Across two studies (N = 4,526), we characterize a taxonomy of spontaneous face impressions by applying artificial intelligence text analyses to thousands of free-response descriptions of computer-generated faces. The taxonomy codes almost 100% of the impressions into Appearance (including Beauty), Sociability, Morality, Ability, Assertiveness, Emotion, Social Group, socioeconomic Status, Uniqueness, Family, Health, Occupation, Geographic origin, and political-religious Beliefs content. Results suggest that dimensions from low-dimensional models (e.g., Communion, Agency facets) are highly prevalent, but that alternative dimensions such as Uniqueness and Health are also prevalent. Most dimensions show high (positive) directions, and their correlational structure supports the clustering of low-dimensional models as separate from the expanded taxonomy dimensions. Finally, the taxonomy improves predictions of general evaluations of faces (how positive/negative the face is evaluated overall) and decision making in hypothetical scenarios (e.g., how much to prioritize a target for health care access or antidiscrimination protections).

Sexual minorities report avoiding disclosure of their sexual orientation to healthcare providers to prevent encounters with bias. The present work explores the unique anticipated stereotypes that sexual minorities in the United States expect from healthcare providers using open-text responses and novel machine learning methods to code anticipated stereotypes into stereotype content dimensions. Sexual minority participants (N = 361) reported the traits or characteristics that they would expect a healthcare provider to believe to be true about them under two conditions; one wherein they just met the provider and one wherein their sexual orientation was disclosed. As expected, the valence of expected stereotypes was more negative in the identity disclosed condition, and, critically, participants expected more morality and deviance-based stereotypes when their sexual orientation was known. Further, these stereotype dimensions differentially predicted healthcare visit expectations (i.e., anticipated treatment quality, anticipated concealment of health behaviors/symptoms, anticipated comfort disclosing sexual orientation) and healthcare avoidance. As sexual minority individuals may vary in the types of stereotypes that they expect from healthcare providers and in the valence of stereotypes that they expect from healthcare providers when their identity becomes known, research on anticipated healthcare stigma needs to capture unique stereotypes that should be targeted to reduce healthcare disparities.

People who are stigmatized along concealable features (e.g., individuals reporting adverse childhood experiences) often experience challenges to the self-concept, which can promote psychological distress. Developing a stigmatized identity might counter these effects, but the internality of concealable features can forestall this process: individuals may look to similarly-stigmatized others, but if these group members remain concealed (i.e., are not "out"), they are less identifiable as guides for development. In two studies (Ntotal = 845), less outness among similarly-stigmatized others in the social environment predicted increased distress-but only for individuals reporting low progress in processes of positive meaning-making (Studies 1 and 2) and exploration (Study 2). The interaction held when controlling for stigmatizing views endorsed by non-stigmatized counterparts (Study 2). Findings highlight similarly-stigmatized others as important constituents of the social environment: low group visibility and accessibility may uniquely contribute to distress for individuals at early phases of developing a positive and clear stigmatized identity.

People perceive social groups along stereotype dimensions. Several models of social evaluation identify so-called horizontal (relational, warmth, communion) and vertical (achievement, competence, agency) judgments, also known as the Big Two. Each has two facets, respectively indicating perceived morality and friendliness for horizontal judgments plus ability and assertiveness for vertical judgments. Perceivers also locate groups within sociopolitical structures, such as socioeconomic status and ideological beliefs. These six commonly used stereotype dimensions (morality, friendliness, ability, assertiveness, status, and beliefs) each predict specific and pragmatic behaviors toward (members of) groups, including approach, investment, cooperation, and inclusion. Overall, the trait dimensions correlate positively (e.g., the two respective facets of each of the Big Two), but contextual goals can override general patterns. For example, when people encounter two unequal groups and strive for social justice, harmony and positive identity, the horizontal and vertical judgments correlate negatively. Contextual goals and transient motives also moderate the importance of the stereotype dimensions. We conclude by suggesting avenues for future research.

Social identity complexity (SIC), the perceived overlap among one’s social group memberships, has been linked to reduced prejudice. We extend this work by investigating how SIC shapes both personal (Study 1) and perceived societal (Study 2) stereotypes of intersectional targets across two pre-registered experiments. Participants described targets who shared either two (Double Ingroup), one (Single Ingroup), or no (Double Outgroup) identities with them. Participants provided four traits they personally associate (Study 1) or believe most Americans would associate (Study 2) with each target. Stereotypes were analyzed using valence and emergence coding. Across both studies, targets sharing fewer identities were stereotyped more negatively and with more emergent traits. Higher SIC was associated with less positive personal stereotypes of Double Ingroup targets but did not significantly buffer against negative stereotypes of Double Outgroup targets. Additionally, individuals high in SIC perceived societal stereotypes as more negative and more emergent across all conditions. These findings suggest that individuals with complex social identities are less prone to ingroup favoritism and may be more attuned to emergent content in stereotypes, reflecting heightened awareness of societal bias without endorsing it personally.

Representativeness is a relevant but unexamined property of stereotypes in language models. Existing auditing and debiasing approaches address the direction of stereotypes, such as whether a social category (e.g. men, women) is associated more with incompetence vs. competence content. On the other hand, representativeness is the extent to which a social category's stereotypes are about a specific content dimension, such as Competence, regardless of direction (e.g. as indicated by how often dimension-related words appear in stereotypes about the social category). As such, two social categories may be associated with competence (vs. incompetence), yet one category's stereotypes are mostly about competence, whereas the other's are mostly about alternative content (e.g. Warmth). Such differentiability would suggest that direction-based auditing may fail to identify biases in content representativeness. Here, we use a large sample of social categories that are salient in American society (based on gender, race, occupation, and others) to examine whether representativeness is an independent feature of stereotypes in the ChatGPT chatbot and SBERT language model. We focus on the Warmth and Competence stereotype dimensions, given their well-established centrality in human stereotype content. Our results provide evidence for the construct differentiability of direction and representativeness for Warmth and Competence stereotypes across models and target stimuli (social category terms, racialized name exemplars). Additionally, both direction and representativeness uniquely predicted the models' internal general valence (positivity vs. negativity) and human stereotypes. We discuss implications for the use of AI in the study of human cognition and the field of fairness in AI.

Dominant models of impression formation focus on two fundamental dimensions: a horizontal dimension of warmth/communion/trustworthiness and a vertical dimension of competence/agency/dominance. However, these models have typically been studied using theory-driven methods and stimuli of restricted complexity. We used a data-driven approach and naturalistic stimuli to explore the latent dimensions underlying >300,000 unconstrained linguistic descriptions of 1000 Facebook profile pictures from 2,188 participants. Via traditional (Exploratory Factor Analysis) and modern (natural language dictionaries, semantic sentence embeddings) approaches, we observed impressions to form with regard to the horizontal and vertical dimensions and their respective facets of sociability/morality and ability/assertiveness, plus the key demographic variables of gender, age, and race. However, we also observed impressions to form along numerous further dimensions, including adventurousness, conservatism, fitness, non-conformity, and stylishness. These results serve to emphasize the importance of high-dimensional models of impression formation, and help to clarify the content dimensions underlying unconstrained descriptions of individuals.

This study introduces a taxonomy of stereotype content in contemporary large language models (LLMs). We prompt ChatGPT 3.5, Llama 3, and Mixtral 8x7B, three powerful and widely used LLMs, for the characteristics associated with 87 social categories (e.g., gender, race, occupations). We identify 14 stereotype dimensions (e.g., Morality, Ability, Health, Beliefs, Emotions), accounting for  90 frequent content, but all other dimensions were significantly prevalent. Stereotypes were more positive in LLMs (vs. humans), but there was significant variability across categories and dimensions. Finally, the taxonomy predicted the LLMs' internal evaluations of social categories (e.g., how positively/negatively the categories were represented), supporting the relevance of a multidimensional taxonomy for characterizing LLM stereotypes. Our findings suggest that high-dimensional human stereotypes are reflected in LLMs and must be considered in AI auditing and debiasing to minimize unidentified harms from reliance in low-dimensional views of bias in LLMs.

Abstract Psychological theories continue to expand our understanding of stereotype content and processes. Stereotype content refers to what people think about social groups’ characteristics. Stereotype processes reflect how people integrate information to navigate social interactions. Advances in artificial intelligence introduce innovative analytical tools to revolutionize how psychologists understand stereotypes. Content research builds on theory-driven surveys (e.g. warmth and competence), to data-driven multidimensional scaling (e.g. belief), to large-scale linguistic analysis (e.g. emotion, appearance) to describe a myriad of dimensions people use for social evaluations. Process research starts from an information-processor metaphor (e.g. decoding, encoding), to the predictive brain (e.g. statistical learning, probabilistic modeling), and now to a feedback loop framework (e.g. reinforcement learning, algorithmic bias), paving the way to understand how and why people evaluate others. Understanding stereotypes is a collective enterprise, as evidenced by the scholarly debate that has helped move the field forward.

Many people who are stigmatized along concealable features (e.g., sexual minorities or people with mental illness) anticipate social rejection due to their features and associated labels, and these beliefs are a prominent predictor of psychological distress. While ecological approaches to stigma research have highlighted the social basis of these two related outcomes, it typically has focused on the impact of non-stigmatized counterparts. Also embedded in the social environment are similarly-stigmatized others who, in concealing, may be less accessible to the individual. Given the centrality of psychological distress and rejection concerns as a relational self-conception in attachment theories, we tested if identity-based rejection sensitivity and distress may emerge from diminished access to similarly-stigmatized others as identity group members. Leveraging the University as a partially-controlled, naturalistic setting, we collected measures of concealment, identity-based rejection sensitivity, and psychological distress from undergraduate students in introductory psychology courses who reported a concealable stigmatized identity (N = 355; k = 15 identity groups). With concealment aggregated to the level of the identity group, multi-level modeling showed that concealment by similarly-stigmatized students was positively associated with both individuals’ identity-based rejection sensitivity and their psychological distress. Moreover, rejection sensitivity mediated the association of group-level concealment and distress. Findings suggest that rejection concerns and distress may emerge from identity group inaccessibility in the social environment, with the association of concerns and distress possibly contextualized by underlying group attachment dynamics. Results reveal the identity group as a novel source of social influence in the lives of individuals with concealable stigmatized identities.

Social impressions from faces have been studied in psychology for over 100 years. These impressions are rapid, efficient, and consequential. Yet, disagreements about the content of face impressions remain. Across two studies (N = 4,526), we develop a taxonomy of spontaneous face impressions content by applying novel interdisciplinary methods from Artificial Intelligence text analysis to thousands of free-response descriptions of computer-generated faces. We identify a taxonomy of face impression dimensions, and describe their coverage, prevalence, directionality, and correlational structure. We characterize a diverse and nuanced taxonomy of content that, when compared to just the content that dominant low-dimensional models focus on, increases the coverage of spontaneous responses from about 50% to almost 100%. Our results describe general patterns of prevalence, indicating that dimensions from low-dimensional models (e.g., Sociability, Morality, Assertiveness) are highly prevalent, but that alternative dimensions such as Uniqueness and Health, among others, are also significantly prevalent in face impressions of naturalistic face photographs. Most dimensions show a positivity bias, and the correlational structure of the dimensions further supports the clustering of low-dimensional model’s content as separate from the expanded taxonomy dimensions. Finally, this expanded taxonomy improves predictions of general evaluations and decision making in various real-world relevant contexts. The derived taxonomy of spontaneous face impressions content is a foundation for further theoretical development and practical applications in an area central to human behavior.

People belong to multiple social groups simultaneously. However, much remains to be learned about the rich semantic perceptions of multiply-categorized targets. Two pretests and three main studies (n = 1,116) compare perceptions of single social categories to perceptions of two intersecting social categories. Unlike previous research focusing on specific social categories (e.g., race and age), our studies involve intersections from a large sample of salient societal groups. Study 1 provides evidence for biased information integration (vs. averaging), such that ratings of intersecting categories were more similar to the constituent with more negative and more extreme (either very positive or very negative) stereotypes. Study 2 indicates that negativity and extremity also bias spontaneous perceptions of intersectional targets, including dimensions beyond Warmth and Competence. Study 3 shows that the prevalence of emergent properties (i.e., traits attributed to intersecting categories but not the constituents) is greater for novel targets and targets with incongruent constituent stereotypes (e.g., one constituent is stereotyped as high Status and the other as low Status). Finally, Study 3 suggests that emergent (vs. present in constituents) perceptions are more negative and tend to be more about Morality and idiosyncratic content and less about Competence or Sociability. Our findings advance understanding about perceptions of multiply-categorized targets, information integration, and the connection between theories of process (e.g., individuation) and content. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

It is crucial to understand why people comply with measures to contain viruses and their effects during pandemics. We provide evidence from 35 countries (Ntotal = 12,553) from 6 continents during the COVID-19 pandemic (between 2021 and 2022) obtained via cross-sectional surveys that the social perception of key protagonists on two basic dimensions-warmth and competence-plays a crucial role in shaping pandemic-related behaviors. Firstly, when asked in an open question format, heads of state, physicians, and protest movements were universally identified as key protagonists across countries. Secondly, multiple-group confirmatory factor analyses revealed that warmth and competence perceptions of these and other protagonists differed significantly within and between countries. Thirdly, internal meta-analyses showed that warmth and competence perceptions of heads of state, physicians, and protest movements were associated with support and opposition intentions, containment and prevention behaviors, as well as vaccination uptake. Our results have important implications for designing effective interventions to motivate desirable health outcomes and coping with future health crises and other global challenges.

The spontaneous stereotype content model (SSCM) describes a comprehensive taxonomy, with associated properties and predictive value, of social-group beliefs that perceivers report in open-ended responses. Four studies (N = 1,470) show the utility of spontaneous stereotypes, compared to traditional, prompted, scale-based stereotypes. Using natural language processing text analyses, Study 1 shows the most common spontaneous stereotype dimensions for salient social groups. Our results confirm existing stereotype models' dimensions, while uncovering a significant prevalence of dimensions that these models do not cover, such as Health, Appearance, and Deviance. The SSCM also characterizes the valence, direction, and accessibility of reported dimensions (e.g., Ability stereotypes are mostly positive, but Morality stereotypes are mostly negative; Sociability stereotypes are provided later than Ability stereotypes in a sequence of open-ended responses). Studies 2 and 3 check the robustness of these findings by: using a larger sample of social groups, varying time pressure, and diversifying analytical strategies. Study 3 also establishes the value of spontaneous stereotypes: compared to scales alone, open-ended measures improve predictions of attitudes toward social groups. Improvement in attitude prediction results partially from a more comprehensive taxonomy as well as a construct we refer to as stereotype representativeness: the prevalence of a stereotype dimension in perceivers' spontaneous beliefs about a social group. Finally, Study 4 examines how the taxonomy provides additional insight into stereotypes' influence on decision-making in socially relevant scenarios. Overall, spontaneous content broadens our understanding of stereotyping and intergroup relations. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Societal threats that face the world today seem overpowering, especially for young generations who will need to develop creative solutions. The present study examined the relationships between societal threats and social motives. Social motives function to orient individuals toward the social world and prepare them to engage socially. This adaptive function of social motives may be particularly useful when threats are looming in the environment. We thus expected that perceived societal threats would correlate positively with activation of social motives, especially among individuals with lower self‐esteem, who tend to show higher interdependency when threatened. Our cross‐cultural samples from Australia, the United States, New Zealand, the Philippines, China (Macao), Malaysia (Sabah), and Austria (N = 1,269) showed evidence to support these expectations. Perceived societal threats correlated positively with all social motives (Belong, Understand, Control, Esteem, and Trust); however, the link was most vital for the Control motive, and especially in the United States and China. In line with our expectations, higher perceived societal threats were associated with more robust social motives, especially among those with low self‐esteem. Potential mechanisms through which social motives assist adaptation to societal threats and country‐specific contents of threats are discussed.

A new scale to measure core social motives was developed based on the BUC(K)ET framework (Belong, Understand, Control, Esteem, and Trust). The scale was completed by 1,516 university students from seven countries: Australia, the United States, New Zealand, the Philippines, Malaysia, China (Macao), and Austria. Multigroup confirmatory factor analysis supported the scale's full scalar invariance between Australia and the United States and between Australia and Austria. Partial scalar invariance was established for all countries after omitting the Understand motive, suggesting that the remaining four subscales can be used to compare levels of social motives across diverse cultural groups with caution. We further established the scale's construct validity by examining its correlations in the nomological networks involving several individual difference variables. The profile of social motives was remarkably similar across countries and gender groups, although three Asian groups showed higher motives to belong than non-Asian groups, and women showed generally stronger core social motives than men, especially the Belong motive. Implications and possible directions of research are discussed.

People gather information about others along a few fundamental dimensions; their current goals determine which dimensions they most need to know. As proponents of competing social-evaluation models, we sought to study the dimensions that perceivers spontaneously prioritize when gathering information about unknown social groups. Because priorities depend on functions, having relational goals (e.g., deciding whether and how to interact with a group) versus structural goals (e.g., getting an overview of society) should moderate dimensional priorities. Various candidate dimensions could differentiate perceivers' impressions of social groups. For example, the Stereotype Content model argues that people evaluate others in terms of their Warmth (i.e., their Sociability and Morality) and Competence (i.e., their Ability and Assertiveness). Alternatively, the Agency-Beliefs-Communion (ABC) model proposes conservative-progressive Beliefs. Five studies (N = 2,268) found that participants consistently prioritized learning about targets' Warmth. However, goal moderated priority: When participants had a relational goal, such as an unknown group increasing in their neighborhood, they showed more interest in targets' Sociability, a facet of Warmth. When participants had a structural goal, such as an unknown group increasing in their nation, they showed more interest in the groups' Beliefs, as well as increased interest in Competence-related facets. Diverse methods reveal interest in all dimensions, reconciling discrepancies among social-evaluation models by identifying how relational versus structural goals differentiate priorities of the fundamental dimensions proposed by current models. Results have implications for fundamental dimensions of social cognition, more generally. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Advances in natural language processing provide accessible approaches to analyze psychological open-ended data. However, comprehensive instruments for text analysis of stereotype content are missing. We developed stereotype content dictionaries using a semi-automated method based on WordNet and word embeddings. These stereotype content dictionaries covered over 80% of open-ended stereotypes about salient American social groups, compared to 20% coverage from words extracted directly from the stereotype content literature. The dictionaries showed high levels of internal consistency and validity, predicting stereotype scale ratings and human judgments of online text. We developed the R package Semi-Automated Dictionary Creation for Analyzing Text (SADCAT; ) for access to the stereotype content dictionaries and the creation of novel dictionaries for constructs of interest. Potential applications of the dictionaries range from advancing person perception theories through laboratory studies and analysis of online data to identifying social biases in artificial intelligence, social media, and other ubiquitous text sources.