Neurodynamic evidence reveals identity top-down influences emotional contagion of race

Abstract

Faces contain important information about emotion, race, identity, and age. A large body of research has illustrated that emotional contagion is influenced by race. The Categorization-Individuation Model (CIM) suggests that situational cues (e.g., authority, subjectively important ingroup–outgroup) cause perceivers to shift their attention to identity-diagnostic facial characteristics, especially for other-race faces. The current study is designed to reveal whether identity can top-down influence emotional contagion across races, and the time course of this influence. We recruited 30 Chinese college students to participate in two experiments. Experiment 1 used dynamic emotional faces of Asians and Whites to assess emotional contagion in different races. Experiment 2, based on experiment 1, employed a minimal group paradigm assigning identity information to the racial faces. We used ERP analysis to predict the potential neural mechanism of the influence of identity on racial emotion contagion, and used representation similarity analysis (RSA) to explore the temporal dynamics of the representation of race, emotion, and identity. Our results showed that (1) in experiment 1, Whites produced stronger P1 amplitudes than Asians; in experiment 2, RSA results showed that the time course of representation of race was about 100 ms. (2) In experiments 1 and 2, Happy produced stronger P200 amplitude than Angry; Asians produced stronger P200 amplitude than Whites; The RSA results showed that the time course of representation of emotion and emotional contagion both began about 200 ms after face appearance. (3) In experiment 2, the P300 amplitudes showed a significant interaction of identity and race, and in different group conditions, the P300 amplitude in Asians was stronger than in Whites; however, in the same group conditions, the difference between the two races was insignificant. Results illustrate that identity information top-down influences the neural mechanisms of racial emotional contagion, and the effects are divided into at least three stages: (1) an early stage bottom-up perceptual categorization of other-race; (2) a middle stage emotional and individualization processing; and (3) a late stage top-down modulation by identity cues. Our study is the first to explain the neurodynamics of emotional contagion processing using the Categorization-Individuation Model.

Introduction

Emotional contagion is a core mechanism of human social interaction, influenced not only by the emotional expressions but also by social context such as race and identity of the expresser. Understanding how these social cues modulate emotional contagion can provide insight into the neural basis of emotion processing. Faces play an important role in nonverbal communication (Barraclough and Perrett 2011). People can extract important information from faces such as age, race, identity, and emotion (Hadders-Algra 2022), and facial information is also an important channel for emotional contagion. Faces are particularly well suited for studying these mechanisms because they convey both emotional and identity-related information.

The process by which a person or a group influences the emotions or behaviors of another person or group by consciously or unconsciously inducing emotional states and behavioral attitudes is called emotional contagion (Schoenewolf 1990). While initially considered an automatic response (e.g. Hatfield et al. 1994), growing evidence shows that emotional contagion is influenced by social context, including perceived similarity, group membership, and interpersonal relationships (Weyers et al. 2009; Wróbel and Królewiak 2017). For instance, the correction hypothesis posits that a top-down social appraisal can be initiated to block, attenuate, or reverse the default emotional contagion response when the sender’s social characteristics are perceived to conflict with the receiver’s goals (Wróbel and Imbir 2019). Furthermore, emotional responses to facial expressions can vary depending on the race of the sender (Peng et al. 2020), with studies showing a same-race advantage in emotional mimicry and differences in early neural responses (90 ~ 170 ms post-stimulation) to same-race versus other-race faces (Zhou et al. 2022a). These findings illustrate that the emotions of different races convey different emotional processing. In recent years, several studies have sought to narrow the differences between races. For example, Zhou et al. (2022b) found that learning from the experiences of other-race individuals enhances the impression of other-race.

Hugenberg et al. (2010) proposed the Categorization-Individuation Model (CIM), which posits that face coding involves two types of processing: categorization and individualization, with motivation serving as the core of the model. Categorization refers to the low-level perceptual characteristics that distinguish among social categories, such as race, sex, and age, which are extracted quickly by the visual system. Individualization, by contrast, refers to the attention to identity-diagnostic facial characteristics, to distinct target faces within a category (Tang et al. 2017). And individualization is more prolonged and effortful (Rollins et al. 2020). When processing faces, the categorization of dimensions occurs quickly. Other-race faces tend to elicit stronger category activation than same-race faces. This should elicit stronger attention to category-diagnostic facial characteristics in other-race faces for categorization, and stronger attention to identity-diagnostic facial characteristics in same-race faces for making individualized judgments. Situational cues (e.g., authority, subjectively important ingroup-outgroup) can enhance the motivation to individualization, causing perceivers to shift their attention to identity-diagnostic facial characteristics, especially for other-race faces, thereby weakening the other-race effect (Young et al. 2012). In sum, the CIM suggests that there may be a priority difference between processing race and identity, and the identity can alter the processing bias of racial faces.

The ERP studies on face processing also indicate that processing the race and the identity of the face may be carried out in two different ways. Processing race appears to be automated, rapidly processed, and is a bottom-up pathway, with attention paid to the early P1 stage, where participants direct their attention to the other-race face rather than the same-race face (de Lissa et al. 2023; Ito and Urland 2003,2005). Bottom-up representations of race are often automatically encoded and have an impact on perceptual processing within a few hundred milliseconds (Kaul et al. 2014). Bruce and Young (1986) proposed that top-down social information, e.g., identity-related information, does not influence early face structure encoding processes. Ratner and Amodio (2013) argued that identity processing is top-down information and that when they used the minimal group paradigm to categorize faces into two groups, they found an N170 effect for in-groups, with a higher degree of in-group faces than out-group. Schindler and Bublatzky (2020) reviewed previous research and proposed a model of facial emotion perception as a function of attention, which showed in the ERP time course is evidenced by the following: in the early stage (P1/N170), low-level analysis occurs with little influence from top-down attentional goal-driven; in the middle stage (P2/EPN), face recognition is based on structural information, and processing emotional information occurs in this stage; and in the late stage (P300/LPP), top-down and bottom-up processing interact with each other.

In summary, we hypothesize that the processing of racial faces may be a “bottom-up” processing strategy. Race categorization may occurs at the early stage (such as P1) and is driven by attention. Bottom-up processing focuses on basic information about the face image, such as depth and orientation (Lee and Lee 2011). Chinese faces have wider noses and smaller mouths than White faces (Le et al. 2002), and these facial differences cause different attentional resource allocation. According to the previous research on the stages of face processing (e.g., Schindler and Bublatzky 2020), we hypothesize that at the early stage of face processing, racial features may elicit bottom-up processing driven by low-level perceptual features. At middle stage (P2/N170), emotion expression and face’s second order spatial configuration as identity-diagnostic facial characteristics, this stage may be more sensitive to happiness (Schweinberger and Neumann 2016) and same-race. In late stages (P300), identity-related information, such as group membership, manipulated through the minimal group paradigm, not driven by perceptual input but by prior knowledge and expectations, exerts a top-down influence that modulates the processing of racial faces.

However, in most studies on emotional contagion or emotional mimicry, the boundary between race and identity has not been well distinguished, and many studies have simply attributed racial differences to the fact that the other-race belongs to a different group (e.g., Hess and Fischer 2013; Peng et al. 2020). Few studies have confirmed the effect of identity on emotional contagion across racial faces. Do race and identity have different priorities for emotional contagion as assumed by the CIM? Race and identity may have different meanings, with race being a categorization that is processed bottom-up and identity being an individuation that is processed top-down, making it necessary to distinguish between race and identity in the context of emotional contagion. In our study, race serves as a low-level perceptual characteristic that elicits early bottom-up categorization, whereas identity, manipulated via the minimal group paradigm, represents a top-down situational cue that emerges at later stages of processing. Group membership can be experimentally manipulated through arbitrary social categorization in the laboratory, such as the minimal group paradigm (Tajfel 1970). People acquire a part of their self-concept from the category and social group to which they belong (Turner et al., 1987). Even in virtual groups, people will identify with the group they are in (Ratner and Amodio 2013; Tajfel 1970). Therefore, by experimentally manipulating group membership, we aim to examine how top-down identity cues interact with bottom-up race processing in shaping emotional contagion.

To investigate the effects of race and identity on emotional contagion, the current study contained 2 experiments. Experiment 1 replicated the method of Peng et al. (2020) and used Asian and White dynamic emotional faces as materials to explore the differences between emotional contagion to the same-race (Asian) and other-races (White). However, in Peng et al. (2020), only the differences in emotional mimicry in viewing the emotions of different racial faces were explored, and the differences in emotional contagion and neural mechanisms were not explored. Experiment 1 supplemented the above two points. Experiment 1 was a 2 × 2 within-subjects design with emotions containing happy and angry, and races containing Asian and White. We also wanted to examine the effect of identity on racial processing and to investigate when this effect occurs. Experiment 2 was based on Experiment 1, using a minimal group paradigm, with the addition of the within-subjects variable of “group membership”. Experiment 2 was a 2 × 2 × 2 within-subjects design, with emotions including happy and angry, race including Asian and White, and group including the same group and the different group. Group means that the faces were categorized into the same group or a different group from the participants.

To investigate how race and social identity (group membership) modulate emotional contagion over time, in the current study, we employed event-related potential (ERP) and representational similarity analysis (RSA). We focused on the P1, N170, P200, and P300 components. We speculate that racial differences will be observed at the P1, emotional differences may be observed at the N170 or P200, and group × race interactions will be observed at the P300. RSA is a powerful multivariate pattern analysis technique that effectively characterizes the relationships between neural representations across different data modalities (Lu and Ku 2020), and it has been used in recent years to explore the time course of face processing (Cauchoix et al. 2014; Li et al. 2022). RSA interprets the neural response patterns associated with each experimental condition as the neural representation of a specific mental state (Wagner et al. 2019). We use RSA to reveal the temporal dynamics of the representation of race, identity (group membership), emotion type, and emotional contagion of the participants. RSA has been used to demonstrate that representing gender and age information of faces precedes the representation of identity information of faces (Dobs et al. 2019). Previous studies have rarely explored the time course representing race and identity simultaneously. The current study is the first to use RSA to investigate this time course to explore whether these representations occur at the early stage of low-level analysis or at the stage of processing the structure of faces. Importantly, RSA enables us to determine whether race and identity information emerge at distinct time points, which allows us to test the Categorization-Individuation Model (Hugenberg et al. 2010).

Experiment 1: neural mechanisms of emotional contagion to different Racial faces

Methods

Participants

The sample size required for the experiment was calculated by the software G.power 3.1 (Faul et al. 2007) with an effect size f of 0.25 and an alpha level of 0.05, which showed that 26 subjects were able to achieve power of 80%. We recruited 32 healthy Han Chinese college students from Renmin University of China, and after excluding 2 participants’ data with fewer trials remaining after preprocessing, data from 30 participants entered the final analysis (21 females, 9 males, M = 20.17, SD = 2.692). They had normal visual or corrected visual, and no history of neurological or psychiatric problems. All participants signed an informed consent form before the study. The study was approved by the Institutional Review Board of the Department of Psychology at the Renmin University of China. Participants were offered course credit or cash as rewards.

Experimental design

The current experiment was a 2-factor within-subjects design: 2 (emotion: angry, happy) × 2 (race: Asian, White). The dependent variable is the emotional contagion self-rating scores (SAM), the amplitude of the P1, N170, P200, and P300 components.

Materials

Referring to Chen and Zhu (2019), we used the FaceGen Modeller software to generate 40 Asian faces (20 males and 20 females) and 40 white faces (20 males and 20 females) aged 20–30 years. We recruited 32 Chinese college students (7 males, M = 20.79, SD = 3.03) to rate the degree of similarity of these faces to Asians/Whites on a 7-point Likert scale, where 1 means ‘this face was definitely Asian’ and 7 means ‘this face was definitely White’.

Some faces with atypical scores were removed, and 36 Asian faces (18 males and 18 females) and 36 White faces (18 males and 18 females) served as experimental stimuli. Independent samples t-tests were used to compare the mean scores of Asian and White faces, and it was found that the scores of Asian faces were significantly lower than the scores of white faces (Asian faces: M = 2.33, SD = 0.236, White faces: M = 6.27, SD = 0.219, t₇₀ = -73.381, p < 0.001), and both of them were significantly different from the median score of 4 (Asian faces: t₃₅ = -42.412, p < 0.001; White faces: t₃₅ = 62.215, p < 0.001). This indicates that the face material used in this study was able to distinguish Asian from White.

We then used the Morph function of the FaceGen Modeller software to generate expressions for each face that gradually shifted from calm to happy/angry, and based on existing emotional contagion/mimicry studies (Achaibou et al. 2008; Kuang et al. 2021), the intensity of the expressions was sequentially 0%, 15%, 30%, 45%, 60%, 70%, 80%, 90%, 100% and 110%. Video clips were synthesized from these images by using Matlab 2013b (MathWorks, MA, USA). In each video clip, the first 9 images (i.e., expressions of 0%-100% intensity) were presented for 40 ms each, and the last expression of 110% intensity was presented for 1100ms, with the entire video clip lasting 1460ms (see Fig. 1a).

Fig. 1.

Materials and procedures used in Experiment 1. Note: a) Emotional contagion stimulus material and b) experimental procedure

Procedure

The current experiment contained four levels (2 × 2), each containing a total of 54 trials in a pseudo-randomized order. There were 5 blocks in the current experiment.

To prevent fatigue, participants rested for one minute after each block. The flow within each trial is shown in Fig. 1b. First there was a 200-ms gaze point, followed by a 1-s blank screen, the video clip lasted for 1460 ms, immediately followed by a self-assessment manikin (SAM) scale (Bradley and Lang 1994), and when the SAM scale was pressed, a randomized black screen of 1–1.2 s was displayed (see Fig. 1b). Subjective feelings of emotion were collected using a nine-point rated SAM, where 1 indicates very angry, 9 indicates very happy, and 5 indicates calm. All experimental stimuli were presented using E-prime 3.0. Participants first filled out the experimental informed consent form. After the participants fully understood the experimental task, the experiment began.

Data acquisition and preprocessing

EEG data were acquired using a NeuroScan SynAmps amplifier and a 64-channel Ag/AgCl nickel electrode Quik-Cap EEG cap, these were distributed over the head surface according to the extended 10–20 EEG system (Oostenveld and Praamstra 2001). The EEG data was sampled at 1000 Hz and referenced online at M1 electrode, and all inter-electrode impedance was maintained below 10 k Ω.

The EEG data were processed offline using the EEGLAB toolbox based on Matlab 2013b. Electrodes of no concern (M1,M2,VEO, etc.) were removed and re-referenced offline to obtain a global average; resampling to 500 Hz filtered out signals above 30 Hz. The epochs of -100–500 ms from the face presentation were created, and the 100 ms before the face presentation as the baseline. Trials with fluctuations exceeding ± 80 µV were removed, and artifacts such as blinks and eye movements were manually removed using the ICA method. Ensure that at least 30 trials per condition per participant were retained, removing 2 participants with fewer remaining trials.

Based on previous studies related to face processing and emotional contagion/mimicry (Kuang et al. 2021; Lomoriello et al. 2021; McCrackin and Itier 2021), we selected the average of three electrode sites, P7,P8,Pz as regions of interest. Based on the ERP waveforms of all conditions and the components of interest in the relevant studies and their time windows, we focused on the following components: P1, the wave peak appeared 122 ms after the face presentation, 112–132 ms; N170, the wave peak was located 172 ms after the face presentation, 162–182 ms; and P200, the wave peak was located 226 ms after the face presentation, 216–236 ms; P300, the wave peak was located 332 ms after stimulus appearance, 322–342 ms.

Data analysis

Manipulation check

To test whether emotional contagion occurred and whether the emotion was successfully manipulated, the means of each participant’s SAM for each condition were calculated, and these means were subtracted from a median score of 5 to obtain an emotional self-report score with a median value of 0, where − 4 means “very angry”, 0 means calm, and 4 means “very happy”. SAM scores for happy and angry were compared with 0 in a one-sample t-test to confirm whether emotional contagion had occurred. A paired samples t-test was conducted to determine whether the SAM scores for happy were higher than the SAM scores for angry to test whether the manipulation of emotion was successful.

Emotional contagion

To explore the degree of emotional contagion, the SAM scores were converted to values between 0 and 4: for happy emotions: SAM-5, and for angry emotions: 5-SAM. Where 0 means “calm” and 4 means “very happy/angry”. Converted SAM were analyzed using a 2-way repeated measures ANOVA for emotion (happy, angry) × race (Asian, White) using SPSS 19.0.

ERP analysis

Our region of interest was designated as the average amplitude of P7, P8, and Pz electrodes. The amplitude means of P1, N170, P200, and P300 within their respective time windows were extracted for each participant, and each condition, and a 2-factor repeated-measures ANOVA was performed for each ERP component for emotion (happy, angry) × race (Asian, White).

Correlation analysis

Pearson correlation analyses were conducted between the mean amplitudes of the P1, N170, P200, and P300 components and the emotional contagion scores for each condition level (Asian happy, Asian angry, White happy, White angry), using SPSS 19.0.

Results

Manipulation check

A paired samples t-test of happy and angry SAM scores revealed a significant difference between happy and angry (Aappy: M = 1.680, SD = 0.416, Angry: M = -2.941, SD = 0.504, t₂₉ = 39.312, p < 0.001) and both happy and angry SAM scores were significantly different from 0 (happy: t₂₉ = 22.118, p < 0.001, angry: t₂₉ = -31.959, p < 0.001). It showed that emotional manipulation is effective and that emotional contagion occurs.

Emotional contagion

A two-way repeated measures ANOVA on converted SAM revealed a significant emotional main effect (F_(1,29) = 108.535, p < 0.001,η²_p = 0.789, Happy: M = 1.680, SD = 0.416, Angry: M = 2.941, SD = 0.504). The main effect of race was significant (F_(1,29) = 28.865, p < 0.001, η²_p = 0.499, Asian: M = 2.163, SD = 0.350, White: M = 2.458, SD = 0.360). In addition, the emotion × race interaction was significant (F_(1,29) = 17.566, p < 0.001, η²_p = 0.377), and White had higher emotional contagion scores than Asian in angry (Asian: M = 2.682, SD = 0.563, White: M = 3.199, SD = 0.475, F_(1,29) = 116.002, p < 0.001, η²_p = 0.800), whereas there was no significant difference between Asian and White in happy (Asian: M = 1.644, SD = 0.380, White: M = 1.716, SD = 0.585, F_(1,29) = 0.555, p = 0.462, η²_p = 0.019). It indicates that emotional contagion is higher for Whites under the angry (see Fig. 2).

Fig. 2.

Results of analysis of variance on two-way measurement of emotional contagion. Note: a) Main effect of Emotion, b) Main effect of Race and c) The interaction effect of Emotion and Race, * p < 0.05, ** p < 0.01, *** p < 0.001, error bars indicate SEM (Standard Error of the Mean)

ERP results

P1

In the P1 amplitude race main effect was significant. In particular, White elicited a larger positive amplitude compared to Asian (Asian: M = 2.158, SD = 1.590, White: M = 2.448, SD = 1.574, F_(1,29) = 7.075, p = 0.013, η²_p = 0.196). The main effect of emotion and interaction effect were not significant (all ps > 0.05)(see Fig. 3a).

Fig. 3.

ERP results. Note: a) Race effect of P1 component, b) Emotion effect of P200 component, c) Race effect of P200 component, d) Emotion effect of P300 component and e) ERP waveforms, * p < 0.05, ** p < 0.01, *** p < 0.001, error bars indicate SEM

N170

Neither the N170 amplitude main effect nor the interaction was significant (all ps > 0.05).

P200

Both emotion main effects and race main effects were significant. In particular, happy elicited a larger positive amplitude compared to angry (Happy: M = 2.352, SD = 2.179, Angry: M = 2.033, SD = 2.242, F_(1,29) = 9.310, p = 0.005, η²_p = 0.243); Asian elicited a larger positive amplitude compared to White (Asian: M = 2.326, SD = 2.106, White: M = 2.059, SD = 2.329, F_(1,29) = 4.311, p = 0.047, η²_p = 0.129). And the interaction between emotion and race was not significant (p > 0.05) (see Fig. 3b, c).

P300

The emotion main effect was significant, with happy eliciting a larger positive amplitude compared to angry (Happy: M = 2.205, SD = 2.858, Angry: M = 1.585, SD = 2.916, F_(1,29) = 21.878, p < 0.001, η²_p = 0.430). The race main effect and the interaction effect were not significant (all ps > 0.05) (see Fig. 3d).

Correlation results

The Pearson correlation results show that: under the Asian Happy condition, emotional contagion scores were significant positive correlated with the amplitudes of P200 (r = 0.473, p = 0.008; Fig. 4a) and P300 (r = 0.362, p = 0.0496; Fig. 4c). Under the White Happy condition, emotional contagion scores were also significant positive correlated with P200 (r = 0.391, p = 0.033; Fig. 4b) and P300 (r = 0.443, p = 0.014; Fig. 4d). In contrast, under the Asian Angry condition, emotional contagion scores were significant negative correlated with P200 amplitude (r = -0.398, p = 0.029; Fig. 4e). The Pearson correlation coefficients in other conditions were not significant (ps > 0.05).

Fig. 4.

Correlation results. Note: Pearson correlation result in emotional contagion score between a) P200 amplitude in Asian Happy, b) P200 amplitude in White Happy, c) P300 amplitude in Asian Happy, d) P300 amplitude in White Happy and e) P200 amplitude in Asian Angry

Experiment 2: the influence of identity on Racial emotion contagion and its neural mechanism

Method

Participant

The participants for Experiment 2 were the same as for Experiment 1.

Experimental design

The current experiment was a 3-factor within-subjects design: 2 (emotion: angry, happy) × 2 (group: same group, different group) × 2 (race: Asian, White). The dependent variable is the emotional contagion self-rating scores (SAM), the average amplitude of P7, P8 and Pz in the P1, N170, P200, and P300 components.

Materials

The stimulus materials for Experiment 2 were the same as for Experiment 1.

Procedure

There were a total of 8 levels (2 × 2 × 2), each containing 54 trials in a pseudo-randomized order. The tasks were divided into a total of 8 blocks, each containing 54 trials. to prevent participant fatigue, all participants took at least a 5-minute break after taking part in Experiment 1 before entering Experiment 2.

Referring to previous research in the minimal group paradigm (Ratner and Amodio 2013), participants were tested for their cognitive style (overestimation type, underestimation type) using the Numerical Estimation Style Test (NEST) task and were assigned to the corresponding subgroups (red team, blue team). Participants’ responses were independent of their final grouping. Before wearing the EEG caps, participants put on the corresponding color team uniforms.

All experimental stimuli were presented using E-prime 3.0. The experimental procedure is shown in Fig. 5. First, there was a 200-ms gaze point, followed by a 500-ms presentation of group cues, which consisted of red or blue clothing, indicating that the faces presented next belonged to either the blue or red team, and that the face that matched the participant’s team was the same group (SG), and vice versa, the different group (DG). This was followed by a 1-s blank screen, the video clip lasted for 1460 ms, immediately followed by a self-assessment manikin (SAM) scale (Bradley and Lang 1994), and when the SAM scale was pressed, a randomized black screen of 1–1.2 s was displayed. In order to conduct a manipulation test for grouping, at the end of the experiment, participants were required to complete the Overlap of Self Ingroup and Outgroup questionnaire (OSIO), which ranges on a scale of 1–7, with higher scores indicating closer relationships. This questionnaire was adapted from the Inclusion of Other Scale (IOS), which is often used as a manipulation test for the minimal group paradigm (Aron et al. 1992; Schubert and Otten 2002; Van Der Schalk et al. 2011).

Fig. 5.

Procedure for Experiment 2

Data acquisition and preprocessing

Refer to Experiment 1 for data acquisition and preprocessing.

Data analysis

Manipulation check

The manipulation check for emotional contagion is the same as in Experiment 1. In addition, to test whether the manipulation was successful for the group, a paired-sample t-test was performed on the scores of OSIO to the SG versus DG.

Emotional contagion

To explore the degree of emotional contagion, the SAM were converted to values between 0 and 4 (see Experiment 1 for details). The converted emotional contagion scores were subjected to a 3-factor repeated measures ANOVA using SPSS 19.0 for group (SG, DG) × emotion (happy, angry) × race (Asian, white).

ERP analysis

Our region of interest was designated as the average amplitude of P7, P8, and Pz electrodes. The amplitude means of P1, N170, P200, and P300 within their respective time windows were extracted for each participant, and each condition, and a 3-factor repeated-measures ANOVA of group (SG, DG) × emotion (happy, angry) × race (Asian, White).

Correlation analysis

Pearson correlation analyses were conducted between the mean amplitudes of the P1, N170, P200, and P300 components and the emotional contagion scores for each condition level (SG Asian happy, SG Asian angry, SG White happy, SG White angry, DG Asian happy, DG Asian angry, DG White happy, DG White angry), using SPSS 19.0.

Representational similarity analysis (RSA)

To determine when neural representations discriminated faces at the level of group, race, and emotion, we used the NeuroRA toolbox (Lu and Ku 2020) based on Python 3.7.16 to perform representational similarity analysis on EEG data. Representational dissimilarity matrix (RDMs), which are the dissimilarity between a pair among the eight experimental levels, were first created, with larger values indicating greater dissimilarity. For the EEG RDMs, we computed time-resolved RDMs using cross-validated Euclidean distance (Ashton et al. 2022). Using Python 3.7.16, we implemented a 4-fold cross-validation procedure at each time bin for each participant and each pair of the 8 experimental conditions. Within each fold, trials were randomly permuted and averaged to generate pseudo-trials. For each condition pair (e.g., Asian happy vs. White happy), we computed the cross-validated Euclidean distance between pseudo-trial patterns from independent folds, which provides a more reliable estimate of representational dissimilarity by enhancing the signal-to-noise ratio (Ashton et al. 2022; Walther et al., 2016). This process was repeated 200 times with different trial permutations to enhance the stability of the estimates. The resulting time-by-time dissimilarity matrices for each participant were then averaged across permutations to form time-resolved RDMs at each time point. Next, model RDMs were constructed for the group, race, emotion, and emotional contagion scores. Model RDMs were 8 × 8 binary matrices, where a value of 1 was assigned to stimulus pairs from different categories (e.g., Asian vs. White for the race model; same group vs. different group for the group model), and 0 to pairs from the same category (e.g., Asian vs. Asian or same group vs. same group). For emotional contagion scores, the model RDM was computed as the absolute difference in contagion scores between each pair of conditions. Since the data in matrices above and below the diagonal were symmetric, referring to the method of Li et al. (2022), we selected only the data on the upper diagonal (see Fig. 6). Then, the Pearson correlation between the EEG RDMs and the model RDMs at every time point and for each subject were calculated.

Fig. 6.

Schematic for model representational dissimilarity matrix (RDM)

Results

Manipulation check

A paired-sample t-test of OSIO for the same and different groups revealed a significant difference between groups (SG: M = 4.167, SD = 1.512, DG: M = 3.167, SD = 1.512, t₂₉ = 3.042, p = 0.005). It indicates that the group manipulation was successful. Paired samples t-tests on happy and angry SAM scores revealed that there was a significant difference between happy and angry (Happy: M = 1.989, SD = 0.394, Angry: M = -2.948, SD = 0.542, t₂₉ = 33.131, p < 0.001), and that both happy and angry SAM scores were significantly different from 0 (happy: t₂₉ = 27.609, p < 0.001, angry: t₂₉ = -29.779, p < 0.001). It showed that emotional manipulation is effective and emotional contagion occurs.

Emotional contagion

A 3-factor repeated-measures ANOVA on emotional contagion scores revealed a significant main effect of emotion, with higher emotional contagion scores for angry than for happy (F_(1,29) = 117.908, p < 0.001, η²_p = 0.803, Happy: M = 1.989, SD = 0.395, Angry: M = 2.946, SD = 0.542), suggesting that angry emotional contagion to a higher degree. The race main effect was significant, with White eliciting higher emotional contagion scores than Asian (F_(1,29) = 92.501, p < 0.001,η²_p = 0.761, Asian: M = 2.298, SD = 0.413, White: M = 2.637, SD = 0.425). In addition, the emotion × race interaction was significant (F_(1,29) = 12.633, p < 0.001, η²_p = 0.303), and it showed higher scores for White compared to Asian under happy (Happy Asian: M = 1.882, SD = 0.424, Happy White: M = 2.096, SD = 0.426, F_(1,29) = 13.704, p < 0.001, η²_p = 0.321); and higher scores were found for White compared to Asian in angry (Angry Asian: M = 2.713, SD = 0.565, Angry White: M = 3.178, SD = 0.541, F_(1,29) = 132.072, p < 0.001,η²_p = 0.820). The emotion × group interaction was significant (F_(1,29) = 8.653, p = 0.006, η²_p = 0.230), it showed that there was no significant difference in emotional contagion scores between the same and different groups in the happy emotion (F_(1,29) = 3.511, p = 0.071, η²_p = 0.108); in the angry emotion, compared to the SG, the different group had higher emotional contagion scores (Angry SG: M = 2.892, SD = 0.534, Angry DG: M = 2.999, SD = 0.585, F_(1,29) = 4.185, p = 0.050, η²_p= 0.126) (see Fig. 7). However, we found no significant interaction between emotion × group × race (F_(1,29) = 0.005, p = 0.943, η²_p < 0.001).

Fig. 7.

Results of analysis of variance on three-way measurement of emotional contagion. Note: a) Main effect of Emotion, b) Main effect of Race, c) The interaction effect of Emotion and Group and d) The interaction effect of Emotion and Race. SG = same group, DG = different group, * p < 0.05, ** p < 0.01, *** p < 0.001, error bars indicate SEM

ERP results

P1

Neither the main effect nor the interaction was significant (all ps > 0.05).

N170

None of the main effects were significant, and the emotion × group interaction was significant (F_(1,29) = 9.577, p = 0.004, η²_p = 0.248). It showed that SG elicited higher negative amplitude in happy compared to the different group in happy (Happy SG: M = -0.394, SD = 2.239; Happy DG: M = -0.129, SD = 2.137, F_(1,29) = 5.701, p = 0.024, η²_p = 0.164), whereas this difference was not significant in angry (F_(1,29) = 1.717, p = 0.200, η²_p = 0.056) (see Fig. 8a).

Fig. 8.

ERP results. Note: a) The interaction effect of Emotion and Group in P1 component, b) Emotion effect of P200 component, c) Race effect of P200 component, d) Emotion effect of P300 component, e) Race effect of P300 component, f) The interaction effect of Race and Group in P300 component and g) ERP waveforms, SG = same group, DG = different group, * p < 0.05, ** p < 0.01, *** p < 0.001, error bars indicate SEM

P200

Emotion main effect was significant, with happy eliciting higher positive amplitude compared to angry (Happy: M = 2.284, SD = 2.059, Angry: M = 2.049, SD = 2.089, F_(1,29) = 5.196, p = 0.030, η²_p = 0.152). The race main effect was significant, with Asian eliciting higher positive amplitude compared to White (Asian: M = 2.358, SD = 2.083, White: M = 1.974, SD = 2.047, F_(1,29) = 25.124, p < 0.001, η²_p = 0.464). The remaining main effects and interactions were not significant (see Fig. 8b and c).

P300

Emotion main effect was significant, with happy eliciting higher positive amplitude compared to angry (Happy: M = 2.572, SD = 2.900, Angry: M = 2.100, SD = 2.944, F_(1,29) = 10.382, p = 0.003,η²_p = 0.264). The race main effect was significant, with Asian eliciting higher positive amplitude compared to White (Asian: M = 2.536, SD = 2.900, White: M = 2.136, SD = 2.915, F_(1,29) = 15.896, p < 0.001,η²_p = 0.354). The interaction between group and race was significant (F_(1,29) = 7.862, p = 0.009,η²_p = 0.213), and it showed that Asian elicited higher positive amplitude compared to White in different group (DG Asian: M = 2.693, SD = 3.050, DG White: M = 2.076, SD = 2.959, F_(1,29) = 19.844, p < 0.001,η²_p = 0.406), whereas this difference was not significant in SG (F_(1,29) = 2.625, p = 0.116,η²_p = 0.083) (see Fig. 8d, e, f).

Correlation results

The Pearson correlation result shows that: under the SG Asian Happy condition, the amplitude of P1 was significantly positively correlated with the emotional contagion score (r = -0.385, p = 0.034; Fig. 9). The Pearson correlation coefficients in other conditions were not significant (ps > 0.05).

Fig. 9.

Pearson correlation result in emotional contagion score between P1 amplitude and emotional contagion score in SG Asian Happy

RSA results

Pearson correlation between the EEG RDMs and the race RDM, group RDM, emotion RDM and emotional contagion score RDM at every time point and for each subject was calculated. The RSA results explain the EEG neural dynamics of the representation of race, group, emotion, and emotional contagion. As Fig. 10 shows the time course of 100–114 ms for the representation race (Orange line), 202–500 ms for the representation of emotion (Purple line), 200–500 ms for the representation of emotional contagion (Green line). The horizontal lines were marked below significant time courses with p < 0.05 (the cluster-based permutation test) (Lu and Ku 2020). There were no significant clusters for the representation group (Pink line).

Fig. 10.

Time course of Pearson correlations between EEG RDMs and Emotional Contagion RDM (green), Emotion RDMs (purple), Race RDMs (orange), and Group RDMs (pink), the bottom lines show the significantly time courses (p < 0.05, cluster-based permutation test)

Discussion

The current study aimed to investigate how and when identity modulates racial processing in emotional contagion. Rather than focusing on racial emotional contagion, we examined the dynamic influence of identity cues on the racial emotional contagion.

Our results revealed these main findings: (1) race-related neural responses emerged early (100 ms), reflecting bottom-up perceptual categorization; (2) emotional contagion began around 200 ms, primarily shaped by happy and same-race, but group membership had not yet affected on race; and (3) identity (group membership) exerted a top-down influence on race processing at 300 ms (P300), reducing neural difference between same-race and other-race faces.

These findings provide evidence that identity has a top-down influence over racial perception during emotional contagion, indicating that emotional contagion is influenced by appraisal of the social context. These neurodynamic evidences also align with the Categorization-Individuation Model (CIM): early race-driven categorization is followed by a later individuation process modulated by identity cues. Our results demonstrate that identity cues can attenuate race distinctions in emotional contagion at later stages, thus extending the CIM framework to the domain of emotional contagion. By integrating ERP analysis and RSA, this study offers neurodynamic evidence for the integration of perceptual and social information during real-time emotional contagion.

Emotional contagion is influenced by social context

The present study verified Hess and Fischer’s view (2014) and Wróbel and Królewiak (2017) that emotional contagion is not a simple reflex, but by the interpretation of expressions in a particular social context. Both Experiment 1 and Experiment 2 found higher levels of emotional contagion in the other-race than in the same-race; Experiment 2 also found higher levels of emotional contagion in the other-race than in the same-race under angry conditions, and higher levels of emotional contagion in the different group than in the same group under angry conditions. While the same-race/group preference for emotional contagion is more in line with our expectations, however, some studies have found a preference for emotional other-race or group. For example, Van der Schalk et al. (2011) conducted a study of emotional contagion and emotional mimicry using major (economics vs. psychology) to categorize groups. It was found that out-group anger made individuals more fearful. Similarly, Rauchbauer et al. (2015) argued that outgroup anger led to higher imitation behavior in individuals, suggesting that outgroup anger is a threat and behavioral imitation is a compromise strategy. Rauchbauer et al. (2016) and Axt (2018) used an implicit joint paradigm to find that positive preferences exist for the same-race, while there are negative preferences for other-race, and that positive self-focus leads people to attribute positive things to their race and negative things to the other-race. In the present study, since the self-report of emotions included only two dimensions, “happy” and “angry,” our results may suggest that strong feelings of angry toward different group or other-race come from a mixture of emotional experiences and that the increased emotional contagion of angry toward other race or group may come from a combination of out race/group and negative valence. However, the results of the present study differ from previous studies in that, unlike the findings of Wróbel et al. (2015), the degree of emotional contagion of different group angry was higher than that of same group angry in our study, i.e., similarity between the expresser and receiver did not result in emotional convergence.

Identity Affects Racial emotional contagion from the top-down

In summary, our findings support the view that identity modulates racial emotional contagion from the top-down, consistent with the predictions of the CIM (Hugenberg et al. 2010) and prior social-cognitive neuroscience frameworks (Hietanen et al. 2014; Schindler and Bublatzky 2020).

First, the early P1 and RSA results (around 100 ms) suggest bottom-up processing of physical facial features such as race. The P1 component is influenced by selective attention (Heinze and Mangun 1995), modulation of emotional content (Batty and Taylor 2003), and attention-attracting stimuli triggering greater P1 amplitude. The physical properties of the stimulus also modulate P1, and the other-race faces were a novel stimulus, which reflects participants’ attention to the White face (Kolassa and Miltner 2006; Leleu et al. 2015). Larger P1 amplitudes to White faces in Experiment 1 and early racial representation in RSA in Experiment 2 reflect other-race faces as novel stimulates attract more attention (Achaibou et al. 2008; Kolassa and Miltner 2006; Leleu et al. 2015). This might be categorization processing for the other-race faces as the CIM.

Second, at around 200 ms (P200), we observed stronger neural responses to happy expressions and Asian faces. This stage likely reflects deeper configural and emotional evaluation, but group membership had not yet affected race, suggesting that emotional contagion begins before identity modulation. The P200 responded to a deeper analysis of a face’s second-order spatial relations (Latinus et al. 2010; Pell et al. 2022). In our study, the larger P200 amplitude of the same-race than that of other-race reflects the processing of identity-diagnostic facial characteristics of the same-race, which is individualization processing. Happy emotions contribute to face recognition, which results from better encoding (Chen et al. 2015).

Third, at around 300 ms (P300), we observed an interaction between group and race, with the same group reducing the difference in processing of Asian vs. White faces. Schindler and Bublatzky’s (2020) model of facial emotion perception as a function of attention suggests that top-down information interacts with bottom-up information during the P300/LPP phase. Top-down attention makes effects on the P3/LPP responses reflecting later processing stages (Hietanen et al. 2014). There is no difference between the faces of the different group and the same group in our study. The identity cue is presented before faces, and the faces seen by the participants are the same. The group differences at the neural level indicate that the effect was not driven by perceptual input but by prior knowledge and expectations, which reflects top-down processing.

The dissociation between early race-driven perceptual attention and later group-modulated individuation aligns with the CIM framework, where other-race faces are initially categorization processes, while same-race faces are individuation processes. Furthermore, the P300 result is consistent with that top-down attention and social identity have an effect at later stages (Schindler and Bublatzky 2020). Thus, the identity cue altered the neural processing of racial emotional faces through top-down integration of social context with affective perception. At this stage, the identity-diagnostic facial characteristics of other-race may draw attention from the top-down.

One noteworthy finding of the present study is the dissociation between behavioral emotional contagion and ERP results. Specifically, participants reported stronger emotional contagion toward White (other-race) faces, particularly when the emotion was anger. However, ERP responses (P200 and P300 components) showed higher positive amplitudes in Asian (same-race) and happy faces. This might be caused by emotional contagion following two different paths of happiness and anger. Our correlation results show that in experiment 1, under the Asian Happy condition, emotional contagion scores were significantly positively correlated with the amplitudes of P200 and P300. Under the White Happy condition, emotional contagion scores were also significantly positively correlated with P200 and P300. In contrast, under the Asian Angry condition, emotional contagion scores were significantly negatively correlated with P200 amplitude. In the absence of identity cues, the strength of emotional contagion is associated with faces’ second-order spatial configuration processing (P200) and top-down attentional engagement (P300), but only for happy faces. For angry faces, the relationship appears reversed: increased second-order facial processing (P200) is associated with lower emotional contagion. This pattern supports the dual-pathway model of emotional contagion proposed by Deng and Hu (2018), which posits that happy faces evoke mimicry-based contagion, while angry faces require social appraisal. The group and race differences in emotional contagion across both experiments reinforce this distinction: contagion to happy faces may result from spontaneous facial processing, whereas contagion to angry faces is influenced by social appraisals, such as race and group membership. This suggests that emotional contagion is shaped not only by early perceptual processes but also by top-down assessments of social meaning (Wróbel and Imbir 2019). This is consistent with the result of RSA, that is, the time course of representing emotional contagion from 200 ms to 500 ms, meaning that the face’s second-order spatial relations represented by P200 and top-down process represented by P300 may indicate the process by which emotional contagion is influenced by appraisal of the social context. The discrepancy between ERP and behavior may reflect the distinction between cognitive processing and social context evaluation.

In experiment 2, identity cues were presented before faces, and group membership may have influenced attentional allocation, potentially obscuring early attention effects, which might be the reason why no significant P1 effect was observed in experiment 2. This early attention processing only shows differences when multivariate analysis (representation similarity analysis) is used. Interestingly, the only condition showing a significant correlation between P1 and emotional contagion was the same group Asian Happy condition. This may suggest that early attention to same group happy faces could decrease emotional contagion.

The current study has some limitations. First, although the use of normalized facial stimuli allowed us to control differences across conditions, such as expression intensity, hairstyle, or accessories (e.g., glasses), this standardization may have limited the generalizability of our findings to real-world emotional contagion. In real-life interactions, emotional contagion often occurs in response to unconstrained facial expressions. Therefore, future research should consider using more naturalistic facial stimuli to simulate emotional contagion in real social contexts. Second, we did not assess or control for participants’ attitudes toward other races. Prior research has shown that attentional and neural biases toward other-race faces are more strongly influenced by social prejudice than by familiarity (Giménez-Fernández et al. 2020). In particular, individuals with higher implicit bias may exhibit enhanced early neural responses (e.g., N170) to other-race faces, regardless of their exposure level. This suggests that attitudes toward the races may play a crucial role in modulating emotional contagion and cognitive responses to other-race faces. Future research should use explicit and implicit measures of racial attitudes to better control individual differences.

Conclusion

Our study used ERP and RSA to explain the time course of top-down effects of identity information on racial emotional contagion. We confirmed that, consistent with Hess and Fischer (2014) and Wróbel and Imbir (2019), emotional contagion is modulated by social context, and the time course of this influence can be characterized by at least three stages: bottom-up attention to racial cues (around 100 ms): other-race faces first attract more attention from the bottom-up for categorization processes; emotion and individuation processing (around 200 ms): process happy emotion and the same-race faces for individuation processes, that is face’s second order spatial configuration processing; and top-down identity-based modulation (around 300 ms): the identity-diagnostic facial characteristics of other-race may draw attention from the top-down. This staged progression highlights how perceptual and social information jointly shape emotional contagion in dynamic and temporally specific ways. Our study explains the different stages of processing of emotional contagion, and also, for the first time, our study explains emotional contagion with the Categorization-Individuation Model.

Funding

The National Social Science Fund of China (19ZDA021).

Data availability

The data and code will be available per request.