Repositório ISCTE-IUL

Individuals feel more empathy for those in their group (i.e. ingroup members) than those who are not (i.e. outgroup members). But empathy is not merely selective to group distinctions, rather it fluctuates according to how groups are perceived. The goal of this research was to determine whether group-based evaluations can drive biases in self-reported empathy as well as in the underlying neural activity. Participants were asked to rate a target’s physical pain while BOLD responses were recorded via functional magnetic resonance imaging (fMRI). The target was either a member of the ingroup or one of two outgroups, one which was more of a rival to the ingroup than the other. Participants reported more empathy for targets experiencing painful compared to innocuous events showing bias only in favour of their ingroup. Neural responses were stronger while observing painful, compared to innocuous, events but only for targets from the ingroup or the less competitive outgroup. The difference was non-significant and trended in the opposite direction when the target was from the more competitive outgroup. This provides evidence that empathy is not merely selective to “us” vs “them” but is more nuanced by whom we refer to by “them”.

Individuals feel more empathy for those in their group (i.e.ingroup members) compared to those who are not (i.e.outgroup members)-this is referred to as empathic bias.Using self-report ratings, researchers have shown that empathic bias can manifest at varying degrees from ambivalence (Stürmer, Snyder, & Omoto, 2005;Tarrant, Dazeley, & Cottom, 2009) to malicious pleasure (i.e.Schadenfreude; Cikara, Botvinick, & Fiske, 2011;Leach, Spears, Branscombe, & Doosje, 2003).Empathic bias can also manifest at the level of neural activity.Xu, Zuo, Wang, and Han (2009), found that participants exhibited stronger blood oxygenation level dependent (BOLD) responses associated with empathy while viewing same-race compared to other-race targets in physical pain.Recent accumulating evidence shows that empathy for others in pain is not just simply always more for ingroup compared to outgroup members, but is dependent on the outgroup and the intergroup context.Indeed, Avenanti, Sirigu, and Aglioti (2010) found that seeing a person being stabbed was associated with a mu suppression effect which was attenuated when the target ostensibly belonged to both a familiar racial outgroup (i.e. a black hand) and an unfamiliar racial outgroup (i.e. a violet-coloured hand).This suggests that biases in processing other-race pain is not only because of visual differences, but appears to be related to prior experiences or knowledge of a group-triggered by visually salient cues to categorization.
At the self-report level, the extent to which we feel ambivalence or joy to an outgroup member's pain is determined by whether or not we perceive the outgroup as a possible rival to the ingroup (Chang, Krosch, & Cikara, 2016;Cikara, Bruneau, Van Bavel, & Saxe, 2014).Researchers identified that BOLD responses are similarly modulated by whether the outgroup is considered to be a rival or not (Cikara, Botvinick, & Fiske, 2011;Hein, Silani, Preuschoff, Batson, & Singer, 2010) or has higher status than the ingroup (Feng et al., 2016).However, one of the limitations of this research is that empathy is measured in relation to an ingroup and single outgroup ("us" vs. "them"), not multiple outgroups to which the ingroup might relate differently ("us" vs. a rival "them" vs. a non-rival "them").Thus to determine whether or not empathic bias emerges according to how relevant an outgroup is to the ingroup, in terms of rivalry, it is necessary to compare empathic responses not only between the ingroup and an outgroup but also between different outgroups.Cikara and Fiske (2011), report one of few studies that has measured empathy in relation to multiple different social groups.While reading about another's experiences of misfortune, researchers found the anterior insula (a region strongly associated with empathy; Lamm, Decety, & Singer, 2011) was least active if the target was from a typically envied social group compared to if they were from a typically pitied, admired, or contemptible social group.On the other hand, they found if the event was positive, and the target typically envied, the insula was more active than it was in any other condition.Given the insula's involvement in processing pain (Peyron, Laurent, & García-Larrea, 2000) the authors interpreted this activation as indicative of a counterempathic response (i.e.knowing that someone we envy experienced something good is somehow painful).With these results, Cikara and Fiske showed for the first time that empathising with another's misfortune is not simply modulated by ingroup-outgroup distinctions, but varies as a function of how the outgroup is evaluated by the individual.
At this point, however, it is unclear to what extent these biases are specific to a concernrelated empathy (i.e. when responding to another's misfortunes) or whether biases of this nature also penetrate more automatic forms of empathic processing (such as when responding to another's physical pain).Previous research indicates that pre-existing experiences with some groups (other-races) modulate one's processing of their pain (Avenanti, Sirigu, & Aglioti, 2010), but no study has yet measured the extent to which The goal of the current study was to determine whether group-based evaluations can drive biases in empathy for pain at the self-report and at the level of brain activation.Brain activation is not only interesting in itself-as it can shed light on the mechanisms through which group biases emerge-but it also allows us to access responses that respondents cannot control, which is especially important when examining socially desirable responses such as empathy (Tarrant et al., 2009).In this study, we asked participants to observe targets from either the ingroup university or one of two neighbouring outgroup universities as they appeared to experience physically painful or innocuous events.Participants were asked to rate the extent to which they felt empathic or compassion-based responses towards the targets while BOLD responses were recorded via functional magnetic resonance imaging (fMRI).Based on a recent meta-analysis, we expected that observing pictures of another's pain would lead to increased activation in the insula-in anterior (associated with affective representation of another's state) but not posterior regions (which is more associated with the direct experience of pain; Lamm & Singer, 2010).We also expected to observe activation in regions that process the felt unpleasantness of physical pain such as the anterior cingulate (ACC) as well as areas engaged in the inference and representation of mental states such as the medial prefrontal cortex (mPFC) and temporoparietal junction (TPJ; Lamm et al., 2011).
We expected activity in these regions would be modulated by the group membership of the target, but more specifically by how participants evaluated the target's group in terms of a perceived rivalry with the ingroup.Indeed, others have previously highlighted the role that context plays in modulating AI activity during social interaction (Lamm & Singer, 2010).In further support of these predictions, a recent study by Feng et al. (2016) demonstrated that participants exhibited more activity in the AI and anterior medial cingulate cortex (aMCC) when a target in pain was a member of a low compared to high status outgroup.While Molenberghs et al. (2014), found that group membership had no effect on activation in the ACC or AI in their study, they did not measure how the outgroup was perceived by the participants.It is not possible, therefore, to rule out whether participants perceived the outgroup as a relevant rival to the ingroup-which previous research has demonstrated is crucial for the emergence of intergroup empathic bias (Cikara et al., 2014).
In this study, we included items to measure participants' perceptions of the target groups on a range of attributes relevant to intergroup relations and that have been shown to be key in driving intergroup empathic bias including stereotype content (Cikara & Fiske, 2011) and perceived rivalry (Cikara et al., 2014).We expected that activation in regions associated with empathy for pain (ACC and AI) and other processes related to empathy (i.e.mentalizing; mPFC) would be attenuated while observing outgroup targets compared to ingroup targets in pain.Crucially, however, we expected that this effect would be qualified by the extent to which the outgroup was perceived to be a rival to the salient ingroup, such that there would be less empathyrelated activity in response to targets from the relatively more competitive outgroup than to targets from the relatively less competitive outgroup.

Method
Design.This study followed a 2 (event: painful vs.  1FigureB).Participants were told to imagine that the events were happening in real time.Information about the target's group membership was presented in advance of the event (1000ms). 2The nature of the event was manipulated using common objects that can cause either pain (Figure 1C, upper row) or not (Figure 1C, lower row).Objects across the two categories were semantically matched as closely as possible, for example a Q-tip and syringe are both used to prepare a part of the body for surgery; to clean the area and provide anaesthetic, respectively.The event animations consisted of 2-frame sequences of photographs (Figure 1A).The first frame (500ms) showed a face and hand/arm in neutral position beside an object.In half of the sequences, the object was one of three items that might cause pain (Figure 1C, upper row) and in the other it was one of three semantically-matched innocuous objects (Figure 1C, lower row).The second frame (1500ms) showed the object stimulating the hand, resulting in either a painful or neutral facial expression from the target.No inter-stimulus interval was used between these frames in order to create apparent motion.This paradigm was inspired by one used in (Morrison, Tipper, Fenton-Adams, & Bach, 2013) however has a number of substantial changes (for example, orientation of the perceived hand and the addition of the target's face) and, for the first time here, is applied to a task involving empathy.
Stimuli were created using 20 photographs of faces (10 male, 10 female) adapted from the pain expressions image set in the Psychological Image Collection (PICS; pics.stir.ac.uk) and photographs of hands and arms (1 male and 1 female) that were created by the researchers of this study.The images of hands and arms were coded to appear in an allocentric perspective to depict a target sitting opposite, and facing, the participant.All photographs were created or edited using GNU Image Manipulation Program version 2.0 (The GIMP team).The photographs were matched for dimension and orientation as well as perceived age and race of the target.The duration of the fixation was randomly jittered (between 1000-1500ms) and inter-stimulus intervals (ISIs) were placed on either side of the animation, given a random duration between 250-1750ms (with the second ISI occupying the remaining time to amount 2000ms for combined ISIs per trial), thus equating a 'stochastic' design (Friston, Zarahn, Josephs, Henson, & Dale, 1999).A total of 96 trials, 50% depicting painful events (48), and 33% per university (32), each lasting 10 seconds, were presented in a single run.A new sequence of trials and timings was randomly generated for each participant.All stimuli were presented using E-Prime 2.0 software (Psychology Software Tools, Pittsburgh, PA) on an 800x600 LCD monitor.
Measures.Participants completed self-report measures of perceived group competence (all alphas > .90),sociability (all alphas > .90),status (all alphas > .80),rivalry (all alphas > .80),and similarity (all alphas > .90) in relation to all three target groups included in the study (for a full account of these measures please refer to Supplementary Materials).Participants responded to 2 items gauging self-focused empathy (i.e."how painful did the event seem to you?") and other-focused compassion (i.e."how bad did you feel for the person?")Using both items allowed us to discern between responses that are shared with the other (i.e.experience/affect-sharing) compared to responses that are felt for the other (i.e.compassion).Responses were made using a 100 point visual analogue scale (VAS) with a response button-box (4channel bimanual).This VAS was anchored in the same way as a 7-point Likert scale (1 = Not at all to 7 = Very much), but allowed participants to score anywhere between Procedure.After providing written informed consent participants were given a bogus news article to read which made salient a longstanding rivalry between the ingroup and one of the two outgroups (see Supplementary Materials).To ensure participants had paid attention to the article, they were asked to complete three simple questions (e.g."According to this article, which of the 2 universities is a long-standing rival to the University of Exeter?") 3 Observing safety protocols, participants were guided to the scanner and presented with standardised instructions via a digital projector system (Epson EMP-74).Participants completed 15 practice trials during the structural scan.At the onset of each trial of the main task, participants were required to memorise the target's group membership (i.e. a student from one of the target universities), to observe that target experiencing either an innocuous or painful event, recall the target's group membership in a simple 1-back task, and finally answer an item gauging their self-or other-focused responses to that target's experience.After this, participants left the scanner and were asked to provide their self-reported impressions of the groups.At completion, participants were debriefed on the background and purpose of the study and 3 Only participants who answered these questions correctly-and thus who processed the information-were included in the statistical analysis.However, participants were given the opportunity to revise the article at any point, resulting in accurate responses to those items, across the test sample.fMRI data analysis.Pre-processing and statistical analysis of functional images was performed using FMRIBs Expert Analysis Tool (FEAT; FMRIB Software Library, release 5.0, 2012).For each individual participant, standard pre-processing steps were performed.These were: Motion correction (Jenkinson, Bannister, Brady, & Smith, 2002), removal of non-brain tissue (Smith, 2002), spatial smoothing (using a Gaussian kernel of FWHM 5mm), normalisation based on grand-mean intensity, and high-pass temporal filtering (Gaussian-weighted least-squares straight line fitting, with sigma=50.0s).Registration of functional data to high-resolution T1 structural images and subsequently to Montreal Neurological Institute standardised space was carried out using FMRIB's Linear Image Registration Tool (MCFLIRT; Jenkinson & Smith, 2001).
Anatomical locations were determined in the first instance by the Harvard-Oxford anatomical atlas provided with FSL and, in the case of ambiguity, a more detailed paper and digital brain atlas (Mai, Voss, & Paxinos, 2008).
A c c e p t e d M a n u s c r i p t

EMPATHY AND INTERGROUP COMPETITION 12
First level, single-participant analyses were performed using a general linear model with local autocorrelation correction (Woolrich, Ripley, Brady, & Smith, 2001).
Each trial was modelled from the onset of the first frame of the animation for a duration of 2 seconds (i.e. until the end of the animation).For a separate, exploratory analysis we also modelled the 4 second response period for the self-and other-focused items, allowing us to investigate whether regional activity differed during empathic vs compassion-based responses.A design matrix was fitted for each participant with 6 task-related regressors, one for each condition of the 2 x 3 factorial design as well as a regressor for the motion parameters of each orthogonal axis (i.e. to control for head movement; Jenkinson et al., 2002).Main effects of event (ExeterPain + SussexPain + CardiffPain > ExeterNoPain + SussexNoPain + CardiffNoPain) were evaluated to demonstrate whether or not the task engaged empathy-related brain areas.We also tested the main effect of target group membership [(ExeterPain + ExeterNoPain > CardiffPain + CardiffNoPain) + (SussexPain + SussexNoPain > CardiffPain + CardiffNoPain)] and interaction effects between event and target group membership [(ExeterPain > ExeterNoPain + CardiffNoPain > CardiffPain) + (SussexPain > SussexNoPain + CardiffNoPain > CardiffPain)] see Supplementary Details, Table S1 for further clarity of the GLM design.For higher level analyses, participants were analysed according to the between groups condition.Higher level analyses were carried out using FLAME stage 1 (Beckmann, Jenkinson, & Smith, 2003;Woolrich, Behrens, Beckmann, Jenkinson, & Smith, 2004).Activations were thresholded at p < .05,familywise error rate (FWER) corrected with a cluster-extent based thresholding method.We chose a more conservative cluster-defining primary threshold of p < .001 in line with recent recommendations that a threshold of .05greatly inflates false-positive rates which are inherent to clusterwise inferences (Eklund, Nichols, & Knutsson, 2016).
A c c e p t e d M a n u s c r i p t

EMPATHY AND INTERGROUP COMPETITION 13
Region of interest analysis.Initially, we examined the effects of target group membership across the brain by running the main and interaction contrasts at the wholebrain level.We tested the a-priori hypotheses regarding activation in the AI, ACC, and mPFC, by conducting planned analyses using functionally defined regions of interests (ROIs).Functional ROIs were defined using the painful > innocuous events contrast, collapsed across target group membership.To look at differences in activation between the target identities, we extracted the mean percentage signal change in BOLD by expanding a 10 mm radial sphere from the peak voxel of each significant region in the pain > innocuous contrast image.To avoid issues of non-independence, we extracted data from the individual regressors from the first level analysis (and not from contrasts between target identities) using the Featquery tool in FSL.With these, we used SPSS (IBM SPSS Statistics for Windows, version 22.0.Armonk, NY: IBM Corp) to conduct within-group analyses comparing the mean percentage signal change in response to the different target identities.

Results
Self-report ratings.For a detailed report of the group perception measures please see supplementary materials.In short, participants rated one of the outgroups as more of a rival of the ingroup than the other outgroup, in spite of the salience manipulation.Specifically, students from Cardiff University were always rated as more competent and similar to the ingroup.They were also evaluated to be of a higher status and as greater rivals to the ingroup than students from the University of Sussex.This led us to expect that responses to the other's pain would be attenuated for Cardiff targets compared to Sussex targets.

Empathy.
A repeated measures analysis of variance revealed a significant main Manipulating the salient rival had no effect on ratings of empathy when measured in this self-focused way, F(1, 67) = .987,MSE = 603.38,p =.324, η 2 =.015.There were also no interaction effects between the target's group membership and the salience manipulation on this measure, F(1, 67) = .157,p = .855,η 2 =.002.
Compassion.There was a strong main effect of the perceived event on ).There were no significant effects of the salience manipulation on regional activity.
There were also significant differences in how participants responded to these items at the level of neural activation.There was significantly more activity in the AI, t ( 68  family-wise error rate (FWER) corrected with a cluster-extent based thresholding method with a cluster-defining primary threshold of p < .001.
Table 1.Results from the full factorial analysis of variance.

Discussion
The goal of this study was to demonstrate whether evaluations about outgroups can drive biases in pain empathy.To do this, we used a between groups manipulation to make salient the belief that a particular outgroup was a rival to the ingroup, more so than another outgroup.We then measured self-reports and BOLD responses while participants watched targets from these groups experiencing physical pain.
The results demonstrated that, in spite of the manipulation, participants always rated targets from one outgroup (students from Cardiff University) as more competent, of higher status, and are greater rivals to the ingroup (students from the University of Exeter) than the other outgroup (students from the University of Sussex).We identified a potential issue in our choice of target universities, in that one outgroup (Cardiff) was geographically more proximal to the ingroup than the other.Recent evidence suggests there is a significant relationship between perceived threat and physical proximity (Xiao & Van Bavel, 2012), and this might explain why, in our study, one outgroup was always evaluated as a greater competitive threat than the other.
We also identified that even after stating Sussex was a longstanding rival to the ingroup university, it had no bearing on how students from Sussex were evaluated in terms of perceived competence, status, and rivalry.There are a number of reasons why this might be.It may be that our participants were simply too familiar with Sussex and thus were resistant to any information that presented them differently.If participants were already aware of university league tables in the UK (which is often salient information when prospective students make decisions about which universities to apply to) they might have felt a rivalry but were unwilling to admit it, as this would imply a level of comparability between a lower status university and the one they chose.
Another possibility is that, consistent with our notions about Cardiff, Sussex was just not relevant to our participants in terms of their geographical proximity to the ingroup.Sussex is in the south of the UK and so in a sense has a shared background of similarity with Exeter and Cardiff, but it is not a part of the South West in the same way that the other two are.If geographical proximity is an important predictor of threat, this would explain why students from Exeter expressed no feelings of rivalry with the more distant outgroup university.These suggestions are drawn in the absence of direct empirical evidence, but with an eye toward guiding future research in the field of intergroup threat.
While there were no effects of the salience manipulation, we found effects of the target's group membership on empathy and compassion.At the self-report level, participants exhibited a bias in favour of the ingroup compared to either of the two outgroups.This emerged for both compassion (i.e. in how bad they felt for the other) and empathy (i.e. in how painful the stimuli appeared to be).At the level of neural activation, however, we found that in line with our hypotheses there were biases in BOLD responses consistent with the specific relations between the target groups.
Specifically, during trials depicting pain we observed activity in regions commonly associated with higher-order representations of another's states (AI).Consistent with existing literature, our empathy task recruited activation in the anterior but not posterior portions of insula-the latter of which is more greatly associated with processing the primary aspects of nociceptive signals (i.e.pain; Craig, 2009;Lamm et al., 2011;Lamm & Singer, 2010).As expected, we also observed activation in regions associated with processing the felt unpleasantness of physical pain (ACC; Rainville, Duncan, Price, & Bushnell, 2011;Singer et al., 2004) and those that support mentalizing processes (mPFC;Frith & Frith, 2006).
Crucially, we found that this difference in neural activation only emerged when the target was from the ingroup or from the less competitive outgroup, Sussex.When the target was from the more competitive outgroup, Cardiff, the difference was not only non-significant, it trended in the opposite direction (i.e.regions were marginally less active during painful events compared to innocuous events).Said another way, if the person being stabbed was from a rival outgroup, their experience of pain was not processed to be any more painful than if they appeared to be lightly touched.This is consistent with recent work that found participants exhibited responses associated with sensorimotor resonance following the observation of painful injections, compared to non-painful touches-but, crucially, only when the target was a member of the ethnic ingroup (Riečanský, Paul, Kölble, Stieger, & Lamm, 2015).
While viewing others in pain, participants exhibited less BOLD activity if the other was from Cardiff compared to if they were from the ingroup university or from the other outgroup, Sussex.There was no difference in how participants responded to targets from the ingroup and from Sussex.This reveals that our processing of another's pain is not simply attenuated to the outgroup when compared to the ingroup.It shows that, instead, empathy and compassion vary according to specific relations between the ingroup and outgroup.We identified that this context-specific bias was not apparent in the self-report responses but was clear at the level of neural activation-which is not subject to social desirability.We highlight, however, that our sample size for this study was powered to detect interaction effects at the self-reported level and was not necessarily powered to detect effects in the fMRI data-this should be considered a limitation of the current study and thus caution should be assumed in the interpretation of our fMRI findings.
These findings are consistent with previous research but also differ in a number of ways.Cikara and Fiske (2011), for example, found that participants exhibit biases in their concern for specific outgroups and that these biases emerged based on how those groups were evaluated by the individual.They also found that participants exhibited a "counter-empathic" response to typically envied targets.That is, activation of the AI went in the opposite direction than expected given the experimental condition.This reflects to a certain extent the findings that we report here-that AI activity was higher when the rival outgroup experienced an innocuous event compared to a painful event.
However, what we found was that empathy was not entirely switched off, nor did participants exhibit joy or reward-related neural activity (as other studies have found; Cikara, Bruneau, & Saxe, 2011).
In this study, participants reported feeling some empathy for a rival's pain and they still exhibited BOLD activity related to empathy, it was merely relatively lower to that reported/exhibited for the ingroup and less competitive outgroup.It is not entirely clear whether this relative difference was due to a dampening of empathy to a rival's pain (as Cikara and colleagues suggest, 2011) or an extraordinary amount of empathy for the ingroup (as suggested by Mathur, Harada, Lipke, & Chiao, 2010).What is clear, however, is that empathy does not simply switch on and off to an ingroup or outgroup's pain, it waxes and wanes according to our group memberships and more specifically to the relations between those groups.
An important consideration, here, is that rivalry in this context was presented somewhat more indirectly than in those previously tested (e.g.sporting competitions)related to status positions rather than realistic conflicts (e.g. for resources).It is also important to note that we did not take measures of the participant's explicit identification with the ingroup and this makes it impossible to determine how relevant or meaningful the categorization of 'student at the University of Exeter' was when participants responded to the task.We derive support, however, from classic Social Identity Theory (Turner, Hogg, Oakes, Reicher, & Wetherell, 1987) which submits that imposing a social categorisation in the absence of other social relations is sufficient to elicit intergroup bias.
In addition to the main analyses, we ran an item-wise analysis to delineate whether the type of self-reported item was supported by activation of distinct brain regions.We found dissociation between regional activations and self-reports: Whereas  Batson et al., 1987).This also supports previous work demonstrating that different facets of empathy and processes closely resembling it rely upon distinct neural architecture in the brain (Kanske et al., 2016; A c c e p t e d M a n u s c r i p t EMPATHY AND INTERGROUP COMPETITION 24 Lamm et al., 2011;Lamm, Rütgen, & Wagner, 2017;Ochsner et al., 2008;Zaki, Ochsner, Hanelin, Wager, & Mackey, 2007).Taken together, these analyses suggest that our task engaged the affect-sharing components of empathy as well as the more other-focused, compassion.These results also advocate for researchers to pay particular attention to what they ask participants at the self-report level-doing so has important implications for how participants respond.
In sum, we believe this study confirms some of what we already knew about empathy-that it is often reserved for "us" and attenuated to "them".This study also supports a recently converging notion that empathy is not always determined by "us" and "them" but rather by whom we refer to by "them".What we show is that groupbased evaluations can influence processing of what is an otherwise unambiguously negative event (i.e.physical pain).More broadly, this study provides further evidence that contextual cues (such as social identity) can trigger expectations (such as a particular group is a competitor) that, in turn, can exert a top-down influence on ostensibly automatic perceptual processing (such as pain perception; Decety & Lamm, 2006;Zaki & Ochsner, 2012).
Indeed, social identity processes can penetrate automatic processing at the earliest components of perception (Ratner & Amodio, 2013).And in the context of pain, this can have potentially disastrous consequences for certain groups in society.For example, researchers have identified that individuals from racial minority groups typically receive less pain relief treatment than white individuals in emergency room settings (Pletcher, Kertesz, Kohn, & Gonzales, 2008).This appears to be tied to the inaccurate beliefs that medical practitioners have over the biological differences between whites and racial minorities (Hoffman, Trawalter, Axt, & Oliver, 2016).Table S1.Design matrices for the fMRI GLM higher analysis.The contrasts denote the main effects for Target Group (A) and Perceived Event (B) as well the interaction effects (A*B).
based evaluations, as a result of those prior experiences, can drive biases in empathy for pain while controlling for visually salient cues to categorization.
Figure 1.(A) Trial structure; (B) Example of the four possible events from the combination of the object types (painful and innocuous) and the target group membership (ingroup and outgroup); (C) Object stimuli depicted as causing either painful or innocuous stimulations to the target.

Figure 2 .
Figure 2. Self-reported group-based evaluations following primes to intergroup rivalry with each of the outgroups (top left panel), activation map for the event*group interaction (top right panel), self-reported empathy in response to painful vs. innocuous Figure 1.(A) Trial structure; (B) Example of the four possible events from the combination of the object types (painful and innocuous) and the target group membership (ingroup and outgroup); (C) Object stimuli depicted as causing either painful or innocuous stimulations to the target.

Figure 2 .
Figure 2. Self-reported group-based evaluations following primes to intergroup rivalry with each of the outgroups (top left panel), activation map for the event*group interaction (top right panel), self-reported empathy in response to painful vs. innocuous events (middle left panel), mean percentage signal change in the Medial Prefrontal Cortex (middle right panel), the Anterior Insula (bottom left panel), and the Anterior Midcingulate (bottom right panel).The presented maps were thresholded at p <.05, family-wise error rate (FWER) corrected with a cluster-extent based thresholding method with a cluster-defining primary threshold of p < .001.
Cardiff University vs. University of Sussex) mixed measures design with the perceived event and target group membership varied within participants, and the salience of rivalry varied between participants.We manipulated the perceived competitiveness of the outgroups by presenting participants with a bogus news article in which the outgroup was either described as a rival to the ingroup or not (refer to Supplementary Materials for further details about the article).

Table 1 .
Results from the full factorial ANOVA.

Main-effect Other-focussed > Self-focussed
Note: The presented activations were thresholded at p < .05,family-wise error rate (FWER) corrected with a cluster-extent based thresholding method with a cluster-defining primary threshold, p < .001.A minimum voxellevel threshold of 10 mm was applied to all contrasts.

Table S2 .
Results from the pilot study showing mean ratings of empathy in response to the outgroups following priming to intergroup rivalry = 64.The rating scale ranged from 1 (not at all) to 100 (very much so).Standard deviations are given in parentheses.Within a row, means that do not share subscripts are significantly different