Determinants of automatic age and race bias: ingroup-outgroup distinction salience moderates automatic evaluations of social groups

People rapidly categorize others into social groups, often using primary categories like age and race that are perceptually salient and processed automatically. Such categorization yields ingroup–outgroup distinctions and intergroup bias—more positive evaluations of ingroup than outgroup members. Understanding determinants of automatic age and race bias is important due to their adverse societal impacts. The present research tests whether making a particular ingroup–outgroup distinction salient (White vs. Black or young vs. old) determines which dimension guides automatic intergroup evaluations for multiply categorizable targets (e.g., young White, old Black). The core hypothesis is that shifting salience between White–Black and young–old will correspondingly increase automatic race bias or age bias, respectively, even when the targets in the automatic evaluation task are novel and unrelated to the manipulation.

Prior work shows that social categorization moderates automatic social evaluation. Many studies manipulated how participants categorized the very stimuli used in implicit measures (e.g., Implicit Association Test or Evaluative Priming Task) by requiring categorization by race vs. another dimension, varying distractors, or blocking stimulus presentation (Mitchell et al., 2003; Gawronski et al., 2010; Jones & Fazio, 2010; Todd et al., 2021). Other work used minimal group paradigms to redefine ingroup/outgroup membership with mixed-race teams, finding automatic favoritism toward team ingroup regardless of race (Van Bavel & Cunningham, 2009), potentially involving cooperation contexts or evaluative learning. Such manipulations often entail practicing categorization rules or learning about specific exemplars, possibly producing procedural priming or evaluative conditioning. It remains unclear whether directly targeting the categorization of the evaluation targets is necessary. The present studies extend this work by manipulating only the salience of participants’ own ingroup identity (White vs. young) and the contrasting outgroup, without categorization practice, cooperation contexts, or learning about the targets to be evaluated.

Overview: Two laboratory experiments with young, predominantly White German participants tested whether making White–Black vs. young–old salient shifts automatic intergroup bias between race and age toward novel face primes in an Evaluative Priming Task (EPT). Materials and data: https://osf.io/7g8va/. Tasks were programmed in DirectRT and MediaLab. Experiment 1: Design: 2 (Ingroup–outgroup salience: White–Black vs. young–old; between) × 2 (Evaluation score: race bias vs. age bias; within). Participants: N=71 young White females (19–38 years; M=23.90, SD=3.95); exclusions: non-White (0), ≥40 years (2), data loss (1). Power analysis indicated adequate sensitivity (observed η²p≈.14). Procedure: Salience manipulation—participants wrote 3–5 sentences describing how they, as a White (vs. young) person, differed from Black (vs. old) people (Haslam et al., 1999). EPT—40 color head-and-shoulder photos of female faces from CAL/PAL (10 each: young White, old White, young Black, old Black); 10 positive and 10 negative German nouns. Trial structure: fixation 500 ms; prime 200 ms; target until response; 20 practice (targets only) + 160 test trials. Responses: T-key=positive (right index), E-key=negative (left index); ITI 500 ms; too-slow feedback if RT>1500 ms; error feedback 500 ms. Outlier handling: exclude errors (4.5%), RT<150 ms (1 resp.), RT>1000 ms (1.7%); validated with alternative cutoffs (1250 ms; log-transform after 1500 ms). Bias score computation: Race bias = average of [RT(positive|Black) − RT(positive|White)] and [RT(negative|White) − RT(negative|Black)]; Age bias = average of [RT(positive|old) − RT(positive|young)] and [RT(negative|young) − RT(negative|old)]. Additional measures: Social identification (Inclusion of Ingroup in the Self; IIS) for White and young (1–7 overlap); self-reported warmth (7-point feeling thermometers) for each of the 40 targets. Analyses: Mixed ANOVAs in SPSS; η²p and CIs computed via scripts. Experiment 2: Design: 2 (Salience: White–Black vs. young–old; between) × 2 (Type of manipulation: with vs. without attribute description; between) × 2 (Evaluation score: race vs. age bias; within). Participants: N=159 (94 female, 65 male), ages 16–31 (M=20.87, SD=2.66). Exclusions: non-White (2), didn’t select required ingroup option (7), refused attribute listing (1), repeated participation (2); compensation: chocolate bar. Procedure: All participants were told that data of White & Black (vs. young & old) participants would be compared and indicated their group membership ("I am White/Black" or "I am young/old"). With-attribute-description condition replicated Experiment 1’s writing task; without-attribute-description skipped writing. EPT identical to Experiment 1 except gender-matched primes for male participants (male faces from CAL/PAL). Additional measures: Frequency of thoughts about being White vs. young during the experiment (1–7); free-response on when these thoughts occurred; self-stereotyping (Lun et al., 2009): 48 traits (12 stereotypic White, 12 Black, 12 young, 12 old; half positive/half negative) rated 1–7. Outlier handling: exclude errors (4.8%), RT<150 ms (2 responses), RT>1000 ms (3.0%). Self-stereotyping indices: Race (White−Black), Age (young−old), computed separately for positive and negative traits. Analyses: Mixed ANOVAs; separate analyses by manipulation type as needed.

Experiment 1 (N=71):

• Automatic evaluation: Significant Salience × Evaluation Score interaction, F(1,69)=11.48, p=.001, η²p=.14 [90% CI: .038, .267]. • Race bias higher when White–Black salient (M=10.42, SD=16.40) vs. young–old salient (M=1.75, SD=13.84), F(1,69)=5.78, p=.019, η²p≈.08. • Age bias higher when young–old salient (M=7.14, SD=11.80) vs. White–Black salient (M=1.57, SD=11.55), F(1,69)=4.04, p=.048, η²p=.06. • Bias vs. zero: White–Black salient—race bias significant, t(35)=3.81, p<.001; age bias ns, t(35)=0.81, p=.421. Young–old salient—age bias significant, t(34)=3.58, p=.001; race bias ns, t(34)=0.75, p=.459.
• Explicit evaluations: Main effect of Evaluation Score, F(1,69)=18.09, p<.001, η²p=.21 (overall age bias > race bias; race bias negative). Salience × Evaluation Score interaction, F(1,69)=5.95, p=.017, η²p=.08. Age bias lower when young–old salient (M=−0.02, SD=0.57) vs. White–Black salient (M=0.30, SD=0.72), F(1,69)=4.41, p=.039; race bias difference ns, F(1,69)=2.00, p=.162.
• Social identification (IIS): Main effect of salience, F(1,69)=5.39, p=.023, η²p=.07; higher identification reports overall when White–Black was salient; no differences by group or interaction. Experiment 2 (N=159):
• Automatic evaluation (3-way mixed ANOVA): Salience × Evaluation Score significant, F(1,155)=17.22, p<.001, η²p=.10; the 3-way interaction with manipulation type ns, F(1,155)=0.54, p=.466. • Simple comparisons (collapsed across type): Race bias higher when White–Black salient (M=11.90, SD=21.79) vs. young–old salient (M=0.88, SD=18.96), F(1,155)=11.28, p<.001, η²p=.07. Age bias higher when young–old salient (M=5.94, SD=16.85) vs. White–Black salient (M=0.01, SD=17.35), F(1,155)=4.72, p=.031, η²p=.03. • With attribute description: Salience × Evaluation Score, F(1,75)=5.70, p=.019, η²p=.07; race bias descriptively higher White–Black salient (M=10.16) vs. young–old (M=2.05), p=.100; age bias descriptively higher young–old (M=6.26) vs. White–Black (M=0.48), p=.132. Bias vs. zero: White–Black—race bias significant, t(38)=2.66, p=.011; age bias ns. Young–old—age bias significant, t(37)=2.45, p=.019; race bias ns. • Without attribute description: Salience × Evaluation Score, F(1,80)=12.23, p<.001, η²p=.13; race bias higher White–Black salient (M=13.92) vs. young–old (M=0.21), F(1,80)=10.01, p=.002; age bias descriptively higher young–old (M=5.63) vs. White–Black (M=−0.44), F(1,80)=2.42, p=.124. Bias vs. zero: White–Black—race bias significant, t(40)=4.38, p<.001; age bias ns. Young–old—age bias marginal, t(40)=2.01, p=.052; race bias ns.
• Frequency of thoughts: Interaction patterns indicate participants thought more often about the salient ingroup (young when young–old salient; White when White–Black salient), with stronger effects in the attribute-description condition; White-identity thought frequency did not increase in the without-attribute condition.
• Self-stereotyping: Main effects of score type and trait valence and their interaction, but no effects of salience or manipulation type on self-stereotyping indices, suggesting the salience manipulations did not change self-ascribed stereotypic traits. Overall: Across both experiments, making a specific ingroup–outgroup distinction salient shifted automatic intergroup bias toward that dimension without requiring categorization practice on the evaluation targets or team-based learning.

Findings support the hypothesis that merely directing attention to one’s ingroup identity relative to an outgroup is sufficient to alter automatic intergroup evaluations of novel, multiply categorizable targets. This extends prior evidence by showing that changes in automatic evaluations do not require manipulating categorization of the targets in the implicit task, prior classification training, cooperative team framing, or evaluative learning. The effects generalized across two operationalizations of salience (writing about group differences and simply indicating group membership with prior information that groups would be compared), and to targets unconnected to the manipulation context. The divergence between automatic and explicit measures (with explicit ratings showing reduced age bias when young–old was salient and overall warmer self-reported feelings toward Black than White targets) suggests possible social desirability or correction processes in self-reports when a category distinction is made salient. Ancillary measures indicated that salience increased thoughts about the corresponding ingroup but did not alter self-stereotyping, arguing against a mechanism based on activating positive ingroup and negative outgroup attributes. Together, these results underscore the sensitivity of automatic intergroup bias to contextual shifts in identity salience and the utility of implicit measures for primary categories like race and age.

Across two experiments, the research shows that a simple shift in the salience of ingroup–outgroup distinctions (White–Black vs. young–old) systematically moderates automatic race and age bias toward multiply categorizable targets, even when targets are novel and unrelated to the manipulation. This advances understanding of determinants of automatic intergroup bias by demonstrating that attention to a salient ingroup identity alone can reconfigure automatic evaluations. Future research should test boundary conditions, including whether subliminal ingroup primes can induce similar shifts, disentangle reduction versus augmentation effects (potentially requiring large samples), include explicit control conditions, and examine moderators such as strength of identification with the salient ingroup.

• No explicit control (neutral salience) condition was included, limiting inferences about absolute reduction vs. augmentation relative to baseline. - Generalizability: Experiment 1 included only young White females; Experiment 2 sampled university students in Germany; non-White participants were excluded. - Some effects (especially age-bias increases in Experiment 2 without attribute description) were descriptive or marginal, indicating modest effect sizes and the need for larger samples for certain contrasts. - Divergence between implicit and explicit measures suggests potential social desirability and correction processes in self-reports, complicating interpretation of explicit attitudes. - The manipulations focused on race (White vs. Black) and age (young vs. old) only; other salient identities and intersectional categories were not tested. - Short-term laboratory context; persistence of effects over time and across contexts was not assessed.