RUNNING HEAD: The BIAS Map
The BIAS Map: Behaviors from Intergroup Affect and Stereotypes
Amy J. C. Cuddy
Susan T. Fiske
Key Words: STEREOTYPES, PREJUDICE, EMOTIONS, DISCRIMINATION,
INTERGROUP BEHAVIOR, COMPETENCE, WARMTH, AMBIVALENT
The present research, consisting of two correlational studies (N = 616), including a representative
U.S. sample, and two experiments (N = 350), investigated how stereotypes and emotions shape
behavioral tendencies toward groups, offering convergent support for the Behaviors from
Intergroup Affect and Stereotypes (BIAS) Map framework. Warmth stereotypes determine active
behavioral tendencies -- attenuating active harm (harassing) and eliciting active facilitation
(helping). Competence stereotypes determine passive behavioral tendencies – attenuating passive
harm (neglecting) and eliciting passive facilitation (associating). Admired groups (warm,
competent) elicit both facilitation tendencies; hated groups (cold, incompetent) elicit both harm
tendencies. Envied groups (competent, cold) elicit passive facilitation but active harm; pitied
groups (warm, incompetent) elicit active facilitation but passive harm. Emotions predict
behavioral tendencies more strongly than stereotypes do and usually mediate stereotype-to-
"Discrimination leads to all sorts of curious patterns" -Allport, 1954, p. 55
Allport noted that groups can be discriminated against in quite different ways, but did not
provide a theoretical rationale. Here, we differentiate types of discriminatory behaviors, as
outcomes of competence-warmth stereotypes and intergroup emotions, by combining various
theories and findings with a model of stereotype content (Fiske, Cuddy, Glick, & Xu, 2002), to
predict specific intergroup behaviors. Both correlational and experimental investigations test this
new framework, which predicts four classes of discriminatory behavioral tendencies, along two
dimensions (active-passive and facilitative-harmful). The proposed Behaviors from Intergroup
Affects and Stereotypes (BIAS) Map systematically links behavioral tendencies to the contents
of stereotypes and emotions about groups, as rooted in underlying structure of intergroup
An Integrative Foundation
This research aims to integrate several principles derived from existing intergroup bias
theory. Consistent with the tripartite view of attitudes, bias has been conceptualized as
comprising three components – cognitive (stereotypes), affective (emotional prejudices), and
behavioral (discrimination) (Esses & Dovidio, 2002; Fiske, 1998). Prior work on the relevant
functional, motivational, and social-cognitive processes suggests three interrelated principles.
First, biases vary qualitatively across groups and situations, often including both
negative and subjectively positive responses (see Mackie & Smith, 2002, for examples). Several
recent approaches illustrate this principle. In Cottrell and Neuberg’s sociofunctional approach
(2005), different groups (e.g., gay men vs. Mexican-Americans) elicit different perceived threats
(e.g., to health vs. property), which evoke functionally-relevant, distinct emotions (e.g.,
respectively: disgust, pity vs. fear, anger; see also Esses, Dovidio, Jackson, & Armstrong, 2001;
Stephan & Stephan, 2002). Intergroup emotions theory (IET; Mackie, Devos, & Smith, 2000)
traces group-based emotions (e.g., anger) and action tendencies (e.g., offensive tendencies) to
situational appraisals of potential harm or benefit. Alexander, Brewer, and Hermann’s functional
model (1999) suggests that appraisals of other groups’ goal compatibility, relative status, and
relative capacity to attain goals combine to elicit specific behavioral inclinations, emotions, and
outgroup “images.” The stereotype content model (SCM; Fiske et al., 2002; Fiske, Xu, Cuddy, &
Glick, 1999) posits that competence and warmth stereotypes respectively stem from the
perceived social status and competitiveness of the target group, and lead to distinct intergroup
emotions (admiration, contempt, envy, and pity).
Second, specific social situations synchronize the cognitive, affective, and behavioral
components of bias. For instance, appraisal theories of emotions link cognitive appraisals to
discrete interpersonal (e.g., Frijda, Kuipers, & ter Schure, 1989) and intergroup emotions
(Mackie et al., 2000). Cognitive appraisals assess implications of others’ behavior for the self:
Will this hurt or help me (Lazarus & Folkman, 1984)? In turn, discrete emotions elicit specific
behavioral inclinations adapted to deal with the potential threat (Frijda et al., 1989; Roseman,
Wiest, & Swartz, 1994). IET, an appraisal-based approach, suggests that when ingroup
identification is salient, appraisals of an outgroup lead to distinct emotions. For example,
appraising the outgroup as weaker elicits anger and offensive tendencies (Mackie et al., 2000).
Third, compared to cognitions, emotions more strongly and directly relate to behavior.
Emotion theorists have long argued for the primacy of affect as preceding and motivating both
cognition and behavior (see Zajonc, 1998, review). Indeed, affect appears superior to stereotypes
in predicting both discrimination and intent (e.g., Dovidio, Brigham, Johnson, & Gaertner, 1996;
Schütz & Six, 1996; Talaska, Fiske, & Chaiken, in press). For example, general affective
reactions to national, ethnic, and religious groups better predicted social distance than did
stereotypes (Stangor, Sullivan, & Ford, 1991). Similarly, focusing on emotions (more than
focusing on thoughts) while viewing an anti-racism video increased willingness for contact with
Black people (Esses & Dovidio, 2002). Moreover, affect appears to mediate the effect of
cognitions on behaviors, a view supported by appraisal theories of emotion, including IET, as
reflected in their cognitive appraisal Æ emotion Æ behavioral intention sequence (Frijda et al.,
1989; Mackie et al., 2000; Roseman et al., 1994).
Building on these three principles, the BIAS Map proposes: (a) differentiated biases,
which include both negative and positive responses, will stem from social structural appraisals of
groups; (b) the contents of stereotypes (i.e., cognition), emotions (i.e., affect) and discriminatory
tendencies (i.e., behavior) will coordinate in systematic, functional, and predictable ways; and (c)
emotions will more strongly and directly predict behavioral tendencies than will stereotypes.
While existing theory offers hope for predicting behaviors, no previous work specifically links
dimensions of specific stereotypes, discrete emotions, and behavioral tendencies, the aim of the
Our approach differs in significant ways. First, it provides theoretical and empirical
support for the importance of specific stereotype contents, which result from perceived structural
relations, in predicting behavioral tendencies. Many of the existing approaches neglect this
component, moving directly from the cognitive appraisal of the structural relation (as a single
variable) to the emotion (Cottrell & Neuberg, 2005) or from the emotion to action tendency
(Mackie et al., 2000). Second, the BIAS Map identifies theoretically-supported underlying
dimensions of behavioral tendencies. Existing work has either tested only one class of behaviors
such as intergroup contact or policy preferences (e.g., Esses & Dovidio, 2002; Stangor et al.,
1991), or has experimentally tested only two classes of behaviors along one dimension (e.g.,
Mackie et al., 2000). We attempt to integrate that previous work in one framework.
The Stereotype Content Model
The proposed BIAS Map evolves from the stereotype content model (SCM; Fiske et al.,
2002, 1999), which diverges from other theories of differentiated biases (reviewed above) in its
emphasis on underlying trait dimensions and its focus on ambivalent stereotypes and emotions.
Based on the premise that different traits are processed in markedly different ways (Rothbart &
Park, 1986) and lead to dramatically different outcomes (Wojciszke, 2005), the SCM focuses on
the two trait dimensions, warmth (e.g., warm, sincere) and competence (e.g., capable,
competent), which consistently emerge as the two central dimensions of social perception, from
impressions of individuals (Judd, Hawkins, & Yzerbyt, 2005; Fiske, Cuddy, & Glick, in press;
Rosenberg, Nelson, & Vivekananthan, 1968; Wojciszke, Bazinska, & Jaworski, 1998;
Wojciszke, 2005) to stereotypes of specific social groups (e.g., Clausell & Fiske, 2005; Cuddy,
Fiske, &Glick, 2004; Cuddy, Norton, & Fiske, 2005; Eckes, 2002; Glick, 2002; Lin, Kwan,
Cheung, & Fiske, 2005; Phalet & Poppe, 1997; Yzerbyt, Provost, & Corneille, 2005). The SCM
proposes that warmth and competence stereotypes respectively stem from appraisals of the (a)
potential harm or benefit of the target group’s goals and (b) degree to which the group can
effectively enact those goals. Groups viewed as competitors are stereotyped as lacking warmth,
while groups viewed as cooperative are stereotyped as warm; groups viewed as high status are
stereotyped as competent, while groups viewed as low status are not. These relationships have
been replicated in diverse U.S. samples (Fiske et al., 2002) and over a dozen international
samples (Cuddy et al., 2006; Eckes, 2002; Fiske & Cuddy, 2006) using widely-varied target
groups, such as occupations, nationalities, races, socio-economic groups, religions, and gender
From these locations defined by stereotypic high vs. low warmth and competence, the
SCM identifies four resulting emotions: admiration, contempt, envy, and pity (Fiske, Cuddy, &
Glick, 2002; Fiske et al., 2002). Four types of interpersonal social comparisons (R. Smith, 2000)
and related outcome attributions (e.g., Weiner, 2005) generate four emotional responses. Upward
assimilative social comparisons – to groups stereotyped as warm and competent (e.g., ingroups)
– elicit admiration and pride (Fiske et al., 2002), emotions linked to dispositional attributions
(i.e., deservingness) for another’s positive outcome (Weiner, 2005). Downward contrastive
comparisons – to groups stereotyped as incompetent and cold – elicit contempt and disgust (e.g.,
poor people; Fiske et al., 2002; Dijker et al., 1996), emotions linked to dispositional attributions
(i.e., deservingness) for another’s negative outcome (Rozin, Lowery, Imada, & Haidt, 1999).
Upward contrastive comparisons – to groups stereotyped as competent but not warm (e.g.,
Asians; Fiske et al., 2002; Lin et al., 2005; e.g., Jews; Fiske et al., 2002; Glick, 2002, 2005) –
elicit envy, an emotion linked to situational attributions (i.e., undeservingness) for another’s
superior outcomes (R. Smith, Parrott, Ozer, & Moniz, 1994). Downward assimilative
comparisons – to groups stereotyped as warm but not competent – elicit pity (e.g., the elderly;
Cuddy & Fiske, 2002; Cuddy et al., 2005; Fiske et al., 2002), an emotion linked to situational
attributions (i.e., undeservingness) for another’s negative outcome (Weiner, 2005).
The Present Research: From the SCM to the BIAS Map
By identifying and mapping the types of discriminatory behaviors that result from each
combination of stereotypic high vs. low competence and warmth (e.g., low-competence/high-
warmth) and its corresponding emotion (e.g., pity), the BIAS Map picks up where the SCM ends,
integrating existing theory and findings along the way.
Identifying Dimensions of Discriminatory Behaviors
Past work suggests that two dimensions capture a wide range of intergroup behaviors:
Active-passive concerns intensity; harm-facilitation concerns valence. The active-passive
distinction runs through various areas of psychology; behaviors tend to be enacted with relatively
more or less effort, directness, engagement, intent, and intensity. This dimension distinguishes
more overt and effortful intergroup behaviors, such as harassment, from more subtle types that
involve less exertion, such as neglect. Active behaviors act either for or against the group;
passive behaviors act either with or without the group, but they do so incidentally and with less
effort, directness, engagement, and intensity. The active-passive dimension classifies a range of
interpersonal behavior: aggression (Buss, 1961), romantic relationship behaviors (Sinclair &
Fehr, 2005), leadership styles (Eagly, Johannesen-Schmidt, & van Engen, 2003), and minority
social influence (Kerr, 2002). Ayduk and colleagues (2003) describe active behaviors as direct,
explicit, overt, confrontational, intense, and high risk, in contrast to passive behaviors, which are
indirect, covert, less intense, and avoidant. “Passive” does not imply a completely inert state
(which would make “passive behavior” an oxymoron); rather, “passive” in psychology often
describes behaviors that require less effort, direction, and intention (e.g., passive aggression)
relative to behaviors that are unambiguously active and goal-directed (e.g. active aggression).
For the intergroup domain, we define active behaviors as those that are conducted with
directed effort to overtly affect the target group; they act for or against the target group. We
define as passive behaviors those that are conducted or experienced with less directed effort, but
still have repercussions for the outgroup; they act with or without the target group. Passive
behaviors may reflect a less deliberate or obvious intention on the part of an actor to bring about
a specific outcome, but can constitute consequential forms of discrimination (e.g., passive
segregation, failure to hire members of a specific group, neglecting an outgroup member’s
welfare, not providing service). On the positive side, passive behaviors represent noncommittal
rapprochement, as when prejudiced people “go along to get along,” patronize businesses owned
by disliked outgroups, or tolerate but neither object to nor endorse the outgroup’s presence.
A second frequent distinction concerns the valence of behavior as determined by its
intended effect on others. We refer to this second dimension as facilitation-harm. This dimension
is basic to distinguishing prosocial/helping behavior from antisocial/aggressive behavior (see
Batson, 1998 and Geen, 1998 for reviews). Similarly, interdependence theorists focus on how
social behavior facilitates or impedes others’ goals (e.g., Thibaut & Kelley, 1959). In the
intergroup context, we define facilitation-harm as follows: Facilitation leads to ostensibly
favorable outcomes or gains for groups; harm leads to detrimental outcomes or losses for
groups. Thus, we consider four classes of behaviors, along two bipolar dimensions:
Active facilitation (i.e., acting for) explicitly aims to benefit a group. Interpersonally,
these behaviors include helping, assisting, or defending others (e.g., opening a door for
someone). Institutionally, these behaviors include assistance programs for the needy, corporate
charitable giving, progressive tax codes, and anti-discrimination policies.
Active harm (i.e., acting against) explicitly intends to hurt a group and its interests.
Verbal harassment, sexual harassment, bullying, and hate crimes all constitute interpersonal
active harm. Institutionally, active harm can range from discriminatory policies to legalized
segregation to mass internment (e.g., Japanese Americans during World War II) to genocide.
Passive facilitation (i.e., acting with) accepts obligatory association or convenient
cooperation with a group. Such behavior is passive because contact is not desired, but merely
tolerated in the service of other goals; facilitation of the group is a mere byproduct. Interpersonal
examples include hiring the services of an outgroup member (e.g., as a domestic) or choosing to
work with a member of a group assumed to be smart (e.g., an Asian American) on a team
project. Institutionally, realpolitik cooperation with a disliked regime illustrates passive
facilitation. Passive facilitation acts with the group for one’s own purposes, but simultaneously
benefits the other group as a tolerated by-product.
Passive harm (i.e., acting without) demeans or distances other groups by diminishing
their social worth through excluding, ignoring, or neglecting. Relational or social aggression
(e.g., Crick & Grotpeter, 1995) and passive negative coping (e.g., withdrawal of social support;
Ayduk et al., 2003) are related concepts. Interpersonal passive harm includes avoiding eye
contact, being dismissive, or ignoring outgroup members. Institutionally, passive harm involves
disregarding the needs of some groups or limiting access to necessary resources such as
education, housing, and healthcare. Passive harm acts without the group, denying its existence,
harming its members by omission of normal human recognition.
Three hypotheses specify how competence-warmth stereotypes and their corresponding
social emotions might predict the four classes of behavioral tendencies.
Hypothesis 1: Stereotypes
Because of its apparent primacy perception of others (reviewed below), we hypothesize
that the warmth dimension will predict active behaviors, both harmful and facilitative, whereas
the competence dimension will predict passive behaviors, both harmful and facilitative. Warmth
stereotypes theoretically derive from the inferred goals of the target group and the potential
benefits or harms caused by these goals (Wojciszke, 2005). The SCM supports this link:
Competitive or exploitative groups (whose goals are perceived as harmful) are stereotyped as
lacking warmth, while non-competitive groups (perceived as not having harmful goals) are
stereotyped as possessing warmth. In social interactions, negative warmth information (e.g.,
dishonest, insincere, unkind) is weighted far more heavily than negative competence information
(e.g., incapable, incompetent, unintelligent) (Kubicka-Daab, 1989; Wojciszke, Brycz, &
Borkenau, 1993). Perceivers are more interested in learning warmth-related traits, which better
predict their evaluations of others, than competence-related traits (Wojciszke et al., 1998).
Moreover, warmth traits are judged more quickly than competence traits (Willis & Todorov,
The primacy of the warmth dimension may occur because of potentially greater costs to
dealing with someone who is not warm versus not competent (Wojciszke, 2005). Cognitively,
negative warmth information is seen as more diagnostic because people who are not friendly are
more dangerous to others than people who are not competent, who are more dangerous to
themselves (Reeder, 1993). Motivationally, being warm is other-profitable, whereas being
competent is self-profitable (Peeters, 1983). Thus, we hypothesize that warmth information
creates a relatively urgent need to react, leading to active behavioral tendencies that act for (i.e.,
active facilitation) or against (i.e., active harm) the other. We predict that groups stereotyped as
warm will elicit active facilitation; groups stereotyped as lacking warmth will elicit active harm.
Perceived competence theoretically derives from the inferred efficacy with which the
target’s goals are enacted (Wojciszke, 2005). The SCM’s parallel analysis shows that groups
high in status (i.e., having the resources or power to carry out goals) are stereotyped as
competent, whereas low-status groups are stereotyped as lacking competence. We hypothesize
that in contrast to the exigency of warmth information in person perception, competence
information is less pressing because it is less self- or ingroup-relevant. As noted, perceivers are
less interested in and influenced by competence (vs. warmth) information (Wojciszke et al.,
1998). Compared with inferred warmth, inferred competence does not create as immediate a
need to react, thus cuing more passive behaviors, which involve acting with (i.e., passive
facilitation) or without (i.e., passive harm) others. We predict that groups perceived as competent
will elicit passive facilitation, whereas groups perceived as incompetent will elicit passive harm.
In sum, the first hypothesis states that the warmth dimension of stereotypes will predict
the valence (i.e., facilitation vs. harm) of active behaviors, and the competence dimension of
stereotypes will predict the valence of passive behaviors. Specifically, we predict that warmth
stereotypes will elicit active facilitation (e.g., helping) and prevent active harm (e.g., attacking);
competence stereotypes will elicit passive facilitation (e.g., associating with) and prevent passive
harm (e.g., excluding). Each combination of competence and warmth stereotypes thus relates to a
distinct pattern of behavioral tendencies (see Figure 1).
Hypothesis 2: Emotions
Assuming that cognitions cue behaviors and emotions activate them (Frijda et al., 1989),
we hypothesize that the distinct emotion linked to each SCM combination of competence-warmth
stereotypes will also predict the hypothesized behavioral tendencies (see Figure 1). A distinct
emotion links to each combination of high/low competence-warmth stereotypes (Fiske et al.,
2002, Study 4), so we hypothesize that, as depicted in Figure 1, two emotions will predict each
behavioral tendency. These specific links are supported by theories that conceptualize discrete
emotions as outcomes of social comparisons (e.g., R. Smith, 2000), outcome attributions (e.g.,
Weiner, 2005), and cognitive appraisals (e.g., Dijker et al., 1996; Mackie et al., 2000).
Admiration (high-competence, high-warmth). Admiration and pride—univalent, upward
assimilative emotions (R. Smith, 2000)—are directed toward others whose positive outcomes do
not detract from the self (Tesser & Collins, 1988). We hypothesize that admiration will lead to
both active and passive facilitation. Admiration and pride motivate contact (Dijker et al., 1996)
and relate to cooperation (Alexander et al., 1999); happiness, a linked emotion, predicts positive
approach behaviors (Neuberg & Cottrell, 2002). People tend to act actively for or passively with
Contempt (low-competence, low-warmth). Contempt and disgust—univalent, downward
contrastive emotions (R. Smith, 2000)—target those with negative outcomes perceived as onset-
controllable (Weiner, 2005). We hypothesize that contempt will cue both active and passive
harm. Contempt-related emotions elicit passively harmful actions such as demeaning
paternalistic behaviors (Brewer & Alexander, 2002); neglect (Weiner, 2005); and distancing,
excluding, or rejecting (Roseman et al., 1994; Rozin et al., 1999). Disgust also motivates
attempts to remove a noxious stimulus from one’s perceptual field, eliciting the desire to
forcefully expel or obliterate the stimulus (Plutchik, 1980, cited in Roseman et al., 1994). People
tend to act actively against or passively without others who elicit contempt.
Envy (high-competence, low-warmth). Envy covets another’s superior outcome and
comprises feelings of injustice or inferiority (R. Smith et al., 1994). Envy is ambivalent,
involving both resentment and respect. We hypothesize that envy cues both passive facilitation
and active harm. Because envy implicitly acknowledges that another group has outdone the
ingroup, it cues cooperation that might enable the ingroup to acquire some of the coveted
outcome. Envy involves begrudging admiration for the other, an ambivalent type of respect that
might produce passive facilitation. Second, envied groups are scapegoated when societies
experience widespread instability, because envied groups are perceived to have ability
(competence) as well as intent to disrupt society (Glick, 2005; Staub, 1996). Scapegoating can
lead to hostile acts against the envied group. People tend to act passively along with but also
actively against envied others.
Pity (low-competence, high-warmth). Pity is also an ambivalent emotion, comprising
both compassion and sadness. Pity results from appraising another’s negative outcome as
uncontrollable (Weiner, 2005). Pity elicits active facilitation and passive harm. Active
facilitation includes help-giving elicited by pity (Weiner, 2005). However, sympathy for the
suffering can distance, not just activate help. Pity involving sadness and depression can lead to
inaction, avoidance, and neglect, such as turning off an appeal to aid starving children (Green &
Sedikides, 1999; Roseman et al., 1994); pity involving disrespect may lead to dismissive
behaviors, such as patronizing speech and poor medical treatment directed at elderly people (e.g.,
Pasupathi & Lockenhoff, 2002). People tend to act actively for but also passively without pitied
Corollary of Hypotheses 1 & 2: Bias Clusters
The first two hypotheses imply coordinated “bias clusters” of specific stereotypes,
distinct emotions, and pairs of behavioral tendencies. Further, if the specific hypothesized links
are supported, ambivalent bias clusters should emerge: Groups with ambivalent competence-
warmth stereotypes (i.e., high on one, low on the other), and ambivalent emotions (i.e., envy,
pity) will be targets of ambivalent patterns of intergroup behaviors – one facilitation behavior
and one harm behavior. We predict that high-competence/low-warmth stereotypes will link to
passive facilitation and active harm, and low-competence/ high-warmth stereotypes will link to
active facilitation and passive harm (see Figure 1).
Hypothesis 3: Emotion Priority
Consistent with the third principle presented earlier – that emotions more strongly and
directly predict behaviors – Hypothesis 3 states that the relationship between emotions and
behavioral tendencies will be stronger than the relationship between stereotypes and behavioral
tendencies, and emotions will mediate the stereotypes→behaviors relationship.
We aim to develop and test an overarching framework for predicting differentiated types
of discriminatory treatment from the contents of stereotypes and the experience of emotions,
building on existing knowledge and also moving in new theoretical directions. We first develop
the behavioral tendencies scales in a preliminary study. Next we present a national correlational
study, and then two experiments that examine the hypothesized causal links. Last, we present a
correlational study that investigates the roles of primary emotions (i.e., anger and fear) in the
BIAS Map framework.
Preliminary Study: Developing Behavioral Tendencies Scales
We conducted a preliminary study to develop scales to measure the behavioral
tendencies. Drawing from a wide range of sources (e.g., Dijker et al. 1996; Roseman et al., 1994;
Weiner, 2005), we identified 31 items to represent an array of behaviors that could fall along the
two dimensions of active-passive and facilitation-harm: help, avoid, follow, compete with,
derogate, imitate, cooperate with, tolerate, assist, neglect, steal from, fight, demean, hinder,
undermine, unite with, accept, criticize, support, exclude, attack, abide by, endure, protect,
ignore, harass, associate with, lead, belittle, sabotage, and aggress against. Participants rated the
same 23 groups used in prior SCM work (Fiske et al., 2002, Studies 2 & 4): women, blue-collar
workers, elderly people, homeless people, young people, Blacks, Jews, Whites, welfare
recipients, Native Americans, educated people, retarded people, professionals, middle-class
people, Hispanic people, poor people, students, Asians, Muslims, gay men, Christians, rich
people, disabled people, and men. Groups were generated in pilot studies in which participants
were asked to list salient groups in American society (Fiske et al., 2002, Pilot Study).
In a classroom, 100 Princeton undergraduates (60% female) completed the questionnaire
along with several unrelated ones for $8 (US). To avoid fatigue, participants were randomly
assigned to rate 11 or 12 of 23 groups. Participants rated "…how you think most Americans
behave toward these groups," on a 5-point scale (1 = not at all; 5 = extremely).
Our hypotheses required the development of behavioral tendencies scales that worked for
each group separately and also overlapped across groups. Thus, we calculated 23 principal
components factor analyses (one per group) using direct oblimin rotation, examining all 31
response items; these yielded 4-7 factors with eigenvalues greater than 1.0. Across groups, four
similar factors emerged consistently; these formed the scales of active facilitation, active harm,
passive facilitation, and passive harm. Items that loaded onto the first factor, passive harm, were
demean, exclude, hinder, and derogate. For the second factor, passive facilitation, items were
cooperate with, unite with, and associate with. For the third factor, active harm, items were fight,
attack, and sabotage. For the fourth factor, active facilitation, items were assist, help, and protect.
Given the time constraints of a national telephone survey, we could choose only two items for
each scale, so we chose two of the three items with the highest average factor loadings: "help"
and "protect" for active facilitation; "fight" and "attack" for active harm; "cooperate with" and
"associate with" for passive facilitation; and "exclude" and "demean" for passive harm.
Study 1: Representative National Telephone Survey
We conducted a nationally representative, random-sample telephone survey to investigate
society's perceptions of how various naturally occurring social groups are perceived and treated
in the United States, testing the BIAS Map's three hypotheses. We aimed to extend existing
theory in several ways. First, including both active and passive, harmful and facilitative
behavioral tendencies in the same study allowed us to differentiate biases more thoroughly. This
made it possible to focus on groups treated ambivalently – those eliciting both harmful and
facilitative responses. Second, in this study we sought to identify bias clusters with distinctive
and qualitatively different patterns of stereotyping, emotions, and behaviors. Methodologically,
this study goes beyond previous research by testing the links among stereotypes, emotions, and
behaviors simultaneously across a broad range of naturally occurring social groups, which
provides a unique intergroup comparative context. Other work may apply theoretically to a broad
range of groups, but so far, most is limited empirically to only one or two (exceptions are
Alexander et al., 1999; and Cottrell & Neuberg, 2005). Also, using a representative national
sample allowed us to overcome the inherent limitations of college student samples in
investigations of societal biases, including our own previous research.
The Princeton University Survey Research Center administered the telephone survey in
the spring of 2003. The sample included English-speaking adults, 18 or older, in the 48
contiguous United States, whose households included at least one telephone. The survey used
nationwide random-digit dialing. Unscreened random telephone numbers in replicates of 100
were created using a method that generates a stratified sample frame of estimated telephone
households from blocks of exchanges containing three or more active telephones. Checking for
active telephones within block occurred prior to the randomization of the last four digits. Phone
numbers within that block were then attempted. If reached, an adult from each household was
selected randomly (adult with the "next birthday" was requested) and interviewed. The response
rate for eligible calls (e.g., residences, English-speaking, not fax, etc.) was 25%. Although this
rate is low, recent evidence indicates that low response rates do not invalidate the sample’s
substantive accuracy (e.g., Curtin, Presser, & Singer, 2000). And, as we do here, weighting can
correct demographic shortcomings. Completion rate for those who agreed to participate was
The total sample size was 571, of which 62% were female, and the average age was 43.5
years (SD = 17.6 years). Most participants (77%) were White; remaining percentages were 6%
Black, 9% Latino; 1.5% Asian or Pacific Islander; and 1.5% Native American. On education, 7%
had not finished high school; 24% had graduated from high school only; 30% had some college
background; 22% were college graduates; and 13% had completed an advanced degree. The
sample was 34% Protestant, 25% Catholic, 2% Jewish, 24% identified with a religion not listed,
and 15% agnostic or atheist. On annual household income, 24% reported less than $25,000, 31%
reported $25,000-$49,999, 18% reported $50,000 to $74,999, 14% reported $75,000 to $99,999;
13% reported greater than $100,000. On region, 20% were from the Northeast, 24% from the
Midwest, 36% from the South, and 21% from the West.
Data were weighted on gender, age, education, census region, and race/ethnicity to match
Census Bureau estimates of the proportion of English-speaking adults, aged 18 or older, residing
in the contiguous United States. The demographic weighting parameters came from a specific
analysis of the most recently available Census Bureau Annual Demographic File (March 2002
Current Population Survey). The weights were derived using an iterative technique that
simultaneously balanced the distributions of all weighting parameters. After an optimum sample-
balancing solution was reached, the weights were constrained to fall within the range of 1.00 to
7.17, ensuring that individuals did not inordinately affect overall results. Because the range of
weights produced an n greater than the actual sample n, an adjusted weight value (.34 to 2.43)
was used in all analyses of weighted data.
The questionnaire listed 20 U.S. social groups, chosen from previous studies (Fiske et al.,
2002; Fiske et al., 1999; Katz & Braly, 1933). We selected five groups likely to represent each of
the four quadrants of the competence-warmth space, resulting in a total of 20 groups, because the
focus of this research was documenting relationships among the BIAS Map variables (as
opposed to studying the contents of stereotypes of specific groups).
Each participant rated 4 of the 20 groups on 12 two-item scales (Appendix A), resulting
in a total of 24 ratings per group. The scales – perceived social structure, traits, emotions, and
behavioral tendencies – were, respectively: competitiveness and status (social structure),
competence and warmth (stereotypes), admiration, contempt, envy, and pity (emotions), and the
four behaviors – active harm, passive harm, active facilitation, and passive facilitation. All but
the behavioral tendencies scales were adapted from previously used scales, and each scale
included the two items with the highest average factor loadings across our previous studies
(Fiske et al., 2002, 1999). The four behavioral tendencies scales came from the preliminary
study. Using 5-point scales (1 = not at all; 5 = extremely), participants rated how the groups “are
perceived by Americans.” As before, this instruction was intended to assess perceived societal
reactions and to reduce participants’ social desirability concerns.
Participants completed the phone-administered questionnaire in approximately 17
minutes. Each participant rated 4 groups on 24 items, resulting in 96 ratings per participant. After
completing the social-groups questions, participants answered the demographic questions.
Analyses aimed to demonstrate that combinations of competence-warmth stereotypes and
related emotions associate with differentiated patterns of behavioral tendencies. All analyses
used the weighted data (described above). We found no systematic, significant effects of
participant sex or any other demographic variables (e.g., income, race, religion, sex).
Reliabilities for the two-item scales were: status α = .87; competitiveness α = .79; competence α
= .79; warmth α = .83; admiration α = .80; contempt α = .60; envy α = .82; pity α = .71; active
facilitation α = .60; active harm α = .59; passive facilitation α = .61; passive harm α = .68.
To be sure that our new emotions and behaviors items were distinct and not redundant
with each other, we conducted principal components factor analyses using varimax rotation on
the emotions and behaviors items. As in previous studies, we conducted a separate factor
analysis for each group examining the 16 emotions and behaviors items. Across groups, the
emotions and behaviors consistently loaded on separate factors. Also, in every case, the two
items included in the scale co-occurred more frequently than any other pairing.
First we present correlation and regression analyses of the hypothesized relationships
among (a) competence-warmth stereotypes and behavioral tendencies, and (b) emotions and
behavioral tendencies. These analyses also address the hypotheses that emotions trump
stereotypes in predicting behavioral tendencies and emotions mediate stereotypes Æ behavioral
We calculated correlations two ways. At the group level we averaged ratings across
participants for each of the 20 groups, and then calculated the correlation coefficients from the
group means. At the participant level we calculated correlations separately for each individual
participant (N = 571), converted them using Fisher's r to z, averaged them, and reverted them to
rs. Each procedure offers an advantage and a disadvantage. The group-level procedure uses a
smaller n, but stable means that mask participant-level variation, thus producing larger rs. The
participant-level procedure lacks stable means but provides more power. Together the estimates
bracket the true effect size (see Table 1)
Hypothesis 1: Stereotypes
We hypothesized that warmth stereotypes would predict the valence of active behaviors,
and competence stereotypes would predict the valence of passive behaviors. As expected,
warmth ratings correlated positively with active facilitation and negatively with active harm.
Competence ratings correlated positively with passive facilitation and negatively with passive
harm. The only unpredicted stereotypes-behaviors relationship to emerge at both the groups- and
participants-levels was between warmth and passive facilitation. In sum, correlations support all
four of the predicted stereotypes Æ behavioral tendencies relationships, in both group and
(see Table 1).
Hypothesis 2: Emotions
We hypothesized that differentiated emotions would predict distinct patterns of
behavioral tendencies (Table 1). Admired groups elicited both higher active facilitation and
higher passive facilitation ratings. Groups high on contempt elicited both active harm and
passive harm. Envied groups elicited higher passive facilitation ratings, but also higher active
harm ratings, although the envy Æ active harm correlation reached significance only at the
participants-level of analysis. Finally, pitied groups elicited higher active facilitation ratings, but
also higher passive harm ratings. In sum, correlations supported all eight of the specific emotions
Æ behaviors predictions at the participant level and seven of eight at the group level.
Hypotheses 1 & 2 Corollary: Bias Clusters
To distinguish the predicted bias clusters (coordinated stereotypes, emotions, and
behaviors), the second analyses compares common patterns of emotions and behavioral
tendencies for groups that share competence-warmth stereotypes. First, cluster analyses
identified the collections of groups with similar competence-warmth stereotypes, following the
same analytic procedure used in our SCM studies (Cuddy et al., 2006; Fiske et al., 2002). Results
pointed to a four-cluster solution, confirming our choice of the groups explicitly to represent the
four quadrants of the competence-warmth space. The clusters spread out in the two-dimensional
space, using both dimensions equally (see Figure 2 for cluster). Focused t-tests of a priori
predictions compared competence and warmth within clusters, and focused independent t-tests
compared competence and warmth between clusters. Results clearly confirmed the four clusters:
high-competence, high-warmth (HC-HW, e.g., middle-class); high-competence, low-warmth
(HC-LW, e.g., Asians); low-competence, high-warmth (LC-HW, e.g., elderly); and low-
competence, low-warmth (LC-LW, e.g., welfare recipients), all ps < .05.
Three groups moved into clusters adjacent to their locations in previous samples (Fiske et
al., 2002, 1999). For two (Black professionals, Whites), the movement reflects a shift in
clustering rather than big differences in ratings (from HC-LW and HC-HW, into adjacent
clusters HC-HW and HC-LW, respectively); change on 5-point scales were only .09-.17
(competence) and .11-.24 (warmth). The most striking difference from previous samples was the
migration of housewives from the LC-HW cluster to the HC-HW cluster. Compared to previous
studies (Fiske et al., 2002, 1999), housewives did not change on warmth, but gained 1.04 points
on competence. Analyses did not support the most obvious explanation – that this sample was
the first to represent housewives. Lacking occupation data, we compared male and female
responses and found no differences on either trait. The shifting standards model provides a more
likely explanation, describing how stereotypes provide references against which group members
are compared (e.g., Biernat & Vescio, 2002). A woman might be subjectively judged as more
financially successful than a man who objectively earns more money, because women are not
stereotyped as high wage-earners. Thus, “competence” for a housewife might shift from the
more typical meaning (e.g., paid work) to the household context (e.g., child-rearing).
Contrasts compared the four clusters on each emotion. Replicating prior findings (Fiske
et al., 2002, Study 4), more admiration went to HC-HW groups (M = 3.54) than to other clusters
(M = 2.62), t(16) = 4.67, p < .001. More contempt went to LC-LW groups (M = 2.65) than to
other clusters (M = 2.30), t(16) = 3.73, p < .01. More envy went to HC-LW groups (M = 2.76)
than to other groups (M = 1.71), t(16) = 3.92, p < .01. Finally, more pity went to LC-HW groups
(M = 3.31) than to other clusters (M = 2.21), t(16) = 5.09, p < .001. Thus, current data replicated
all of the previously established, fundamental links between competence-warmth stereotypes and
Each of the four competence-warmth stereotypes and its emotions was hypothesized to
carry a unique signature of behavioral tendencies (Figure 1). First, groups stereotyped as warm
were expected to receive more active facilitation than other groups. Indeed, groups in the two
high-warmth clusters (n = 9) did differ from groups in the two low-warmth clusters (n = 11),
t(16) = 6.46, p < .001. Groups stereotyped as lacking warmth were expected to receive more
active harm than other groups, which they did, t(16) = 2.98, p < .01. (See Table 2 for means.)
We hypothesized that groups stereotyped as competent would receive more passive
facilitation than other groups. As predicted, groups in the two high-competence clusters (n = 11)
differed from groups in the two low-competence clusters (n
= 9), t(16) = 5.32, p < .001. We also
hypothesized that groups stereotyped as lacking competence would receive more passive harm
than other groups, which they did, t(16) = 3.64, p < .01. In sum, supporting the Hypotheses 1-2
Corollary, the four predicted bias clusters emerged.
Hypothesis 3: Emotion Priority
Using regression analyses, we compared the relative boosts to the percent variance
explained when (a) adding the two predictor emotions to the predictor stereotype (i.e.,
competence or warmth) in predicting each behavioral tendency versus (b) adding the predictor
stereotype to the two predictor emotions in predicting each behavioral tendency (See Figure 1 for
a pictorial depiction of these hypotheses). For each behavioral tendency, adding the emotions to
the models significantly improved the R
(range of improvement to R
= .203 to .577, all Fs > 10,
ps < .001), but adding the stereotype did not (range of improvement to R
= .009 to .027).
A series of regressions examined the proposed dual-mediation (via the two predictor
emotions) of the effect of the predictor stereotype (i.e., competence or warmth) on each
behavioral tendency by: (a) regressing the behavioral tendency (the criterion) on the stereotype
(the predictor), (b) regressing the two emotions (the mediators) on the stereotype, and (c)
simultaneously regressing the behavioral tendency on the stereotype and the two hypothesized
emotions. In all analyses, we controlled for the non-predictor trait by including it in analyses; for
example, when testing for mediation by contempt and pity of the effect of perceived competence
on passive harm, we also included warmth as an independent variable. We calculated Sobel tests
to check for full mediation. Figure 3 presents the results of these analyses.
In all cases, at least one emotion significantly mediated the direct effect of the stereotype
on the behavioral tendency. For active facilitation, the pattern of results suggested that both
admiration and pity mediated the direct effect of warmth (Figure 3a). For active harm, the results
indicated that contempt mediated the direct effect of warmth (Figure 3b). For passive facilitation,
admiration mediated the direct effect of competence (Figure 3c). For passive harm, pity mediated
the direct effect of competence (Figure 3d).
In sum, as hypothesized, emotions more strongly predicted behavioral tendencies than
did stereotypes, and emotions generally mediated the stereotypes Æ behavioral tendencies link.
Results of our national sample survey document four hypothesized patterns of
discriminatory behavioral tendencies, based on competence-warmth stereotypes and related
emotions. These results converge with existing research in the following ways: (a) differentiated
biases, which included both negative and positive responses, stemmed from appraisals of groups
(e.g., Alexander et al., 1999; Cottrell & Neuberg, 2005); (b) the contents of stereotypes,
emotions, and behavioral tendencies were coordinated (e.g., Mackie et al., 2000); and (c)
emotions trumped stereotypes in predicting behavioral tendencies (e.g., Dovidio et al., 1996;
Esses & Dovidio, 2002).
Study 1 makes several new contributions. First, the findings provide theoretical and
empirical support for the significance of specific stereotype contents, namely competence and
warmth, in predicting specific discriminatory behavioral tendencies, active-passive, and harmful-
facilitative. Groups stereotyped as possessing warmth elicited more active facilitation and less
active harm than groups stereotyped as lacking warmth. Groups stereotyped as competent
elicited more passive facilitation and less passive harm than groups stereotyped as lacking
competence. Unexpectedly, stereotypically warm groups also elicited more passive facilitation
than stereotypically low-warmth groups, a finding that we discuss in greater detail in Study 2.
Study 1 also supports the hypothesized relationships between specific positive and
negative social emotions (admiration, contempt, envy, and pity) and unique patterns of
intergroup behavioral intentions. This is the first study to simultaneously link these four
theoretically-derived emotions to specific patterns of intergroup behavioral intentions.
Correlational data strongly supported seven of eight of the specific predicted links, but the envy
to active harm link was significant only at the individual level of analysis. Study 4 addresses this
The four combinations of competence-warmth stereotypes formed bias clusters, linking
with predicted patterns of emotions and behavioral tendencies. By including positive and
negative stereotypes and emotions, as well as active and passive, harmful and facilitative
behavioral tendencies in the same study, we were able to investigate ambivalent patterns of bias.
Indeed, the results support the hypothesized ambivalent bias clusters – those comprising mixed-
valence stereotypes, emotions, and behaviors. Groups stereotyped as high on competence but
low on warmth elicited envy and passive facilitation but active harm. Groups stereotyped as low
on competence but high on warmth, on the other hand, elicited pity and active facilitation but
Study 1 compared the relative strengths of stereotype and emotions in predicting
intergroup behavioral tendencies. Consistent with previous research (e.g., Dovidio et al., 1996;
Talaska et al., in press), in general, emotions more strongly and directly predicted behavioral
tendencies than did stereotypes. Following an appraisal Æ emotion Æ behavior sequence (e.g.,
Mackie et al., 2000), for each behavioral tendency, at least one emotion mediated the stereotype
Æ behavior link. However, some emotions took priority over others. Admiration fully mediated
the relationship between warmth stereotypes and active facilitation, and partially mediated the
relationship between competence stereotypes and passive facilitation. Contempt fully mediated
the relationship between warmth and active harm. And pity fully mediated the relationship
between competence stereotypes and passive harm. Only envy did not mediate any relationships
of stereotypes to behavioral tendencies (discussed below).
Although a strength of Study 1 is its support for the hypothesized relationships across a
range of real social groups and with a nationally representative sample of participants, its
correlational design prevents establishing causality. Studies 2 and 3 were designed to provide
experimental tests of the hypothesized causal relations between stereotypes and behavioral
intentions and between emotions and behavioral intentions.
Studies 2 and 3: Testing Causality of BIAS Map Links
Although Study 1 correlations support the BIAS Map, they did not test the hypothesized
causal relations. Studies 2 and 3 test causality of links between stereotypes and behavioral
tendencies (Hypothesis 1), and emotions and behavioral tendencies (Hypothesis 2).
To test the hypotheses more cleanly, we held constant the target group, varying only
competence and warmth stereotypes (Study 2) and the emotions elicited by the group (Study 3).
Both experiments described a fictitious ethnic group expected to immigrate soon in large
numbers to the United States. Study 2 manipulated the extent to which the immigrant group was
allegedly perceived as competent or incompetent, and warm or not warm, in their society of
origin. Study 3 manipulated the distinct emotions (admiration, contempt, envy, pity) elicited by
the immigrant group in their society of origin. Participants responded to longer versions of
behavioral tendencies scales used in Study 1.
Study 2: Causal Test of Hypothesis 1(Stereotypes to Behavioral Tendencies)
Participants. One-hundred fifty Princeton University undergraduates (59% female)
voluntarily completed the questionnaire as part of a larger packet. Participant sex had no effect.
Questionnaire and Procedure. The questionnaire described a fictitious ethnic group
expected to immigrate to the United States in the near future. The 2 x 2 between-subjects design
manipulated two perceived traits of the immigrant group: warmth (high/low) and competence
(high/low). Participants were randomly assigned to one of four conditions, and read:
“Due to political and economic circumstances, demographers predict waves of
immigration in the next few years from an ethnic group outside our borders called
Wallonians. Members of this group are viewed by their society as competent (or
incompetent) and intelligent (or unintelligent), and as warm (or not warm) and good-
natured (or not good-natured). When people of this ethnic group arrive, to what extent
will people here behave in each of the following ways toward them?”
Using Likert-type scales (1 = extremely unlikely to 7 = extremely likely), participants rated four
3-item behavioral tendencies scales: active facilitation (assist, help, protect), active harm (attack,
fight, harass), passive facilitation (associate with, cooperate with, unite with), and passive harm
(exclude, ignore, neglect).
Results and Discussion
Participants' responses to all four 3-item scales were reliable, active facilitation α = .84,
active harm α = .82, passive facilitation α = .74, passive harm α = .71, so responses to the three
items for each scale were averaged, resulting in four scale means. Because of main effect
variations in the degree to which participants will endorse the different behavioral tendencies
(e.g., participants seem more comfortable endorsing active facilitation than active harm, across
conditions), resulting in significantly different behavioral tendencies means, the means were
standardized to Z scores for comparison across groups, regardless of endorsement baselines.
Active Behaviors. To test whether warmth affected the valence of active behaviors, we
entered the behavior ratings into a 2 (Competence: high vs. low) × 2 (Warmth: high vs. low) × 2
(Active behavior valence: facilitate vs. harm) ANOVA, with repeated measures on the last
factor. Results revealed a significant Warmth × Active behaviors interaction, F(1, 146) = 30.93,
p < .001, η
= .18. There were no main effects. Planned comparisons helped to further interpret
the interaction. As expected, high warmth groups elicited more active facilitation than low
warmth groups, F(1, 148) = 22.37, p < .001, η
= .13; and low warmth groups elicited more
active harm than high warmth groups, F(1, 148) = 18.86, p < .001, η
= .11. The relevant
means are presented in Table 2.
Passive Behaviors. To test whether competence affected the valence of passive
behaviors, we entered the behavior ratings into a 2 (Competence: high vs. low) × 2 (Warmth:
high vs. low) × 2 (Passive behavior valence: facilitate vs. harm) ANOVA, with repeated
measures on the last factor. The Competence × Passive behaviors interaction was significant,
F(1, 146) = 26.00, p < .001, η
= .15. There were no main effects.
Planned comparisons revealed that competent groups elicited more passive facilitation
than incompetent groups, F(1, 148) = 17.71, p < .001, η
= .11; and incompetent groups
elicited more passive harm than competent groups, F(1, 147) = 19.47, p < .001, η
Results also revealed a significant Warmth × Passive behaviors interaction, F(1, 146) = 14.15, p
< .001, η
= .09. High warmth groups elicited more passive facilitation (M = .293) than low
warmth groups (M = -.278), F(1, 148) = 13.23, p < .001, η
= .08; and low warmth groups
elicited slightly more passive harm (M = .169) than high warmth groups (M = -.160), F(1, 148) =
4.15, p = .043, η
Replicating an unpredicted Study 1 finding, warmth also increased passive facilitation
tendencies and decreased passive harm tendencies, although the former effect was much larger
= .09) than the latter (η
= .02). The relationship is not entirely surprising, and is
consistent with research on the drive to affiliate with similar (i.e., liked) others (e.g., Newcomb,
1956). However, it also could have resulted from our broad operationalization of passive
facilitation (e.g., associating, uniting), which may have been interpreted by some participants as
more communal than agentic. Nonetheless, effect sizes in both Studies 1 and 2 consistently show
competence to be the stronger predictor than warmth of both passive behaviors.
Study 2 supports causal links between competence and warmth stereotypes and,
respectively, active and passive behavioral tendencies. These results fit the general notion that
cognitive appraisals predict action tendencies (Mackie et al., 2000), but tailored to our two-
dimensional space .They go beyond the experimental specification of distinct cognitive images
(Alexander et al., 1999), by adding behavioral tendencies. The next study examines emotion-
Study 3: Causal Test of Hypothesis 2 (Emotions
Study 2 provided a causal test of Hypothesis 1 (the proposed relationships between
societal competence-warmth stereotypes and behavioral tendencies) and therefore did not
manipulate the emotions. Study 3 provides a causal test of Hypothesis 2, that the four
qualitatively distinct emotions (i.e., admiration, envy, pity, contempt) associated with the four
competence-warmth stereotypes predict specific combinations of behavioral tendencies.
Two hundred Princeton undergraduates (63% female) completed the questionnaire in
small group sessions in exchange for course credit. Participant sex had no effects.
The questionnaire was the same as the Study 2 questionnaire, describing a fictitious
group expected to soon immigrate to the United States. The four-cell between-subjects design
manipulated the type of emotion (admiration, envy, contempt, pity) the group allegedly elicited
from others in their native society. Participants were randomly assigned to condition and read:
“…Members of this group are generally admired/envied/hated
/pitied by others in their
society. When people of this ethnic group arrive, to what extent will people here behave
in each of the following ways toward them?”
Participants rated the same four 3-item behavioral tendencies scales as in Study 2.
Results and Discussion
Participants' responses to all four 3-item scales were reliable, active facilitation α = .91,
active harm α = .86, passive facilitation α = .83, passive harm α = .72. As in Study 2, responses
to the three items for each scale were averaged, then standardized to Z scores for analyses.
We sought to demonstrate that each emotion causes a unique pattern of behavioral
tendencies. Specifically, we predicted admiration would increase both active and passive
facilitation; contempt would increase both active and passive harm; envy would increase passive
facilitation and active harm; and pity would increase active facilitation and passive harm.
We conducted a 4 (Emotion: admire, envy, hate, pity) × 4 (Behavior: active and passive
facilitation and harm) ANOVA on the behavior ratings, with repeated measures on the behavior
factor. The analysis revealed a significant Emotion × Behavior interaction, which supported the
general hypothesis that distinct intergroup emotions lead to unique patterns of behavioral
tendencies, F(9, 546) = 17.94, p < .001, η
= .23. There were no main effects.
Contrasts tested more focused predictions for each of the four behavioral tendencies. For
each behavioral tendency, we assigned weights of 1 to both of the putative predictor emotions,
and weights of -1 to both of the non-predictor emotions. For example, for active facilitation as a
DV, we assigned weights of 1 to admiration and pity, and weights of -1 to contempt and envy.
As predicted, admiration and pity elicited higher active facilitation (Ms = .456 and .403,
respectively), compared to contempt and envy (Ms = -.654 and -.218, respectively), t(182) =
6.58, p < .001. Contempt and envy elicited higher active harm (Ms = .556 and .115,
respectively), compared to admiration and pity (Ms = -.432 and -.246, respectively), t(182) =
5.01, p < .001. Admiration and envy elicited higher passive facilitation (Ms = .748 and .282,
respectively), compared to contempt and pity (Ms = -.815 and .217, respectively), t(196) = 8.86,
p < .001. Contempt and pity elicited higher passive harm (Ms = .551 and .046, respectively),
compared to admiration and envy (Ms = -.544 and -.053, respectively), t(196) = 4.54, p < .001.
The data thus supported the hypothesized causal links between each group’s typical
emotion and the behavioral tendencies toward that group, replicating the four different patterns
of behavioral tendencies documented in Study 1. Active facilitation was higher for admired and
pitied groups, compared with envied and hated groups, who elicited higher active harm. Passive
facilitation was higher for admired and envied groups, compared with hated and pitied groups,
who elicited higher passive harm. Effects for the ambivalent emotions, envy and pity, were
weaker than effects for the univalent emotions, admiration and contempt, albeit all significantly
followed the hypothesized patterns. This study fits several previous contributions but goes
beyond each: It fits IET’s emotions-behavior link (Mackie et al., 2000) and provides some
specific examples based on our framework; it also fits the functional idea that emotion enters
into intergroup behavior (Alexander et al., 1999) and specifies which emotions predict which
behaviors; it likewise fits the socio-functional idea that specific emotions resulting from threat
will predict approach-avoidance (Cottrell & Neuberg, 2005) and specifies differentiated emotion
links to differentiated behavior. Finally, it goes beyond the SCM specification of social structure
leading to stereotypes and emotions (Fiske et al., 2002; Fiske et al., 1999), by linking the
emotions to behavioral tendencies. Thus, these compatible results integrate previous intergroup
Study 4: Anger and Fear in the BIAS Map
All of the BIAS Map emotions considered thus far – admiration, contempt, envy, and pity
– are secondary, or “uniquely human,” emotions (Demoulin et al., 2005). But much of the
existing intergroup emotions literature research has focused upon the primary (i.e., “non-
uniquely human”) emotions of anger and fear (e.g., Dijker et al., 1996; Mackie et al., 2000).
Because one goal of this work is to integrate prior research on intergroup emotions, we
conducted a fourth study to examine the roles of these more primary emotions in the BIAS Map
framework. Additionally, appending these more basic emotions might clarify the relatively weak
link between envy and active harm, as discussed below.
Both anger and fear are activated by the perception that another person or group is, in
some way, unfriendly. Anger is elicited by the perception that another’s behavior is unfair (i.e.,
immoral; see Frijda et al., 1989) and by appraisals of unwelcome competition (i.e., low warmth)
from outgroups (Alexander et al., 1999; Mackie et al., 2000). Fear is elicited by perceived threat
(i.e., low warmth) from another individual (Frijda et al., 1989) or outgroup (e.g., Stephan &
Stephan, 2000). In short, fear and anger occur toward groups viewed as hostile. That the warmth
dimension alone is likely to drive the primary emotions of anger and fear is consistent with the
evidence already presented concerning the primacy of the warmth dimension. Therefore, we
hypothesize that, regardless of perceived competence, groups perceived as lacking warmth
(compared to groups perceived as warm) will be more likely to elicit anger and fear.
If anger and fear are driven by the warmth dimension in intergroup perception, then these
primary emotions may predict active, rather than passive behaviors. Past research indeed
suggests that anger leads to antagonistic and offensive actions toward others, such as verbal or
physical assault, but not to defensive or passive actions, such as neglecting or ignoring (Dijker et
al., 1996; Frijda et al., 1989; Mackie et al., 2000). So, we hypothesize that anger will correlate
positively with active harm and negatively with active facilitation, but will not correlate with
either of the passive behaviors. Although fear has been theoretically linked to defensive action
tendencies toward others (Frijda et al., 1989; Mackie et al., 2000), such as avoiding and
excluding, empirically this link has received mixed support (Mackie et al., 2000; see also Devos
et al., 2002). Thus, we were agnostic about the relationship of fear to the behavioral tendencies.
Our final prediction identifies anger as a possible mediator of the relatively weak
relationship of envy to active harm. Envy has been linked to anger (Hareli & Weiner, 2002), and
as discussed, anger leads to offensive actions toward others. Envy may elicit active harm only
when a society is under great stress, or under circumstances that heighten intergroup competition
(Glick, 2002, 2005; Staub, 1996), which thereby increase anger. In particular, Glick (2002, 2005)
has suggested that when a society experiences difficult life conditions (Staub, 1996), groups
perceived as competent competitors (i.e., envied groups) are most likely to be scapegoated. For
example, the Nazis viewed the Jews as powerful, competent manipulators who had engineered
Germany’s defeat in World War I and the subsequent economic crisis. In Rwanda, the Tutsi, also
a high-status minority, were similarly blamed for the nation’s economic problems. Active harm
(at its most extreme, genocidal attack) can be justified and motivated when an outgroup is
viewed as a powerful and competent competitor or exploiter. We therefore hypothesize that
anger, sometimes elicited by the circumstances just described, mediates the link between envy
and active harm.
We used a similar methodology to that used in Study 1, although the questionnaire was
administered by computer, not by telephone interview. Forty-two Rutgers University
undergraduates (62% female) rated a list of eight groups (Asians, disabled, elderly, homeless,
middle-class, rich, students, welfare recipients) presented in random order, on a total of 30 items
measuring: (a) competence and warmth (single items); (b) admiration, contempt, envy, and pity;
(c) anger (angry, mad) and fear (afraid, anxious)
; and (d) active facilitation, active harm, passive
facilitation, and passive harm. Using 5-point scales (1 = not at all; 5 = extremely), participants
rated how the groups “are perceived by Americans.” One participant was omitted for answering
fewer than 50% of the questions. The anger and fear scales were new, but the other scales were
derived from Studies 1-3, with one exception: “resentful,” identified by emotion theorists as a
critical component of envy (e.g., R. Smith et al., 1996), was added to the envy scale.
Results & Discussion
Scale reliabilities follow: admiration α = .79, contempt α = .77, envy α = .86, pity α = .87,
anger α = .92, fear α = .71, active facilitation α = .86, active harm α = .83, passive facilitation α =
.86, and passive harm α = .87.
Our analyses focused on the role of the primary emotions, anger and fear, in the BIAS
Map framework, so we report only results relevant to those predictions. As hypothesized,
warmth correlated negatively with both anger (participant r = -.43, p < .01; group r = -.58, p =
.12) and fear (participant r = -.48, p < .01; group r = -.66, p < .08). Also as expected, competence
did not correlate with either anger or fear at the group-level (both ps > .50), and correlated only
slightly with fear (r = -.15, p = .05) but not with anger (p > .50) at the participant-level.
We next examined correlations between the new emotions and the behavioral tendencies.
As hypothesized, anger correlated negatively with active facilitation (participant r = -.40, p <
.05; group r = -.82, p = .01) and positively with active harm (participant r = .64, p < .01; group r
= .93, p = .001). Fear correlated positively with active harm (participant r = .40, p < .05; group r
= .68, p = .08), but did not correlate with active facilitation. Neither anger nor fear correlated
with the passive behaviors at the group-level, ps > .60 and .30, respectively. However, at the
participant-level, anger and fear did correlate with passive facilitation (rs = -.23 and -.31, ps <
.05, respectively) and passive harm (rs = .22 and .29, respectively, ps < .05).
Our next set of analyses involved showing that the relationship of envy to active harm
would be mediated by anger. A series of analyses regressed (a) active harm (the criterion) onto
envy (the predictor); (b) anger (the mediator) onto envy; and (c) simultaneously active harm onto
both anger and envy. Figure 4 presents the results of those analyses. As predicted, anger fully
mediated the envy to active harm relationship.
We also conducted a post-hoc investigation of the possibility that competence might have
moderated the effects of warmth on fear and anger, such that people may have experienced more
anger toward low-warmth, low-competence groups and more fear toward low-warmth, high-
competence groups. These links have been suggested by appraisal theories, which contend that
anger is elicited by the perception that the self is stronger or more powerful (i.e., more
competent) than a threatening (i.e., not warm) other, while fear is elicited by the perception that
the self is weaker or less powerful (i.e., less competent) than a threatening other (Frijda et al.,
1989; Mackie et al., 2000)
. Post-hoc analyses of the present data do not support such a pattern.
First, correlations between warmth and anger and warmth and fear did not differ for high-
competence groups (rs = -.80 and -.71, respectively) versus low-competence groups (rs = -.76
and -.68, p = .32, respectively). Second, high-competence and low-competence groups did not
differ on anger (Ms = 2.12 and 1.89, respectively) or fear (Ms = 2.00 and 2.01, respectively),
both ts < 1, both ns. Although competence did not moderate the effects of warmth on fear and
anger in the present study, these analyses were post-hoc and should be interpreted with caution.
Results from Study 4 suggest that the inclusion of two relatively primary emotions –
anger and fear – adds valuable information to the BIAS Map framework. As predicted, both
anger and fear correlated with warmth, but not competence (except for a small negative
correlation between fear and competence at the participant-level). Anger correlated with both of
the active behaviors; fear correlated with active harm but not active facilitation. Anger and fear
did not correlate with either of the passive behaviors at the group-level, but did correlate with the
passive behaviors at the participant-level. Perhaps most importantly, anger fully mediated the
relationship of envy to active harm, which helps to resolve the weak relationship between these
variables in Studies 1 and 3. Overall, these results fit previous intergroup emotions research by
identifying roles for primary emotions, fear and anger (e.g., Cottrell & Neuberg, 2005; Mackie et
al., 2000); these findings integrate that prior work within the overall framework.
Together these four studies address a fundamental question in the psychology of
intergroup relations: How do stereotypes and emotions shape behavioral tendencies toward
groups? We identify specific patterns of stereotypic traits, distinct emotions, and related
behavioral responses. Grounded in the structure of intergroup relations, the BIAS Map provides
an integrative theoretical approach that: (a) identifies underlying dimensions of intergroup
behavior (active/passive, facilitative/ harmful), and their roots in (b) dimensions of stereotypes
(competent/incompetent, warm/cold) and (c) corresponding discrete emotions; and (d) identifies
both univalent and ambivalent clusters of stereotypes, emotions, and discriminatory behaviors.
Lacking the distinctions among stereotype traits, specific social emotions, and dimensions of
behavior, past research may have underestimated the relationships among stereotypes, emotions,
This work is unique in its theoretical focus on stereotype trait dimensions as significantly
determining the nature of discriminatory treatment. The BIAS Map theoretically links the
proposed dimensions of behavior to the two traits that consistently emerge as the most central in
social perception – competence and warmth. This allows us to separate cognitive appraisals of
structural relations (i.e., perceived status and competitiveness) from cognitive beliefs about a
group’s traits (i.e., competence and warmth), in turn linking both to behavioral tendencies.
Although both appraisals and stereotypes are cognitive, one likely precedes the other.
While stereotypes affected intergroup behavioral tendencies, the relationship of
stereotypes to behavioral tendencies was typically indirect, mediated by emotions. Consistent
with earlier SCM research, competence and warmth combined to produce distinct intergroup
emotions. The emotions, in turn, were strongly related to distinct behavioral tendencies and
(either partially or more often fully) mediated the stereotype Æ behavioral tendency links.
The BIAS Map focuses not on personal stereotypes, but on stereotypes as culturally
shared knowledge. Even when individuals personally reject stereotypes that are prevalent in their
cultures, they know and often cannot help be affected by them. Thus, the study of cultural
stereotypes has gained momentum in recent years (e.g., Devine, 1989; Jost, Pelham, & Carvallo,
2002; Glick & Fiske, 2001). For example, aversive racism theory proposes that exposure to
cultural stereotypes leads White people who genuinely desire to be egalitarian nonetheless to
have automatic negative associations with Black people (Gaertner & Dovidio, 1986). In other
words, exposure to (even without endorsement of) cultural stereotypes considerably affects
reactions to outgroups. Analyses of demographic sub-groups from Study 1’s representative
national sample indicate high agreement about the contents of the stereotypes, emotions, and
behavioral tendencies toward a range of salient social groups, regardless of the social location of
the perceivers’ group. Altogether, the present research illuminates an apparently consistent
representation of the contents of emotional prejudices and intergroup behaviors elicited by
stereotypic competence and warmth. Still, societal prejudices do not always equal personal
prejudices. We do not yet know how the perspective of the perceiver will affect the BIAS Map’s
relationships at the personal level, a central question for future research.
The relationships among competence-warmth stereotypes, specific emotions, and
intergroup behaviors may represent a lay theory of cultural bias. Personal lay theories are
organized knowledge structures that interpret people’s social worlds, significantly helping to
direct their social behaviors (e.g., Heider, 1958; Hong, Levy, & Chiu, 2001). Recent analyses
have examined the role of lay theories in group perception (e.g., Hong et al., 2001; Yzerbyt &
Rocher, 2002). If the BIAS Map represents a cultural lay theory, then the links could be activated
from any point in the sequence. For example, manipulating the behavioral tendencies might
activate the linked competence-warmth stereotypes and discrete emotions. Certainly, this
possibility in no way rules out the idea that the BIAS Map also reflects real intergroup
phenomena. In fact, given that lay theories often direct social behaviors, the BIAS Map might
both reflect and shape intergroup phenomena. Again, this is a focal question for future research.
Study 4 added anger and fear to the BIAS Map framework. Both emotions were strongly
linked to perceptions of low warmth, regardless of perceived competence, and to the active
behavioral tendencies. These links are consistent with the notion that the primary emotions of
anger and fear are driven by the perceived friendliness or hostility of groups, which we have
argued to have greater primacy than perceived competence. Exactly how such primary emotions
fit into the SCM is an important matter for future investigation. Perhaps the most important
contribution of Study 4 is clarification of the relationship of envy to active harm, showing that it
is mediated by anger. Future research will be needed to identify and understand in more detail
the conditions under which envy transforms into anger. Glick’s (2002, 2005) model of
scapegoating offers one possibility: that envied groups elicit anger when they are believed to
have intentionally caused harm to the rest of society.
Directions of Future Research
Although the current study suggests that each ambivalent prejudice (envious and
paternalistic) can potentially produce helpful or harmful behavioral responses, it does not
identify when one or the other will be triggered. Which pole of the ambivalence guides responses
to groups in the envied (HC-LW) and pitied (LC-HW) clusters may depend on which stereotypic
dimension activates. For example, if their putative lack of competence is salient, pitied groups
may evoke passive harm (avoid, demean), but if their warmth is salient, they may elicit active
facilitation (help, protect). Notably, however, even when the ostensibly “positive” pole of an
ambivalent bias is activated, the consequences may not be wholly beneficial. Active facilitation
promoted by pity and stereotypic incompetence includes over-helping or over-protectiveness,
which implicitly reinforce a pitied group’s lower status.
Situational context may also play an important part in determining whether the positive
or negative pole of an ambivalent bias is activated. For example, a professional context likely
primes competence. In an experiment comparing behavioral intentions toward consultants
(female/male, parents/not parents) at a high-status firm (i.e., a professional context), participants
expressed significantly more passive harm (i.e., failure to hire, promote, or train) toward the
mom (Cuddy et al., 2004). Moreover, competence ratings negatively related to the passive harm
intentions. In a context that makes salient moms’ stereotypic warmth (e.g., an elementary school
function), the mom may be “preferred” (e.g., offered a better seat). Similarly, Hebl, Kazama,
Singletary, and Glick (2005) found that apparently pregnant (vs. nonpregnant) women were
treated with greater benevolence when posing as store customers, but greater hostility when they
posed as job applicants. Stereotype priming might have direct effects or (as the current data
suggest) be mediated by emotions.
If emotions more directly determine behavior, situational factors that prime the positive
versus the negative components of the emotions toward targets of ambivalent prejudice could
have powerful effects on behavior. The SCM suggests that the underlying questions that
determine people’s reactions to other groups are whether they are perceived as friend or foe and
capable of helping or harming one’s own group. Situations that prime inclusive orientations
toward target groups (e.g., as a friend) may elicit positive, and situations that prime an exclusive
orientation (e.g., identity politics) may elicit negative, emotional responses to targets of
ambivalent prejudice. For example, Allport (1954) describes a veteran’s admiration toward the
Jewish lieutenant of his platoon, who “took good care of his men” and was adept at getting
scarce supplies; “‘That’s the Jew in him—he was good at getting things like that’” (p. 191, italics
in original). Because the Jewish lieutenant’s stereotypical cleverness (i.e., competence) benefited
a common in-group (the platoon), he elicited subjectively positive emotions and facilitative
behavior from a biased perceiver. Probably, in an exclusionary context (e.g., competition for
civilian jobs), the biased perceiver would exhibit negative emotions and behaviors toward Jews.
On a methodological note, Studies 2 and 3 may suffer external validity shortcomings
inherent to most scenario studies. We chose to use the scenarios to isolate the effects of the
predictor traits and emotions on the behavioral tendencies, stripping away potential confounds of
pre-existing beliefs about real groups. For similar reasons other intergroup researchers have also
used scenario studies (e.g., Alexander et al., 1999; Castano & Giner-Sorolla, 2006). However,
the tradeoff gain on control can come at the cost of external validity. Future studies should
address this issue by manipulating the critical information in a different format, such as a
For targets of bias, it is the behavioral consequences (i.e., discriminatory treatment) of
group stereotypes and emotions that count. The BIAS Map charts how a group’s location in the
competence-warmth map of stereotypes predicts the “bias climate” that group is likely to
experience. Specifically, competence-warmth stereotypes and four distinct patterns of emotions
(admiration, pity, envy, and contempt) predict facilitative versus harmful and active versus
passive behavioral tendencies. The map provided here sketches a general structure, for which
some details (e.g., factors that elicit the positive versus negative response potentials of
ambivalent prejudices) remain to be filled in by further investigation. If the general framework is
sound, however, the blueprint offered here differentiates distinctive patterns of discriminatory
behavioral tendencies across a broad spectrum of groups, offering new insight into how
stereotypes and emotions relate to discriminatory behaviors.
Alexander, M. G., Brewer, M., & Hermann, R. (1999). Images and affect: A functional analysis
of out-group stereotypes. Journal of Personality and Social Psychology, 77, 78-93.
Allport, G. W. (1954). The nature of prejudice. Reading, MA: Addison-Wesley.
Ayduk, O., May, D., Downey, G., & Higgins, E. T. (2003). Tactical differences in coping with
rejection sensitivity: The role of prevention pride. Personality and Social Psychology
Bulletin, 29, 435-448.
Batson, C. D. (1998). Altrusim and prosocial behavior. In D. T. Gilbert, S. T. Fiske, & G.
Lindzey (Eds.), Handbook of social psychology (4
ed., Vol. 2, pp. 282-316). Boston:
Biernat, M., & Vescio, T. K. (2002). She swings, she hits, she's great, she's benched:
Implications of gender-based shifting standards for judgment and behavior. Personality
and Social Psychology Bulletin, 28, 66-77.
Brewer, M. B., & Alexander, M. G. (2002). Intergroup emotions and images. In D. M. Mackie &
E. R. Smith (Eds.), From prejudice to intergroup emotions: Differentiated reactions to
social groups (pp. 209-225). New York: Psychology Press.
Buss, A. H. (1961). The psychology of aggression. New York: Wiley.
Castano, E., & Giner-Sorolla, R. (2006). Not quite human: Infra-humanization as a response to
collective responsibility for intergroup killing. Journal of Personality and Social
Psychology, 90, 804-819.
Clausell, E., & Fiske, S. T. (2005). When do the parts add up to the whole? Ambivalent
stereotype content for gay male subgroups. Social Cognition, 23, 157-176.
Cottrell, C. A., & Neuberg, S. L. (2005). Different emotional reactions to different groups: A
sociofunctional threat-based approach to "prejudice." Journal of Personality and Social
Crick, N. R., & Grotpeter, J. K. (1995). Relational aggression, gender, and social-psychological
adjustment. Child Development, 66, 710-722.
Cuddy, A. J., & Fiske, S. T. (2002). Doddering, but dear: Process, content, and function in
stereotyping of older persons. In T. D. Nelson (Ed.), Ageism (pp. 3-26). Cambridge, MA:
Cuddy, A. J. C., Fiske, S. T., & Glick, P. (2004). When professionals become mothers, warmth
doesn't cut the ice. Journal of Social Issues, 60, 701-718.
Cuddy, A. J. C., Fiske, S. T., Kwan, V., Glick, P., Demoulin, S., Leyens, J-Ph., Bond, M. H., et
al. (2006). Is the Stereotype Content Model culture-bound? Cross-cultural universalities
and differences of stereotyping principles. Invited revision under review at British Journal
of Social Psychology.
Cuddy, A. J. C., Norton, M. I., & Fiske, S. T. (2005). This old stereotype: The pervasiveness and
persistence of the elderly stereotype. Journal of Social Issues, 61, 265-283.
Curtin, R., Presser, S., & Singer, E. (2000). The effects of response rate changes on the index of
consumer sentiment. Public Opinion Quarterly, 64, 413-428.
Demoulin, S., Leyens, J-Ph., Paladino, M-P., Rodriguez-Torres, R., Rodriguez-Perez, A., &
Dovidio, J. F. (2004). Dimensions of “uniquely” and “non-uniquely” human emotions.
Cognition and Emotion, 18, 71-96.
Devine, P. (1989). Stereotypes and prejudice: Their automatic and controlled components.
Journal of Personality and Social Psychology, 56, 5-18.
Devos, T., Silver, L. A., Mackie, D. M., & Smith, E. R. (2002). Experiencing intergroup
emotions. In D. M. Mackie & E. R. Smith (Eds.), From prejudice to intergroup emotions:
Differentiated reactions to social groups (pp. 111-134). New York: Psychology Press.
Dijker, A. J., Koomen, W., van den Heuvel, H., & Frijda, N. H. (1996). Perceived antecedents of
emotional reactions in inter-ethnic relations. British Journal of Social Psychology, 35, 313-
Dovidio, J. F., Brigham, J. C., Johnson, B. T., & Gaertner, S. L. (1996). Stereotyping, prejudice,
and discrimination: Another look. In C. N. Macrae, C. Stangor, & M. Hewstone (Eds.),
Stereotypes and stereotyping (pp. 276-319). New York: Guilford.
Eagly, A. H., Johannesen-Schmidt, M. C., & van Engen, M. L. (2003). Transformational,
transactional, and laissez-faire leadership styles: A meta-analyses comparing women and
men. Psychological Bulletin, 129, 569-591.
Eckes, T. (2002). Paternalistic and envious gender stereotypes: Testing predictions from the
stereotype content model. Sex Roles, 47, 99-114.
Esses, V. M., & Dovidio, J. F. (2002). The role of emotions in determining willingness to engage
in intergroup contact. Personality and Social Psychology Bulletin, 28, 1202-1214.
Esses, V. M., Dovidio, J. F., Jackson, L. M. & Armstrong, T. L. (2001). The immigration
dilemma: The role of perceived group competition, ethnic prejudice, and national
identity. Journal of Social Issues, 57, 389-412.
Fiske, S. T. (1998). Stereotyping, prejudice, and discrimination. In D. T. Gilbert, S. T. Fiske, &
G. Lindzey (Eds.), Handbook of social psychology (4
ed., Vol. 2, pp. 357-411). Boston:
Fiske, S. T., & Cuddy, A. J. C. (2006). Stereotype content across cultures as a function of group
status. In S. Guimond (Ed.), Social comparison processes and levels of analysis. New
York: Cambridge University Press.
Fiske, S. T., Cuddy, A. J. C., & Glick, P. (2002). Emotions up and down: Intergroup emotions
result from perceived status and competition. In D. M. Mackie & E. R. Smith (Eds.). From
prejudice to intergroup emotions: Differentiated reactions to social groups ( pp. 247-264).
Philadelphia, PA: Psychology Press.
Fiske, S. T., Cuddy, A. J. C., & Glick, P. (in press). First judge warmth, then competence:
Fundamental social dimensions. Trends in Cognitive Science.
Fiske, S. T., Cuddy, A. J. C., Glick, P. S., & Xu, J. (2002). A model of (often mixed) stereotype
content: Competence and warmth respectively follow from perceived status and
competition. Journal of Personality and Social Psychology, 82, 878-902.
Fiske, S. T., Xu, J., Cuddy, A. J. C., & Glick, P. (1999). (Dis)respecting versus (dis)liking: Status
and interdependence predict ambivalent stereotypes of competence and warmth. Journal of
Social Issues, 55, 473-489.
Frijda, N. H., Kuipers, P., & ter Schure, E. (1989). Relations among emotion, appraisal, and
emotional action readiness. Journal of Personality and Social Psychology, 57, 212-228.
Gaertner, S. L., & Dovidio, J. F. (1986). The aversive form of racism. In S. L. Gaertner & J. F.
Dovidio (Eds.). Prejudice, discrimination, and racism. (pp.61-89). San Diego: Academic.
Geen, (1998). Aggression and antisocial behavior. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.),
Handbook of social psychology (4
ed., Vol. 2, pp. 317-356). Boston: McGraw-Hill.
Glick, P. (2002). Sacrificial lambs dressed in wolves' clothing: Envious prejudice, ideology, and
the scapegoating of Jews. In L. S. Newman & R. Erber (Eds.) Understanding genocide:
The social psychology of the Holocaust. (pp. 113 142). London: Oxford University Press.
Glick, P. (2005). Choice of scapegoats. In J. F. Dovidio, P. Glick, & L. Rudman (Eds.),
Reflecting on the nature of prejudice. (pp. 244-261). Malden, MA: Blackwell Press.
Glick, P., & Fiske, S. T. (2001). Ambivalent stereotypes as legitimizing ideologies:
Differentiating paternalistic and envious prejudice. In B. Major & J. Jost (Eds.), The
psychology of legitimacy: Emerging perspectives on ideology, justice, and intergroup
relations (pp. 278-306). New York: Cambridge University Press.
Green, J. D., & Sedikides, C. (1999). Affect and self-focused attention revisited: The role of
affect orientation. Personality and Social Psychology Bulletin, 25, 104-119.
Hareli, S., & Weiner, B. (2002). Dislike and envy as antecedents of pleasure at another's
misfortune. Motivation and Emotion, 26(4), 257-277.
Hebl, M. R., Kazama, S. M., Singletary, S. L., & Glick, P. (2005). Hostile and benevolent
behavior toward pregnant women: Interpersonal discrimination in a retail setting. Rice
University. Manuscript submitted for publication.
Heider, F. (1958). The psychology of interpersonal relations. Oxford, England: Wiley.
Hong, Y. Y., Levy, S. R., & Chiu, C. Y. (2001). The contribution of the lay theories approach to
the study of groups. Personality and Social Psychology Review, 5, 98-106.
Jost, J. T., Pelham, B. W., & Carvallo, M. R. (2002). Non-conscious forms of system
justification: Implicit and behavioral preferences for higher status groups. Journal of
Experimental Social Psychology, 38, 586-602.
Judd, C. M., Hawkins, L. J., Yzerbyt, V., & Kashima, Y. (2005). Fundamental dimensions of
social judgment: Understanding the relations between judgments of competence and
warmth. Journal of Personality and Social Psychology, 89, 899-913.
Katz, D., & Braly, K. (1933). Racial stereotypes of one hundred college students. Journal of
Abnormal and Social Psychology, 28, 280-290.
Kerr, N. (2002). When is a minority a minority? Active versus passive minority advocacy and
social influence. European Journal of Social Psychology, 32, 471-483.
Kubicka-Daab, J. (1989). Positivity and negativity effects in impression formation: Differences
in processing information about morality. Polish Psychological Bulletin, 20, 295-307.
Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal, and coping. New York: Springer.
Lin, M. H., Kwan, V. S. Y., Cheung, A., & Fiske, S. T. (2005). Stereotype content model
explains prejudice for an envied outgroup: Scale of anti-Asian-American stereotypes.
Personality and Social Psychology Bulletin, 31, 34-47.
Mackie, D. M., Devos, T., & Smith, E. R. (2000). Intergroup emotions: Explaining offensive
action tendencies in an intergroup context. Journal of Personality and Social Psychology,
Mackie, D. M., & Smith, E. R. (Eds.) (2002). From prejudice to intergroup emotions:
Differentiated reactions to social groups. New York: Psychology Press.
Neuberg, S. L., & Cottrell, C. A. (2002). Intergroup emotions: A sociofunctional approach. In D.
Mackie & E. R. Smith (Eds.), From prejudice to intergroup emotions: Differentiated
reactions to social groups. New York: Psychology Press.
Newcomb, T. M. (1956). The prediction of interpersonal attraction. American Psychologist, 11,
Pasupathi, M. & Lockenhoff, C. E. (2002). Ageist behavior. In T. Nelson (Ed.), Ageism:
Stereotyping and prejudice against older persons (pp. 201-246). Cambridge: MIT Press.
Peeters, G. (1983). Relational and informational pattern in social cognition. In W. Doise & S.
Moscovici (Eds.) Current issues in European social psychology (pp. 201-237). Cambridge:
Cambridge University Press.
Phalet, K., & Poppe, E. (1997). Competence and morality dimensions of national and ethnic
stereotypes: A study in six eastern-European countries. European Journal of Social
Psychology, 27, 703-723.
Reeder, G. D. (1993). Trait-behavior relations and dispositional inference. Personality and
Social Psychology Bulletin, 19, 586-593.
Roseman, I., Wiest, C., & Swartz, T. (1994). Phenomenology, behaviors, and goals differentiate
discrete emotions. Journal of Personality and Social Psychology, 67, 206-221.
Rosenberg, S., Nelson, C., & Vivekanathan, P. (1968). A multidimensional approach to the
structure of personality impressions. Journal of Personality and Social Psychology, 9, 283-
Rothbart, M., & Park, B. (1986). On the confirmability and disconfirmability of trait concepts.
Journal of Personality and Social Psychology, 50, 131-142.
Rozin, P., Lowery, L., Imada, S., & Haidt, J. (1999). The CAD triad hypothesis: A mapping
between 3 moral emotions (contempt, anger, disgust) and 3 moral codes (community,
autonomy, divinity). Journal of Personality and Social Psychology, 76, 574-586.
Schütz, H., & Six, B. (1996). How strong is the relationship between prejudice and
discrimination? A meta-analytic answer. International Journal of Intercultural Relations,
Sinclair, L., & Fehr, B. (2005). Voice versus loyalty: Self-construals and responses to
dissatisfaction in romantic relationships. Journal of Experimental Social Psychology, 41,
Smith, R. H. (2000). Assimilative and contrastive emotional reactions to upward and downward
social comparisons. In J. Suls & L. Wheeler (Eds.), Handbook of social comparison:
Theory and research (pp. 173-200). New York: Kluwer Academic/Plenum Publishers.
Smith, R. H., Parrott, W. G., Ozer, D., & Moniz, A. (1994). Subjective injustice and inferiority
as predictors of hostile and depressive feelings of envy. Personality and Social Psychology
Bulletin, 20, 705-711.
Smith, R. H., Turner, T. J., Garonzik, R., Leach, C. W., Urch-Druskat, V., & Weston, C. M.
(1996). Envy and schadenfreude. Personality and Social Psychology Bulletin, 25, 158-168.
Stangor, C., Sullivan, L. A., & Ford, T. E. (1991). Affective and cognitive determinants of
prejudice. Social Cognition, 9, 359-380.
Staub, E. (1996). Cultural-societal roots of violence: The examples of genocidal violence and of
contemporary youth violence in the United States. American Psychologist, 51, 117-132.
Stephan, W. G., & Stephan, C. W. (2000). An integrated threat theory of prejudice. In S. Oskamp
(Ed.), Reducing prejudice and discrimination (pp. 23-45). Mahwah, NJ: Erlbaum.
Talaska, C. A., Fiske, S. T., & Chaiken, S. (in press). Predicting discrimination: A meta-analysis
of the racial attitude-behavior literature. Social Justice Research.
Tesser, A., & Collins, J. E. (1988). Emotion in social reflection and comparison situations:
Intuitive, systematic, and exploratory approaches. Journal of Personality and Social
Psychology, 55, 695-709.
Thibaut, J. W., & Kelley, H. H. (1959). The social psychology of groups. New Brunswick, NJ:
Weiner, B. (2005). Social motivation, justice, and the moral emotions: An attributional
approach. Lawrence Earlbaum: Mahwah, NJ.
Willis, J., & Todorov, A. (2006). First impressions: Making up your mind after a 100-Ms
exposure to a face. Psychological Science, 17, 592-598.
Wojciszke, B. (2005). Affective concomitants of information on morality and competence.
European Psychologist, 10, 60-70.
Wojciszke, B., Bazinska, R., & Jaworski, M. (1998). On the dominance of moral categories in
impression formation. Personality and Social Psychology Bulletin, 24, 1245-1257.
Wojciszke, B., Brycz, H., & Borkenau, P. (1993). Effects of information content and evaluative
extremity on positivity and negativity biases. Journal of Personality and Social
Psychology, 64, 327-336.
Yzerbyt, V. Y., Provost, V., & Corneille, O. (2005). Not competent but warm…really?
Compensatory stereotypes in the French-speaking world. Group Processes and Intergroup
Relations, 8, 291-308.
Yzerbyt, V. Y., & Rocher, S. (2002). Subjective essentialism and the emergence of stereotypes.
In V. Yzerbyt & C. McGarty (Eds.), Stereotypes as explanations: The formation of
meaningful beliefs about social groups. (pp. 38-66). New York, NY, US: Cambridge
Zajonc, R. B. (1998). Emotions. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), Handbook of
social psychology (4
ed., Vol. 1, pp. 591-634). Boston: McGraw-Hill.
As in previous studies (Cuddy et al., 2006; Fiske et al., 2002; 1999), in the current study,
perceived status correlated with competence ratings, group-level r = .93, p <. 001, participant-
level r = .83, p < .001; and perceived competitiveness correlated negatively with warmth ratings,
group-level r = -.70, p < .001, participant-level r = -.43, p < .001. The opposite social structure-
traits correlations were not significant, as predicted.
We had not hypothesized negative correlations between emotions and behavioral tendencies
because emotions—the “hot” components of prejudice—are less likely to thwart than to enable a
behavior. Four unpredicted negative correlations emerged; all retrospectively fit the theoretical
model. Contempt inhibits both facilitation tendencies, as befits feeling repelled. Pity inhibits
passive facilitation; feeling sorry makes one avoid, not associate. Admiration inhibits passive
harm; one does not avoid the object of assimilative emotions.
To maintain parallel structure among the four emotions conditions, we used “hated” in place of
“contempted,” an unnatural construction.
Participants also rated the groups on affection and fondness, together representing another
emotion of theoretical interest. We do not report those results here, but they are available upon
Empirical support has been mixed for the prediction that fear is elicited by outgroups that are
perceived to be stronger or more powerful than the ingroup (Devos et al., 2002; Mackie et al.,
Amy J. C. Cuddy, Kellogg School of Management, Northwestern University; Susan T.
Fiske, Department of Psychology, Princeton University; Peter Glick, Department of Psychology,
The preliminary study and Studies 1-3 were conducted as part of the first author’s
doctoral dissertation at Princeton University. Study 1 was supported by funds from Princeton
University, granted to the second author. We are grateful to Marc Weiner and Ed Freeland at the
Princeton Survey Research Center for their helpful advice. We thank Mindi Rock and Katie
Dover-Taylor for their help with data collection and editing, respectively. We thank Virginia
Kwan, Sam Glucksberg, and Terri Vescio and four anonymous reviewers for their valuable
feedback on drafts of this paper, and Charles Judd for advice about data analysis.
Correspondence should be directed to Amy Cuddy (firstname.lastname@example.org),
Kellogg School of Management, 2001 Sheridan Rd., Northwestern University, Evanston, IL
Table 1: Correlations of Behavioral Tendencies with Stereotypes and Emotions, Study 1
Predictor Active facilitation Active harm Passive facilitation Passive harm
Competence .08 -.20 .77*** -.68***
Warmth .73*** -.55*** .45* -.24
Admiration .59** -.35 .95** * -.69**
Contempt -.63** .93*** -.46* .48*
Envy -.06 .22 .57** -.39
Pity .51* -.10 -.48* .65**
Competence .17*** -.10** .64*** -.50***
Warmth .47*** -.34*** .53*** -.24***
Admiration .49*** .31*** .74*** -.58***
Contempt -.24*** .54*** -.33*** .48***
Envy .00 .21*** .43*** -.25***
Pity .40*** .00 -.26*** .41***
Note: Bolded correlations were predicted to be significant (23/24 were). But 15/24 others were also significant, although they were theoretically consistent, most in the participant-level analyses, which
had high power (participant df = 569, group df =18) * p < .05, **p < .01, ***p < .001. As in previous studies (Cuddy et al., 2006; Fiske et al., 2002; 1999), perceived status correlated with competence,
group-level r = .93, p <. 001, participant-level r = .83, p < .001; and perceived competitiveness correlated negatively with warmth, group-level r = -.70, p < .001, participant-level r = -.43, p < .001.
Opposite structure-traits correlations were not significant, as predicted.
Table 2: Behavioral Tendencies Standardized Means by Competence and Warmth Stereotypes, Studies 1 and 2
High Low High Low
Study 1 (measured stereotypes)
Active facilitation .331
Passive facilitation .277
Active harm -.305
Passive harm -.270
Study 2 (manipulated stereotypes)
Active facilitation .352
Passive facilitation .345
Active harm -.325
Passive harm -.310
Note: Within study, within trait (i.e., warmth, competence), means not sharing a subscript differ at p < .01. All predicted differences are significant.
Figure 1: Schematic representation of BIAS Map Hypotheses 1 & 2. Competence and warmth
stereotypes are represented outside the figure along the X and Y axes, respectively. Emotions are
represented by gray arrows, within the figure on diagonal axes. Behavioral tendencies are
represented by black arrows, within the figure on horizontal and vertical axes.
Figure 2: Scatter plot and cluster analysis of groups on competence and warmth ratings. HC-HW
= high-competence, high-warmth; HC-LW = high-competence, low-warmth; LC-HW = low-
competence, high-warmth; LC-LW = low-competence, low-warmth.
Figures 3a-3d: Regression analyses testing mediation by emotions of the direct effect of
stereotypes on behavioral tendencies. For each analysis, we controlled for the non-predictor trait
(i.e., warmth when competence was the predictor; competence when warmth was the predictor).
The coefficient in parentheses represents the direct effect of the stereotype trait on the behavioral
tendency, whereas the adjacent coefficient was observed when emotions were added to the
model. Broken lines indicate non-significant effects. *p < .05, **p < .01. Sobel test results:
Active facilitation (panel 3a) Z = 1.94, p = .05, Active harm (panel 3b) Z = 1.67, p < .10, Passive
facilitation (panel 3c) Z = 2.30, p < .05, Passive harm (panel 3d) Z = 2.19, p < .05.
Figure 4: Regression analyses showing that anger mediated the effect of envy on active harm. *p
= .07, **p < .05, ***p < .01. Sobel test results: Z = 2.21, p < .05.
Interview Script and Items for National Survey, Study 1
Live interviewers introduced the study as follows:
"Hi, my name is ___ and I'm calling from Princeton University to conduct a survey about how
Americans view different social groups. It's an opinion survey only; we are not selling anything."
After an adult household member agreed to participate, the interviewer explained:
"We are studying how different groups are perceived by Americans. We are interested in how
you think other people in general view these groups. We are not asking how you personally view
these groups, but how you think most people view them."
After receiving instructions about how to rate the groups using the five-point scale (1 = not at all; 5 =
extremely), participants began making ratings, answering all questions -- traits, social structure, emotions,
and behaviors -- about one group before moving on to the next group.
Questions about perceived traits were phrased as follows:
"Consider how [group, e.g., the elderly] are viewed by Americans in general. As viewed by most
Americans, how [e.g., competent] are [group]?"
For the social structure items, the interviewer read the following four items:
"Again, as viewed by Americans, how economically successful have [group] been?"
"… how prestigious are the jobs generally held by [group]?"
"…how much does special treatment given to [group] make things more difficult for other groups
"…if resources go to [group], to what extent does that take resources away from the rest of
For emotion items, the interviewer read:
"Now I'm going to ask you about some feelings that people in America have toward [group] as a
group. To what extent do people tend to feel [emotion, e.g., pity] toward [group]?"
For behavior items, the interviewer read:
"Finally, I am going to ask you about the ways people in America generally behave toward [name
of group] as a group? Do people tend to [behavior, e.g., help] [group]?"
Social structure scales
status: economic success, prestigious jobs
competitiveness: special breaks, resources
competence: competent, capable
warmth: warm, friendly
contempt: contempt, disgust
admiration: admire, proud
pity: pity, sympathy
envy: envious, jealous
Behavioral tendencies scales
active facilitation: help, protect
active harm: fight, attack
passive facilitation: cooperate with, associate with
passive harm: exclude, demean
-.64** .96 **