Extracting intersectional stereotypes from embeddings: Developing and validating the Flexible Intersectional Stereotype Extraction procedure.

Social group-based identities intersect. The meaning of "woman" is modulated by adding social class as in "rich woman" or "poor woman." How does such intersectionality operate at-scale in everyday language? Which intersections dominate (are most frequent)? What qualities (positivity, competence, warmth) are ascribed to each intersection? In this study, we make it possible to address such questions by developing a stepwise procedure, Flexible Intersectional Stereotype Extraction (FISE), applied to word embeddings (GloVe; BERT) trained on billions of words of English Internet text, revealing insights into intersectional stereotypes. First, applying FISE to occupation stereotypes across intersections of gender, race, and class showed alignment with ground-truth data on occupation demographics, providing initial validation. Second, applying FISE to trait adjectives showed strong androcentrism (Men) and ethnocentrism (White) in dominating everyday English language (e.g. White + Men are associated with 59% of traits; Black + Women with 5%). Associated traits also revealed intersectional differences: advantaged intersectional groups, especially intersections involving Rich, had more common, positive, warm, competent, and dominant trait associates. Together, the empirical insights from FISE illustrate its utility for transparently and efficiently quantifying intersectional stereotypes in existing large text corpora, with potential to expand intersectionality research across unprecedented time and place. This project further sets up the infrastructure necessary to pursue new research on the emergent properties of intersectional identities.

(when) do counterattitudinal exemplars shift implicit racial evaluations? Replications and extensions of Dasgupta and Greenwald (2001).

Dasgupta and Greenwald (2001) demonstrated that exposure to positive Black exemplars (e.g., Colin Powell) and negative White exemplars (e.g., Jeffrey Dahmer) can reduce implicit pro-White/anti-Black evaluations, as measured by an Implicit Association Test. Here, we report seven preregistered online experiments conducted with volunteer U.S. participants (N = 6,953) that sought to replicate and probe the boundary conditions of this finding. Contrary to expectations, we found no shift in implicit racial evaluations in two close replication attempts (Experiments 1-2). Experiments 3-4 ruled out the possibility of insufficiently strong exemplar valence and subtyping as explanations for the failures to replicate. In Experiment 5, implicit racial evaluations did exhibit malleability in response to two different procedures relying on repeated evaluative pairings and evaluative statements, suggesting that they are capable of change. With insight from these studies, Experiments 6-7 were mounted with modifications to the Dasgupta and Greenwald (2001) procedure. Significant reductions in implicit pro-White/anti-Black evaluations were now observed when race, valence, and the contingency between the two were highlighted. In addition, across all experiments, the magnitude of shift in implicit racial evaluations was significantly predicted by participants' ability to recall the Black-positive and White-negative contingencies experienced during the exemplar exposure task. Together, these data suggest that exposure to counterattitudinal exemplars can shift implicit racial evaluations toward neutrality, but such malleability strongly depends on contingency awareness. We discuss implications for social cognitive theory, theoretically informed debiasing interventions, and different paths toward resolving initial replication failures. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Identifying and predicting stereotype change in large language corpora: 72 groups, 115 years (1900-2015), and four text sources.

The social world is carved into a complex variety of groups each associated with unique stereotypes that persist and shift over time. Innovations in natural language processing (word embeddings) enabled this comprehensive study on variability and correlates of change/stability in both manifest and latent stereotypes for 72 diverse groups tracked across 115 years of four English-language text corpora. Results showed, first, that group stereotypes changed by a moderate-to-large degree in manifest content (i.e., top traits associated with groups) but remained relatively more stable in latent structure (i.e., average cosine similarity of top traits' embeddings and vectors of valence, warmth, or competence). This dissociation suggests new insights into how stereotypes and their consequences may endure despite documented changes in other aspects of group representations. Second, results showed substantial variability of change/stability across the 72 groups, with some groups revealing large shifts in manifest and latent content, but others showing near-stability. Third, groups also varied in how consistently they were stereotyped across texts, with some groups showing divergent content, but others showing near-identical representations. Fourth, this variability in change/stability across groups was predicted from a combination of linguistic (e.g., frequency of mentioning the group; consistency of group stereotypes across texts) and social (e.g., the type of group) correlates. Groups that were more frequently mentioned in text changed more than those rarely mentioned; sociodemographic groups changed more than other group types (e.g., body-related stigmas, mental illnesses, occupations), providing the first quantitative evidence of specific group features that may support historical stereotype change. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

All human social groups are human, but some are more human than others: A comprehensive investigation of the implicit association of "Human" to US racial/ethnic groups.

All human groups are equally human, but are they automatically represented as such? Harnessing data from 61,377 participants across 13 experiments (six primary and seven supplemental), a sharp dissociation between implicit and explicit measures emerged. Despite explicitly affirming the equal humanity of all racial/ethnic groups, White participants consistently associated Human (relative to Animal) more with White than Black, Hispanic, and Asian groups on Implicit Association Tests (IATs; experiments 1-4). This effect emerged across diverse representations of Animal that varied in valence (pets, farm animals, wild animals, and vermin; experiments 1-2). Non-White participants showed no such Human=Own Group bias (e.g., Black participants on a White-Black/Human-Animal IAT). However, when the test included two outgroups (e.g., Asian participants on a White-Black/Human-Animal IAT), non-White participants displayed Human=White associations. The overall effect was largely invariant across demographic variations in age, religion, and education but did vary by political ideology and gender, with self-identified conservatives and men displaying stronger Human=White associations (experiment 3). Using a variance decomposition method, experiment 4 showed that the Human=White effect cannot be attributed to valence alone; the semantic meaning of Human and Animal accounted for a unique proportion of variance. Similarly, the effect persisted even when Human was contrasted with positive attributes (e.g., God, Gods, and Dessert; experiment 5a). Experiments 5a-b clarified the primacy of Human=White rather than Animal=Black associations. Together, these experiments document a factually erroneous but robust Human=Own Group implicit stereotype among US White participants (and globally), with suggestive evidence of its presence in other socially dominant groups.

Patterns of Implicit and Explicit Attitudes: IV. Change and Stability From 2007 to 2020.

Using more than 7.1 million implicit and explicit attitude tests drawn from U.S. participants to the Project Implicit website, we examined long-term trends across 14 years (2007-2020). Despite tumultuous sociopolitical events, trends from 2017 to 2020 persisted largely as forecasted from past data (2007-2016). Since 2007, all explicit attitudes decreased in bias between 22% (age attitudes) and 98% (race attitudes). Implicit sexuality, race, and skin-tone attitudes also continued to decrease in bias, by 65%, 26%, and 25%, respectively. Implicit age, disability, and body-weight attitudes, however, continued to show little to no long-term change. Patterns of change and stability were generally consistent across demographic groups (e.g., men and women), indicating widespread, macrolevel change. Ultimately, the data magnify evidence that (some) implicit attitudes reveal persistent, long-term change toward neutrality. The data also newly reveal the potential for short-term influence from sociopolitical events that temporarily disrupt progress toward neutrality, although attitudes eventually return to long-term homeostasis in trends.

Historical representations of social groups across 200 years of word embeddings from Google Books.

Using word embeddings from 850 billion words in English-language Google Books, we provide an extensive analysis of historical change and stability in social group representations (stereotypes) across a long timeframe (from 1800 to 1999), for a large number of social group targets (Black, White, Asian, Irish, Hispanic, Native American, Man, Woman, Old, Young, Fat, Thin, Rich, Poor), and their emergent, bottom-up associations with 14,000 words and a subset of 600 traits. The results provide a nuanced picture of change and persistence in stereotypes across 200 y. Change was observed in the top-associated words and traits: Whether analyzing the top 10 or 50 associates, at least 50% of top associates changed across successive decades. Despite this changing content of top-associated words, the average valence (positivity/negativity) of these top stereotypes was generally persistent. Ultimately, through advances in the availability of historical word embeddings, this study offers a comprehensive characterization of both change and persistence in social group representations as revealed through books of the English-speaking world from 1800 to 1999.

Two thousand years after Archimedes, psychologist finds three topics that will simply not yield to the experimental method.

Cesario argues that experiments cannot illuminate real group disparities because they leave out factors that operate in ordinary life. But what Cesario calls flaws are, in fact, the point of the experimental method. Of all the topics in science, we have to wonder why racial discrimination would be uniquely unsuited for investigating with experiments. The argument to give up the most powerful scientific method to study one of the hardest problems we confront is laughable.

An implicit gender sex-science association in the general population and STEM faculty.

We investigated implicit associations between social categories female or male and the attributes sex or science. In six experiments, Implicit Association Tests (IATs) showed female + sex/male + science associations. The bias was observed (a) in both men and women; (b) in participants who reported sexual attraction to both females and males (greater for the former); (c) in members of the general population as well as among STEM faculty from the highest ranked U.S. STEM universities; (d) even when both gender categories were clearly presented as scientists, via photos and words, (e) using both the standard IAT and a single category variation; and (f) hardly at all on explicit measures in contrast to implicit measures. By introducing the attribute of sexuality, these studies bring to light a robust if unintended mental association of women as sexual beings. The automaticity and surprising generality of the effect suggests that this association may be an unintentional yet potent barrier to women's lower representation and success in STEM.

Patterns of implicit and explicit attitudes II. Long-term change and stability, regardless of group membership.

Over the past decade, implicit attitudes about sexual orientation, race, and age have revealed both change toward neutrality (sexuality and race attitudes) and stability (age attitudes). But how consistently have such patterns of change and stability unfolded across U.S. society? Are the trends widespread, with most demographic groups changing or remaining stable in parallel, at the same rate and in the same direction? Or are the trends more idiosyncratic, with groups moving at different rates and/or directions, revealing nonparallel change? The answer can reveal whether the sources of change are unfolding at the collective, macrolevel of society, or at the mezzo-level of demographic group memberships. Results from over 2.5 million tests of sexuality, race, and age attitudes, collected continuously in the United States over 10 years (2007-2016) show that attitude trends are largely parallel across most demographic groups (e.g., respondents' gender, race, education). Parallel trends are more strongly evident in implicit social group attitudes, with explicit attitudes showing relatively more nonparallel trends. Two demographics, respondent age and political orientation, are exceptions: younger and politically liberal groups are generally changing faster toward implicit attitude neutrality than older and conservative groups. Nevertheless, the surprising consistency in trends across demographic groups points to the role of macrolevel societal variables as the most likely sources of widespread reductions in implicit and explicit social group attitudes over the past decade. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Systemic racism: individuals and interactions, institutions and society.

Systemic racism is a scientifically tractable phenomenon, urgent for cognitive scientists to address. This tutorial reviews the built-in systems that undermine life opportunities and outcomes by racial category, with a focus on challenges to Black Americans. From American colonial history, explicit practices and policies reinforced disadvantage across all domains of life, beginning with slavery, and continuing with vastly subordinated status. Racially segregated housing creates racial isolation, with disproportionate costs to Black Americans' opportunities, networks, education, wealth, health, and legal treatment. These institutional and societal systems build-in individual bias and racialized interactions, resulting in systemic racism. Unconscious inferences, empirically established from perceptions onward, demonstrate non-Black Americans' inbuilt associations: pairing Black Americans with negative valences, criminal stereotypes, and low status, including animal rather than human. Implicit racial biases (improving only slightly over time) imbed within non-Black individuals' systems of racialized beliefs, judgments, and affect that predict racialized behavior. Interracial interactions likewise convey disrespect and distrust. These systematic individual and interpersonal patterns continue partly due to non-Black people's inexperience with Black Americans and reliance on societal caricatures. Despite systemic challenges, Black Americans are more diverse now than ever, due to resilience (many succeeding against the odds), immigration (producing varied backgrounds), and intermarriage (increasing the multiracial proportion of the population). Intergroup contact can foreground Black diversity, resisting systemic racism, but White advantages persist in all economic, political, and social domains. Cognitive science has an opportunity: to include in its study of the mind the distortions of reality about individual humans and their social groups.

Specificity and incremental predictive validity of implicit attitudes: studies of a race-based phenotype.

Four studies involving 2552 White American participants were conducted to investigate bias based on the race-based phenotype of hair texture. Specifically, we probed the existence and magnitude of bias in favor of Eurocentric (straight) over Afrocentric (curly) hair and its specificity in predicting responses to a legal decision involving the phenotype. Study 1 revealed an implicit preference, measured by an Implicit Association Test (IAT), favoring Eurocentric over Afrocentric hair texture among White Americans. This effect was not reducible to a Black/White implicit race attitude nor to mere perceptual preference favoring straight over curly hair. In Study 2, the phenotype (hair) IAT significantly and uniquely predicted expressions of support in response to an actual legal case that involved discrimination on the basis of Afrocentric hair texture. Beyond replicating this result, Studies 3 and 4 (the latter preregistered) provided further, and even more stringent, evidence for incremental predictive validity: in both studies, the phenotype IAT was associated with support for a Black plaintiff above and beyond the effects of two parallel explicit scales and, additionally, a race attitude IAT. Overall, these studies support the idea that race bias may be uniquely detected by examining implicit attitudes elicited by group-based phenotypicality, such as hair texture. Moreover, the present results inform theoretical investigations of the correspondence principle in the context of implicit social cognition: they suggest that tailoring IATs to index specific aspects of an attitude object (e.g., by decomposition of phenotypes) can improve prediction of intergroup behavior.

Gender Stereotypes in Natural Language: Word Embeddings Show Robust Consistency Across Child and Adult Language Corpora of More Than 65 Million Words.

Stereotypes are associations between social groups and semantic attributes that are widely shared within societies. The spoken and written language of a society affords a unique way to measure the magnitude and prevalence of these widely shared collective representations. Here, we used word embeddings to systematically quantify gender stereotypes in language corpora that are unprecedented in size (65+ million words) and scope (child and adult conversations, books, movies, TV). Across corpora, gender stereotypes emerged consistently and robustly for both theoretically selected stereotypes (e.g., work-home) and comprehensive lists of more than 600 personality traits and more than 300 occupations. Despite underlying differences across language corpora (e.g., time periods, formats, age groups), results revealed the pervasiveness of gender stereotypes in every corpus. Using gender stereotypes as the focal issue, we unite 19th-century theories of collective representations and 21st-century evidence on implicit social cognition to understand the subtle yet persistent presence of collective representations in language.

Meta-Analytic Use of Balanced Identity Theory to Validate the Implicit Association Test.

This meta-analysis evaluated theoretical predictions from balanced identity theory (BIT) and evaluated the validity of zero points of Implicit Association Test (IAT) and self-report measures used to test these predictions. Twenty-one researchers contributed individual subject data from 36 experiments (total N = 12,773) that used both explicit and implicit measures of the social-cognitive constructs. The meta-analysis confirmed predictions of BIT's balance-congruity principle and simultaneously validated interpretation of the IAT's zero point as indicating absence of preference between two attitude objects. Statistical power afforded by the sample size enabled the first confirmations of balance-congruity predictions with self-report measures. Beyond these empirical results, the meta-analysis introduced a within-study statistical test of the balance-congruity principle, finding that it had greater efficiency than the previous best method. The meta-analysis's full data set has been publicly archived to enable further studies of interrelations among attitudes, stereotypes, and identities.

How effectively can implicit evaluations be updated? Using evaluative statements after aversive repeated evaluative pairings.

Implicit evaluations (attitudes) are often described as resistant to change, especially when they were initially formed in a seemingly associative manner, such as via repeated evaluative pairings (REP), and new learning is created via propositional material, such as evaluative statements (ES). The present research (total N = 2,124) tested the responsiveness of implicit evaluations instantiated via REP to updating via different types of ES. In Experiment 1, initial learning was created via repeatedly pairing a novel target with strongly negative stimuli (screams) in an aversive REP (A-REP) task. Subsequent ES of opposing valence providing diagnostic information about the target's behavior substantially updated implicit (IAT) evaluations. In Experiment 2, behavioral ES resulted in successful updating after A-REP whether or not they provided an explanation for the initial A-REP learning. A previously unobtained result emerged in Experiment 3 showing that updating was durable even after 1 day. Finally, in Experiment 4, implicit evaluations were updated via diagnostic behavioral ES, but not via an ES instruction to suppose that different pairings had occurred during A-REP. Taken together, these experiments challenge associative theories of implicit evaluation by demonstrating that diagnostic behavioral statements can durably override the effects of initial learning on implicit evaluations, even if such initial learning is aversive and involves direct experience with stimulus pairings. Moreover, by showing that verbal manipulations based on diagnostic behavior but not a mere supposition instruction had impact, the present project advances theory by starting to identify the nature of learning that can adaptively update social impressions. (PsycInfo Database Record (c) 2020 APA, all rights reserved).

Children's implicit attitude acquisition: Evaluative statements succeed, repeated pairings fail.

From the earliest ages tested, children and adults show similar overall magnitudes of implicit attitudes toward various social groups. However, such consistency in attitude magnitude may obscure meaningful age-related change in the ways that children (vs. adults) acquire implicit attitudes. This experiment investigated children's implicit attitude acquisition by comparing the separate and joint effects of two learning interventions, previously shown to form implicit attitudes in adults. Children (N = 280, ages 7-11 years) were taught about novel social groups through either evaluative statements (ES; auditorily presented verbal statements such as 'Longfaces are bad, Squarefaces are good'), repeated evaluative pairings (REP; visual pairings of Longface/Squareface group members with valenced images such as a puppy or snake), or a combination of ES+REP. Results showed that children acquired implicit attitudes following ES and ES+REP, with REP providing no additional learning beyond ES alone. Moreover, children did not acquire implicit attitudes in four variations of REP, each designed to facilitate learning by systematically increasing verbal scaffolding to specify (a) the learning goal, (b) the valence of the unconditioned stimuli, and (c) the group categories of the conditioned stimuli. These findings underscore the early-emerging role of verbal statements in children's implicit attitude acquisition, as well as a possible age-related limit in children's acquisition of novel implicit attitudes from repeated pairings.

Pupillometric decoding of high-level musical imagery.

Humans report imagining sound where no physical sound is present: we replay conversations, practice speeches, and "hear" music all within the confines of our minds. Research has identified neural substrates underlying auditory imagery; yet deciphering its explicit contents has been elusive. Here we present a novel pupillometric method for decoding what individuals hear "inside their heads". Independent of light, pupils dilate and constrict in response to noradrenergic activity. Hence, stimuli evoking unique and reliable patterns of attention and arousal even when imagined should concurrently produce identifiable patterns of pupil-size dynamics (PSDs). Participants listened to and then silently imagined music while eye-tracked. Using machine learning algorithms, we decoded the imagined songs within- and across-participants following classifier-training on PSDs collected during both imagination and perception. Echoing findings in vision, cross-domain decoding accuracy increased with imagery strength. These data suggest that light-independent PSDs are a neural signature sensitive enough to decode imagination.

Inferring an unobservable population size from observable samples.

Success in the physical and social worlds often requires knowledge of population size. However, many populations cannot be observed in their entirety, making direct assessment of their size difficult, if not impossible. Nevertheless, an unobservable population size can be inferred from observable samples. We measured people's ability to make such inferences and their confidence in these inferences. Contrary to past work suggesting insensitivity to sample size and failures in statistical reasoning, inferences of populations size were accurate-but only when observable samples indicated a large underlying population. When observable samples indicated a small underlying population, inferences were systematically biased. This error, which cannot be attributed to a heuristics account, was compounded by a metacognitive failure: Confidence was highest when accuracy was at its worst. This dissociation between accuracy and confidence was confirmed by a manipulation that shifted the magnitude and variability of people's inferences without impacting their confidence. Together, these results (a) highlight the mental acuity and limits of a fundamental human judgment and (b) demonstrate an inverse relationship between cognition and metacognition.

Gender in Science, Technology, Engineering, and Mathematics: Issues, Causes, Solutions.

The landscape of gender in education and the workforce has shifted over the past decades: women have made gains in representation, equitable pay, and recognition through awards, grants, and publications. Despite overall change, differences persist in the fields of science, technology, engineering, and mathematics (STEM). This Viewpoints article on gender disparities in STEM offers an overarching perspective by addressing what the issues are, why the issues may emerge, and how the issues may be solved. In Part 1, recent data on gaps in representation, compensation, and recognition (awards, grants, publications) are reviewed, highlighting differences across subfields (e.g., computer science vs biology) and across career trajectories (e.g., bachelor's degrees vs senior faculty). In Part 2, evidence on leading explanations for these gaps, including explanations centered on abilities, preferences, and explicit and implicit bias, is presented. Particular attention is paid to implicit bias: mental processes that exist largely outside of conscious awareness and control in both male and female perceivers and female targets themselves. Given its prevalence and persistence, implicit bias warrants a central focus for research and application. Finally, in Part 3, the current knowledge is presented on interventions to change individuals' beliefs and behaviors, as well as organizational culture and practices. The moral issues surrounding equal access aside, understanding and addressing the complex issues surrounding gender in STEM are important because of the possible benefits to STEM and society that will be realized only when full participation of all capable and qualified individuals is guaranteed.

Children use targets' facial appearance to guide and predict social behavior.

Humans possess a tendency to rapidly and consistently make character evaluations from mere facial appearance. Recent work shows that this tendency emerges surprisingly early: children as young as 3-years-old provide adult-like assessments of others on character attributes such as "nice," "strong," and "smart" based only on subtle variations in targets' face shape and physiognomy (i.e., latent face-traits). The present research examined the behavioral consequences of children's face-trait judgments by asking whether, and if so when in development, the appearance of face-traits also (a) shapes children's judgments of targets' behaviors and (b) guides children's behavior toward targets. Experiments 1 and 2 showed that, by 3 years of age, children used facial features in character evaluations but not in judgments of targets' behavior, whereas by 5 years of age, children reliably made both character and behavior judgments from face-traits. Age-related change in behavior judgments was also observed in children's own behaviors toward targets: Experiments 3 and 4 showed that, by age 5 (but not earlier), children were more likely to give gifts to targets with trustworthy and submissive-looking faces (Experiment 3) and showed concordance between their character evaluations and gift-giving behaviors (Experiment 4). These findings newly suggest that, although children may rapidly make character evaluations from face-trait appearance, predicting and performing social behaviors based on face-traits may require more developed and specific understanding of traits and their relationships to behaviors. Nevertheless, by kindergarten, even relatively arbitrary and subtle face-traits appear to have meaningful consequences in shaping children's social judgments and interactions. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

Attitude change via repeated evaluative pairings versus evaluative statements: Shared and unique features.

When tested immediately, evaluative statements (ES; verbal information about upcoming categories and their positive/negative attributes) surprisingly shift implicit (IAT) attitudes more effectively than repeated evaluative pairings (REP; actual pairing of category members with positive/negative attributes). The present project (total N = 5,317) explored the shared and unique features of these two attitude change modalities by probing (a) commonalities visible in the extent to which propositional inferences created by ES infiltrate REP learning and (b) differences visible in performance of ES and REP learning over time. In REP, the number of stimulus pairings (varied parametrically from 4 to 24) produced no effect (Study 1), but verbally describing stimulus pairings as diagnostic versus nondiagnostic did modulate learning (Study 2), suggesting that even REP give rise to some form of propositional representation. On the other hand, learning from ES decayed quickly, whereas learning from REP remained stable over time both within an immediate session of testing (Study 3) and following a 15-min delay (Study 4), revealing a difference between these two forms of learning. Beyond their theoretical import, these findings may inform interventions designed to produce short- and long-term change in implicit attitudes. (PsycINFO Database Record (c) 2019 APA, all rights reserved).