some of the fit indices were beyond the strict criteria we chose (CFI = 0.95,
TLI = 0.91, RMSEA = 0.10, and SRMR = 0.04; but see L. Rutkowski & Svetina, 2014 for an
argument for less stringent criteria). Constraining the factor loadings to be constant across sites,
the CFI of the metric invariance model was 0.91 (ΔCFI = -0.04) and constraining the intercepts in
the scalar invariance model decreased CFI to 0.31 (ΔCFI = -0.64). These results suggest that the
individual items may function differently across our sites in predicting the overall willingness to
fight score and that the average levels of this willingness substantially varied across our sites. To
improve the measurement invariance, we removed one item (“I would help others get revenge on
someone who insulted my country”), which improved the overall fit of the configural model (CFI
= 0.99, TLI = 0.97, RMSEA = 0.05, and SRMR = 0.02; see Tab. S3 for loadings and intercepts by
country) and the ΔCFI was positive (ΔCFImetric-configural = 0.01). However, comparing the scalar and
metric invariance of this new latent variable again revealed decrease in fit (ΔCFIscalar-metric = -0.81).
Further exploration of this measure revealed that Japan was driving most of the scalar variance
and removing this site from the MG-CFA decreased scalar variance (ΔCFIscalar-metric = -0.33),
although this variance remained beyond acceptable levels.
Assessing the configural invariance of the identity fusion measure revealed that the CFA
model with group structure did not fit well our data (CFI = 0.88, TLI = 0.82, RMSEA = 0.17, and
SRMR = 0.07), suggesting that the individual scale items may have not been understood/translated
equally well across our sites. To assure the feasibility of cross-cultural use of this scale, we
eliminated two items from the seven-item scale (“I will do for my country more than any of the
other group members would do”; “I make my country strong”). These two items had the lowest
mean of correlation coefficients across our sites (mean Pearson’s r = 0.32 and 0.30, respectively)
with the visual identity fusion scale used in previous cross-cultural research (Purzycki & Lang,
2019). The configural invariance model of this abbreviated scale showed a better fit to the data
(CFI = 0.95, TLI = 0.89, RMSEA = 0.15, and SRMR = 0.04; see Tab. S3 for loadings and intercepts
by country) and the ΔCFI between the metric and configural invariance models were within
acceptable boundaries (ΔCFImetric-configural = -0.01). However, constraining the item intercepts to be
7
constant across sites in the scalar invariance model revealed a substantial decrease in model fit,
suggesting between-site variability in fusion levels (ΔCFIscalar-metric = -0.18).
8
Tab. S3. Multi-Group Confirmatory Factor Analysis Accounting for Group Structure of Survey Data (Configural Invariance).
Willingness to fight
Identity fusion
Brazil
Japan
Mauritius
New
Zealand
Singapore
Spain
Brazil
Japan
Mauritius
New
Zealand
Singapore
Spain
Loading
F1
0.59
0.64
0.68
0.42
0.74
0.55
0.77
0.86
0.75
0.87
0.82
0.84
F2
0.56
0.74
0.69
0.72
0.55
0.82
0.62
0.79
0.82
0.83
0.90
0.81
F3
0.47
0.33
0.51
0.52
0.24
0.61
0.64
0.88
0.82
0.77
0.84
0.85
F4
0.47
0.42
0.52
0.37
0.46
0.55
0.54
0.63
0.73
0.83
0.58
0.72
F5
-
-
-
-
-
-
0.46
0.37
0.48
0.72
0.48
0.57
Intercepts
F1
2.46
1.77
2.89
1.83
2.54
1.92
1.98
2.64
3.96
3.26
5.17
1.85
F2
1.67
1.78
3.38
1.58
2.65
1.44
2.42
2.44
3.98
3.43
5.19
2.45
F3
1.13
3.06
2.06
1.71
2.47
1.37
2.47
2.46
3.58
3.05
4.78
2.32
F4
1.70
1.85
3.02
2.05
3.74
1.57
1.75
2.04
3.41
2.45
3.58
1.64
F5
-
-
-
-
-
-
1.66
2.10
3.14
2.21
3.71
1.61
CFI
0.99
0.95
TLI
0.97
0.89
RMSEA
0.05
0.15
SRMR
0.02
0.04
Note. Factor loadings and intercepts are standardized. F1-F5 refer to individual items of the two scales after excluding the problematic items (see Supplementary R
code for further information).
Tab. S2. Means (SD) of Variables Used as Simple Effects
Site
Fusion
Conflict
Disaster
Extraversion
Neuroticism
Conservatism
Brazil
3.34 (1.16)
4.01 (1.86)
3.43 (1.9)
5.11 (1.63)
3.46 (1.86)
2.89 (1.26)
Japan
3.22 (1.06)
5.21 (1.59)
6.55 (0.81)
3.69 (1.69)
3.13 (1.41)
4.01 (0.85)
Mauritius
4.76 (1.05)
3.89 (1.71)
5.03 (1.51)
5.19 (1.41)
3.73 (1.76)
4.83 (1.36)
New Zealand
4.27 (1.29)
3.57 (1.57)
6.33 (0.8)
4.77 (1.51)
4.38 (1.57)
3.15 (1.09)
Singapore
4.89 (0.88)
4.76 (1.50)
3.29 (1.63)
4.50 (1.52)
3.81 (1.49)
3.70 (1.16)
Spain
3.20 (1.31)
4.38 (1.90)
4.03 (1.8)
5.45 (1.49)
4.04 (1.74)
2.84 (1.24)
Grand M/Total
3.93 (1.35)
4.29 (1.79)
4.79 (1.95)
4.82 (1.64)
3.75 (1.70)
3.60 (1.39)
9
1.3. Analysis
As a first robustness check, we fit each variable with appropriate distribution and compared these
results with the results obtained from LMMs. Specifically, we used generalized linear mixed
models (GLMMs) to: A) fit the beta distribution to our measure of movement mirroring, which
should account for the specific distributional assumption of percentage data (Smithson &
Verkuilen, 2006); B) fit the gamma distribution to our measure of activity to account for the fact
that activity cannot assume negative values and is censored at 0 (Ng & Cribbie, 2017); and C) fit
the negative binomial models to the latent variables of willingness to fight and make costly
sacrifices. The usage of the negative binomial models in this case is warranted by the fact that our
data are bounded at 0 and can be summed as counts of individual variables. While a more
traditional approach to ordinal dependent variables would be to use the cumulative link mixed-
effects model (CLMM), our latent variables have more than 10 items and the usage of CLMM in
such a context would be impractical. Thus, we decided to relax the assumption that our data should
be discrete counts and compare the LMM results with results from negative binomial models.
The second robustness check involved letting the slopes of prime credibility, conflict
salience, natural disaster salience, and individual conservatism to vary by site, that is, including
these variables as random slopes at the site level. We suspected that some of these variables may
differentially affect the outcome variable across our sites and, therefore, we let them vary across
sites.
In the third robustness check, we built the same LMMs as in our main analyses but
excluded the New Zealand site. The rationale for this exclusion was the presence of multiple
nationalities in the sample recruited in New Zealand (46 participant out of 144 were not New
Zealanders), which may be problematic when manipulating in/outgroup items and measuring the
willingness to fight for a country. While we excluded participants who were not native speakers
of the dominant language at each site in all our analyses (assuming that language is a stronger
identifier of belonging to a nation than nationality), the higher number of non-nationals in New
Zealand compared to other sites may cause unexpected problems in the analyses of sociometric
data. Since the proximity and mirroring analyses always relied on data that emerge by interaction
between two badges, excluding only non-nationals from our analyses does not exclude their
interaction data that are part of dyadic interactions with other members of a particular session who
were nationals. Thus, we decided to perform our analyses without the New Zealand site in order
to minimize the impact that non-nationals may have had on our results.
The final robustness check for the sociometric results involves analysis of behavior during
free-interaction task that followed immediately after the Hidden Profiles task. That is, participants
waited together for five minutes in front of the experimental room and were free to interact with
each other. Rather than enforcing interaction schemas on participants (as in the Hidden Profiles
task), the free-interaction task was designed to assessed unconstrained spontaneous interaction
between our participants.
10
The LMMs were built using the command lme from the nlme package (Pinheiro et al.,
2014); the GLMMs with the glmmTMB command from the glmmTMB package (Brooks et al.,
2017). Model fit was assessed using the function simulateResiduals from the DHARMa package
(Hartig, 2019). This package is designed specifically to test model fits of LMMs and GLMMs,
taking into account the random-effects structure of mixed models and various distributional
assumptions of GLMMs, which are often misinterpreted when tested with model-fit procedures
designed for normally distributed residuals. Figures were plotted using the ggplot2 package
(Wickham, 2016), the ggridges package (Wilke, 2017) or Matlab, version 2013a (MathWorks Inc.,
2017).
11
2. Site Descriptions
In this section, we present descriptions of each experimental site with a specific focus on
environmental disaster, terrorist attacks, and similar country-level variables that may have
affected our main measures. Furthermore, we also detail data collection procedures and any
problems that occurred during testing.
2.1 Brazil
To date, Brazil has had only few environmental disasters such as earthquakes and hurricanes, with
hurricane Catarina in 2004 being the largest in recent history. The most common environmental
disasters are floods and flesh floods resulting from heavy rain, drought and water pollution. The
South, like Santa Catarina, generally suffers from flooding (Marcelino et al., 2006), but it also
affects other states like Rio de Janeiro, Pernambuco and Bahia, with many victims each year.
Droughts occur mainly in the Northeast, including Rio Grande do Norte. During data collection,
Brazil experienced the worst drought in decades (Marengo et al., 2017). In addition, during data
collection in November 2015, a significant environmental disaster occurred in Brazil, when a
mining dam broke, polluting water sources and causing enormous irreparable environmental,
economic, and social damage.
Regarding the terrorism threat, defined as the intentional use of the threat or violence
against civilians for political purposes not involving general crime (Ganor, 2002), Brazil has one
of the lowest global rates of terrorism (1.74 on a scale that varies up to 10). Brazil never suffered
attacks as seen in Europe, the Middle East or the USA. During the World Cup (2014) and the
Fig. S4. Map of our experimental sites.
12
Olympic Games (2016), Brazilian media speculated about the possibility of terrorist attacks, which
caused anxiety amongst some of the population, but no attacks were recorded. Although terrorism
was, at the data-collection period, an exogenous threat (Suarez, 2012), some Brazilian authorities
have worked to elaborate strategies and laws to combat it (Lasmar, 2015). However, Brazilians
generally do not worry about this type of violence inside the country.
The data were collected by the Laboratory of Human Behavior and Evolution (LECH) at
the Federal University of Rio Grande do Norte. Participants were recruited randomly from the
corridors and through online advertising. At the Brazilian site, we completed 36 sessions. Thirteen
sessions had to be transferred into another room, but were completed according to standard
specifications. In one session, a research assistant did not show up and the experiment was
administered by a researcher familiar with our hypotheses. Two sessions were discarded due to
having an insufficient number of participants.
2.2 Japan
In general, Japan is a very safe country and has not suffered any major terrorist attack in recent
decades. The most recent attack that reached national salience was the sarin gas attack on the
Tokyo Subway in the mid-90s by the New Religious Movement Aum Shinrikyo. However, the
experiment was conducted only a few months after the capture and eventual beheading of two
Japanese nationals by ISIS in Syria. These events were national news in Japan and made the topic
of terrorism, particularly by Islamic extremists, more salient. Moreover, Japan’s comparable low
levels of risk are not reflected in personal threat perception; hence, concern about potential terrorist
incidents amongst the public is significantly higher than that reported in the US (see Vosse, 2014
(Vosse, 2014)). In terms of environmental threats, in 2011, following the Tōhoku earthquake and
resultant tsunami, Japan suffered the Fukushima Daiichi nuclear disaster. This was national and
international news and dramatically increased concerns amongst the general population about the
potential danger of nuclear energy (Iwai & Shishido, 2015). After the incident Japan had shut down
all nuclear reactors but at the time of the study a limited number of plants were being reactivated.
As a result, there were prominent anti-nuclear demonstrations and a lot of media coverage related
to the issue.
All experimental sessions were conducted in the ‘Group Experiment’ laboratory facilities
provided by the Center for Experimental Research in Social Sciences (CERSS) at Hokkaido
University. Participants were all students of Hokkaido University recruited through online
advertisements sent to all those registered to the CERSS laboratory mailing list (approx. 3,000-
4,000 students). Advertisements provided only general study information and the estimated length
of the study. There was a loss of all Qualtrics data in one session due to the program’s
malfunctioning.
2.3 Mauritius
Mauritius is an isolated island in the Indian Ocean. Thanks to its geographical remoteness and a
smooth transition from colonialism to independence, it has no military and enjoys good relations
with other nations. The country has never witnessed any international terrorist incident. Moreover,
13
the numerous ethnic and religious groups that comprise its population co-exist peacefully, and
there have not been any violent acts of domestic terrorism in recent history (Xygalatas et al., 2018).
Thus, terrorism is not a major concern to the local population, although news coverage of
international terrorism does reach the island.
Natural disasters, on the other hand, are a bigger concern for Mauritians. The island lies on
the South-West Indian Ocean basin, which is regularly hit by severe tropical cyclones. In recent
years, a warning system has helped reduce fatalities, but significant devastation is often caused by
cyclones. In addition, torrential rain often causes floods that result in extensive damages and
casualties. In 2013, a flash flood in the capital of Port Louis claimed the lives of eleven people
(Lang et al., 2020).
Participants were randomly recruited on the University of Mauritius campus, due to the
absence of a participant pool. We invited participants to the lab at a specific time to eliminate
acquaintance among participants in the sessions; however, since we conducted the experiment
during school holidays, the availability of students was limited, and participants often knew each
other. Nevertheless, we controlled for acquaintance in our analyses. We ran the experiment in two
identical classrooms where tables and chairs grouped at one wall formed natural barriers between
participants, functioning as provisional cubicles. Furthermore, due to the absence of computers at
this field site, all materials were printed and responded to by writing/ticking responses. All
questionnaires were presented in the exact time intervals as in the Qualtrics version of our survey.
The transcription from paper sheets to excel sheets was checked by an independent auditor blind
to our hypotheses.
2.4 New Zealand
New Zealand is an island in the Pacific Ocean sharing maritime borders with American Samoa,
Australia, Fiji, French Polynesia, Samoa, and Tonga. Its unspoiled scenery and friendly people
mark it as a remote, but popular tourist destination. Notably, New Zealand has a strong
commitment to peace and peacekeeping--especially through the UN.
Terrorism in New Zealand is of little concern. Acts of terrorism are covered in the media,
but always in relation to other nations; most often presented as a problem for America and Europe.
A report from the New Zealand Security and Intelligence Service places the actual risk of a terrorist
attack as low. The report further stipulates that the government’s objective is that “New Zealand
should be neither the victim nor the source of an act of terrorism (Report of the New Zealand
Security Intelligence Service: Report to the House of Representatives for the Year Ended 30 June
2006, 2006).
Media coverage of natural disasters is more prominent. As a country on the Pacific Rim,
New Zealand has a history of earthquakes. Most recently a 6.3 magnitude earthquake in the city
of Christchurch (22 Feb. 2011; 185 dead, approx. 1750 injured) and a 7.3 magnitude earthquake
in Kaikoura (14 Nov. 2016; 2 dead, 57 injured). Both earthquakes claimed lives and both
dominated the media cycle during the study. Christchurch was a particularly notable natural
disaster and New Zealanders are still recovering physically and emotionally from the damage.
14
Participants were predominantly psychology undergraduates recruited at the University of
Otago in exchange for course credit from a standard participant pool and online advertising. To
our knowledge there are no significant deviations in protocol. With regard to the room layout
diagramed in Fig. S1, the camera was located in the Northeast (rather than the Southeast) corner
of the room. Visibility of the participants’ sessions was not obstructed by this change.
2.5 Singapore
Singapore is a small island nation nestled between the two large Muslim majority countries of
Indonesia and Malaysia. With the growth of radical Islamic terror groups in the region and in the
world, the government of Singapore has been increasingly concerned about the risk of terrorism
on home soil. This has culminated in the creation of ‘SG Secure’ – a Government program to
“sensitize, train and mobilize the community to play a part to prevent and deal with a terrorist
attack”. Although SG Secure was launched in September 2016, about a year after our experiment
was run, it is indicative of the growing fear of terrorism within the Government. The general public
is probably not as concerned as the government with terrorism, with Singapore being a very safe
and stable country to live. The only incident in Singapore that has been described as a terrorist
attack is a bombing that occurred in 1965. However, a few plots of terrorist attacks on Singapore
by terrorist groups such as by Jemaah Islamiyah have been discovered by Singapore authorities in
recent years which were high profile news in the local media. Fatal terrorist attacks have also
occurred in neighboring countries such as the Bali bombing of 2002.
Singapore was established as a trading city due in part to its sheltered position from natural
disasters. The nearest known fault line is 300km away in eastern Sumatra in Indonesia with the
risk from earthquakes and tsunamis being minimal. Indonesia and Malaysia protect Singapore
from severe storms. Singapore has been experiencing an annual ‘haze’ from forest burning in
Indonesia and Malaysia which is severe enough to adversely affect the health of the vulnerable.
The biggest concern from people and the government with regards to natural disasters is how such
disasters in other countries, particularly neighboring ones, may adversely affect the Singapore
economy.
Participants were recruited through an advertisement placed on a student learning website
accessible to all students, information flyers passed out to students, posters placed around campus,
and through class email lists. Compared to the standard protocol, participants did not sit in
individual cubicles as the lab was open plan. However, they could not readily see what was on
other participants’ screens and could only see each other through peripheral vision. In two sessions
the group discussion was briefly interrupted by students coming into the lab. Due to occasional
problems with the Qualtrics survey, some participants were given a hard copy version of the article
(noted in the data set). Two sessions were excluded from the analyses of the free-interaction task
because other students occupied the hall before the experimental room or participants left the
waiting area.
15
2.6 Spain
Spain is one of the European countries that has suffered most due to terrorist attacks. On March
11 2004, ten explosions on four trains killed 192 people. These bombings were, at the time, the
largest terrorist attack on European soil. A jihadist cell perpetrated the attack in revenge for the
Spanish participation in the war in Iraq. In August 2017 a series of terrorist attacks perpetrated by
the Islamic State took place in the cities of Barcelona and Cambrils (Catalonia) killing 16
people. These events were not the first jihadist attacks in Spain. In 1982 an attack in a restaurant
killed 18 people.
Besides jihadist terrorism, a socialist and separatist organization called Euskadi Ta Askatasuna
(ETA; Basque Country and Freedom) has killed at least 829 people (343 civilians) in numerous
attacks perpetrated from 1975 to 2011, when they announced a definitive cessation of armed
activity. While ETA was active, terrorism was a very prominent issue for Spaniards. In fact, in
September 2000 and in September 2006 domestic terrorism was the most important concern for
Spaniards (CIS, 2000, 2006 (“Barómetros. Numbers between Septembter 1985 and March 2017,”
2017)) ahead of other concerns as unemployment–a structural problem in Spain–and immigration.
Nowadays, terrorism is not as big of a concern for Spaniards according to the last survey of the
Spanish Centre of Sociological Research (“Barómetros. Numbers between Septembter 1985 and
March 2017,” 2017).
Natural disasters are also not a big concern for Spaniards, probably because few devastating
natural events have affected Spain during the last decades. Due to its geographical location,
hurricanes, tornadoes, earthquakes, volcanic eruptions, etc. are very rare and the effects of climate
change are less visible there than in other parts of the world. Most Spaniards (75%) believe that
climate change is happening now (Cartea et al., 2013), but only 8% of Spanish respondents in a
special Eurobarometer on Climate change (2014) (“Climate Change (Special Eurobarometer 409),
Comisión Europea,” 2014) consider it as the most serious problem facing the world–the second
lowest proportion following Portugal (6%). In fact, concerns about the environment have
decreased in Spain during the last decade.
Participants for this study were recruited from a student database at the National University
of Distance Education (UNED) or randomly from a corridor. The Spanish sample is exceptional
in its higher age compared to the other samples in this study: the UNED is a public university
allowing students to study from their homes and without a strict schedule, hence most of the
students are adults. We limited the age for participation to 35 to match the samples in other
countries and attempted to match similarly aged participants within each session. However, in
some instances, participants were older than 35 (see the ‘age’ variable in the Supplementary Data
Set).
16
3. Supplementary Results
In this section, we analyzed each of our dependent variables in more detail, supplementing the
results presented in the main text. Our modeling strategy comprised five steps, sequentially adding
control and moderator variables:
1) Basic models comprised main treatment effects, accounting for the nesting of participants
within sessions and sites.
2) We added control variables, holding constant the effects of sex; pre-treatment identity
fusion with participants’ country; international conflict salience; natural disaster salience
extraversion; neuroticism; and conservatism.
3) In the third step, we adjusted our models for perceived credibility of our priming material,
number of mistakes made during an attention check, and whether participants were
acquainted with someone from their session.
4) To investigate the male-warrior hypothesis (McDonald et al., 2012), we added a
Treatment*Sex interaction.
5) To investigate whether salience of international conflict would influence participants’
behavior differently in the outgroup threat condition, we added a Treatment*Conflict
interaction.
6) To investigate whether pre-treatment levels of identity fusion would influence participants’
behavior differently in the outgroup threat condition (Gómez et al., 2017), we added the
Treatment*Fusion interaction.
We first analyzed participants’ behavior during the Hidden Profiles task with LMMs, then
performed three robustness checks: performing the same analysis with GLMMs, on a sample
excluding the New Zealand site, and on data from the free-interaction task. The detailed results are
reported below.
3.1 Manipulation Checks
To check the internal validity of our measurements before conducting the planned analyses, we
first examined whether our threat manipulation was successful by asking participants how the
article made them feel. In both the outgroup and environmental threat conditions, participants
indicated feeling more threatened compared to the no-threat baseline condition (outgroup: β-
estimate = 1.49, 95% CI = [1.21 – 1.78]; environmental: β-estimate = 1.52, 95% CI = [1.23 – 1.81].
These estimates represent robust support for our manipulation, and also indicate that both
conditions elicited threat equally well. Importantly, the threat manipulation did not affect our
crucial moderator variable: the salience of international conflict. There was no difference between
conditions in conflict salience (outgroup threat: β-estimate = -0.16, 95% CI = [-0.46 – 0.15];
environmental threat: β-estimate = 0.10, 95% CI = [-0.21 – 0.41]), indicating that participants
responded to the former question with our manipulation in mind while the latter question was
answered more generally.
17
As a second manipulation check, we examined whether participants in the outgroup threat
condition were more likely to choose candidates with parochial profiles. On a five-point scale,
participants in the outgroup threat condition chose roughly 0.31 more conservative candidates
compared to the baseline condition (95% CI = [0.05 – 0.58]). The same effect was not observed in
the environmental threat condition (β-estimate = -0.02, 95% CI = [-0.28 – 0.24]), indicating that
only the prime with outgroup threat increased participants’ choice of candidates expressing more
suspicion or hostility toward foreigners. Using a cumulative link mixed model provided similar
results (see supplementary R code). Contrary to our expectations, participants in the outgroup
threat condition did not chose military candidates more often (β-estimate = -0.04, 95% CI = [-0.19
– 0.11]), although this might have been due to a weak manipulation of the military factor (e.g.,
three profile images wore military uniforms).
As a check of the importance of our moderator variables in group-conflict dynamics, we
examined whether they predicted willingness to fight, thus indicating their possibly important role
in moderating the effects of our treatment on mirroring, proximity, and activity in the outgroup
threat condition. First, on a scale from one to seven, males expressed 0.18 higher willingness to
fight compared to females (95% CI = [0.03 – 0.33]). While this difference appears small, it is the
result of generally low means and variance in the measure of willingness to fight, bordering with
floor effects at some sites (see Tab. S1). Salience of international conflict positively predicted
willingness to fight (β-estimate = 0.08, 95% CI = [0.04 – 0.12]) as did identity fusion with one’s
country, supporting previous findings (Gómez et al., 2011; Swann et al., 2009). For identity fusion,
the upper bound of the 95% CI was as high as 0.45, indicating that an increase in one point on the
fusion scale is associated with increase of roughly 0.4 on the dependent scale. These effects
remained robust after including all three predictors into one model, suggesting independent roles
for each of them in predicting willingness to fight for one’s country (see Tab. S4).
Tab. S4. Beta-Estimates with 95% CI for the Effects of Moderator Variables on
Willingness to Fight for One’s Country.
Willingness to Fight Models
(1)
(2)
(3)
(4)
Sex (0/1)
0.18*
0.26***
(0.03, 0.33)
(0.12, 0.40)
Conflict (1-7)
0.08***
0.06**
(0.04, 0.12)
(0.02, 0.09)
Fusion (1-7)
0.38***
0.38***
(0.32, 0.45)
(0.31, 0.44)
Constant
2.95***
2.70***
3.03***
2.67***
(2.38, 3.53)
(2.10, 3.30)
(2.46, 3.60)
(2.07, 3.28)
N Participants
824
823
824
823
Note. Ϯ p<.1; * p<.05; ** p<.01; *** p<.001
18
3.2 Mirroring
We measured movement mirroring between participants by analyzing their activity during the
Hidden Profiles and free interaction tasks using the in-built accelerometer in the Sociometric
Badges. Within a sliding 5-sec window, we assessed how much two participants mirrored each
other’s acceleration patterns when they were in proximity and facing each other. We hypothesized
that both threat conditions would bring participants closer together, corresponding to the predicted
increase in affiliative behaviors under threat.
LMMs with varying intercepts for groups and sites revealed a positive mirroring effect in
the environmental threat condition (β-estimate = 0.71; 95% CI = [0.07 – 0.35]), although after
adjusting our models for control variables this effect weakened (see Tab. S5). The 0.71 increase
in mirroring corresponds to an average 0.7% more mirroring detected for each 5 seconds of
proximal interaction. There was no main effect of outgroup threat on movement mirroring (β-
estimate = -0.06; 95% CI = [-0.69 – 0.58]). Investigating the possible reasons for the lack of the
main effect in the outgroup threat condition, we observed a significant interaction between the
outgroup threat and sex. Specifically, while females mirrored less each other in the outgroup threat
condition compared to the no-threat baseline condition (28.3% vs. 29%), males mirrored each
other around 0.65% more in the outgroup threat condition compared to the baseline condition,
reaching the environmental threat levels of mirroring (see Tab. S5 for specific estimates and Fig.
2 for a plot of this interaction effect). The opposite behaviors that the outgroup threat elicited
between males and females may account for the lack of the outgroup threat main effect in this
measure.
These results suggest that while the environmental threat increased affiliative behaviors for
both sexes, the outgroup threat increased affiliative behaviors only for males while decreased for
females. Furthermore, we observed similar positive coefficients for the Condition*Conflict
salience interaction as in the proximity models (see section 3.3). Participants who scored lowest
on the salience of international conflict measure were estimated to mirror each other less in the
outgroup threat condition compared to the baseline condition (28.4% vs. 29.2%), but this
difference was reversed for participants scoring the highest on the conflict salience scale (29% vs.
28.6%). In other words, we detected a difference between the slopes of conflict salience in the
outgroup threat and no-threat baseline conditions (βdifference = 0.20, 95% CI = [-0.01 – 0.40]) but
this slope difference was not detected for the comparison of the environmental threat and baseline
conditions (βdifference = 0.09, 95% CI = [-0.12 – 0.29]). There were no moderating effects of identity
fusion.
To test these results for their robustness, we first fit a GLMM to the mirroring data in order
to account for the fact that these data are percentages and, therefore, bounded on the interval 0-1.
Note that while we multiplied the mirroring data by 100 in the LMM for easier interpretation, the
GLMM analysis of percentage data requires the data on the interval 0-1 (i.e., the LMM and GLMM
coefficients are comparable when divided by 100). Looking at the density plot in the
supplementary R code (section 2.2.2), the data indeed suggest that the assumption of normality is
violated. This is further confirmed by a significant deviation from the normal distribution detected
19
by the Kolmogorov-Smirnov test (D = 0.16, p < .001). While the Kolmogorov-Smirnov test may
reach significant values with large samples (Hartig, 2019), the additional goodness-of-fit indices
indeed suggest that the LMM is misspecified. Thus, we opted for beta regression as a suitable
alternative for modeling percentage data (Smithson & Verkuilen, 2006), and this choice was
supported by a substantial decrease in AIC when compare to LMM (AICdif = 88.74, Δdf = 1).
However, in terms of qualitative differences from the results reported in Tab. S5, the beta
regression results did not differ from the LMM results and lend support for the same interpretation
(see Tab. S6).
In our second robustness check, we fit four models that varied the slopes of conflict
salience, natural disaster salience, conservatism, and credibility of our manipulation across sites.
However, none of these varying slopes substantially changed the detected effects (Tab. S7).
Furthermore, we also built LMMs while excluding the New Zealand site. As with the previous
robustness checks, the results did not qualitatively differ from our main LMMs (see Tab. S8). The
site-specific analysis of our manipulation revealed that the coefficients were in a similar direction
across our sites except for Singapore where the outgroup threat condition had substantially lower
rates of mirroring (see Tab. S9). Finally, during free interaction, there was a similar trend for the
interaction between sex and outgroup threat condition, however, these results were much more
variable than during the Hidden Profiles task and do not afford broader conclusions (see Tab. S10).
20
Tab. S5. Beta-Estimates from Linear Models with 95% CI for the Measure of Mirroring.
Mirroring: Hidden Profile Task
(1)
(2)
(3)
(4)
(5)
(6)
Treat: Outgroup
-0.06
-0.05
-0.15
-0.73x
-0.97x
-0.15
(-0.69, 0.58)
(-0.68, 0.58)
(-0.81, 0.51)
(-1.58, 0.12)
(-2.07, 0.13)
(-0.81, 0.51)
Treat: Enviro
0.71*
0.75*
0.63x
0.67
0.26
0.63x
(0.07, 1.35)
(0.12, 1.38)
(-0.03, 1.28)
(-0.18, 1.51)
(-0.84, 1.36)
(-0.02, 1.28)
Sex (0/1)
0.14
0.14
-0.34
0.15
0.15
(-0.38, 0.66)
(-0.41, 0.69)
(-1.32, 0.64)
(-0.40, 0.71)
(-0.40, 0.70)
Fusion (1-7)
-0.07
-0.09
-0.09
-0.09
-0.2
(-0.21, 0.08)
(-0.24, 0.07)
(-0.24, 0.07)
(-0.24, 0.07)
(-0.47, 0.07)
Conflict (1-7)
-0.004
0.003
0.01
-0.09
0.003
(-0.09, 0.09)
(-0.09, 0.10)
(-0.09, 0.10)
(-0.25, 0.06)
(-0.09, 0.10)
Natural disaster (1-7)
-0.004
-0.01
-0.01
-0.01
-0.01
(-0.11, 0.10)
(-0.11, 0.10)
(-0.11, 0.10)
(-0.11, 0.10)
(-0.12, 0.10)
Extraversion (1-7)
0.06
0.08
0.07
0.07
0.08
(-0.03, 0.16)
(-0.02, 0.17)
(-0.02, 0.17)
(-0.02, 0.17)
(-0.02, 0.17)
Neuroticism (1-7)
-0.03
-0.01
-0.01
-0.01
-0.01
(-0.11, 0.06)
(-0.10, 0.08)
(-0.11, 0.08)
(-0.10, 0.09)
(-0.11, 0.08)
Conservatism (1-7)
0.08
0.07
0.07
0.07
0.07
(-0.04, 0.19)
(-0.06, 0.19)
(-0.06, 0.19)
(-0.05, 0.19)
(-0.06, 0.19)
Prime credibility (1-9)
0.09x
0.08x
0.09x
0.09x
(-0.01, 0.18)
(-0.01, 0.18)
(-0.01, 0.19)
(-0.01, 0.18)
Mistakes (0-3)
0.26*
0.27*
0.26*
0.26*
(0.01, 0.51)
(0.02, 0.52)
(0.01, 0.51)
(0.01, 0.51)
Acquaintance (0-2)
-0.14
-0.14
-0.12
-0.14
(-0.38, 0.11)
(-0.39, 0.10)
(-0.37, 0.12)
(-0.38, 0.11)
Outgroup *Sex
1.38*
(0.06, 2.71)
Enviro*Sex
-0.02
(-1.32, 1.28)
Outgroup*Conflict
0.20x
(-0.01, 0.40)
Enviro*Conflict
0.09
(-0.12, 0.29)
Outgroup*Fusion
0.15
(-0.23, 0.52)
Enviro*Fusion
0.19
(-0.18, 0.55)
Constant
28.82***
28.30***
27.77***
27.96***
28.12***
27.77***
(26.87, 30.77)
(26.15, 30.46)
(25.53, 30.00)
(25.69, 30.23)
(25.83, 30.41)
(25.52, 30.01)
N Participants
761
741
691
691
691
691
Note. The baseline condition is the reference category for the treatment variable. Interactions compare the slopes of moderating
variables across the outgroup threat and environmental threat conditions with the baseline condition. Sex is a difference between
females and males. Mistakes indicate how many questions from attention check of our manipulation participants answered
incorrectly. Acquaintance indicates how well participants knew other members of their group. Outgroup = Outgroup threat; Enviro
= Environmental threat.
x p<.1; * p<.05; ** p<.01; *** p<.001
21
Tab. S6. Beta-Estimates from Beta Regression Models with 95% CI for the Measure of Mirroring. Coefficients are transformed
using the logit link.
Mirroring: Hidden Profile Task
(1)
(2)
(3)
(4)
(5)
(6)
Treat: Outgroup
-0.004
-0.003
-0.01
-0.04x
-0.05x
-0.01
(-0.04, 0.03)
(-0.03, 0.03)
(-0.04, 0.02)
(-0.08, 0.004)
(-0.10, 0.005)
(-0.04, 0.02)
Treat: Enviro
0.04*
0.04*
0.03x
0.03
0.01
0.03x
(0.003, 0.07)
(0.01, 0.07)
(-0.002, 0.06)
(-0.01, 0.07)
(-0.04, 0.07)
(-0.001, 0.06)
Sex (0/1)
0.01
0.01
-0.02
0.01
0.01
(-0.02, 0.03)
(-0.02, 0.03)
(-0.07, 0.03)
(-0.02, 0.04)
(-0.02, 0.04)
Fusion (1-7)
-0.003
-0.004
-0.004
-0.004
-0.01
(-0.01, 0.004)
(-0.01, 0.004)
(-0.01, 0.004)
(-0.01, 0.004)
(-0.02, 0.004)
Conflict (1-7)
0
0
0
-0.005
0
(-0.005, 0.004)
(-0.005, 0.005)
(-0.004, 0.005)
(-0.01, 0.003)
(-0.005, 0.005)
Natural disaster (1-7)
0
0
0
0
-0.001
(-0.01, 0.005)
(-0.01, 0.005)
(-0.01, 0.005)
(-0.01, 0.005)
(-0.01, 0.005)
Extraversion (1-7)
0.003
0.004
0.004
0.004
0.004
(-0.001, 0.01)
(-0.001, 0.01)
(-0.001, 0.01)
(-0.001, 0.01)
(-0.001, 0.01)
Neuroticism (1-7)
-0.001
-0.001
-0.001
0
-0.001
(-0.01, 0.003)
(-0.005, 0.004)
(-0.005, 0.004)
(-0.005, 0.004)
(-0.005, 0.004)
Conservatism (1-7)
0.004
0.003
0.003
0.004
0.003
(-0.002, 0.01)
(-0.003, 0.01)
(-0.003, 0.01)
(-0.003, 0.01)
(-0.003, 0.01)
Prime credibility (1-9)
0.004x
0.004x
0.005x
0.004x
(-0.001, 0.01)
(-0.001, 0.01)
(0.00, 0.01)
(-0.001, 0.01)
Mistakes (0-3)
0.01*
0.01*
0.01*
0.01*
(0.00, 0.03)
(0.001, 0.03)
(0.00, 0.03)
(0.00, 0.03)
Acquaintance (0-2)
-0.01
-0.01
-0.01
-0.01
(-0.02, 0.01)
(-0.02, 0.01)
(-0.02, 0.01)
(-0.02, 0.01)
Outgroup *Sex
0.07*
(0.004, 0.13)
Enviro*Sex
0
(-0.06, 0.06)
Outgroup*Conflict
0.01x
(-0.001, 0.02)
Enviro*Conflict
0.004
(-0.01, 0.02)
Outgroup*Fusion
0.01
(-0.01, 0.03)
Enviro*Fusion
0.01
(-0.01, 0.03)
Constant
-0.91***
-0.93***
-0.96***
-0.95***
-0.94***
-0.96***
(-1.00, -0.82)
(-1.03, -0.83)
(-1.07, -0.86)
(-1.06, -0.84)
(-1.05, -0.84)
(-1.07, -0.86)
N Participants
761
741
691
691
691
691
Note. For the beta regression model, mirroring is on a scale from 0-1, as opposed to 0-100 as in the other models. The baseline
condition is the reference category for the treatment variable. Interactions compare the slopes of moderating variables across the
outgroup threat and environmental threat conditions with the baseline condition. Sex is a difference between females and males.
Mistakes indicate how many questions from attention check of our manipulation participants answered incorrectly. Acquaintance
indicates how well participants knew other members of their group. Outgroup = Outgroup threat; Enviro = Environmental threat.
x p<.1; * p<.05; ** p<.01; *** p<.001
22
Tab. S7. Beta-Estimates from Linear Models with 95% CI for the Measure of Mirroring. Each
model varies the effect of a different variable across sites (see column names).
Mirroring: Hidden Profile Task
(CONFLICT)
(NATURAL)
(CONSERV)
(CREDIBLE)
Treat: Outgroup
-0.15
-0.15
-0.15
-0.15
(-0.81, 0.50)
(-0.16, -0.15)
(-0.81, 0.50)
(-0.81, 0.50)
Treat: Enviro
0.63x
0.63
0.63x
0.63x
(-0.02, 1.27)
(0.62, 0.63)
(-0.02, 1.27)
(-0.02, 1.27)
Sex (0/1)
0.14
0.14
0.14
0.14
(-0.40, 0.69)
(0.14, 0.15)
(-0.40, 0.69)
(-0.40, 0.69)
Fusion (1-7)
-0.08
-0.08
-0.08
-0.08
(-0.24, 0.07)
(-0.09, -0.08)
(-0.24, 0.07)
(-0.24, 0.07)
Conflict (1-7)
0.002
0.002
0.002
0.002
(-0.09, 0.10)
(0.001, 0.003)
(-0.09, 0.10)
(-0.09, 0.10)
Natural disaster (1-7)
-0.005
-0.005
-0.005
-0.005
(-0.11, 0.10)
(-0.11, 0.10)
(-0.01, -0.004)
(-0.11, 0.10)
Extraversion (1-7)
0.08
0.08
0.08
0.08
(-0.02, 0.17)
(0.08, 0.08)
(-0.02, 0.17)
(-0.02, 0.17)
Neuroticism (1-7)
-0.01
-0.01
-0.01
-0.01
(-0.10, 0.08)
(-0.01, -0.01)
(-0.10, 0.08)
(-0.10, 0.08)
Conservatism (1-7)
0.07
0.07
0.07
0.07
(-0.06, 0.19)
(0.06, 0.07)
(-0.06, 0.19)
(-0.06, 0.19)
Prime credibility (1-9)
0.08x
0.08
0.08x
0.08x
(-0.01, 0.18)
(0.08, 0.09)
(-0.01, 0.18)
(-0.01, 0.18)
Mistakes (0-3)
0.26*
0.26
0.26*
0.26*
(0.01, 0.51)
(0.26, 0.26)
(0.01, 0.51)
(0.01, 0.51)
Acquaintance (0-2)
-0.14
-0.14
-0.14
-0.14
(-0.38, 0.11)
(-0.14, -0.13)
(-0.38, 0.11)
(-0.38, 0.11)
Constant
27.76***
27.76
27.76***
27.76***
(25.67, 29.85)
(27.74, 27.78)
(25.67, 29.85)
(25.67, 29.85)
N Participants
692
692
692
692
µint Session
2.62
2.62
2.62
2.62
µint Site
4.65
4.65
4.65
4.65
µslope
0
0
0
0
Resid var
2.98
2.98
2.98
2.98
Note. For the beta regression model, mirroring is on a scale from 0-1, as opposed to 0-100 as in
the other models. The baseline condition is the reference category for the treatment variable.
Sex is a difference between females and males. Mistakes indicate how many questions from
attention check of our manipulation participants answered incorrectly. Acquaintance indicates
how well participants knew other members of their group. Outgroup = Outgroup threat; Enviro
= Environmental threat. CONFLICT is salience of international conflict; NATURAL is
salience of natural disaster threat; CONSERV is individual conservatism; CREDIBLE is the
credibility of our manipulation. µint Session is the variance explained by varying intercepts by
session id; µint Site is the variance explained by varying intercepts by site; µslope is the variance
explained by varying the slopes of particular variables (in columns) by site. Note that we
rounded the estimated varying slopes to three decimal places. Resid var is the residual variance
after fitting varying intercepts for sessions and sites and varying slopes by sites.
x p<.1; * p<.05; ** p<.01; *** p<.001
23
Tab. S8. Beta-Estimates from Linear Models with 95% CI for the Measure of Mirroring. New Zealand Excluded.
Mirroring: Hidden Profile Task
(1)
(2)
(3)
(4)
(5)
(6)
Treat: Outgroup
-0.1
-0.07
-0.19
-0.89x
-1.14x
-0.18
(-0.82, 0.61)
(-0.78, 0.63)
(-0.93, 0.55)
(-1.87, 0.08)
(-2.38, 0.09)
(-0.92, 0.56)
Treat: Enviro
0.66x
0.73*
0.57
0.62
0.11
0.58
(-0.06, 1.38)
(0.02, 1.44)
(-0.17, 1.31)
(-0.35, 1.59)
(-1.17, 1.38)
(-0.15, 1.32)
Sex (0/1)
0.04
0.04
-0.48
0.05
0.06
(-0.54, 0.62)
(-0.57, 0.65)
(-1.55, 0.59)
(-0.56, 0.67)
(-0.55, 0.67)
Fusion (1-7)
-0.08
-0.11
-0.1
-0.11
-0.31x
(-0.24, 0.09)
(-0.28, 0.07)
(-0.28, 0.08)
(-0.28, 0.07)
(-0.62, 0.005)
Conflict (1-7)
-0.02
-0.01
-0.005
-0.12
-0.01
(-0.12, 0.08)
(-0.11, 0.10)
(-0.11, 0.10)
(-0.30, 0.05)
(-0.11, 0.10)
Natural disaster (1-7)
-0.003
-0.005
-0.004
-0.01
-0.01
(-0.12, 0.11)
(-0.12, 0.11)
(-0.12, 0.11)
(-0.12, 0.11)
(-0.13, 0.10)
Extraversion (1-7)
0.06
0.08
0.07
0.07
0.08
(-0.04, 0.17)
(-0.03, 0.18)
(-0.04, 0.18)
(-0.04, 0.18)
(-0.03, 0.19)
Neuroticism (1-7)
-0.03
-0.01
-0.02
-0.01
-0.02
(-0.13, 0.07)
(-0.12, 0.09)
(-0.12, 0.09)
(-0.12, 0.10)
(-0.13, 0.09)
Conservatism (1-7)
0.06
0.05
0.05
0.05
0.06
(-0.07, 0.20)
(-0.09, 0.19)
(-0.09, 0.19)
(-0.09, 0.19)
(-0.08, 0.20)
Prime credibility (1-9)
0.10x
0.10x
0.11*
0.10x
(-0.004, 0.21)
(-0.01, 0.21)
(0.001, 0.22)
(-0.005, 0.21)
Mistakes (0-3)
0.33*
0.34*
0.33*
0.33*
(0.04, 0.61)
(0.06, 0.63)
(0.04, 0.61)
(0.05, 0.62)
Acquaintance (0-2)
-0.18
-0.18
-0.17
-0.18
(-0.46, 0.09)
(-0.45, 0.10)
(-0.44, 0.11)
(-0.46, 0.10)
Outgroup *Sex
1.58*
(0.12, 3.04)
Enviro*Sex
-0.1
(-1.54, 1.34)
Outgroup*Conflict
0.22x
(-0.01, 0.45)
Enviro*Conflict
0.11
(-0.13, 0.34)
Outgroup*Fusion
0.2
(-0.22, 0.62)
Enviro*Fusion
0.37x
(-0.05, 0.79)
Constant
28.56***
28.17***
27.55***
27.78***
28.00***
27.54***
(26.28, 30.84)
(25.67, 30.67)
(24.94, 30.15)
(25.13, 30.43)
(25.33, 30.67)
(24.93, 30.15)
N Participants
655
635
587
587
587
587
Note. The baseline condition is the reference category for the treatment variable. Interactions compare the slopes of moderating
variables across the outgroup threat and environmental threat conditions with the baseline condition. Sex is a difference between
females and males. Mistakes indicate how many questions from attention check of our manipulation participants answered
incorrectly. Acquaintance indicates how well participants knew other members of their group. Outgroup = Outgroup threat; Enviro =
Environmental threat.
x p<.1; * p<.05; ** p<.01; *** p<.001
24
Tab. S9. Beta-Estimates from Linear Models with 95% CI for the Measure of Mirroring. Sites-specific models.
Mirroring: Hidden Profile Task
(Brazil)
(Japan)
(Mauritius)
(New Zeal.)
(Singapore)
(Spain)
Treat: Outgroup
0.85
0.2
0.09
-0.16
-1.53x
0.03
(-0.46, 2.16)
(-0.28, 0.69)
(-1.86, 2.05)
(-1.57, 1.25)
(-3.13, 0.08)
(-1.69, 1.74)
Treat: Enviro
0.45
0.65*
0.31
0.72
-0.05
2.08*
(-0.84, 1.75)
(0.15, 1.14)
(-1.68, 2.29)
(-0.65, 2.09)
(-1.66, 1.56)
(0.34, 3.82)
Sex (0/1)
-0.19
-0.2
1.27
0.86
-0.79
-0.16
(-1.05, 0.66)
(-0.59, 0.19)
(-0.33, 2.87)
(-0.37, 2.09)
(-2.41, 0.82)
(-1.63, 1.31)
Fusion (1-7)
-0.06
0.03
0.05
0.02
0.07
-0.35
(-0.20, 0.07)
(-0.12, 0.18)
(-0.28, 0.38)
(-0.25, 0.29)
(-0.36, 0.49)
(-0.90, 0.21)
Conflict (1-7)
-0.08x
0.04
-0.14
0.08
0.08
0.09
(-0.16, 0.00)
(-0.05, 0.13)
(-0.35, 0.07)
(-0.08, 0.24)
(-0.24, 0.39)
(-0.24, 0.42)
Natural disaster (1-7)
0.02
0.08
0.15
-0.02
-0.11
-0.08
(-0.06, 0.09)
(-0.10, 0.25)
(-0.08, 0.38)
(-0.41, 0.36)
(-0.40, 0.17)
(-0.44, 0.27)
Extraversion (1-7)
0.04
-0.01
-0.05
0.06
0.1
0.24
(-0.04, 0.11)
(-0.09, 0.07)
(-0.28, 0.18)
(-0.10, 0.22)
(-0.20, 0.40)
(-0.12, 0.61)
Neuroticism (1-7)
0.004
-0.01
0.08
0.004
-0.40**
-0.06
(-0.07, 0.08)
(-0.11, 0.10)
(-0.12, 0.29)
(-0.15, 0.16)
(-0.67, -0.14)
(-0.39, 0.26)
Conservatism (1-7)
0.05
0.01
0.14
0.17
-0.12
0.18
(-0.05, 0.16)
(-0.16, 0.18)
(-0.09, 0.37)
(-0.05, 0.38)
(-0.47, 0.24)
(-0.27, 0.63)
Constant
30.50***
29.79***
28.21***
29.05***
27.16***
24.98***
(29.35, 31.64)
(28.08, 31.49)
(25.20, 31.22)
(26.21, 31.89)
(23.92, 30.41)
(21.58, 28.38)
N Participants
122
128
143
106
96
146
Note. The baseline condition is the reference category for the treatment variable. Interactions compare the slopes of moderating
variables across the outgroup threat and environmental threat conditions with the baseline condition. Sex is a difference between
females and males. Mistakes indicate how many questions from attention check of our manipulation participants answered
incorrectly. Acquaintance indicates how well participants knew other members of their group. Outgroup = Outgroup threat; Enviro
= Environmental threat.
x p<.1; * p<.05; ** p<.01; *** p<.001
25
Tab. S10. Beta-Estimates from Linear Models with 95% CI for the Measure of Mirroring during Free Interaction.
Mirroring: Free Interaction