Increased sensitivity to social hierarchy during social competition versus cooperation

Abstract

Social hierarchy is a typical feature of social organization. The ability to quickly recognize social hierarchy information is crucial for adapting to social contexts. Here, we adopted fast periodic visual stimulation with electroencephalography to assess the neural responses to social hierarchy during social competition and cooperation, respectively. Participants first learned hierarchical faces from a competitive game versus a cooperative game. We then sequentially presented the learned hierarchical faces with a specific frequency in a set of faces. Results showed that participants rated the inferior player as lower in the social hierarchy in the cooperative context compared to the competitive context, indicating that social context affects the judgment of others’ rank. Moreover, higher neural responses to high and low-hierarchy faces versus medium-hierarchy faces were observed, suggesting rapid discrimination of social hierarchy from faces. Interestingly, rank-specific neural responses were more pronounced in the competitive context than in the cooperative context, indicating increased sensitivity to social hierarchy during social competition versus social cooperation. This study provides behavioral and neural evidence for rapid, automatic processing of social hierarchy information and for an increased sensitivity to such information in competitive versus cooperative social contexts.

Introduction

Social hierarchy is an essential component of social organization (Yang et al. 2014, Zhou et al. 2018, Su et al. 2021), which is ubiquitous across species, including ants (Cole 1981), mice (Desjardins et al. 1973), fish (Grosenick et al. 2007), and primates (Pineda et al. 1994). Even in interactions between two people, an informal hierarchy can exist in which one person is dominant, and the other is submissive (Markey et al. 2003). Although this hierarchical relationship involves inequality, it can reduce the frequency and intensity of individual aggression within the society, helping maintain the group’s relative stability and keep the population healthy and reproducing (Sapolsky 2005). Surviving in such a hierarchical environment for a long time enables individuals to quickly recognize each other’s hierarchical relationships and guide their behavioral decisions accordingly (Qu et al. 2017).

There are two general ways in which individuals can obtain information about the social rank of others: through explicit cues and through implicit knowledge. Physical external characteristics are the most apparent cues of social hierarchy. Previous studies have suggested that people can assess the social rank of others by their body posture (Marsh et al. 2009), facial features (Chiao et al. 2008), and body size (Thomsen et al. 2011). On the other hand, individuals can also learn others’ social rank through implicit social information, including observational learning (Kumaran et al. 2016) and/or direct social interaction (Ligneul et al. 2016, Janet et al. 2022). Competition and cooperation are the most common ways of interaction. Through these types of interactions, individuals can gather information about their own abilities and the competencies of others. Previous studies suggested that in a competitive situation, where social rank is achieved through a direct contest, viewing a superior player elicited a significantly higher N170 electroencephalogram (EEG) component (Santamaria-Garcia et al. 2015). Furthermore, neural activity in the prefrontal cortex, encoding, a.o. the social status of opponents (Ligneul et al. 2016), was significantly increased when facing superior players (Ligneul et al. 2017). These findings suggest that social status during competition in humans is linked to specific electrophysiological and neural responses.

Studies investigating neural responses to the perception of social hierarchy are partly inconsistent. Previous event-related potential (ERP) studies indicated higher activity in early ERP components in response to high-rank cues. For example, when required to make gender judgements, higher N200 amplitudes over occipitotemporal regions were observed when participants viewed faces with dominant facial expressions compared to submissive facial expressions (Chiao et al. 2008). Moreover, higher N170 activity was observed when participants faced a superior player in a visual discrimination task, revealing stronger N170 responses to implicit social hierarchy perception (Santamaria-Garcia et al. 2015). Other studies, however, found an influence of social rank on the late positive potential (LPP) rather than the early N170 component (Breton et al. 2014, 2019, Miao et al. 2022). These discrepancies may be attributable to variations in experimental designs and task requirements, given that task demand can affect the N170 amplitudes of face processing (Goffaux et al. 2003, Senholzi and Ito 2013). Thus, whether individuals can process social hierarchy information automatically remains a question worth exploring.

Social context influences how people perceive others. For example, individuals perceive others as more similar to themselves when in a cooperative context than in a competitive context (Toma et al. 2010). In addition, individuals’ performance can be affected by whether they cooperate or compete with others (Wittmann et al. 2016), suggesting that social contexts such as competition versus cooperation have a different impact on the perception of others’ abilities.

However, little is known about the neural processing of social hierarchy information acquired in cooperative and competitive contexts, nor the potential difference between them. In order to shed light on this question, an objective and implicit measure that is unaffected by decision-making is required. To that end, we used the EEG combined with the Fast Periodic Visual Stimulation (FPVS) method to identify implicit neural processing of social hierarchy in competitive versus collaborative contexts.

The FPVS approach is based on the principle that a periodic, repetitive stimulus produces a periodic change of electrical activity in the human brain (Liu-Shuang et al. 2014, Rossion 2014). This paradigm has been primarily used to study low-level visual processes (Regan 1966, 1974, Tyler and Kaitz 1977, Norcia et al. 2002). More recently, it has been used to reveal the neural signature of face perception (Liu-Shuang et al. 2014, Vettori et al. 2018, 2020), and of high-level implicit processing of facial expressions (Dzhelyova et al. 2017, Van der Donck et al. 2020, Matt et al. 2021), facial attractiveness (Luo et al. 2019), trustworthiness (Sutherland et al. 2020, Verosky et al. 2020a), and social dominance discrimination (Su et al. 2021). By using the combined FPVS-EEG method, we aim to reveal implicit neural responses to social hierarchy perception in the context of competition versus cooperation.

In the current study, participants learned the social rank of others through either a competitive or a cooperative game, while EEG was recorded. Subsequently, they viewed oddball sequences of face pictures. In such sequences, a new face picture served as the base stimulus and was presented repeatedly at 6 Hz (6 images appeared in 1 s). Using this approach, a clear frequency-tagging marker of 6 Hz can be captured in the EEG spectrum. Importantly, a face picture with the rank information participants had previously learned was inserted at every fifth picture, which served as the oddball stimulus and thus generated a periodic repeated change at 1.2 Hz (6 Hz/5 = 1.2 Hz). If participants encode the base stimuli and oddball stimuli in distinct patterns, a clear oddball response in the EEG spectrum will occur at exactly 1.2 Hz, reflecting the automatic recognition of the rank-related face. The periodic emergence of hierarchical faces, thus allowed us to track participants’ sensitivity to social hierarchy. We hypothesized that (i) participants would detect the hierarchical information automatically and that (ii) sensitivity to social hierarchy would differ between cooperative and competitive contexts.

Materials and Methods

Participants

Sixty-seven participants were recruited and randomly assigned to the competitive context or cooperative context. Ten participants had to be excluded because they failed to learn the hierarchical information, and one was excluded for missing data, leaving a final sample of 56 participants. Twenty-nine (16 males, mean age = 20.29, s.d. = 1.76) of them were in the competitive context, and the other 27 participants (13 males, mean age = 20.24, s.d. = 1.82) were in the cooperative context. This study was approved by the ethics committee of Shenzhen University.

Materials

Four neutral male faces and four neutral female faces were selected from CAFPS (Gong et al. 2011). Participants were only presented with faces that matched their own gender to avoid the influence of gender factors. Besides, to minimize the influence of picture differences, each specific face was presented as either a superior, medium, or inferior face (in terms of hierarchy) or served as the base stimulus in the FPVS task in pseudorandom order. The face used for the base stimuli in the FPVS task did not appear in the learning task.

Procedure

Learning task

First, participants were asked to play an arrow game. In this game, a grey circle with 14 arrows was displayed on the screen for 2 s. Five arrows were directed towards one side, while nine pointed towards the other side. Participants were asked to judge the orientation of the majority of the arrows as quickly and accurately as possible by pressing either ‘F’ or ‘J’, with ‘F’ being left and ‘J’ being right, followed by feedback indicating “Correct” or “Wrong”. They were informed that they would play alone and that the response and reaction time of each trial would be recorded and uploaded to our database. They were notified that 300 players had been recruited to do this task, who were classified into five performance levels (from 1 to 5, with 1 being the lowest and 5 being the highest) based on their performance compared to the other players. After finishing 40 trials, participants would receive their performance ranking using the same criteria rather than their overall accuracy. Unbeknownst to the participants, they were always informed that their rank was 3 if their accuracy was >60% (24 out of 40 trials). Otherwise, they were requested to repeat the game.

Next, participants were instructed that they were going to play either a competitive or a cooperative game with three other players, who had completed the same arrow tasks earlier and obtained their rank information. The participant’s task was to learn their rank information through the feedback of the arrow game. The procedure and the setup were the same in both contexts, while the instructions differed.

In the competitive context, participants were instructed that they were going to compete one-on-one with these players. The computer would randomly select one player to be presented on the screen as their opponent and compare their performance on a given trial with the matching trial of this opponent. The task that the participants needed to do was to guess the rank of this opponent (1—lowest; 5—highest). The results of the comparison from the arrow game, which were pre-established, would then appear after their decision, informing the participants whether they had won or lost. Note, that this feedback referred to the performance in the arrow task, not to whether participants guessed the rank information correctly or incorrectly. The positive feedback meant participants performed better than the given opponent in the arrow task, while the negative feedback meant a poorer performance than this opponent. Considering that the rank of the participants had already been determined, more positive feedback implied a lower ranking of this opponent, while more negative feedback implied a higher ranking of this opponent.

In the cooperative context, participants were informed that they were going to team up with each of these players in a two-person team against another team. Their teammate would be assigned randomly by the computer. Next, their team performance will be compared with another team. During this phase, participants needed to guess the rank of this teammate from one to five. Next, the results of the comparison would be displayed on the screen, indicating which team performed better. Positive feedback indicated that the participant’s team performed better than the other team; negative feedback indicated that they performed worse. Because the members of the other team and the participants themselves stayed the same during the whole game, the only factor that affected the result of the team comparison was their teammate. Thus, positive feedback also indicated a higher ranking of this teammate, while negative feedback implied a lower ranking of this teammate.

The learning task consisted of 120 trials, with each of the three players appearing for 40 trials at random. One of the players would have a 75% probability resulting in positive feedback, indicating he/she was an inferior opponent in the competitive context while a superior teammate in the cooperative context. Another player had a 50% chance of positive feedback (medium-hierarchy condition) in both contexts. A further player had a 25% chance of positive feedback, suggesting a superior opponent in the competitive context while an inferior teammate in the cooperative context (Fig. 1a). The feedback of each player was predetermined and probabilistic, independent of participants’ performance in the arrow task. Participants were expected to learn others’ ranking information through the feedback.

After 120 trials, participants needed to answer two questions: who belonged to the highest rank and who to the lowest. These two questions served as the criterion for whether participants had correctly accessed social hierarchy information. Participants were considered to have learned the hierarchical information if they answered both questions correctly. Otherwise, their data would be excluded from further analysis.

To establish the ranking context of the social hierarchy, they would perform a money allocation task with these players. In this task, the one with a higher rank had primary access to get more money as their extra payoff. It was introduced to participants before the learning task to make sure they understood the concept of a dominant hierarchy relationship and to make it more tangible (see Supplementary Materials: Allocation task).

FPVS task

After finishing the learning task, participants were asked to complete the FPVS task. Stimuli were presented using psychtoolbox (Kleiner et al. 2007) running in the Matlab environment (R2016a, Math Works). Again, there were three conditions: superior, medium, and inferior. Each condition was repeated four times to produce a total of 12 sequences, all of which appeared randomly. In all sequences, each picture appeared at the frequency of 6 Hz, i.e. six pictures appeared within one second. For every five pictures, the face changed from the base stimulus to an oddball stimulus (i.e. BBBBOBBBBOBBBBO…). Therefore, the frequency of interest was 1.2 Hz (6 Hz/5). In this task, a new face served as the base stimulus, and a hierarchical face served as a possible oddball stimulus in the corresponding condition. As a result, EEG amplitude at 1.2 Hz and its harmonics (i.e. 2.4 Hz, 3.6 Hz, etc.) were considered as the oddball response distinguishing it from a normal face. Different responses between conditions were considered as rank-specific responses.

Each sequence lasted for 74 s, which began with a fixation cross presented for 2–5 s. Next, a face stimulus gradually emerged and reached full contrast after 2 s (fade-in). The formal testing sequence lasted for 70 seconds, at which time the image gradually disappeared after 2 s (fade-out) (Fig. 1b). The fade-in and fade-out phase was implemented to prevent ocular artifacts (Liu-Shuang et al. 2014) and was excluded from analyses. In addition, each image within the sequence was also presented by gradually increasing and decreasing the contrast, following a sinusoidal modulation. According to previous studies, sinusoidal modulation is a smoother stimulation pattern when compared to square wave stimulation, thereby making it more comfortable for participants to view (Liu-Shuang et al. 2014). Moreover, in order to reduce low-level adaptation, the size of the image varied randomly from 90% to 110% of the original stimulus each time it appeared (Dzhelyova et al. 2017, Luo et al. 2019).

Participants were asked to pay attention to the center of the screen and press the spacebar when the central fixation cross changed color. For ten times during each sequence, the fixation cross randomly changed color briefly for 333 ms. This sub-task was meant to ensure the continuous engagement of participants throughout the experiment. Participants were offered a short break between each sequence, the duration of which was determined by themselves.

EEG acquisition

EEG data were recorded using 64 Ag–AgCl active electrodes sampling at 1000 Hz with the standard 10–20 layout (Brain Products, Gilching, Germany). The data were online referenced to the FCz electrode. The impedance of all electrodes was kept below 10 kΩ. Eye movements were monitored by an electrode that was placed below the right eye and two electrodes on the outer canthi of the eyes.

EEG analysis

Preprocessing

EEG data were preprocessed using EEGlab toolbox in the Matlab environment (R2016a, Math Works). Firstly, individual EEG data were resampled to 500 Hz and then segmented for each condition, including 1 s before the beginning of the fade-in and 1 s after the end of the fade-out, resulting in 76 s segments. A fourth-order Butterworth band-pass filter between 0.1 and 100 Hz and a notch-pass between 48 and 52 Hz were applied to each segment. Noisy channels were identified by visual inspection and interpolated using the spherical algorithm implemented in EEGLAB. In sum, 31 participants were interpolated, with the total number of interpolated channels was 79. No participant had more than six channels interpolated. Next, we performed an independent component analysis (ICA) (James and Gibson 2003) to obtain the first 40 components, and components corresponding to eye movement were identified by visual check and removed. Finally, all channels were re-referenced to a common average excluding the ocular channels and bilateral mastoid channels, which were also excluded from further analysis. All segmentations were included in further analysis.

Frequency-domain analysis

The preprocessed data were submitted to Letswave 6 in the Matlab environment for the frequency-domain analysis. A second segmentation was performed from the onset to 70 s and then averaged for all segmentations in each condition. Next, we performed a Fast Fourier Transform (FFT) on the individual averaged data using the default setting implemented in Letswave 6. The data were transformed from the time domain to the frequency domain, resulting in amplitude outputs. Subsequently, signal-to-noise ratios (SNR), z-score, and baseline-corrected amplitude were calculated separately from these data, as detailed in the next sections.

For better visualization, we computed the SNR as the ratio of the amplitude at each frequency to the average of the baseline, which was set by the 20 surrounding bins of the frequency of interest (10 on each side; the immediately adjacent bin was not included, and one maximum and one minimum value was excluded to take out the extremes).

To determine the significance of the response at the frequencies of interest, we calculated the z-score by subtracting the average of the baseline from the amplitude of each frequency of interest and then dividing it by the standard deviation of the baseline. We considered the response at a given frequency point to be significant if Z > 1.65, which is equal to P < .05 (one-tailed test). We averaged the Z-score for all of 60 channels and all the participants. Then we obtained a Z-score value for each harmonic in each condition. We included harmonics in the statistical analysis until they were no longer significant in any condition. Based on this criterion, up to the 6th harmonic for the oddball frequency (Supplementary Table S1) and up to 3rd harmonic for the base frequency were included into statistical analyses.

Finally, to compare the difference between conditions and statistical analysis, we computed the baseline-corrected amplitude by subtracting the baseline average from the amplitude at each frequency point and then summed the baseline-corrected value of all significant harmonics. We summed the baseline-corrected amplitude of 1.2 Hz and its four harmonics (2.4, 3.6, 4.8, and 7.2 Hz) for comparing condition difference of oddball response. The value of 6 Hz was excluded because it corresponds to the base frequency. Similarly, we summed the value of 6 Hz and its two harmonics (12 and 18 Hz) for each condition for base response.

We selected three regions of interest (ROI) based on the topographical maps and previous studies (Liu-Shuang et al. 2014, Van der Donck et al. 2020). These ROIs were comprised of the following channels: the left occipito-temporal (LOT: P7, PO7), the medial-occipital (MO: POz, Oz), and the right occipito-temporal (ROT: P8, PO8) region. The LOT and ROT were selected to reflect the oddball response, while the MO was selected to indicate the synchronization of the visual system to the stimulation at 6 Hz and its harmonics.

All the statistical analysis were conducted in R statistical package (version 4.3), using the ‘bruceR’ (Bao 2022). P-value was adjusted for multiple testing using the Bonferroni method.

Results

Learning task

In the arrow task, the timing and distribution of arrows allowed most people to make a judgment. Four participants were asked to repeat it because their accuracy was <60%. All of them reached the goal and obtained their ranking of medium level. This task was meant to enable participants to get their own rank information and make sure all of them were aware that their rank was medium. Thus, the real performance would not be discussed further. Neither their response nor the feedback related to the learning task.

In the learning task, 56 participants correctly identified both the superior and the inferior players. Additionally, four participants correctly recognized the superior player but failed to identify the inferior player, while two participants correctly recognized the inferior player but failed to identify the superior player. Moreover, two participants incorrectly identified both superior and inferior players. Thus, those participants were excluded from further analysis.

We divided 40 trials into eight blocks of five trials each and calculated the mean rating for each block. The group-level curves showed that participants exhibited distinct trends after the first block (see Fig. 2a). Next, we performed separate mixed ANOVAs for reaction time (RT) and mean rating in the learning task, with social context (competition, cooperation) as a between-subject factor and rank (superior, medium, and inferior) as a within-subject factor. No significant difference was found in the results of RT, P’s >.05. The result of the mean rating showed a significant main effect of rank [F (1.621, 87.552) = 282.360, P < .001, η_p² = 0.839]. Pairwise comparisons suggested that the mean rating was highest for the superior player (M = 4.180, SE = 0.062), followed by the medium player (M = 3.169, SE = 0.053), and the inferior player (M = 1.979, SE = 0.070), with all P-values <.001. The interaction effect of rank and social context was also significant [F (1.162, 87.552) = 4.021, P = .029, η_p² = 0.069]. Post-hoc tests revealed the simple main effect was significant for the inferior player (P = .006). Pairwise comparisons indicated that the rating of the inferior player was lower in the cooperative context (M = 1.777, SE = 0.101) than in the competitive context (M = 2.180, SE = 0.098; Fig. 2b). No other main effect or interactions reached significance (P’s > .05).

FPVS task

Behavioral results

All participants accomplished the fixation color detection task with an average response time (RT) of 0.34 s and an average accuracy rate (ACC) of 96.2%. The mixed ANOVA with social context (competition or cooperation) as the between-subject factor and rank (superior, medium, and inferior) as the within-subject factor showed no significant main effects or interaction effects in ACC or RT (P’s > .05).

Oddball response

All pictures were presented at the speed of 6 Hz, and a rank-based face was inserted in every fifth picture. Thus, the response corresponding to 1.2 Hz indicated the discrimination of the rank-based face from the normal face. Results showed a clear 1.2 Hz response and its harmonics in the frequency domain for all conditions (Fig. 3a). The distinct peak of oddball response confirmed that the social hierarchical information was implicitly perceived.

We perform a mixed ANOVA to the oddball response, with social context (competition, cooperation) serving as a between-subjects factor, rank (superior, medium, and inferior), and ROI (LOT, MO, and ROT) as within-subjects factors.

Results showed a significant main effect of ROI [F (1.743, 94.107) = 12.570, P < .001, η_p² = 0.189]. Pairwise comparisons revealed that the response of ROT (M = 0.465, SE = 0.041) was significantly stronger than that of LOT (M = 0.325, SE = 0.030) and MO (M = 0.319, SE = 0.025), P’s <.001. The main effect of rank also reached significance [F (1.996, 107.767) = 21.951, P < .001, η_p² = 0.289], which reflected that the response corresponding to the superior (M = 0.405, SE = 0.032) and inferior (M = 0.409, SE = 0.030) face was significantly stronger than to the medium face (M = 0.295, SE = 0.024), P’s <0.001. Importantly, there was a significant interaction effect between social context and rank [F (1.996, 107.767) = 3.534, P = .033, η_p² = 0.061]. The simple main effects were all significant for the competitive (P < .001) and cooperative contexts (P = .024). In the competitive context, the amplitude of the superior (M = 0.433, SE = 0.044) and inferior (M = 0.455, SE = 0.042) conditions was significantly larger than the medium condition (M = 0.290, SE = 0.033), P’s <.001. In the cooperative context, the amplitudes of the superior condition (M = 0.377, SE = 0.046) were significantly stronger than that of the medium condition (M = 0.300, SE = 0.034), P = .031, while there was no significant difference found between the response of the inferior (M = 0.363, SE = 0.043) and medium conditions, P = .081 (Fig. 3b). No other significant effect was found.

Rank-specific response

To better compare participants’ sensitivity to social hierarchy, the difference in response between conditions was measured. Specifically, the rank-specific response to the superior condition was calculated by subtracting the data of the medium condition from the superior condition. Similarly, the response of the inferior condition was measured by the difference between the medium condition and itself. We defined the medium condition as the comparison baseline because all participants were assigned medium rank. Next, we conducted a mixed ANOVA on the rank-specific response, with social context (competition, cooperation) serving as a between-subjects factor, rank (superior and inferior), and ROI (LOT, MO, and ROT) as within-subjects factors.

Results showed a significant main effect of context [F (1, 54) = 6.110, P = .017, η_p² = 0.102], where the rank-specific response in the competitive context (M = 0.154, SE = 0.024) was significantly stronger than in the cooperative context (M = 0.070, SE = 0.025) (Fig. 3c). The main effect of ROI was also significant [F (1.965, 106.095) = 4.177, P = .018, η_p² = 0.072], where the amplitude in the ROT (M = 0.159, SE = 0.027) was significantly larger than the region of MO (M = 0.091, SE = 0.019), P = .043. No other significant difference was found (P’s > .05).

To explore whether gender had an impact on the sensitivity to social hierarchy, we performed a mixed ANOVA with an additional factor of biological sex (male, female) serving as a between-subject factor on the oddball response and the rank-specific response, respectively (see Supplementary Materials: Gender difference).

Base frequency responses

The base responses simply reflected the contrast between the background and the stimuli. Therefore, it was not expected to yield any differences among conditions. The sum of the baseline-corrected amplitudes at 6 Hz and its two harmonics (12  and 18 Hz) was submitted to the mixed ANOVA with social context (competition, cooperation) as the between-subject factor, rank (superior, medium, and inferior) and ROI (LOT, MO, and ROT) as within-subject factors. The results showed a significant main effect of ROI [F (1.852, 100.011) = 17.718, P < .001, η_p² = 0.247]. Pairwise comparisons revealed that the response in the region of the MO (M = 0.492, SE = 0.044) was significantly larger than the region of ROT (M = 0.384, SE = 0.034, P = .027) and the region of LOT (M = 0.281, SE = 0.027, P < .001). Besides, the response in the ROT was significantly greater than the LOT, P = .003. No other significant main effect or interaction was found (P’s > .05).

Discussion

In the present study, we investigated how individuals process social hierarchy in competitive versus cooperative contexts using EEG and FPVS techniques. First, participants played a game to learn the social rank of others. We found that participants rated an inferior player lower in the cooperative context than in the competitive context. This indicates that social context affects the judgment of others’ rank. Next, we presented face pictures with and without rank information in oddball sequences at 6 Hz. Our results showed that (i) significant neural markers were observed at the frequencies corresponding to the rank-specific face, and that (ii) superior and inferior players elicited a greater brain response than medium players, suggesting that participants could rapidly and implicitly detect social hierarchy information. The rank-specific response occurred over the right occipitotemporal region and was more pronounced in the competitive context compared to the cooperative context, suggesting that participants were more sensitive to social hierarchy information in the competitive situation.

These findings show that implicit social hierarchy information can be processed automatically in the human brain, as demonstrated by the consistent oddball response in the EEG spectrum. Previous EEG and functional magnetic resonance imaging (fMRI) studies presented social hierarchy information explicitly, requiring participants to make judgments regarding the hierarchical face. However, this approach may have allowed interference from other processes, such as emotional and motivational processing (Zink et al. 2008, Breton et al. 2014, 2019, Feng et al. 2015, Santamaria-Garcia et al. 2015). In contrast, by using FPVS, in which an image only appeared for 167 ms and was “overwritten” by the subsequent pictures, we were able to constrain the potential interference of face processing (Liu-Shuang et al. 2014, Dzhelyova et al. 2017, Luo et al. 2019). This task design allowed us to isolate the neural signature induced by social hierarchy without requiring participants to make any face-related judgments. The clear and stable oddball response confirmed that individuals processed social hierarchy automatically and implicitly.

Although the oddball response is likely to be a mixture of familiarity and rank information, we argue that hierarchical rank exerts a significant influence. The pictures used as the oddball stimuli were viewed before the FPVS task, resulting in participants being more familiar with these pictures than the base stimuli. Previous studies using the FPVS-Oddball method demonstrated that the more familiar picture would elicit a greater oddball response (Yan and Rossion 2020, Verosky et al. 2020b, Yan et al. 2023), which suggested that familiarity could be one cause of this oddball effect. However, in the current study, all pictures appeared the same number of times, therefore the oddball effect caused by familiarity should be nondifferent. Yet, we observed that participants exhibited stronger neural responses to images ranked as superior and inferior compared to the medium-ranked images. This finding revealed the presence of a rank effect beyond mere familiarity, suggesting that hierarchical rank plays a crucial role in the observed oddball effect.

One possible explanation for the heightened neural response to the superior and inferior face is that individuals are sensitive to hierarchy differences. In the current study, all participants were assigned a medium ranking, establishing the medium player to be similar to themselves and could be considered a comparative baseline, while others diverged from themselves on social rank. Being alert to those in higher positions can help people avoid unnecessary flight in a competitive context and seek protection in a cooperative context, while awareness of those in lower positions offers opportunities for accessing resources (Cheng et al. 2010, Piff et al. 2012, Maner and Case 2016). Consequently, this enhanced neural response reflects the cognitive mechanism of selective attention to social hierarchy.

Moreover, to better isolate the rank effect, we calculate the rank-specific response by subtracting the response of the medium condition. This approach allows us to eliminate the influence of familiarity and focus on the impact of rank. We found that the resulting difference, i.e. the rank-specific response, was more pronounced in the competitive context than in the cooperative context, indicating higher sensitivity to rank-specific information in a competitive situation. One explanation is that in the competitive context, individuals need to face conflict directly, while there is no confrontation in the cooperative context. Competition is concerned with one person trying to outperform another person in a zero-sum situation (Liu et al. 2021). In any social environment, those who are high-ranking are more likely to have better resources and greater competence (Qu et al. 2017). Getting into a conflict with people of higher rank implicates risk and elicits avoidance tendencies, whereas challenging people of a lower rank means that the individual is in an advantageous, potentially profitable position. As a result, people are more sensitive to hierarchical information when in a competitive context (Qu et al. 2017).

We further observed that participants rated the low-ranking player lower in the cooperative than in the competitive context. This can perhaps be explained by the self-serve bias, which means that people are inclined to take credit for positive outcomes and make external attributions for negative ones (Miller and Ross 1975, Duval and Silvia 2002, Li et al. 2018, Zhou et al. 2022). In the competitive context, their level of performance is the primary determinant of success or failure. In comparison, cooperation refers to a group of individuals working together to achieve a common goal (Liu et al. 2021), in which both the participant and partner are responsible for winning or losing outcomes. Previous studies found that the self-serve bias often occurs in interpersonal environments without explicit feedback on individual achievement (Deffains et al. 2016, Wang et al. 2017, Zhou et al. 2022). For example, participants rated their partner less favorably when facing failure during a collaborative video game (McGloin et al. 2016). It thus seems plausible that the lower rating for the inferior players in the cooperative context in the current experiment was caused by participants attributing their failure to the low ability of their teammates.

The rank-specific response was observed over right occipitotemporal regions, which are usually implicated in high-level visual processing in FPVS designs, with right hemisphere dominance in face coding (Van der Donck et al. 2020, Verosky et al. 2020a). It has also been demonstrated that the occipitotemporal region is associated with the discrimination of social dominance hierarchy (Su et al. 2021), and the processing of social dominance cues correlated with responses within the occipitotemporal region (Chiao et al. 2008). When facing a superior player, significantly greater brain activity in the occipitotemporal area was observed (Zink et al. 2008). The current results suggested that facial identity processing in the FPVS approach occurs primarily in the right occipitotemporal regions, with ranking cues being encoded with greater saliency. Caution is needed, though, when localizing EEG results, given the “inverse problem” that arises from the uncertainty of underlying neural sources.

Furthermore, our study revealed a gender difference in the sensitivity of social rank in the rank-specific response, indicating that males exhibit a greater sensitivity to hierarchical information than females. This heightened sensitivity in males may stem from biological and cultural norms, both of which have shaped the behavior of males over time and drive them to compete for status and resources within social hierarchy (Redhead and Power 2022). However, given the limited sample size for each gender, further investigation is needed to fully understand this observed gender difference.

While our study offered informative results, there are limitations to consider. Firstly, we excluded 10 participants from the formal analysis. They were removed because they failed to learn the rank information, implying that the learning task could be difficult to understand. Alternatively, it could be a consequence of the way social hierarchy was conceptualized in our task. This may be somewhat artificial and not identical to social hierarchies one encounters in daily life. Secondly, there were only two players in the competitive context (participants and their respective opponent), while in the cooperative context, four people were involved (participants and their teammate in one team, and another two players in another team). This could potentially lead to some imbalance between social contexts. Further research should carefully consider and manipulate the social context settings. Third, it is essential to consider that this study was conducted in China and all participants were Chinese. Previous research demonstrated that socioeconomic status (SES) differently affects self- and other-oriented attributes across different cultures (Miyamoto et al. 2018). It is crucial to be cautious when extending our findings to other cultural groups. Further research could explore cultural differences in social hierarchies.

In summary, we provided neurocogntive evidence for the automatic processing of implicit social hierarchy information. The neural responses corresponding to rank-specific faces revealed that the emergence of social hierarchy could be detected at a glance. More importantly, higher neural responses were observed to the superior- and the inferior-ranked faces than to the medium-ranked faces, suggesting differential neural coding of social rank. Moreover, the observed higher sensitivity to social hierarchy in the competitive versus the cooperative contexts highlights the modulating influence of social context on the processing of social hierarchy information. Importantly, our results demonstrate the applicability of the FPVS paradigm to isolate and capture the implicit processing of social hierarchy from passively viewed faces. Overall, the current approach thus provides a promising new avenue for expanding our understanding of how people process social hierarchy.

Supplementary data

Supplementary data is available at SCAN online.

Conflict of interest

The authors declare no conflict of interest.

Funding

This study is funded by the National Natural Science Foundation of China (31920103009 and 32371104), the Major Project of the National Social Science Foundation (20&ZD153), Space Medical Experiment Project of China Manned Space Program (HYZHXMN01012), the Fundamental Research Funds for the Central Universities, and Shenzhen-Hong Kong Institute of Brain Science-Shenzhen Fundamental Research Institutions (2023SHIBS0003).

References