An overwhelming amount of evidence indicates that humans are by default cooperative, moral, and deeply averse to harming others. And yet, by some counts, over 200 million civilians have been killed in acts of genocide, war, and other forms of group conflict over the last century alone. How do we reconcile humanity’s tendency toward good with its capacity for the unspeakable?

When people’s psychology shifts from ‘me and you’ to ‘us vs. them’ it fundamentally reshapes their views of what is acceptable or fair. Intergroup dynamics are therefore a critical boundary condition on our most cherished theories of morality, justice, and human nature. Yet, a deep understanding of “groups” as a general concept still eludes us. What is a group? How do we know to which groups we belong? How do we assign others to groups? This is where our work takes aim.

The contemporary intergroup literature emphasizes static category-memberships, usually in conflict dyads—for instance, Black vs. white people. However, the category-based approach is limited because it is inflexible. Social categories are not fixed, monolithic entities. The associations with specific categories change over time (e.g., when Italians in the U.S. became racialized as white in the early 20th century, despite being marked as ethnic outsiders for decades prior); allegiances between categories can change (e.g., Americans’ relationships to Germans in 2022 versus 80 years ago); and not all categories carry with them the psychological potency of purposive groups (e.g., societies have never been concerned about an uprising of Brunettes). Moreover, many conflicts include more than two groups. Thus, studying only pairs of categories is unlikely to get us very far in the pursuit of understanding the general concept of ‘groups’ and hampers our ability to generalize from one intergroup context to another. In other words, most intergroup researchers study what might be best characterized as group geometry—how similar individuals or collectives are to one another, whether people share an identity, what shape the resulting networks take, and so on. But a static framework cannot account for the fact that group boundaries contract and expand over time and contexts, and that these boundaries are constantly being redrawn and negotiated by those within and outside those groups (Cikara, 2021).

This is why we study what we characterize as the physics of groups: the psychologically meaningful forces that regulate the formation, organization, and dissolution of purposive groups. Our coalition-based framework accounts for groups’ dynamic nature and builds in the assumption that groups are best understood as distributions of members who will continually reorganize themselves, subject to these forces which bind them together and rend them apart. Specifically, we investigate: (i) antecedents of coalitional cognition: what generalized features and conditions turn sets of people into meaningful groups, including observed coordination with self and others, competition, and perception of threat; (ii) consequences of coalitional cognition: how these features and conditions alter our emotions, brains, and behaviors; and (iii) interventions: how we can use these forces to reduce intergroup conflict. To address these questions we employ the tools most appropriate for each question: including survey, experimental, behavioral, physiological, computational, and cognitive neuroscience methods.

ANTECEDENTS OF COALITIONAL COGNITION: HOW DO the mind and brain DISCOVER "US" and "THEM"

Social groups inform with whom we cooperate (Everett, Ingbretsen, Cushman, & Cikara 2017), from whom we learn (Vives, Cikara, & FeldmanHall, 2021), and how we allocate our resources (Cetron, Haque, Mair, & Cikara, under review). But before any consequences of coalitional cognition can be rendered, people have to identify others as in-group or out-group members.

How do we figure out who is “us” and who is “them” in any given place or time? One thread of our research program investigates how people accumulate group structure information from their environments to discover social groups (especially in the absence of overt cues to peers’ group membership). We adopt a computational model of latent structure learning to move beyond explicit category labels and mere similarity between agents as the sole inputs to social group discovery (Gershman & Cikara, 2020). We find that people integrate information about how agents in the environment coordinate not only with themselves but also with one another to infer a posterior distribution over possible latent social groupings (Lau, Pouncy, Gershman, & Cikara, 2018; Lau, Gershman, & Cikara, 2020). For example, we find that when two agents agree with participants on political issues to the same extent, participants’ decisions about with which of those two agents to align is influenced by the presence of a third agent that alters the coalitional structure. These alliances influence character attributions (e.g., how moral or competent participants judge these agents) and persist even when participants have alternative cues to group membership. The major strengths of this framework are that, like the human mind, it: (i) is highly flexible—new environments and agents will prompt new inferences that include more complex structures than just an in-group and single out-group); (ii) does not require observation of overtly helpful or harmful behavior, which is relatively rare—mere coordination (or its absence) on any task or issue is sufficient to begin building latent structures; and (iii) does not require knowledge of category membership but can incorporate pre-existing categories as priors (Cikara, 2022).

Building on these findings, my coalitional framework argues that there are generalizable cues driving group discovery and representation on which the details of the relevant categories and groups get overlaid. These cues include expectations of coordination given observed behavior, absence or presence of competition, and perhaps most importantly, threat to one’s self and allies (Chang, Krosch, & Cikara, 2016). If this were the case, then we should expect that generalized threat cues mark others as likely foes and predict consequential intergroup behavior above and beyond the specific groups under consideration.

One such threat feature that has garnered a great deal of attention, particularly with increases in shifting demographics, is group size; the larger groups become, the less they are liked. However this relationship is not so straightforward. What is the threshold at which a new group becomes threatening given that there may be many other out-groups already present in a community? To answer this question we introduce the social group reference dependence hypothesis: rather than being sensitive to the absolute size of any one minority group, majority group members are sensitive to minority groups’ relative rank in size, being most discriminating against the largest local minority group (Cikara, Fouka, Tabellini, 2022). As demographics shift and groups change in relative size discriminatory behavior and attitudes should change accordingly.

We test the social group reference dependence hypothesis focusing on the United States and exploiting variation in group size rank across counties for four minority groups – Black, Hispanic, Asian, and Arab – between 1990 and 2010. As predicted, we find that members of the largest minority group in a county are significantly more likely to be targeted with hate crimes relative to when their own group ranks second or lower in the group size distribution in the same county. We always condition on group size relative to county population and perform dozens of robustness checks (including features of the groups as well as regional features) to isolate the predictive role of rank. The effect always holds. We also go one step further and only exploit variation driven by rank switches within the same county over time. Intuitively, this strategy compares the change in victimization suffered by two minority groups whose relative size, in a given decade, grows by the same amount, but that experience a different “rank change” (e.g. from second to first in a county vs no change). Even when applying this very stringent empirical test, we document that, given the same change in a minority's relative size, it experiences more victimization in counties where its rank moves from second to first than in counties where its rank does not change. How does a representation of rank ordering inside a person’s head translate into discrimination? We conjecture that people begin from the premise that they have finite resources with which to defend their groups, and that they take as input a rank ordering of threat from greatest to least urgency. In principle, people could maintain staunchly negative attitudes and behavior towards all out-groups as new out-groups arrive or grow, but they would end up, in practice, entirely surrounded by competitors. Thus, one strategy is to become relatively more inclusive toward less threatening out-groups (i.e., to infer new groupings). In line with this interpretation, we find that minoritized groups that drop down in rank are substantially less likely to be targeted with hate crimes relative to when their rank remains the same or increases. These findings replicate for self-reported attitudes in the U.S. and the U.K. and the effect is more pronounced for smaller units of regional analysis (neighborhood vs. state) suggesting that rank is most likely encoded via local experience.

Notably, we have documented a similar social reference dependence phenomenon in several other contexts, including voting and hiring decisions (Chang & Cikara, 2018; Chang, Gershman, & Cikara, 2019). This was the focus of our second NSF grant, a CAREER award. Together this slate of projects reveals new insights into human psychology that stem from understanding the fundamental inputs to coalitional psychology rather than particular instances of intergroup conflict and discrimination.

CONSEQUENCES OF COALITIONAL COGNITION: How do we overcome harm AVERSion?

Once coalitional psychology has been activated there is a whole cascade of consequences for our social preferences, emotions, and attributions. The moral prohibitions against harm that guide most people’s behavior in interpersonal interactions may be replaced with a different set of goals and values in intergroup contexts. However, the nature of these goals and consequences will hinge on the current relationship between coalitions: are they cooperative, competitive, or neutral with respect to one another?

We find that the absence of cues signaling likely coordination yields indifference between groups, but the overt presence of cues to threat (e.g., competition over resources and incompatibility between groups’ goals) gives way to conflict and emotions like fear, hatred, and disgust. Even when coalitions are not explicitly engaged in competition, categories merely stereotyped as competitive may elicit hostile attributions, emotions, and behaviors. We find that these attributions and emotions are then marshalled to justify overt discrimination against and persecution of groups and their members.

Empathy and Schadenfreude in intergroup conflict. Empathy is generally recognized as a central component of the human condition; it facilitates social functioning by promoting altruism, even among strangers. Despite its early origins and adaptive functions, people often fail to experience empathy for others who are socially distant, such as members of different social or cultural groups (Cikara, Bruneau, & Saxe, 2011). However, these lapses in empathy alone cannot explain intergroup violence. After all, the absence of empathy is mere apathy, which is generally not a strong predictor of aggressive behavior.

Thus, more than the absence of empathy, we’re interested in the conditions under which people experience the exact opposite of empathy—specifically, pleasure in response to others’ misfortunes (Schadenfreude)—again because we think these emotions are better predictors of harm than apathy. If the suffering of an out-group member is consistently accompanied by the experience of pleasure, that pleasure may teach people over time that out-group harm is not only acceptable, but desirable. In line with this prediction, we find that counter-empathic emotions like Schadenfreude are better predictors than empathy or apathy of willingness to harm out-group members.

What predicts which emotions we experience—empathy, apathy, or Schadenfreude—when we see or learn of another person’s suffering? People are least likely to experience empathy and most likely to experience Schadenfreude when they see out-groups as both competitive with their own interests and high-status: not only are “their” goals at odds with “ours,” they also pose a legitimate threat (Chang, Krosch, & Cikara, 2016). In our first study testing the link between Schadenfreude and harm (Cikara, Botvinick, & Fiske, 2011), Red Sox and Yankees fans reported how much they felt pleasure, anger, and pain after watching baseball plays in which their team and their rival scored or failed. Not surprisingly, participants reported feeling pleasure when watching their own team succeed and watching a rival team’s player fail, even against the Orioles (a relatively less competitive team in the same league). We simultaneously collected fMRI data to test whether empathy and pleasure in response to out-group pain relied on separable neural circuitry. As we predicted, pleasurable baseball plays, including rivals failing to score against the Orioles (the pure Schadenfreude condition), increased responses in the ventral striatum (VS). Weeks later, those participants who exhibited greater VS activation in response to watching their rivals fail also reported an increased likelihood of aggressing against rival team fans (relative to Orioles fans). In fact, VS activation in response to watching rivals fail was a better predictor of harm than even participants’ subjective reports of pleasure in response to watching rivals fail. No such correlation emerged with dorsal anterior cingulate or anterior insula (mirroring the absence of a relationship between reduced empathy and aggression).

Of course, it is part of the script of sports rivalries that people are allowed, even encouraged to express emotions like Schadenfreude. We’ve also tested whether people experience Schadenfreude in more subtle contexts: when targets of misfortunes were merely stereotyped as competitive and high-status (e.g., Asians, female professionals, investment bankers). We presented participants with unlabeled images of people who were rated by a separate sample as appearing competitive/cooperative and high/low status; each image was paired with negative, positive, or neutral events (e.g., “Won a $5 bet,” “Got soaked by a taxi”). As predicted, people smiled more (as measured by facial electromyography recording from the cheek muscles) when misfortunes befell competitive, high-status targets as compared to other social targets, for whom they felt bad (Cikara & Fiske, 2012). However, we attenuated Schadenfreude even for high-status competitive targets by manipulating competition and status-relevant information; participants felt less Schadenfreude for the same targets from the previous study when we described them as either less competitive or lower status. In a third study, participants confirmed they were most willing to subject competitive high-status targets to receive painful electric shocks, further bolstering the relationship between Schadenfreude and harm (Cikara & Fiske, 2011; for reviews see Cikara & Fiske, 2013, 2014; Van Dijk, Ouwerkerk, Smith, & Cikara, 2015, 2018). More recent research indicates that his phenomenon is robust: it replicates with race-based groups (but only when White participants are primed with race-related threat) and is moderated by participants’ own Social Dominance Orientation (Hudson, Cikara, & Sidanius, 2019)—people for whom hierarchy-related competition is chronically accessible feel more Schadenfreude for marginalized out-groups when they suffer and are more likely to support harmful policies targeting those groups. Critically, Schadenfreude mediates the SDO to policy preference relationship (Hudson, Cikara, & Sidanius, under review).

One limitation of these previous studies is that rivalries and stereotypes are confounded in that they provide more than just competition and status-relevant information. In another series of experiments, we randomly assigned people to teams online and manipulated whether these teams were competitive, cooperative, or independent. People exhibited Schadenfreude for novel out-group members, even in the absence of stereotypes or rivalry, so long as that out-group posed an immediate competitive threat (Cikara, Bruneau, Van Bavel, & Saxe, 2014). In a follow-up study, participants reported the same degree of empathy (and lack of Schadenfreude) for unlabeled and in-group targets, indicating that the difference between in-group and out-group emotions is best characterized as out-group antipathy than extraordinary in-group empathy.

Beliefs about out-group emotions in intergroup conflict. We find that people exhibit systematic biases in their judgments of (Lazerus, et al., 2016) and beliefs about out-group emotion which has important implications for conflict. For example, people expect out-group members to be more upset in general than out-group members actually report feeling (Lau, Morewedge, & Cikara, 2017). People also expect out-group members to be more upset than they actually are, specifically toward their in-group in intergroup competition (Lees & Cikara, 2020). For example, Democrats’ and Republican’s judgments of how disapproving out-partisans would be if their own party acted strategically is inflated by 40% relative to how out-partisans actually felt. These misperceptions matter because they foreclose on the possibility of coordination: if I believe ‘they’ hate us and everything we do, there is little use in trying to work together (see Lees & Cikara, 2022, Finkel et al., 2021 for reviews in the context of political polarization). We find that the more pessimistic a person’s meta-perceptions, the more they say they think out-partisans are driven by purposeful obstructionism. The good news is these misperceptions are amenable to correction: people who received “corrective feedback” about how the other side actually feels were less likely to say than the control group that the other side is driven by spiteful obstructionism. Recently, we successfully replicated this overly pessimistic distortion in how we think the out-group thinks about ‘us’ (as well as our intervention) in a sample of > 10,000 participants spread across 26 countries (Ruggieri et al., 2021). This massive replication project demonstrates that inaccurate ‘group meta-perceptions’ are not unique to the U.S. nor to political party conflict. These distortions in judgments of out-group emotions may feed into the escalation of intergroup conflict, but happily, we also find that they can be corrected. Insofar as such interventions scale effectively and are implemented widely, they could potentially strengthen our democracies at a time when political sectarianism is rendering them increasingly fragile. An intriguing possibility is that people perceive and predict out-group members’ emotions differently than in-group members, which could contribute to different empathic and counter-empathic emotions in response to their misfortunes.

INTERVENING ON cOALITIONAL COGNITION: HOW DO WE REDUCE INTERGROUP BIAS?

One of the primary benefits of a coalitional account is that it highlights the specific cues that drive the cleaving of collectives into “in-groups” and “out-groups,” irrespective of whether those inferences come from collectives’ size, actual functional relations between collectives, or even divisive political rhetoric. This orientation gives us greater purchase on what levers to pull to try to attenuate inter-coalitional conflict.

Manipulating experienced or implied competition and coordination. If ideologies such as Social Dominance Orientation, which imbue people with a chronic tendency to see the world as a competitive place, increase inter-coalitional antagonism, reducing chronic experiences of threat might mitigate said antagonism. In line with this proposition, we found that priming people with secure attachment schemas significantly decreased negative out-group emotions and aggressive inter-coalition behavior (Saleem, Prot, Cikara, Anderson, & Lam, 2013). Across a series of studies employing variants of a guided imagination task, we randomly assigned participants to write either about solving a life problem with the support of close others or a trip to the grocery story. American participants who recalled a time when someone close to them was available, supportive, and loving reported feeling significantly less intense negative emotions (e.g., fear, anger, disgust) toward Arab people relative to participants in the control condition. More important, participants primed with a secure attachment schema were less likely to support disproportionately aggressive measures against ISIS members compared with those in the positive mood induction and control conditions. One of the benefits of an approach like this is that it does not harbor the same potential for backfiring, that, say recategorization might (i.e., telling people that they suddenly belong to the same group as “the other side”). Recategorization can threaten group distinctiveness, eliciting reactance, and dampen motivation for progressive social change amongst marginalized groups if they are suddenly on the same side as the group that is marginalizing them. Because priming attachment security doesn’t reference the other group at all, it sidesteps these risks.

A more intuitive manipulation is simply to provide people with information that reduces perceptions of other categories as a threat. In a recent suite of experiments, we attempted to leverage positive, achievement-oriented narratives, which emphasize broader contributions to society (and therefore coordination with all collectives) to reduce prejudice and discrimination, specifically toward immigrants (Martinez, Feldman, Feldman, & Cikara, 2021). Across two experiments we manipulated participants’ exposure to criminal, achievement, or struggle-oriented descriptions of immigrants in order to assess how they impact participants’ latent representations of four politically salient immigrant groups—Germans, Russians, Syrians, and Mexicans. We included struggle narratives as a control condition to isolate the effect of criminalized narratives above and beyond being negatively-valenced. We then applied a novel analytic technique borrowed from cognitive neuroscience—representational similarity analysis—to extract participants’ latent cognitive representations of these immigrant groups and their members. We found that criminal and struggle-oriented narratives fostered racialized immigrant representations (i.e., creating clusters of white versus non-white immigrants in trait space). In other words, thinking about a criminal from Russia or Germany activated a distinct representation from thinking about a criminal from Mexico or Syria even though all four groups were associated with the same crimes. The clusters could have been organized along political ally lines (Germany and Mexico clustering together) or even into four unique clusters, but instead we reliably found white vs. non-white clusters. Notably, criminal narratives exerted the racialization effect even among our most egalitarian respondents. Achievement narratives, by contrast, made immigrants from different backgrounds more similar to one another (generating a single unified representation of “immigrants” from all four countries). Greater homogenization across immigrants in this manner was correlated with greater support for immigrants and immigration.

Of course, in many cases we cannot simply “prime” away competitiveness or force coordination when there are, or have been, real zero-sum resources at stake (nor would it be desirable to do away entirely with conflict; Cikara & Paluck, 2013). So what are the alternatives to manipulating the experience or perception of other-coalition competition and status?

Reducing perceptions of coalitional cohesion via individuation. From a coalitional perspective, the capacity and desire to coordinate (or not) are imbuing collectives with the quality of ‘groupiness.’ Analogous to the way that magnetic fields in an fMRI scanner align hydrogen nuclei, competition gives rise to the inference that other-coalition members are united in their purpose against ‘us.’ Thus directly targeting this quality by dismantling perceptions of coalitional cohesion carries the potential to change attitudes, emotions, and behavior. For example, we found that we could significantly attenuate the intergroup empathy bias by providing participants with visual cues to reduced in-group and out-group distinctiveness—specifically, integrated (as opposed to segregated) social network depictions (Cikara et al., 2014).

With this goal of dismantling perceptions of coalitional cohesion in mind, we have shown that embedding good and bad fortune information about same-team versus competitor-team targets in a larger narrative individuated the targets and thereby attenuated the intergroup empathy bias (Bruneau, Cikara, & Saxe, 2015). Moreover, we found that narratives which highlighted the target’s mind via descriptions of their mental states were more effective at decreasing empathy bias than narratives focused on their physical descriptions. Critically, poorer memory for group membership (but not memory for other aspects of the scenarios) mediated the relationship between the narrative manipulation and the empathy bias. These findings suggest that the narratives exerted their effect by shifting focus away from each target’s group membership toward individuating information.

In another line of research also related to shifting focal attention away from group membership, we have found that asking people to vividly imagine a scene in which they helped someone in need increased how much people helped opposing political party members (Gaesser, Shimura, & Cikara, 2019). Particularly surprising was that it was how vividly participants reported imagining the scene around the person (more than the person in the scene) that drove increases in helping. This suggests that the sensory properties of imagined scenarios, especially when they pull attention away from person-specific features like group membership, may play a much larger role in intergroup behavior than we have previously recognized (Vollberg & Cikara, 2018). We have recently replicated this increase in empathy for both same and opposing political party members using an incidental manipulation of episodic simulation—that is, our ability to mentally transport ourselves beyond our present experience and imagine distant events in time and place. This task, borrowed from memory research, does not direct people to simulate helping at all; instead they watch a video and then report as many details as possible about the setting, actions, and people in the video. We found that this incidental manipulation of episodic simulation, unrelated in content and structure to the empathy judgment task, increased overall empathy for both in-group as well as out-group members relative to a control task. Critically, this relationship was mediated by participant-generated episodic detail of in-group and out-group victim’s surroundings (Vollberg, Gaesser, & Cikara, 2021).   

All this said, simply increasing empathy for other-coalition members is not a panacea. One common assumption is that exercises designed to increase empathy should eliminate empathic failures and their assumed consequences (e.g., harm), but our findings strongly challenge that assumption (for reviews, see Zaki & Cikara, 2015; Weisz & Cikara, 2021). We find that neglectful behavior and policy preferences are better predicted by the size of the gap between in-group and out-group empathy, rather than participants’ absolute levels of empathy. For example, across three studies we found that American respondents cared more about fellow Americans suffering misfortunes relative to Arabs; Hungarians were more empathic toward their countrymen than Muslim refugees; and the same with Greeks toward Germans. And across all three studies, people’s empathy gaps—controlling for participants’ trait empathic concern—predicted their perspectives on policy: how much Americans approved of Arab immigration, how many asylum seekers Hungary should accept, and how much Greece should help if a natural disaster hit Germany (Bruneau, Cikara, & Saxe, 2017). Therefore, interventions or programs aimed at increasing overall empathy (e.g., generalized compassion training) may have little or no effect on increasing inter-coalition harmony or so long as the gap between in-group empathy and out-group empathy is maintained.

Yet another intriguing possibility highlighted by our work is that we can fight a group-level cognitive bias with an individuating cognitive bias. People exhibit a robust, invariant tendency to believe that inside every individual there is a ‘good true self’ calling every person to behave in morally virtuous ways and that this essence is separate from a person’s superficial features. This good-true-self bias is present across cultures, perspectives (first versus third), and individual differences, and appears to be rooted in the basic cognitive tendency to assume that all entities have deep, unobservable, inherent properties that comprise their true nature (DeFreitas, Cikara, Grossman, & Schlegel, 2017, 2018). If this is the case, then even threatening category members should be judged as having good true selves, deep down, when we think about their individual essences. A positive bias that falls out of thinking about the essence of an individual person could be leveraged to reduce a negative bias that falls out of thinking about the nature of a threatening coalition.

Across three experiments we tested whether Americans believed that an American, an Arab, and an Arab in the U.S. all contain good true selves, deep down, to equal extents (DeFreitas & Cikara, 2018). Not only did our participants attribute good true selves across these targets in equal measure, we found that asking Americans to reflect on individual Arab targets’ true selves made them subsequently less prejudiced toward and less threatened by Arab people in general and more likely to donate money to the Syrian Arab Red Crescent charity (relative to people who completed prejudice/donation judgments prior to making true self judgments). We submit that thinking about whether an agent's behavior reflected their true versus surface-self led to a particularly strong form of individuation, which in turn led to more nuanced representations of all people as possessing multiple layers (i.e., surface self could go either way but the true self is good) rather than merely characterizing people as “us” (good) versus “them” (bad).