Social conformity updates the neural representation of facial attractiveness

Social conformity—the tendency to align behaviors and thoughts with perceived group norms—is pervasive, spanning mundane choices to consequential decisions. Although conformity is adaptive for learning in uncertain environments and is observed across species and development, the degree to which people internalize social influence (i.e., genuine changes in internal attitudes and beliefs) remains debated. Behavioral compliance can be confounded by impression management and demand characteristics, making it difficult to infer internal attitude change from self-reports alone. Neuroscientific evidence linking explicit rating changes to activity in valuation-related regions (e.g., nucleus accumbens, orbitofrontal cortex) suggests potential internalization, yet such effects have largely been examined during explicit evaluations that remain susceptible to strategic responding. It is unknown whether neural activity reflects internalized social influence in the absence of overt evaluation. Further, the source of influence may matter: from an evolutionary and social-cognitive perspective, people often prioritize in-group over out-group information, but empirical findings on selective in-group conformity are mixed. This preregistered EEG study asked whether learning about facial attractiveness from in-group versus out-group members changes explicit ratings and spontaneous neural representations of attractiveness during a task without explicit evaluation, and whether individual differences in perceived social norm tightness-looseness modulate these effects. Competing hypotheses tested whether internalization would occur selectively for in-group influence or for both in- and out-group influence.

Prior work shows conformity across species and development and links social influence to neural valuation systems. fMRI studies report that post-influence attractiveness ratings co-vary with activity in valuation regions, implying internalization; however, these findings arise during explicit judgments potentially influenced by impression management. ERP components like the face-sensitive N170 and the late positive component (LPC) have been associated with attractiveness perception, but results are mixed, and individual differences in attractiveness perception can obscure univariate effects. Multivariate representational similarity analysis (RSA) can capture distributed neural patterns and individual-specific representations, offering improved sensitivity. The social source of influence is also critical: theory and evidence suggest in-group advantages in cognition and affect, with some studies showing greater conformity to in-group norms and even divergence from disliked out-groups, while other studies report no difference between in- and out-group (or even human vs. computer) sources. Minimal group paradigms can induce group distinctions but may be less motivationally salient than real-world identities, potentially moderating in-group advantage effects.

Design and preregistration: A preregistered EEG study (https://osf.io/5e7kr/) used a within-subject 2 (Affiliation: in-group, out-group) × 3 (Feedback: higher, lower, consistent) social learning design to examine effects on explicit ratings and spontaneous neural representations of attractiveness. Participants: Forty-eight university participants (37 females; 43 heterosexuals; mean age = 23.98, SD = 3.13) completed the study; 45 met EEG inclusion criteria following artifact rejection and correct group recall. All were right-handed native Chinese speakers with normal or corrected vision and no neurological/psychological disorders. Ethics approval: HREC, University of Hong Kong (EA1912003). Stimuli: 70 medium-attractive East Asian female faces (original photos with hair/ears removed), and 10 prototypical attractive faces (morphed from database exemplars). Pilot ratings (n=18) confirmed prototype faces were more attractive than medium-attractive faces (mean 5.84 vs. 3.77 on 1–7 scale; t(17)=11.91, p<0.001, d=3.09). Procedure: Two lab visits separated by 7 days. Visit 1 included questionnaires (PANAS-SF, IRI, TLQ, SDR, SPIN) and three phases: (1) pre-learning, (2) learning, (3) post-learning. Pre- and post-learning phases each included: (a) face perception task with EEG and (b) explicit rating task. In the face perception task, participants viewed 480 faces intermixed with 144 objects across 6 blocks; they pressed a button only when objects appeared (target hit rate > 0.99), minimizing explicit attractiveness evaluation. In the explicit rating task, participants rated 80 faces (70 experimental, 10 prototypes) on attractiveness, confidence, competence, and warmth (1–11). The learning phase comprised minimal group formation (random assignment to Green or White groups, framed as preference-based), an associative name–group learning task to strengthen group identity, followed by the social learning task wherein each to-be-learned face was presented with feedback ostensibly from in-group or out-group members indicating attractiveness ratings that were higher, lower, or consistent with the participant’s initial rating. Each of the six Affiliation × Feedback conditions had 10 faces; 10 additional faces served as no-learning controls. Post-learning tasks mirrored pre-learning; participants also completed a cued recall of face–feedback memory and group identification questions. Visit 2 (7-day delay) included only the explicit rating task (no EEG). Behavioral measures: Attractiveness ratings were mean-corrected within phase to control for regression-to-mean and phase-wise shifts. Immediate update = post-learning minus pre-learning mean-corrected rating; delayed update = delayed minus pre-learning. EEG acquisition and preprocessing: 64-channel ANT Neuro system, 500 Hz sampling (downsampled to 250 Hz offline), CPz reference, AFz ground; 0.05–30 Hz FIR band-pass, 50 Hz CleanLine; EOG/M1/M2 removed; bad channels interpolated; data epoched for ICA (−1000 to 2000 ms) with 1 Hz high-pass; ICA components reflecting ocular/muscle artifacts removed (ICLabel + visual inspection); artifact threshold ±100 µV (preregistered ±75 µV examined in supplements); trials with false alarms excluded. On average, 475.5 (SD=34.1) pre-learning and 453.0 (SD=44.0) post-learning trials were retained. ERP analysis: ERPs epoched −200 to 1000 ms with baseline −200–0 ms. Preregistered components: N170 at bilateral occipitotemporal sites (left: T7, TP7, P7, PO7; right: T8, TP8, P8, PO8) in 120–220 ms; LPC at frontal-central (Fz, FCz, F1/2, FC1/2) and central-parietal (CPz, CP1/2, Pz, P1/2) sites in 300–800 ms. Mean amplitude and adaptive means (50 ms for N170; 100 ms for LPC) were analyzed as post–pre changes. Multivariate RSA: EEG downsampled to 100 Hz; z-transform per time point within participant to reduce ERP amplitude effects. For 0–1000 ms, overlapping 200 ms windows advanced in 10 ms steps. Spearman correlation across all 61 scalp channels computed between each experimental-face trial and all prototype-face trials, restricting comparisons to trials separated by >4 trials to control temporal proximity. Experimental–prototype similarity (EPS) per face was averaged across correlations. EPS updates were computed as post- minus pre-learning. Cluster-based nonparametric permutation tests (FieldTrip; 5000 permutations; cluster alpha 0.05; two-tailed) assessed feedback and affiliation effects; repeated-measures ANOVAs tested factors where applicable. Statistics and preregistration: Behavioral repeated-measures ANOVAs (afex in R) with Bonferroni-corrected post hocs (emmeans) and Bayes factors (BayesFactor). Moderation by TLQ analyzed via regression with feedback direction (higher vs. lower) and interaction. RSA analyses used cluster-based permutation. Deviations from preregistration and additional analyses are reported in the supplements/OSF.

Behavioral conformity: A 2 (Affiliation: in-group, out-group) × 3 (Feedback: higher, lower, consistent) ANOVA on immediate attractiveness updates (post minus pre) showed a significant main effect of Feedback (F(1,91,89.75)=9.17, p<0.001, η²=0.07, BF10=720.31). Pairwise: lower < higher (t(47)=3.79, p=0.001, d=0.60, BF10=109.59) and lower < consistent (t(47)=3.56, p=0.003, d=0.53, BF10=47.22); higher vs. consistent not significant (t(47)=0.89, p=1.000, d=0.14, BF10=0.17). No main effect of Affiliation (F(1,47)=0.31, p=0.579, η²=0.001, BF10=0.15) and no Affiliation×Feedback interaction (F(1.93,90.52)=0.66, p=0.513, η²=0.005, BF10=0.20). Delayed updates (7-day) showed no significant effects (ps>0.430, BF10<0.09). Individual differences: Perceived tightness–looseness (TLQ) moderated immediate updates by feedback direction (β=−0.24, SE=0.09, p=0.009). Simple slopes indicated opposite tendencies across directions (higher: t(188)=1.97, p=0.050; lower: t(188)=−1.77, p=0.078); the difference between slopes was significant (t(188)=2.65, p=0.009). Individuals perceiving tighter norms showed stronger immediate explicit updates consistent with feedback direction. ERP results: Pre- to post-learning changes in N170 (OT sites) and LPC (FC and CP sites) revealed no significant main effects of Affiliation or Feedback, nor interactions (ps>0.056; η²<0.009), indicating no univariate ERP evidence for attractiveness updating. Multivariate RSA (EPS updates): Across affiliations, Feedback significantly affected EPS update (cluster permutation Pcluster=0.013). Post hoc: higher > consistent (Pcluster=0.005); higher vs. lower and consistent vs. lower not significant (Pclusters>0.076). By affiliation, a significant effect of Feedback on EPS update emerged in the in-group condition (Pcluster=0.048): higher > consistent (Pcluster=0.016), with a trend for higher > lower (Pcluster=0.070); consistent vs. lower not significant (Pclusters>0.260). No significant feedback effect was found in the out-group condition (Pclusters>0.211). An Affiliation×Feedback ANOVA on EPS updates showed no significant interaction (Pclusters>0.521). Exploratorily, among female participants, in-group showed greater EPS update difference (higher vs. lower) than out-group (Pclusters<0.041). TLQ and neural representations: Splitting by median TLQ, the high-TLQ group showed significant in-group EPS update effects: higher > lower (Pcluster=0.028) and higher > consistent (Pcluster=0.028); consistent vs. lower not significant. The low-TLQ group showed no significant contrasts (Pclusters>0.101). These effects were specific to in-group learning and not attributable to differences in in-group favoritism ratings. Overall: Participants conformed behaviorally to social feedback immediately, regardless of affiliation, and multivariate EEG revealed spontaneous neural representation updating of attractiveness, particularly for in-group feedback and among individuals perceiving tighter social norms.

The study addressed whether social influence on facial attractiveness is internalized, and whether the source of influence (in-group vs. out-group) and individual norm perceptions modulate this process. Behaviorally, participants updated attractiveness ratings following both in-group and out-group feedback, with clear effects of feedback direction but no affiliation advantage—consistent with the mixed literature and potentially explained by the minimal group manipulation’s lower motivational salience compared to real-world identities. Critically, multivariate RSA of EEG recorded during a task without explicit evaluation showed that social influence updated spontaneous neural representations of facial attractiveness. This supports internalization beyond overt evaluation and demand characteristics. The neural updating was most evident when learning from in-group members and among participants perceiving tighter social norms, though the preregistered affiliation-by-feedback interaction on RSA updates was not significant, tempering strong claims of selective in-group internalization. Together, the findings suggest that social feedback can reshape perceptual–evaluative representations of faces, providing a mechanistic account of how social norms can continue to influence behavior even when not salient. Individual differences in perceived tightness-looseness appear to modulate both explicit and neural susceptibility to social influence, aligning with theories that tighter norm perceptions increase motivation to adhere to group norms.

This preregistered EEG study demonstrates that social conformity can update spontaneous neural representations of facial attractiveness in the absence of explicit evaluation, indicating internalization of social influence. Behaviorally, participants adjusted attractiveness ratings in line with social feedback from both in- and out-group members. Neural representation updating was particularly evident for in-group feedback and among individuals who perceive social norms as tighter. These findings advance understanding of how social influence shapes internal evaluative representations and highlight the roles of group affiliation and individual norm perceptions. Future research should test stronger, real-life group identities, include non-social control feedback (e.g., computer sources) to dissociate social from general learning, assess the persistence of neural internalization over longer delays with EEG/fMRI, extend to attitudes central to values (e.g., moral or political beliefs), examine potential gender effects, and conduct cross-cultural studies on tightness-looseness.

• No non-social control condition was included, limiting the ability to separate social from non-social learning effects.
• EEG was not recorded at the delayed test, preventing assessment of the longevity of neural internalization.
• The study focused on facial attractiveness—a relatively malleable attitude—so generalization to core values (e.g., moral or political attitudes) is uncertain.
• Sample characteristics and stimuli limit generalizability: most participants were heterosexual females, and only Asian female faces were used; potential gender and cross-ethnicity effects were not tested.
• Although in-group RSA effects were observed, the preregistered affiliation-by-feedback interaction on neural updates was not significant, limiting strong conclusions about selective in-group internalization.
• Minimal group manipulation may provide weaker motivational salience than real-world group identities, potentially attenuating in-group advantage effects.