Prejudice

Abstract

Prejudice is a form of social behavior: a negative attitude toward people based on their group membership. This article develops the construct as cognitive and social psychology now treat it — the output of ordinary categorization and attitude processes rather than a mark of pathology. It separates prejudice from the stereotype that supplies its beliefs and the discrimination that is its expression, sets out the dual-process architecture in which an automatic association can be regulated by controlled processing, and surveys the stereotype content model, the Implicit Association Test, and the neural systems behind biased responding. It then asks why prejudice persists — aversive racism, the justification-suppression model, social dominance orientation — and what reduces it, where intergroup contact has the strongest meta-analytic support. A worked example reconstructs Implicit Association Test scoring, and three interactive demonstrations make the core ideas concrete.

Keywords: prejudice, implicit bias, stereotype content model, intergroup contact, social dominance orientation

Prejudice is a negative attitude or affective response directed at people because they belong to a social group, classically defined as an antipathy based upon a faulty and inflexible generalization that may be felt or expressed and that is directed toward a group or an individual as a member of that group (Allport, 1954). Three features of that definition still organize the field: prejudice is an *attitude* rather than an act, it is *group-based* rather than a response to the individual, and it rests on a *generalization* that resists disconfirming evidence. The lay word blurs a feeling, a belief, and a behavior that the research literature keeps apart.

The modern treatment is cognitive. Where mid-century theory often cast prejudice as a symptom of a disordered or authoritarian personality, the dominant frameworks now model it as a byproduct of normal mental operations, the categorization of people into groups, the attachment of learned associations to those categories, and the automatic activation of those associations when a group member is encountered (Devine, 1989). The sections below follow that logic from the structure of a prejudiced response through the reasons it endures to the interventions that can loosen it.

What Prejudice Is

Prejudice is operationally a negative evaluation of a social group and of individuals because of their membership in it (Allport, 1954). It is useful to hold three allied terms apart. A *stereotype* is the cognitive component, a belief about the attributes of a group; *prejudice* is the affective and evaluative component, a feeling about the group; and *discrimination* is the behavioral component, differential treatment of people on the basis of group membership. The three are correlated but dissociable: a person can hold a stereotype without the negative affect of prejudice, feel prejudice without acting on it, and even discriminate through institutions without harboring personal animus.

Prejudice is organized by social categorization. Placing a person into a group, by race, sex, age, or any salient dimension, is a rapid and often involuntary act of perception, and it brings with it the evaluations learned for that category. Tajfel's minimal group studies showed how little is needed: assigning people to arbitrary groups on a trivial basis is enough to produce ingroup favoritism in the allocation of rewards, so that a mere ingroup-outgroup distinction, absent any history or conflict, can seed differential evaluation (Tajfel et al., 1971). This is why the cognitive account treats prejudice as continuous with ordinary category-based perception rather than as an aberration confined to extreme individuals, a shift that reframes the question from *who is prejudiced* to *when and how category associations shape judgment*.

Types of Prejudice

The Medical Subject Headings vocabulary files *Prejudice* (D011287) under social behavior and social psychology, and lists eight narrower descriptors directly beneath it, shown in Table 1. These subtypes are not mutually exclusive: an older worker may face ageism and disability discrimination at once, and implicit bias cuts across every target domain rather than naming a target of its own. Nor is the list a theory of the kinds of prejudice; it is an *indexing classification*, a controlled vocabulary built so that the published literature can be retrieved, not a claim that these eight categories carve prejudice at its natural joints. It records how the literature is indexed, not what the true types are. Four of the eight have their own article on this site and are linked; the remaining four are named in plain text and glossed from their MeSH sense.

The Cognitive Architecture of Prejudice

The decisive move in the modern study of prejudice was to separate what is activated automatically from what a person deliberately endorses. Devine proposed a dual-process account in which a cultural stereotype is learned early and becomes automatically activated in the presence of a group member for everyone who knows it, so that high-prejudice and low-prejudice people share the same automatic knowledge; they differ at a second, controlled stage, where low-prejudice individuals actively inhibit the activated stereotype and replace it with their egalitarian beliefs (Devine, 1989). Prejudice, on this view, is not the mere presence of an association but the failure to override it, which is why it can feel involuntary and why reducing it requires effortful, practiced control rather than simple goodwill. Figure 1 sets out the two stages.

If the architecture is dual, the *content* of the activated stereotype is systematic. The stereotype content model holds that group stereotypes vary along two fundamental dimensions, warmth and competence, whose values follow from the group's structural position: competition with the ingroup lowers perceived warmth, and high status raises perceived competence (Fiske et al., 2002). The two dimensions define four quadrants, each tied to a distinct emotion and a distinct form of prejudice: admiration for high-warmth, high-competence ingroups and allies; contempt for low-warmth, low-competence groups; envy for competent but cold groups; and paternalistic pity for warm but incompetent groups. Prejudice, in this account, is not one undifferentiated hostility but a set of qualitatively different affective reactions mapped onto social structure. The StereotypeContentMap demonstration lets the reader place groups in that space and read off the predicted emotion.

The dual-process account also motivated a measurement problem: if the automatic component operates outside deliberate control, self-report cannot reach it. The Implicit Association Test addressed this by measuring the relative speed with which people pair group concepts with evaluative attributes, on the logic that two concepts sharing a strong association are easier, and so faster, to map onto a single response key (Greenwald et al., 1998). A participant who responds faster when one group shares a key with 'good' than when the other does reveals a relative automatic preference, and that latency difference, suitably standardized, is the test's score. The neural counterpart of this architecture is now partly mapped: Amodio reviewed evidence that the amygdala tracks the automatic evaluation of outgroup cues, while regions of prefrontal and anterior cingulate cortex implement the detection of conflict and the regulation that the dual-process model posits, giving the automatic-controlled distinction a biological basis rather than a purely functional one (Amodio, 2014).

From Attitude to Action

Why does prejudice persist when explicit endorsement of it has fallen so sharply? One answer is that its expression has gone underground rather than away. The theory of aversive racism holds that many people who sincerely endorse egalitarian values nonetheless harbor negative associations they find aversive to admit, and so discriminate only when a situation supplies a non-racial justification that lets them preserve a non-prejudiced self-image. In a selection study spanning a decade, White participants did not discriminate against a Black candidate when qualifications were unambiguously strong or weak, but did so when qualifications were ambiguous and a merit-based rationale was available (Dovidio & Gaertner, 2000). The justification-suppression model generalizes the mechanism: people experience genuine prejudices that prevailing norms lead them to suppress, and those prejudices are released into expression whenever a justification, an ideology, an attribution, or a norm, makes expression feel legitimate (Crandall & Eshleman, 2003). Prejudice is thus doubly regulated, by the automatic-controlled architecture within the person and by the suppression-justification dynamic between the person and the social norm.

Some of the variance is dispositional. Social dominance orientation is an individual's general preference for hierarchy among social groups and for the dominance of some groups over others; it predicts a wide range of prejudices and support for hierarchy-enhancing policies, and it functions as a stable attitudinal orientation rather than a response to any single target group (Pratto et al., 1994). The behavioral stakes are clearest when a biased association must be resolved under time pressure. In the first-person-shooter task, participants decide whether to 'shoot' armed targets and 'not shoot' unarmed ones; they are faster and more accurate to shoot armed Black targets and to withhold fire from unarmed White targets than the reverse, a bias in the speed and threshold of a split-second decision rather than in a considered judgment (Correll et al., 2002). The ImplicitAssociationDemo in the worked example below makes the latency logic that underlies such measures concrete.

Reducing Prejudice

The oldest and best-supported intervention is intergroup contact. Allport proposed that contact reduces prejudice when four conditions hold: the groups have equal status within the situation, pursue common goals, cooperate rather than compete, and have the support of authorities, law, or custom (Allport, 1954). Pettigrew reformulated the hypothesis as a theory, specifying the processes, learning about the outgroup, changed behavior, affective ties, and ingroup reappraisal, through which contact has its effect, and arguing that the optimal conditions are facilitating rather than strictly necessary (Pettigrew, 1998). The empirical verdict came from a meta-analysis of more than 500 studies: contact was associated with lower prejudice in the large majority of samples, the optimal conditions strengthened but were not required for the effect, and the reduction generalized beyond the specific individuals encountered (Pettigrew & Tropp, 2006). A later review traced two further decades of work extending contact to indirect and imagined forms and sharpening the account of when and why it works (Dovidio et al., 2017). The ContactEffectExplorer lets the reader toggle Allport's conditions and watch the predicted effect move.

A second family of interventions targets the automatic component directly. Devine and colleagues framed implicit bias as a habit that can be broken, and taught participants a set of strategies, stereotype replacement, counter-stereotypic imaging, individuation, perspective taking, and increased contact, to use whenever they noticed bias; the intervention produced a reduction in implicit race bias that persisted for two months, well beyond the duration of most laboratory manipulations (Devine et al., 2012). The broader picture is more sobering. A large study that pitted seventeen interventions against one another found that although many could shift an Implicit Association Test score in the moment, none produced a change that survived even a short delay, implying that implicit associations are malleable but not durably so by brief manipulation (Lai et al., 2016). A systematic re-evaluation of the field's foundational contact studies concluded that the strongest causal evidence comes from well-designed contact interventions while much of the broader literature rests on weaker designs, a caution about which claims the evidence can actually bear (Paluck et al., 2019).

Worked Example

The Implicit Association Test scores a relative automatic preference from the difference in response speed between two combined sorting tasks. In a race test, the *congruent* block pairs one group with 'good' and the other with 'bad' on shared keys, and the *incongruent* block reverses the pairing. If the first pairing matches a participant's automatic associations, responses in the congruent block are faster. The modern scoring procedure expresses that difference as the *D* measure: the difference between the two block means divided by the standard deviation of all the response latencies in the two blocks, which standardizes the effect in units of the participant's own variability (Greenwald et al., 1998).

Consider a participant whose correct-trial latencies, in milliseconds, are as follows:

The congruent block mean is (470 + 560 + 700 + 850 + 520 + 640 + 790 + 610) / 8 = 642.5 ms, and the incongruent block mean is 720.0 ms, so the participant is 77.5 ms slower when the group must share a key with 'good' against their automatic association. To standardize, pool all sixteen latencies and compute their standard deviation, which is 126.85 ms. The *D* measure is therefore the difference of means divided by this inclusive standard deviation, 77.5 / 126.85 = 0.61. By the conventional benchmarks, a *D* near 0.15 is slight, 0.35 moderate, and 0.65 strong, so 0.61 indicates a moderate-to-strong automatic preference for the group paired with 'good' in the congruent block. The measure is relative, not absolute: it records which pairing is easier for this participant, not a quantity of animus, and a single administration is a noisy estimate. The ImplicitAssociationDemo lets the reader adjust the block latencies and watch the means, the pooled standard deviation, and *D* update; the numbers here are illustrative of the scoring procedure, not drawn from a specific dataset.

Discussion

Across a century of theory the center of gravity has moved from character to cognition. The mid-century search for the prejudiced personality gave way to Allport's insight that prejudice grows out of normal categorization, which Devine sharpened into a dual-process architecture that separates an automatically activated association from the controlled processing that can override it (Allport, 1954; Devine, 1989). The stereotype content model added that the activated content is structured, not arbitrary, and the Implicit Association Test and its neural correlates gave the automatic component a measure and a substrate (Fiske et al., 2002; Greenwald et al., 1998; Amodio, 2014). The practical payoff of this cognitive turn is that it names mechanisms an intervention can target: an association to be re-learned, a regulatory process to be strengthened, a situational justification to be removed.

The open problems are substantive. The relationship between implicit measures and behavior is weaker and less stable than early enthusiasm assumed, and the failure of brief interventions to produce durable change in those measures has forced a reappraisal of what they index (Lai et al., 2016). The evidence base for prejudice reduction is uneven, strong for contact, thinner for much else, and the field has grown more careful about distinguishing a measurable momentary shift from a lasting change in how people treat one another (Paluck et al., 2019). The integration of the biological substrate with the social account likewise remains schematic: naming an amygdala-prefrontal balance is not yet the same as predicting, for a given person in a given moment, whether an activated association will govern an act (Amodio, 2014).

Current Directions

The most striking recent result is that intergroup attitudes do change at the population scale. Charlesworth and Banaji analyzed millions of implicit and explicit attitude tests collected over a decade and found that some implicit attitudes, notably toward sexual orientation, moved substantially toward neutrality, while others, such as those involving age and disability, barely budged, demonstrating that implicit bias is neither fixed nor uniformly malleable but follows attitude-specific trajectories (Charlesworth & Banaji, 2019). The finding reframes the durability question raised by the intervention studies: implicit associations can shift over years even where they resist a laboratory manipulation (Lai et al., 2016).

A second front is methodological humility about what works. Paluck and colleagues' review of the prejudice-reduction literature catalogued how much of it relies on light-touch manipulations and short-term outcomes measured in the lab, and called for interventions tested in the field, over time, against behavioral outcomes that matter (Paluck et al., 2021). Their earlier re-evaluation of contact made the same point for that specific literature, showing that its strongest designs support a causal effect while weaker ones cannot (Paluck et al., 2019). Together these lines move the field toward longitudinal, field-based, and behavioral evidence rather than one-off laboratory effects, and toward a candid accounting of which claims the data can and cannot support.

Common Misconceptions

Prejudice is confined to a prejudiced minority of bad people.

The dual-process account holds that cultural stereotypes are automatically activated for nearly everyone who knows them; people differ not in whether the association is triggered but in whether controlled processing overrides it (Devine, 1989).

A high Implicit Association Test score proves a person is a racist who will discriminate.

The test yields a relative, noisy index of an automatic association, not a measure of conscious belief or a reliable prediction of any individual's behavior; a single administration is an unstable estimate (Greenwald et al., 1998).

Implicit bias is easily and permanently erased by a brief training.

A large comparison of interventions found that many could move an implicit score momentarily but none produced change that survived even a short delay, so durable reduction requires sustained effort rather than a one-off session (Lai et al., 2016).

Glossary

Aversive racism.

A form of bias in which a person endorsing egalitarian values nonetheless holds negative associations and discriminates only where a non-prejudicial justification is available.

Contact hypothesis.

Allport's proposal that contact between groups reduces prejudice under conditions of equal status, common goals, cooperation, and institutional support.

Discrimination.

The behavioral component of intergroup bias: differential treatment of people on the basis of their group membership.

Dual-process model.

The account separating the automatic activation of a group association from the controlled processing that can inhibit or override it.

Implicit Association Test.

A reaction-time measure that infers an automatic association from the relative speed of pairing group concepts with evaluative attributes.

Implicit bias.

An attitude or stereotype that operates automatically and outside awareness to influence judgment and action.

Ingroup favoritism.

The tendency to evaluate and reward members of one's own group more favorably than members of other groups, seen even for arbitrary minimal groups.

Justification-suppression model.

The account in which genuinely held prejudices are suppressed by norms and released into expression when a justification makes expression feel legitimate.

Prejudice.

A negative attitude or affective response toward people because of their membership in a social group.

Social categorization.

The rapid, often involuntary placement of a person into a social group, which brings the evaluations learned for that category to bear on perception.

Social dominance orientation.

A general individual preference for hierarchy among social groups, predicting a range of prejudices and support for hierarchy-enhancing policies.

Social identity theory.

Tajfel and Turner's account in which part of the self-concept derives from group memberships, motivating favoritism toward the ingroup to sustain a positive social identity.

Stereotype content model.

The framework placing group stereotypes on dimensions of warmth and competence, predicting distinct emotions and forms of prejudice.

Stereotype.

The cognitive component of intergroup bias: a belief about the attributes shared by members of a group.

Key Researchers

Gordon W. Allport

(1897–1967). Professor at Harvard University whose 1954 book *The Nature of Prejudice* defined the construct, grounded it in ordinary categorization, and set out the contact hypothesis that still anchors prejudice reduction.
Wikipedia - Wikidata

Mahzarin R. Banaji

(Harvard University). Richard Clarke Cabot Professor of Social Ethics whose work on implicit social cognition and long-term attitude change mapped how intergroup bias operates and shifts.
ORCID - Wikidata

Patricia G. Devine

(University of Wisconsin–Madison). Professor of psychology whose dual-process account separated automatic from controlled components of prejudice and whose habit-breaking intervention produced durable reductions in implicit bias.
ORCID - Wikidata

Susan T. Fiske

(Princeton University). Eugene Higgins Professor of Psychology and Public Affairs who developed the stereotype content model and the BIAS map linking warmth and competence to emotions and discrimination.
Wikipedia - Wikidata

Anthony G. Greenwald

(University of Washington). Professor emeritus of psychology and co-creator of the Implicit Association Test, which made automatic intergroup attitudes measurable.
ORCID - Wikipedia - Wikidata

Elizabeth Levy Paluck

(Princeton University). Professor of psychology and public affairs whose field experiments and systematic reviews reset the evidentiary standard for prejudice-reduction research.
ORCID - Wikidata

Thomas F. Pettigrew

(University of California, Santa Cruz). Research professor who turned the contact hypothesis into a process theory and, with Linda Tropp, provided its decisive meta-analytic test.
Wikipedia - Wikidata

Henri Tajfel

(1919–1982). Professor at the University of Bristol whose social identity theory and minimal group experiments showed that mere categorization into arbitrary groups is enough to produce ingroup favoritism.
Wikipedia - Wikidata

Frequently Asked Questions

What is prejudice in psychology?

Prejudice is a negative attitude or affective response toward people because of their membership in a social group, classically defined by Allport as an antipathy based on a faulty and inflexible generalization (Allport, 1954).

What is the difference between prejudice, a stereotype, and discrimination?

A stereotype is the belief about a group, prejudice is the negative feeling toward it, and discrimination is the differential behavior directed at its members; the three are related but can occur independently (Allport, 1954).

Is everyone prejudiced?

Cultural stereotypes are automatically activated for nearly everyone who has learned them, so the relevant difference between people is not whether the association is triggered but whether controlled processing overrides it (Devine, 1989).

What is implicit bias?

Implicit bias is an attitude or stereotype that operates automatically and outside awareness; the Implicit Association Test infers it from the relative speed of pairing group concepts with good and bad (Greenwald et al., 1998).

Does a high score on the Implicit Association Test mean I am racist?

No. The test gives a relative and fairly noisy index of an automatic association, not a measure of conscious belief, and it does not reliably predict how any individual will behave (Greenwald et al., 1998).

What is the stereotype content model?

It is the framework that places group stereotypes on two dimensions, warmth and competence, which together predict distinct emotions such as admiration, envy, pity, and contempt, and distinct forms of prejudice (Fiske et al., 2002).

What is the most effective way to reduce prejudice?

Intergroup contact has the strongest meta-analytic support, especially under conditions of equal status, common goals, cooperation, and institutional support, and its benefits generalize beyond the people directly encountered (Pettigrew & Tropp, 2006).

Can implicit bias be changed?

Implicit associations are malleable but not easily made to change for long; brief interventions rarely produce lasting shifts, though sustained strategies and population-level trends show that durable change is possible (Lai et al., 2016; Charlesworth & Banaji, 2019).

References