ArticlePDF Available

Who Is This "We"? Levels of Collective Identity and Self Representations

Authors:

Abstract

Cross-cultural perspectives have brought renewed interest in the social aspects of the self and the extent to which individuals define themselves in terms of their relationships to others and to social groups. This article provides a conceptual review of research and theory of the social self, arguing that the personal, relational, and collective levels of self-definition represent distinct forms of self-representation with different origins, sources of self-worth, and social motivations. A set of 3 experiments illustrates how priming of the interpersonal or collective "we" can alter spontaneous judgments of similarity and self-descriptions. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Journal of Personality
and
Social Psychology
1996,
Vol.
71, No.
1,83-93
Copyright 1996
by
Ihe American Psychological Association,
Inc.
0022-35H/96/S3.00
Who Is This "We"? Levels of Collective Identity and Self Representations
Marilynn
B.
Brewer and Wendi Gardner
Ohio State University
Cross-cultural perspectives have brought renewed interest
in the
social aspects
of
the self
and
the
extent
to
which individuals define themselves
in
terms of their relationships
to
others and
to
social
groups. This article provides
a
conceptual review
of
research
and
theory of the social
self,
arguing
that
the
personal, relational,
and
collective levels
of
self-definition represent distinct forms
of self-
representation with different origins, sources of self-worth, and social motivations.
A
set
of
3
exper-
iments illustrates haw priming
of
the interpersonal
or
collective "we"
can
alter spontaneous judg-
ments of similarity and self-descriptions.
Until recently, social psychological theories of the self focused
on
the
individuated self-concept—the person's sense
of
unique
identity differentiated from others. Cross-cultural perspectives,
however,
have
brought
a
renewed interest in the social aspects
of
the self and the extent to which individuals define themselves in
terms
of
their relationships
to
others
and to
social groups
(Markus
&
Kitayama, 1991; Triandis, Bontempo, Villareal,
Asai,
&
Lucca, 1988). Central to
this new
perspective
is the
idea
that connectedness and belonging are
not
merely affiliations
or
alliances between
the
self
and
others
but
entail fundamental
differences
in
the way the self is construed (Brewer,
1991;
Mar-
kus
&
Kitayama, 1991; Singelis, 1994; Trafimow, Triandis,
&
Goto,
1991;
Triandis,
1989;
Turner, Oakes, Haslam,
&
McGarty, 1994).
Some
of
these theories
of
the social self focus
on
cross-cul-
tural differences in whether the self is typically construed as
in-
dividuated
or
interpersonal. However,
all
recognize that these
different self-construals may also coexist within
the
same indi-
vidual, available to be activated
at
different times or in different
contexts. Furthermore,
in
several theories, achieving
an ex-
tended sense of self has the status of a fundamental human mo-
tivation (Baumeister
&
Leary, 1995; Brewer, 1991).
In
other
words, individuals seek
to
define themselves
in
terms
of
their
immersion
in
relationships with others
and
with larger collec-
tives
and
derive much
of
their self-evaluation from such social
identities (Breckler
&
Greenwald,
1986;
Greenwald
&
Breckler,
1985).
The
motivational properties
of
collective identities
are
systematically documented
in
Baumeister
and
Leary's (1995)
comprehensive review
of the
evidence
in
support
of a
funda-
mental "need to belong" as an innate feature of human nature.
All
of
the theories mentioned above draw some kind
of
dis-
tinction between
the
individuated or personal self (those aspects
of the self-concept that differentiate the self from all others) and
a relational
or
social
self (those aspects of the self-concept that
Marilynn
B.
Brewer and Wendi Gardner, Department of Psychology,
Ohio State University.
Correspondence concerning this article should be addressed to Mari-
lynn
B.
Brewer, Department of Psychology, Ohio State University,
1885
Neil
Avenue,
Columbus,
Ohio
43210-1222.
Electronic mail
may be
sent
via the Internet to mbbrewer@magnus.acs.ohio-state.edu.
reflect assimilation to others or significant
social
groups).
How-
ever, implicit in
a
comparison across
these
different theories
is
a
further distinction between
two
levels
of
social selves—those
that derive from interpersonal relationships
and
interdepen-
dence with specific others and those that derive from member-
ship in larger, more impersonal collectives or social categories.
Both interpersonal
and
collective identities
are
social exten-
sions of the self but differ
in
whether the social connections
are
personalized bonds of attachment or impersonal bonds derived
from common identification with some symbolic group
or so-
cial category. Prototypic interpersonal identities
are
those
de-
rived from intimate dyadic relationships such as parent-child,
lovers,
and
friendships,
but
they also include identities derived
from membership
in
small, face-to-face groups that
are
essen-
tially networks
of
such dyadic relationships. Collective social
identities,
on the
other hand,
do not
require personal relation-
ships among group members.
As
Turner, Hogg, Oakes, Reicher,
and Wetherell (1987) put
it,
social identity entails
a
depersonal-
ized sense
of
self,
"a
shift towards
the
perception
of
self
as
an
interchangeable exemplar
of
some social category
and
away
from the perception of self as a unique person" (p. 50). Consis-
tent with this view, Prentice, Miller,
and
Lightdale (1994) dis-
tinguished between group identities that
are
based
on
common
bonds (attachment
to
other group members)
and
those based
on
common identity
{collective
identities).
The distinction between interpersonal
and
collective identi-
ties is
not simply
a
matter of
the
difference between attachments
that
are
based
on
affect
and
attachments that
are
cognitively
based. Both levels involve affective and cognitive categorization
processes.
The
difference
is a
matter
of
level
of
inclusiveness.
Some social identities
can be
construed either
as
interpersonal
relationships
or as
collective identities. Many social roles
and
professions, for
instance,
can
be
experienced
in
terms of specific
role relationships (e.g., parent-child, doctor-patient)
or in
terms of membership
in
a general social category (e.g., parents,
medical professionals). Consistent with this reasoning, Mill-
ward (1995) recently demonstrated
a
distinction between
nurses who construed their career identity
in
terms of commu-
nal-interpersonal relationships with patients
and
those whose
representation
of
nurse
was construed
in
terms
of
professional
intergroup distinctions.
Some other theorists have also made explicit
the
distinction
83
84BREWER AND GARDNER
between interpersonal and collective
selves.
Triandis (1989) and
Greenwald and Breckler
(1985;
Breckler
&
Greenwald, 1986),
for
instance,
distinguished among
"private,"
"public,"
and "col-
lective" facets of the
self.
The public
sW/represents
those
aspects
of the self-concept most
sensitive
to the evaluation of significant
others and consists of cognitions about the self that reflect in-
teractions and relationships with those others. The
collective
self,
on the other hand, reflects internahzations of the norms
and characteristics of important reference groups and consists
of cognitions about the self that are consistent with that group
identification.
Recent evolutionary models of human social behavior also
call attention to functional distinctions between social attach-
ments at different levels of organization. Caporael (1995;
Brewer & Caporael, 1995), for instance, has developed a hier-
archical model of group structure as
a
comprehensive theory of
social coordination. According
to
this
model,
four fundamental
configurations dyads (two-person relationships), teams
(small face-to-face social and working groups), bands (small,
interacting communities), and
tribes
(macro-bands character-
ized by shared identity and communication but without contin-
ual face-to-face interaction)—have been "repeatedly assem-
bled" throughout human evolutionary history. Each level rep-
resents different forms of functional interdependence and
different types of coordination, with associated differences in
construals of self and
others.
These configurations also are rep-
resented ontogenetically. Bugental's (1995) review of the litera-
ture on social development suggests that the development of at-
tachment relationships and group-oriented relationships repre-
sent functionally distinct domains of social competence.
Levels of Self Representation
Table 1 presents one attempt to characterize systematically
the differences among the three levels of self-construal that are
represented in the current literature on the social
self.
At the
individual
level,
the
personal
self
is
the differentiated, individu-
ated self-concept most characteristic of studies of the self in
Western psychology (e.g., Pelham, 1993). At the interpersonal
level,
the
relational
self
is the
self-concept derived from connec-
tions and role relationships with significant others. This corre-
sponds most closely to the interdependent self as defined by
Markus and ICitayama (1991) in their analysis of the difference
between American and Japanese self-construals. Finally, at the
group level is the
collective
self,
which corresponds to the con-
cept of social identity as represented in social identity theory
and self-categorization theory (Hogg & Abrams, 1988; Turner
etal., 1987).
These different aspects of
the
self refer to different levels of
inclusiveness of the conceptualization of the self—the shift
from
"1"
to "we" as the locus of self-definition. This shift in
inclusiveness of self-representations is postulated to be associ-
ated with corresponding transformations of the bases for
content of the self-concept, the frame of reference for evalua-
tions of self-worth, and the nature of social motivation.
The
Extended
Self-Concept
The idea of
the
socially extended self
goes
beyond perceived
similarity and other relational connections between self and
others. An extended self means that the boundaries of the self
are redrawn, and the content of the self-concept is focused on
those characteristics that make one a "good" representative of
the group or of the relationship. As Brewer (1991) put it, when
collective identities are activated, the most salient features of
the self-concept become those that are shared with other mem-
bers of the in-group. The idea that close relationships involve a
blurring of the boundaries between the self and
a
partner
also
is
represented by Aron and Aron's (1986) self-expansion model
of motivation and cognition in close relationships. Operation-
ally,
this
concept
is
captured in the Inclusion of Other in the Self
Scale (Aron, Aron,
&
Smollan, 1992), which has recently been
extended to collective identities (Tropp
&
Wright, 1995).
Support for the idea that salient interpersonal relationships
are incorporated into the self-concept was obtained in a series
of studies by Aron, Aron, Tudor, and Nelson (1991) on the par-
allels between cognitive effects of self-referencing and referenc-
ing to close relationship partners. The method and results of
Aron et al.'s Experiment 3 are particularly relevant In this
study, married graduate students completed a questionnaire in
which they rated themselves and their spouses on a set of 90
diverse traits. The ratings were used to identify aspects of each
individual's self-concept for which self-ratings matched ratings
of their partner and those for which self-ratings and partner rat-
ings were mismatched. Later, the same individuals made yes-
no self-descriptiveness judgments on the same 90 traits on a
computer with reaction times recorded. Mean reaction times
for matching traits were significantly faster than those for mis-
matched
traits,
suggesting that shared characteristics
were
more
salient or accessible aspects of the self-concept.
Smith and Henry (1996) adapted Aron et al.'s (1991)
method to assess the influence of salient in-group characteris-
tics on judgments of the
self.
College student participants were
asked to describe themselves and then each of
two
groups (an
in-group based on college major or fraternity, and
a
correspond-
ing
out-group)
on
the
90
traits used by Aron et
al.
The research-
Table 1
Levels
of
Representation
of
the Self
Level of analysis
Individual
Interpersonal
Group
Self-concept
Personal
Relational
Collective
Basis of
self-evaluation
Traits
Roles
Group prototype
Frame of
reference
Interpersonal comparison
Reflection
Intergroup comparison
Basic social
motivation
Self-interest
Other's benefit
Collective welfare
LEVELS
OF
IDENTITY85
ers,
having made these group identities
salient,
then had respon-
dents make yes-no judgments
of
these same traits
on a com-
puter. Analyses
of
response time data essentially replicated
Aron
et al.'s
earlier findings
for
spouses. Response times were
facilitated
for
self-descriptive traits that matched those
of
the
relevant in-group
and
were slower
for
mismatching traits.
For
both levels of identity, then, there is evidence that identification
with others enhances
the
accessibility
of
shared characteristics
in the working self-concept.
Self-Concept
and Social
Comparison
Most theories of personal self-esteem assume that global
self-
worth
at the
individual level
is
derived from self-evaluation
of
personal traits and characteristics based
on
interpersonal
com-
parisons
to
relevant others (Pelham, 1995; Pelham
&
Swann,
1989;
Suls & Wills, 1991). By contrast,
the
interdependent
or
relational self-concept
is
defined
in
terms
of
relationships with
others
in
specific contexts,
and
self-worth
is
derived from
ap-
propriate role behavior (Markus
&
Kitayama, 1991; Stryker,
1991).
(In
Table
1.
we refer
to
this process as
reflection,
in the
sense that the self
is
derived from the responses and satisfaction
of the other person
in the
relationship.) Finally,
the
collective
self-concept
is
determined
by
assimilation
to the
prototypic
representation of the in-group, with self-worth derived from the
status of the in-group
in
intergroup comparisons (Turner et al.,
1987).
Evidence
for
changes
in the
bases
of
self-worth
at
different
levels of self-construal come from efforts to measure global
self-
esteem separately
at the
personal
and
collective levels
(Luhtanen
&
Crocker, 1992).
In
general, self-esteem
at the
two levels
are
positively correlated,
but
only moderately
so
(Crocker, Luhtanen, Blaine,
&
Broadnax, 1994; Luhtanen
&
Crocker, 1992).
It should
be
noted here that in-group membership plays
different roles
in the
formation
and
maintenance
of the self-
concept
at
different levels.
On the
one hand, in-groups provide
the frame
of
reference
for
self-evaluation
at
the individual level
and
for
selection of significant others
at the
interpersonal level.
Shared in-group membership
is
one important basis
for
deter-
mining relevant sources of social comparison. For instance,
de-
fining ourselves as social psychologists means that we are more
likely
to
assess
our
academic qualifications
and
research abili-
ties
in
comparison
to
other social psychologists than
to
other
types of behavioral
or
social scientists. Furthermore, confirma-
tion of our self-assessment from other in-group members
is
re-
lated
to the
certainty with which we make trait attributions
to
ourselves (Pelham
&
Swann, 1994).
The other
role
that in-groups play in defining the individual's
self-concept derives from comparisons between characteristics
shared
by
in-group members
in
comparison
to
relevant
out-
groups. This is the essence of social identity. When we think
of
ourselves
as
social psychologists in this
sense,
we are most likely
to attribute traits and characteristics to ourselves that we share
with other social psychologists
and
that make
us
distinct from
other social
and
behavioral scientists.
The
focus
on
intragroup
differences versus intragroup similarities
(and
intergroup
differences) serves
as a
main indicator
of
people's relative
em-
phasis
on
their personal
or
collective selves (McFarland
&
Buehler,
1995;
Simon, Pantaleo,
&
Mummendey, 1995).
The distinction between interpersonal comparison
and in-
tergroup comparison
as
determinants
of
self-evaluations
was
demonstrated
in a
recent experiment
by
Brewer
and
Weber
(1994).
In
this experiment, participants were randomly
as-
signed
to one
of two artificial social categories. Social identifi-
cation with in-group assignment
was
manipulated indirectly by
varying
the
salience
and
distinctiveness
of
the in-group cate-
gory. After being assigned
to a
category, participants viewed
a
videotaped interview with another research participant that
provided exposure
to an
upward
or
downward social compari-
son target.
In the
upward social comparison conditions, the
in-
dividual
on the
videotape was exceptionally high
in
academic
achievement
and
intellectual ability;
in the
downward social
comparison conditions,
the
same individual played
the
role
of
a poor student with relatively
low
academic accomplishment.
Furthermore,
the
individual
on the
video was identified either
as
a
member of the viewer's own social category (in-group)
or
as
a member of the contrasting category (out-group).
The predicted outcomes of the experiment
were
based on
the
assumption that participants
who
had been assigned
to
the non-
distinctive social category would
be
oriented toward intra-
group, interpersonal social comparison
and
would evaluate
their own academic abilities
in
contrast
to
those of another in-
group member
but
would
be
unaffected
by
comparison infor-
mation about an out-group member Members of
the
distinctive
social category, on the other hand,
were
expected to be oriented
toward intergroup social comparison
and
would evaluate their
own academic abilities
by
assimilating
to
another in-group
member but exhibiting contrast to
an
out-group member.
Participants' self-evaluations of academic aptitude following
exposure
to
the videotapes corresponded exactly to predictions.
Participants
in the
nondistinctive in-group condition
had sig-
nificantly lower self-ratings following exposure
to an
upward
comparison target than to
a
downward comparison target when
the person
on the
videotape was
an
in-group member,
but self-
evaluations were
not
differentially affected
by
exposure
to up-
ward
or
downward out-group comparison targets.
In
contrast,
participants
in the
distinctive in-group condition were more
positive
in
their self-evaluations following exposure
to the up-
ward comparison in-group member and more negative follow-
ing exposure
to the
downward comparison in-group member.
However, exposure
to an
out-group comparison target
had the
opposite effect—lowering self-evaluations when
the
target
was
high
in
ability and raising them when the target
was low
in
aca-
demic ability. Thus, exposure
to the
same social comparison
information
had
very different effects
on
self-evaluations,
de-
pending on the individual's relationship to the in-group and
fo-
cus on personal versus social identity.
Social Motivation
Another important transformation associated with different
levels
of
self-construal
is a
change
in the
basic goals
of
social
interaction. There
is a
fundamental difference between social
motives derived from personal self-interest
and
those derived
from concern
for
the interests of others (McClintock, 197
2).
As
Brewer( 1991) postulated, "when the definition of self changes,
86BREWER
AND
GARDNER
the meaning
of
self-interest
and
self-serving motivations also
changes accordingly" (p. 476).
Both Markus
and
Kitayama (1991)
and
Baumeister
and
Leary (1995) stressed that interdependent relationships
are
characterized by mutual concern
for
the interests and outcomes
of the other. Batson (1994) defined this concern as the basis
of
altruistic motivation, which
he
stressed
is not to be
confused
with self-sacrifice (which concerns costs
to
self)
but
as the
mo-
tivation
to
benefit
the
other.
At the
collective
level,
group welfare
becomes
an end in itself.
Experimental research
on
social
di-
lemmas
has
demonstrated
the
powerful effect
of
group identi-
fication on participants' willingness
to
restrict individual gain
to preserve
a
collective good (Brewer
&
Kramer, 1986; Capo-
rael, Dawes, Orbell, &
van de
Kragt, 1989; Kramer & Brewer,
1984).
Identification with in-groups
can
elicit cooperative
be-
havior even
in the
absence
of
interpersonal communication
among group members. Within
the
in-group category, individ-
uals develop
a
cooperative orientation toward shared problems.
Consequences
of
Shifts
in
Levels of Identity
Shift From Personal
to
Collective Self
The consequences of shifting from personal identity
to
social
identity
in
levels
of
self-categorization have been
a
continuing
focus
of
research derived from social identity theory (Turner,
1982).
Although Deaux (1992,
1993) has
argued that social
identities
are
integrated into personal identities, there
is evi-
dence of discontinuities between self-descriptions and social
be-
havior associated with
the
two levels
of
construal (Hogg
&
Ab-
rams,
1988).
As
the results of Smith
and
Henry's (1996) experiment dem-
onstrate, when
a
particular social identity is made salient, indi-
viduals
are
likely
to
think
of
themselves
as
having characteris-
tics that
are
representative
of
that social category. Social iden-
tity,
in
other words, leads
to
self-stereotyping (Simon
&
Hamilton, 1994). This effect
was
demonstrated
in an
experi-
ment by Hogg and Turner (1987) that involved gender identity.
In this study, male and female college students participated
in a
discussion under one of two conditions.
In
the personal identity
condition,
the
discussion
was
between
two
people
of
the same
sex,
and
the two
discussants held different positions on the issue
under consideration. In the social identity condition,
the
discus-
sion group consisted
of
four people—two
men and two
women—and the
sexes
differed on the
issue.
The latter arrange-
ment
was
intended
to
make categorization
by sex
particularly
salient
in
the setting and to increase the probability that partic-
ipants would think of themselves
in
terms of their gender iden-
tity. Following
the
social interaction, participants
in the
social
identity condition characterized themselves
as
more typical
of
their
sex and
attributed more masculine
or
feminine traits
to
themselves than those
in
the personal identity condition.
Consistent with this perspective, other experimental research
has demonstrated that retrieval cues designed
to
activate
the
"private" self-representation increase generation
of
self-cogni-
tions that
are
quite different from
the
self-cognitions retrieved
when
the
"collective" self-aspect
is
activated (Trafimow
et al.,
1991).
These results
led
Trafimow
and
colleagues
to
speculate
that private
and
collective self-concepts
are
stored
in
separate
locations
in
memory.
Shift
From Personal to
Relational Self
Many
of
the cognitive, emotional,
and
motivational conse-
quences associated with
the
cross-cultural differences
in self-
construal reviewed
by
Markus
and
Kitayama (1991) would
be
expected
to
hold
for
shifts in
levels
of self-representation within
the same
individual.
Indeed, Cheek (1989) argued that personal
and social (relational) identities
are
enduring properties of the
self-concept, representing separate sources of individual differ-
ences
in
self-definition. Also, Cousins (1989) demonstrated
that both American
and
Japanese participants produced
different types
of
self-descriptors
in
response
to the
Twenty
Statements Test (Hartley, 1970) when the instructions were
al-
tered from the
generic
("I
am
•") prompt
to
prompts situated
in specific interpersonal contexts.
Shift From Relational
to
Collective Self
Less research
has
been devoted
to
direct comparisons
be-
tween interpersonal
and
collective levels
of
self-categorization
and associated
behaviors.
One exception is
Hogg's
(1992,1993)
work
on
the distinction between interpersonal liking
and
social
identity as sources of attraction to others.
In
the
research literature
on
interpersonal attraction, liking
between two individuals is strongly related to the similarity
be-
tween them (Byrne, 1971), People are likely
to
become friends
or lovers
to
the extent that they perceive that they are similar
to
each other in preferences, attitudes, and
values.
At this
interper-
sonal level, attraction seems to be
a
function of the two individ-
uals'
personal traits and the degree of match between their indi-
vidual identities.
On the
other hand, research
on
social catego-
rization
and
in-group preference suggests that positive
evaluations and liking
for
other individuals can be induced sim-
ply by the knowledge that they share
a
common group identity.
In-group members tend
to
be liked more than out-group
mem-
bers even when
we
know nothing about their personal charac-
teristics.
In
general,
we
tend
to
assume that fellow in-group
members
are
similar
to
each other,
but in
this case liking
and
similarity seem
to be a
consequence of group formation rather
than
its
cause (Hogg
&
Turner, 1985).
As a
consequence,
in-
group favoritism
can
occur
in the
absence
of
interpersonal
at-
traction or its antecedents.
To
represent the idea that liking
is
sometimes based on group
membership alone, Hogg (1992,
1993;
Hogg & Hardie,
1991)
has drawn
a
distinction between idiosyncratic
personal
attrac-
tion
and
depersonalized
social
attraction.
Personal attraction
is
based
on
personal identities
of the
individuals involved,
and
similarity
of
personal interests, attitudes,
and
values is the
pri-
mary
basis
for
this form of
liking.
Social attraction, on the other
hand,
is
based on preferential liking for in-group
over
out-group
members.
To the
extent that
a
particular group member
ex-
emplifies the characteristics that are distinctive or important
to
that
group,
that individual will be socially attractive to other
in-
group members, regardless of interpersonal similarity.
Because these two forms
of
attraction have different origins,
it
is
possible
to
display preference
for an
in-grouper
we
don*t
like very much and to discriminate against
a
member of an
out-
group even if we like that individual personally. Because of this
distinction
in
sources
of
attraction,
it is
possible
for
groups
to
LEVELS
OF
IDENTITY87
work together
as
cohesive units even when members do not like
each other interpersonally,
a
phenomenon that
has
been
dem-
onstrated
in
laboratory groups (Hogg
&
Turner, 1985)
and in
real-life groups such as sports teams
(e.g.,
Lenk, 1969).
Research
by
Prentice
et
al. (1994) also
has
verified
the dis-
tinction between group identification that is based on direct
at-
tachments to the social category and identification based on in-
terpersonal attachments among group members.
In
studies
of
various campus groups, they found that members
of
groups
based
on a
common identity were more attached
to the
group
than
to
fellow group members, whereas members
of
groups
based
on
interpersonal bonds were more attached
to
members
of
the
group overall and showed a stronger relationship between
identification with
the
group and evaluation of individual group
members.
Collective Self-Representations: Some
Preliminary Studies
The idea of the "social
self" as
represented
by the
interper-
sonal and collective self-concepts
is
that of
a
more inclusive
self-
representation
in
which relations and similarities
to
others be-
come
central.
This
is
symbolically represented by the shift from
/to
we
as
a
term
of
self-reference (Taylor
&
Dube, 1986).
Ex-
periments
by
Perdue, Dovidio, Gurtman,
and
Tyler (1990)
have demonstrated that
the
pronouns
we
and us carry positive
emotional significance that
is
activated automatically
and un-
consciously. We suggest further that
the
concept
we
primes so-
cial representations of the self that are more inclusive than that
of
the
personal self-concept.
In a
preliminary investigation
of
the implications of different levels of the social self-concept, we
conducted
a set of
three experiments
to
explore
the
effects
of
priming various
we
schemas on individual judgments and
self-
descriptions.
Experiment
1
The initial experiments were premised on the idea that social
identities should lower the threshold for perceived similarity be-
tween the self and others and direct attention toward judgments
of agreement rather than disagreement (Turner
et al.,
1994).
We reasoned that this change of threshold should affect the
in-
terpretation
and
acceptance
of
ambiguous attitude state-
ments—statements that could be viewed as either supportive of
or opposed
to a
particular attitude position. Whereas neutral
and ambiguous statements
are
often contrasted away from
the
perceiver's own attitude (Eiser &
van der
Pligt, 1984), when
a
social self-concept has been activated such statements should be
more likely
to be
assimilated
and
perceived
as
similar
to the
perceiver's own position.
To test this idea, we used
a
standard priming task
to
activate
the concepts
we
or
they
and then tested respondents* judgments
of similarity/dissimilarity
for
ambiguous attitude statements.
The experimental design consisted
of
five
different priming con-
ditions.
In one
condition we-us pronouns were activated,
and
in
the
contrasting condition ihey-them pronouns were acti-
vated, and these
were
both compared with a condition
in
which
the neutral pronoun
it
was activated.
For
further comparison
purposes, two additional priming conditions were included
in
which either positive
or
negative adjectives were used
in the
priming materials. These adjective conditions served
as a con-
trol
for the
effects
of
positive
and
negative affective priming,
respectively.
Method
Participants. One hundred male and female introductory psychol-
ogy students participated
in the
experiment
for
partial course credit.
Twenty participants were randomly assigned
to
each of the
five
experi-
mental conditions.
Procedure.
In
the priming task, participants read a descriptive par-
agraph
(a
story about
a
trip
to the
city) with instructions
to
circle
all
the pronouns that appeared
in the
text,
as
part
of
a
proofreading
and
word search task.
In the
three pronoun priming conditions
the
para-
graph contained 19 pronouns,
but
the text was varied so that the same
materials were presented with almost
all of
the pronouns referring
to
we or
us,
or to they
or
them,
or to
it.
Participants in the adjective priming
conditions proofread
a
different paragraph (describing residents
of a
city) with
the
instructions
to
circle all of the adjectives
in
the text.
The
lext
was
varied so that the
19
adjectives
were
either predominantly pos-
itive concepts (e.g.,
lucky,
humorous)
or
predominantly negative (e.g.,
dissatisfied,
gossipy).
After completing this word search task, participants were escorted
to
another room to take part in
an
ostensibly different experiment involv-
ing judgments
of
attitude statements. They were presented
on a
com-
puter monitor with
a
series
of
16
attitude statements
on
various issues
and were asked
to
judge,
as quickly as possible, whether the statements
were similar
or
dissimilar
to
their own views by pressing a number key
on
the
keyboard, ranging from
1
(very
dissimilar)
to 4
(very
similar).
Eight
of
the items were selected
to be
unambiguously
pro or con in
regard
to
the attitude
issue,
and the other eight were ambiguous in their
implications (e.g., "Abortion should be available
to
victims
of
rape").
The unambiguous
and
ambiguous items were intermixed
in
random
order.
At
the
conclusion
of
both tasks, participants were probed
for
suspi-
cion (none reported
seeing any
connection between the
two
experimen-
tal tasks), debriefed, and dismissed with thanks.
Results
Preliminary analyses
of
responses
to the
unambiguous atti-
tude statements revealed
no
effects of experimental conditions
on responses
to
these items. Judgments
of
similarity
to
these
items were approximately normally distributed with
a
mean
rating of 2.75 on the
4-point
scale.
Mean response time was 7.49
s for similarity judgments and 8.19
s
for dissimilarity judgments,
with no significant differences
across
priming conditions.
Similarity ratings and response latencies for ambiguous items
served
as the
primary measure
of
threshold
for
judgments
of
agreement
or
similarity to the
self.
Of the items included
in the
judgment task, only those that
met
specific empirical criteria
for ambiguity were used
to
test
the
primary hypothesis. Items
selected
for
analysis were those that
(a)
evoked approximately
half similar and half dissimilar judgments
across
all respondents
in the experiment
and (b)
were responded to consistently (i.e.,
any
one
respondent judged each
of
the items
to be
similar
or
dissimilar).
In
Experiment
1,
two of the ambiguous
items1
met
1 The two statements were "The existence
of
famine
and
disease
in
the world causes doubt
in
some religious doctrines"
and
"Abortion
could
be
made unnecessary with appropriate
sex
education." Most
of
the unselected items
were so
neutral in content that virtually
all
respon-
dents judged them as similar.
88BREWER AND GARDNER
these criteria. For each participant, ratings and response times
were averaged across these two items as the primary dependent
measures.
Similarity judgments. Results of a one-way analysis of
vari-
ance (ANOVA) i ndicated that there were marginally significant
differences in the mean similarity ratings for the ambiguous
items across the five priming conditions,
_F(4,
95) = 2.04, p <
.10.
Both the
we
prime condition (M
=
3.30) and the positive-
adjective condition (M = 3.35) were significantly higher than
the neutral {it) condition {M = 2.90), f(95) = 1.79, p = .07,
and f(95) =
2.02,
p < .05, respectively. Neither the
they
prime
nor the negative-adjective prime conditions differed signifi-
cantly from the neutral prime or from each other {Ms - 2.95
and 2.90, respectively).
Excluding the neutral condition, results of a 2 (positive vs.
negative valence) X 2 (pronouns
vs.
adjectives) ANOVA indi-
cated
a
main effect for
valence,
F{
1,7 6) = 6.16, p
=
.01,
but no
significant interaction between valence and word type. Both of
the positively valenced primes increased judgments of sim-
ilarity for ambiguous items compared to the negatively va-
lenced primes, but
we
did not result in higher similarity judg-
ments than did positive adjectives.
Response latencies. Intraindividual reaction times were
standardized across all 16 attitude statements (Fazio, 1990),
and the z scores for the
2
ambiguous items
were
averaged
as
the
measure of response latency for each participant. Higher values
on this measure represent longer times to make judgments rela-
tive to the individual's own mean response time.
Results of a 2 (positive or negative valence)
X
2 (pronoun or
adjective) X 2 (similar or dissimilar judgment) between-sub-
jeets ANOVA revealed a significant three-way Valence
X
Word
Type
X
Judgment interaction, F{
1,
72) = 11.89, p = .001 (see
Table
2) ?
Analysis of the simple effects underlying this interac-
tion revealed that only the pronoun primes affected response
latencies. For the pronoun conditions there was a significant
two-way interaction between the we-they effect and the type of
judgment,
7^1,
36) =
9.76,
p <
.01.
Judgments of dissimilarity
took significantly longer for those in the we prime condition
than in the
they
prime condition, whereas similarity judgments
were facilitated by the
we
prime.
For the adjective priming con-
ditions, however, there were no significant effects of
positive
or
negative valence on either similarity or dissimilarity judgments,
and there was no significant interaction effect.
Table 2
Response Latency
as
a Function
of
Valence
of Prime. Word
Type,
and
Judgment:
Experiment
1
____
Judgment
PrimeSimilarDissimilar
Pronoun
We
They
Adjective
Positive
Negative
.11
.35
.16
-.09
.41
.05
.09
.11
Discussion
Results from this first experiment confirmed the prediction
regarding the priming effect of the we schema on similarity
thresholds. It further demonstrated that the we-they effect is
distinct from the effect of
positive
versus negative affect per se.
Although both positive adjectives and
we
pronouns had parallel
effects on the propensity to rate ambiguous attitude statements
as similar to one's own, only the
we
prime facilitated similarity
judgments and interfered with dissimilarity judgments as indi-
cated by the reaction time data. It is this differential response
time that indicates a change in threshold for judgments of sim-
ilarity. Whereas participants made similarity judgments readily
following the we prime, judgments of dissimilarity required
more extensive processing to override an initial, opposing re-
sponse tendency. In the adjective priming conditions both sim-
ilarity and dissimilarity judgments of ambiguous items took
about the same amount of time as responses to unambiguous
items.
Although these results provide some evidence for a unique
effect of the
we
prime on judgment thresholds, it is not clear
from this first experiment whether the we schema primed by
our word search task would correspond to the interpersonal or
collective level of self-representation. The paragraph in which
the critical pronouns were embedded was entitled "A Trip to
the City" and contained a running narrative of scenes and
events associated with
a drive
to and through a large
city.
In this
context, it is very likely that the concept
we
was interpreted to
refer to a small face-to-face group of friends or partners. Evi-
dence for an expanded self-concept should be obtained for both
forms of the social
self,
but it is unclear whether the results ob-
tained from Experiment 1 reflect only the assumed similarity
primed by reference to close, interpersonal relationships. For
this reason, we conducted a second experiment to provide a
conceptual replication of Experiment 1, in which the we-they
pronouns were embedded in different social contexts.
Experiment 2
The procedures of the second experiment followed closely
those of Experiment
1,
except that
the
adjective conditions were
dropped from the design.
Two
different versions of the pronoun
word search paragraph were created. In one version, the pro-
nouns were embedded in a story that implied a small group
social context; in the other version the context involved a very
large collective. We assumed that during the word search task,
incidental learning of the story context would imbue the pro-
nouns
we
or
they
with different group size connotations, corre-
sponding to interpersonal and collective meanings, respectively.
Method
Participants. Sixty-one male and female introductory psychology
students participated in this experiment in exchange for credit toward
2 Mean reaction time in the neutral priming condition did not differ
significantly from the
they
or negative-adjective conditions, so this con-
trol condition
was
omitted from further analyses.
LEVELS
OF
IDENTITY89
course requirements. Participants
were
randomly assigned to
one
of
five
different priming conditions prior
to
the attitude judgment task.
Procedure.
As
in Experiment
1,
participants
first
completed a word
search task
in
which they were asked
to
locate and circle all of the pro-
nouns
in a
story text. Two different context stories were used. One was
the same
"A
Trip
to the
City" story used
in
Experiment
I and
consti-
tuted the small-group context. The second contained
a
narrative about
attending
and
watching
a
football game
at a
large stadium—the large-
group context. For each story,
one
version contained primarily
we
and
us pronouns,
and the
other contained primarily they
and
them
pro-
nouns.
A
neutral control condition consisted
of
the
it
version
of
the
small-group paragraph from Experiment
1. All
five story paragraphs
contained
19
pronouns.
After completing
the
word search task, participants moved
to a
different room and completed the attitude judgment task used in Exper-
iment
I. At the
completion
of
this
task, participants were probed
for
suspicion about
the
purposes
of
the
experiment (none reported seeing
any connection between
the two
experimental tasks), debriefed,
and
dismissed with thanks.
Results
As
in
Experiment
1,
there
were
no effects of the priming con-
ditions on judgments
of
the eight unambiguous attitude state-
ments. Applying
the
same criteria
as in the
first study,
we se-
lected two ambiguous items
for
analysis.3 Mean ratings and
re-
sponse latencies
for
these
two
items constituted
the
primary
dependent variables.
Similarity
ratings.
Results of a one-way
ANOYA
revealed
a
significant main effect
of
priming
on
similarity judgments
of
the ambiguous attitude items, >"(4,
56) =
2.63,
p <
.05.
The
two
we
prime conditions
(M -
2.83) were significantly different
from
the
neutral condition
(M
=
2.29)
and
from
the
they
con-
ditions
(M
=
2.50),
F( I, 42) =
4.02,
p
<
.05.
However, the two
(large
vs.
small) we primes did not differ significantly from each
other. Similarity ratings
for
participants
in the
they
prime con-
ditions
did not
differ significantly from those
of the
control
condition.
Response
latencies.
We computed response latencies
as in
Experiment
1
for
reactions
to the
selected ambiguous attitude
statements. Results
for
the we-they conditions
are
reported
in
Table
3. An
ANOVA revealed
a
significant Pronoun
X
Judg-
ment interaction effect, F{
1,
38)
=
5.176,
p
<
.05,
which paral-
leled that obtained
in
Experiment
1.
As
in
the first experiment,
we-primed respondents were significantly faster in making sim-
ilarity judgments and significantly slower than they-primed
re-
spondents
in
making dissimilarity
judgments.
The effect of the
we prime
was the
same regardless
of
context group size.
Re-
sponse times
for the
small (interpersonal) we primes
and the
large (collective)
we
primes were not significantly different.
Table
3
Response Latencies
as a
Function
of
Prime and
Judgment:
Experiment
2
Discussion
Results from this second experiment replicated
the
differ-
ences between
we
and they prime effects
on
readiness
to
make
similarity judgments and inhibition of dissimilarity judgments.
This effect did
not
differ as
a
function of the incidental context
in which
the
concept
we
was embedded. This
was
consistent
with our expectation that any form of activation of a social
self-
concept would entail an expanded sense of self that would lower
thresholds
for
agreement and assimilation.
Experiment
3
Although
the
small
and
large
we
concepts primed
in
Experi-
ment
2 had
similar effects
on
judgments
of
attitudes,
we did
assume that the two contexts activated somewhat different con-
struals
of
the social
self. The
we concept associated with
the
small group should have primed access
to the
relational
self-
concept—those aspects
of
the self associated with close inter-
personal relationships with significant others.
The
large-group
we
schema should have activated collective self-identities asso-
ciated with membership
in
large social groups
or
categories.
Our third experiment was designed
to
test this assumption
di-
rectly by assessing the effects
of
we-they primes
on the
content
of freely generated self-descriptions.
In the third experiment
we
used the same
five
priming condi-
tions
as
in Experiment
2,
but
this
time the second task consisted
of completion
of the
Twenty Statement Test
(TST;
Hartley,
1970) as
a
method of generating spontaneous self-descriptions.
When the TST
is
used with White college students in the United
States,
the
preponderance of responses
are
trait terms
or
other
personal attributes, in keeping with the individualistic construal
of
the
self typical
in
this culture (Markus
&
Kitayama, 1991).
For instance, Cousins (1989) found that with nonconstrained
instructions, 58%
of
responses generated by White college
stu-
dents were personal traits
and
attributes,
and
less than 10%
of
responses referred
to
social relations
or
social category
mem-
berships. Against this baseline,
we
were interested
in
whether
priming different
we
concepts would increase the proportion
of
social self-descriptions on the TST.
Method
Participants. One hundred and forty-six male and female introduc-
tory psychology students participated in the experiment in exchange for
credit toward course requirements. Participants were randomly
as-
signed
to
one of the
five
priming conditions, with proportionally more
distributed between the two
we
prime conditions
(N
= 15) than across
the other three conditions
(N= 51).
Procedure The initial priming task was the same word search task
used in Experiment
2,
with the same
five
versions of the story (m?-small
Pronoun
prime
We
They
Similar
.17
.29
Judgment
Dissimilar
.31
.06
3 The
two
statements were "The existence
of
famine
and
disease
in
the world causes doubt
in
some religious doctrines*'' (same
as
Experi-
ment 1) and "Churches that are beautiful and emphasize the aesthetic
side
of
life
are most interesting."
We
also conducted analyses using the
same
two items as
had been used in Study
1
(dropping
a
few participants
in order to meet the consistency criterion), and results
for
the reaction
time measure were essentially
the
same
as
that reported here
for the
selected items.
90BREWER AND GARDNER
group,
H'f-Iarge
group, iheysmaW group, itey-large group,
(/-small
group).
After finishing the word search, participants went on to a sepa-
rate task in which they were given the TST, completing 20 sentence
stems beginning with "I am
" After completing both tasks, partici-
pants responded to a brief questionnaire that included a manipulation
check item on how many people they believed were being referred to in
the paragraph they read.
At the completion of the experimental session, the experimenter
probed for suspiciousness concerning the priming task and its possible
influence on the self-description task. No participant guessed
the
exper-
imental hypothesis or connected the priming with the TST in any way.
Participants
were
then debriefed, thanked, and released.
Table 4
Proportion
of
Social Self-Descriptions
as a
Function
of
Type
of
Prime:
Experiment 3
Descriptive code
Type of primeInterpersonal
We
They
It
.07
.05
.02
Collective
.18
.09
.09
Results
Manipulation
check.
Results of an ANOVA of the size esti-
mates revealed
only
the expected main effect of the small-versus
large-group versions of the priming paragraphs. Perceptions of
the size of the group described in the stadium scenario para-
graphs (M =
7,332.37)
were significantly larger than were esti-
mates of the
size
of the group described in the city visit scenario
(M
=
10.64), F(
1,
94) =
5.21,/><
.05.
This difference
was
not
affected by whether the paragraph contained we or they pro-
nouns. Although most estimates of group size for the small-
group condition were larger than a dyad, they were clearly
within the range of an interacting, face-to-face social
group.
All
of the estimates for the large-group condition were beyond this
range.
Self-descriptions. Responses to the "I am" completions
were coded by two independent raters who were blind to prim-
ing condition. Self-definitions were coded, on the basis of a
modification of Hartley's (1970) coding
scheme,
into three cat-
egories.
A personal
self-description was one that contained ref-
erence to individual physical
qualities,
traits, attitudes, or activ-
ities (e.g., "I am very athletic").
Interpersonal
self-descriptions
were those that referred to a specific relationship (e.g., "I am
happily married").
Collective
self-definitions were those that
implied membership in a social category (e.g., "I am a Black
woman"). Interrater reliability of codings was .92. (Any dis-
agreements in coding of
the
same item were reconciled before
data analysis.) The proportion of interpersonal descriptions and
the proportion of collective descriptions in each participant's
protocol were the primary dependent measures of the study.
The initial analysis
was
a one-way ANOVA of the three levels
of priming: the we prime (combining the two group size
versions), the they prime, and the it prime. The results of this
analysis revealed a marginally significant main effect of the
primed pronoun for proportion of interpersonal self-descrip-
tions,
F(4,
121 )
= 2.37, p = .06, and a significant main effect
for proportion of collective self-descriptions, /"\4, 121) = 6.25,
p
<
.001.
(See Table 4 for mean values.)
The number of interpersonal descriptions was significantly
greater in the we conditions than in the neutral condition,
((121) -
2.51,
p =
.01,
and marginally different from the they
conditions, /(121) = 1.72, p < .10. The they conditions were
not significantly different from the neutral condition in produc-
tion of interpersonal self-references, t( 121) = 1.07, p > .25.
The
we
conditions did differ significantly from both the they
and the neutral conditions in the proportion of collective
self-
descriptions, f(12l) = 3.13, p <
.01,
and
*(
121) = 2.68, p =
.01,
respectively. Again, the they conditions did not differ sig-
nificantly from the neutral priming group on this self-descrip-
tion measure, /(121) = 0.24, p > .50.
To
assess the effects of priming context,
we
subjected the data
from the four versions of the paragraphs in the
we
and they con-
ditions to a 2 (pronoun) X 2 (group size) ANOVA. Means for
these
interactions are presented in Table
5.
For the interpersonal
self-descriptions there was a marginally significant main effect
of pronoun, F(
1,105)
=
2.88,
p
< .
10,
but no significant effect
of context or a Context
X
Pronoun interaction. For the propor-
tion of collective self-descriptions, however, the main effect of
pronoun was qualified by a marginally significant interaction
between pronoun and context group size, F{
1,
105) = 3.47, p
< ,10, For the
we
paragraphs, the difference between the small-
and large-group contexts was significant, ;(105) = 3.98, p <
.01,
but the effect of group size was not significant for the they
conditions.
Discussion
The overall production of social self-descriptions
was
greatly
enhanced by the
we
prime, compared to the baseline propor-
tions found in the they and it conditions. Furthermore, the re-
sults from the TST provide partial support for the hypothesized
difference between interpersonal and collective
we
schemas, at
least with respect to collective self-descriptions.
The data with respect to interpersonal self-construals are
weak. Relational self-descriptions were relatively infrequent in
all conditions and did not differ as a function of the group size
implied in the priming condition. The failure to find clear evi-
dence of activation of
a
distinctive interpersonal self-represen-
tation may reflect a failure of the priming condition or insensi-
tivity of the measure. There is no doubt that the group envi-
sioned in the trip-to-the-city scenario was smaller than that in
the football stadium story, as revealed by results from the ma-
nipulation check. Nonetheless, a mean estimate of more than
10 does not connote an intimate group of close relationships.
For at least some of the participants in the small-we condition,
the context may not have activated the interpersonal meaning
of
we.
Furthermore,
the
TST format (with the "I am
"stem)
may be particularly unsuited to prompting relational terms.
Collective self-identifications, on the other hand, were clearly
greater in the we-primed conditions, and this effect was partic-
ularly pronounced for
the large-we
context,
for which social cat-
egory memberships appeared in respondents' spontaneous
self-
descriptions to a degree not typically found in studies with
American college
students.
Thus,
this
study produced initial ev-
LEVELS
OF
IDENTITY91
Table
5
Proportion
of
Social Self-Description
as a
Function
of Pronoun
and Prime
Context:
Experiment
3
Size of group
Small
Large
Small
Large
Pronoun
We
Interpersonal descriptions
.07
.07
Collective descriptions
.12
.24
They
.03
.06
.09
.10
idence that priming terms associated with collective identities
affect
the
accessibility
of
different levels
of
self-representa-
tions.
Admittedly, these studies are preliminary, and further re-
search will
be
needed
to
determine whether
the
interpersonal
and collective levels
of
self-construal
can be
operationally
distinguished.
General Discussion
The idea that different levels
of
inclusiveness define concep-
tually distinct construals of the self raises
a
number of interest-
ing questions that are just beginning
to
be addressed in research
on self-representations. One such question
is the
nature
of
the
interrelationships among these different self systems. Turner
and
his
colleagues (Turner
et al., 1987,
1994) have contended
that there
is a
functional antagonism between self-categoriza-
tions
at
different levels
of
inclusiveness.
Taken
to the
extreme
this would imply that social identities
at
different
levels
of orga-
nization
are
essentially independent
and
may include
not
only
different
but
possibly incompatible representations
of
the
self.
At
the
other extreme, Deaux (1992, 1993) has argued that so-
cial identities
are
integrated into
the
personal self-concept.
In
between these
two
extremes
is the
idea that
the
different self
systems interact
as
complementary components
of
an overall
self-concept and self-esteem.
A second important question is what determines which level
of self-representation
is
activated
at any