Polarization debates often presume two sorted “camps” in American politics, yet we lack a map of how citizens perceive the coalitions that constitute these camps—perceptions that structure partisan animus, information processing and vote choice, and political behavior. Across five studies using historical and novel survey data, we inductively discover this cognitive map. Analyses reveal a stable, asymmetric, tripartite structure: a unified conservative coalition alongside a fractured liberal wing composed of distinct “Ideological” and “Demographic” coalitions. This structure is organized along two dimensions: a primary ideological axis capturing left–right conflict, and a secondary common fate axis distinguishing groups perceived as motivated by tangible, widespread welfare concerns versus narrower “special” interests. Critically, tangible-priority perceptions predict partisan hostility towards groups, beyond groups’ perceived ideological and partisan extremity. This perceptual architecture explains the entanglement of social-group stereotypes with partisanship, reconciles rising affective polarization with stable demographic sorting, and identifies limits on persuasion and mobilization.

The belief that humans belong to distinct racial categories pervades our modern world and deeply shapes our social and political life. This perception exists despite the fact that racial categories are both evolutionarily novel and biologically unsound. A candidate explanation for why humans categorize others into racial groups can be found in the "alliance hypothesis of race" which proposes that modern racial categorization is a byproduct of a cognitive system designed for ancestral alliance detection: coalitional psychology. Compelling evidence from memory confusion studies has supported this hypothesis by demonstrating that redirecting coalitional psychology can suppress racial categorization. However, the capacity for a coalitional psychology to generate racial categories from scratch has not yet been explored. Using agent-based models, we show that agents endowed with a coalitional psychology that attempts to detect patterns of alliance based on available cues can hallucinate and then reify correlations between phenotype and allegiance, leading to the emergence of social groups that vary systematically by phenotype. This occurs even when phenotype is in reality distributed continuously and has no true connection to behavior. These models provide evidence that a coalitional psychology alone can be sufficient to create beliefs in phenotype-based social categories even when no such categories truly exist.

What aspects of prejudicial attitudes toward social out-groups predict behavior toward those groups? Social attitudes research typically links favorability toward social out-groups to engagement in intergroup behaviors: the stronger your (un)favorability toward a group, the more costly behaviors you should engage in, concerning that group. But many people make strongly-valenced favorability statements about minoritized out-groups without engaging in corresponding costly actions. We investigate subjective attitude importance as a competing predictor of costly intergroup behaviors. When white respondents rate their attitudes toward minoritized racial out-groups as more personally important to them, they are more likely to give up real money to: (1) prevent biased attitude signaling, (2) preserve their reputations, and (3) increase charitable donation to an out-group-supporting nonprofit. Finally, we find that attitude measurements that centered personal relevance better explained variance in costly behavior than generic favorability or importance measurements. These results indicate that measuring subjective attitude importance, with emphasis on the personal relevance of the attitude, improves prediction of supportive and discriminatory behaviors toward minoritized racial out-groups.

To evaluate whether large language models (LLMs) can be leveraged to predict the results of social science experiments, we built an archive of 70 pre-registered, nationally representative, survey experiments conducted in the United States, involving 469 experimental effects and 119,330 participants. We prompted advanced, publicly-available LLMs, including GPT-4 and three open-weight LLMs, to simulate how representative samples of Americans would respond to the stimuli from these experiments. Predictions derived from simulated responses correlate strikingly with actual treatment effects (r = 0.83), equaling the predictive accuracy of human forecasters. Accuracy remained high for unpublished studies that could not appear in the model’s training data (r = 0.90). In a secondary archive of 15 recent megastudies featuring an additional 606 treatment effects, GPT-4 predictions matched the accuracy of experts. We conducted an original survey of 460 social scientists, measuring likelihood of use of AI predictions for various scientific applications and concerns about risks, and then used our two archives to directly assess several prominent scientific applications (pilot testing, intervention selection, identifying effects needing replication) and risks (bias, misuse). Together, our results suggest LLMs can augment experimental methods in science and practice, but also highlight important limitations.

To effectively address intergroup violence, we must accurately diagnose the psychological motives driving it. Dehumanization--the explicit and blatant denial of an outgroup's humanity--is widely considered one such driver, informing both scholarly theory and social policy on intergroup violence. Nonetheless, dehumanization is often intertwined with intense negative affect, raising concerns that dehumanization's explanatory power is much more restricted than widely assumed. In the extreme, "dehumanization" is merely another way to express intense dislike. If so, then theories of dehumanization distort our understanding of the true motives driving violence. Here, we test dehumanization's reality and explanatory power through three complementary research streams that employ diverse methods and samples. First, we meta-analyze existing studies on dehumanization and dislike to establish their independent effects on violence (k = 120; N = 128,022). We then test the generalizability of these effects across four violent conflicts in the United States, Russia and Ukraine, Israel and the Palestinian diaspora, and India (N_Total = 3,773). In these studies, we also test whether individuals' dehumanizing responses are mere metaphor or intended literally. Finally, we isolate dehumanization's causal impact on violence in another US sample (N = 753). Our results converge to demonstrate that dehumanization is (a) distinct from dislike and often intended literally, (b) a particularly strong predictor of support for violence, and (c) can causally facilitate such support. Collectively, these studies clarify our understanding of the psychology driving violence and can inform efforts to address it.

Is everyone equally justified in blaming another's moral transgression? Across five studies (four pre-registered; total N = 1,316 American participants), we investigated the perception of moral standing to blame--the appropriateness and legitimacy for someone to blame a moral wrongdoing. We propose and provide evidence for a moral commitment hypothesis--a blamer is perceived to have low moral standing to blame a moral transgressor if the blamer demonstrates weak commitment to that moral rule. As hypothesized, we found that when blamers did not have the chance or relevant experience to demonstrate good commitment to a moral rule, participants generally believed that they had high moral standing to blame. However, when a blamer demonstrated bad commitment to a moral rule in their past behaviors, participants consistently granted the blamer low moral standing to blame. Low moral standing to blame was generally associated with perceiving the blame to be less effective and less likely to be accepted. Moreover, indirectly demonstrating moral commitment, such as acknowledging one's past wrongdoing and feeling/expressing guilt, modestly restored moral standing to blame. Our studies demonstrate moral commitment as a key mechanism for determining moral standing to blame and emphasize the importance of considering a blamer's moral standing as a crucial factor in fully understanding the psychology of blame.

A frequently invoked explanation for the sharing of false over true political information is that partisans are motivated by their reputations. In particular, it is often argued that by indiscriminately sharing news that is favorable to one's political party, regardless of whether it is true--or perhaps especially when it is not--partisans can signal loyalty to their group, and improve their reputations in the eyes of their online networks. Across three survey studies (total N = 3,038), and an analysis of over 26,000 tweets, we explored these hypotheses by measuring the reputational benefits that people anticipate and receive from sharing different content online. In the survey studies, we showed participants actual news headlines that varied in (a) veracity, and (b) favorability to their preferred political party. Across all three studies, participants anticipated that sharing true news would bring more reputational benefits than sharing false news. Critically, while participants also expected greater reputational benefits for sharing news favorable to their party, the perceived reputation value of veracity was no smaller for more favorable headlines. We found a similar pattern when analyzing engagement on Twitter: among headlines that were politically favorable to a user's preferred party, true headlines elicited more approval than false headlines.

Scholars warn that partisan divisions in the mass public threaten the health of American democracy. We conducted a megastudy (n = 32,059 participants) testing 25 treatments designed by academics and practitioners to reduce Americans' partisan animosity and anti-democratic attitudes. We find many treatments reduced partisan animosity, most strongly by highlighting relatable sympathetic individuals with different political beliefs, or by emphasizing common identities shared by rival partisans. We also identify several treatments that reduced support for undemocratic practices--most strongly by correcting misperceptions of rival partisans' views, or highlighting the threat of democratic collapse--showing anti-democratic attitudes are not intractable. Taken together, the study's findings identify promising general strategies for reducing partisan division and improving democratic attitudes, shedding new theoretical light on challenges facing American democracy.