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

Significance

All humans belong to the species Homo sapiens. Yet, throughout history, humans have breathed life into the Orwellian adage that “All [humans] are equal, but some [humans] are more equal than others.” Here, participants staunchly rejected this adage, with the overwhelming majority of over 61,000 participants reporting that all humans are equally human. However, across 13 experiments, US White participants (and White participants abroad) showed robust evidence of an implicit Human=Own Group association. Conversely, Black, Latinx, and Asian participants in the United States did not demonstrate this bias. These results highlight the tendency among socially dominant groups to reserve the quality Human for their own kind, producing, even in the 21st century, the age-old error of pseudospeciation.

Abstract

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.

By all relevant biological criteria (e.g., ability to interbreed and produce viable offspring; refs. 1 and 2), all humans—regardless of group membership—belong to a single species. And yet, history provides ample evidence of cultural pseudospeciation (3)—the tendency to treat culturally manufactured social groups (e.g., based on religion, language, ethnicity) akin to distinct biological species. These perceptions are psychologically potent and can drive intergroup differentiation. For example, Leyens et al. (4, 5) and Kteily et al. (6) demonstrated that this differentiation shapes perceived capacity for secondary emotions (e.g., pride and guilt) and judgments of a group’s evolutionary progress, respectively. Haslam (7) further suggested that groups can be differentiated based on their possession of traits central to “human” nature (e.g., imaginative and ambitious) or traits that distinguish humans from other animals (e.g., analytical and artistic).

As these examples illustrate, the field has not coalesced around a single definition of what it means to be more or less human (for a review, see ref. 8). However, data from these and related measures (e.g., mind attribution; ref. 9) converge to show that individuals tend to view three groups as more human: a) the self (e.g., refs. 10 and 11); b) ingroup members (e.g., refs. 12 and 13); and c) high-status groups (e.g., high-SES; ref. 14).

Related streams of research using more indirect methods (e.g., reverse correlation and fMRI; refs. 15–18) have similarly documented differences in affording the quality human to all human groups (for a review, see ref. 8). For instance, using a Go/No-Go association task (19), Saminaden et al. (20) observed a tendency to associate “Modern people” more with Human and “Traditional people” more with Animal; and using an Implicit Association Test (IAT; ref. 21), Bernard et al. (22) observed an implicit thin–human/obese–animal stereotype. The existence of such associations may be consequential, as Goff and colleagues (23) demonstrated; participants with stronger implicit associations between Black Americans (versus White Americans) with nonhuman primates also more strongly endorsed violence against Black suspects.

Taken together, research has demonstrated that the perceived humanity of social groups is not in line with biological or moral imperatives. Indeed, unlike judgments of preference, which are inherently subjective and equivocal, variability in the ascription of “human” to different social groups is a violation of a biological fact, making the existence of such stereotypes additionally noteworthy.

Overview of the Present Report

Building upon this work, we conducted thirteen experiments (including seven supplemental experiments; SI Appendix, Appendix 2) to investigate the presence and strength of human–animal implicit stereotypes toward prominent racial/ethnic groups in the United States. Specifically, harnessing data from 61,377 participants (67% American residents), we aimed to a) establish whether racial/ethnic human–animal stereotypes can be detected on measures of implicit cognition and whether they diverge or align with explicit cognition; b) determine whether these stereotypes, if they exist, are robust across various contrasts (e.g., White–Black, White–Latinx, and White–Asian) and demographic subgroups (e.g., White and non-White groups); and c) explore the mechanisms underlying these stereotypes to clarify the meaning of the obtained results (e.g., clarifying the role of semantics, beyond evaluations).

Cultural pseudospeciation could theoretically occur across any socially manufactured category. We focused on race/ethnicity because individuals of European descent have been historically depicted as being more human (24–26). Further, race/ethnicity in the US context is sui generis given its multiethnic makeup, its relatively recent history of slavery, and explicit ascriptions of Black Americans as nonhuman (e.g., refs. 26–29). As such, participants’ stereotypes may refract these false beliefs or align with the fact that all racial/ethnic groups are equally human.

IATs were used to capture the strength of associations between pairs of racial/ethnic groups (e.g., White–Black) and the attributes Human and Animal. In most studies, a contrastive attribute of Animal was selected because a predominant way in which humans have opted to express the belief that outgroups are less human is to liken them to nonhuman categories of animal, and much of the experimental research to date has adopted it. Here, to ensure that the emergence of any group stereotype was not limited to specific representations of Animal, the attribute was represented within and across experiments by diverse representations of Animal (e.g., pets, farm animals, wild animals, and vermin) as well as generic terms like “beast” and “brute” (SI Appendix, Appendix 2). In yet other experiments, we compared Human to attributes besides Animal, including suprahuman (e.g., God, Gods) and other highly positive (e.g., Dessert, Flower) attributes.

Moreover, to determine the generality or specificity of these stereotypes, we examined whether human–animal stereotypes were confined to the White–Black contrast or reflected a more general tendency to associate certain groups (e.g., socially advantaged groups; ref. 14) with Human over Animal. Specifically, we measured human–animal stereotypes toward four prominent racial/ethnic groups that vary in their history and stereotype content within the US society: White, Black, Latinx, and Asian. In every test in the main text, White served as the comparison category. This provided the opportunity to explore whether implicit stereotypes emerge in all three contrasts (i.e., White–Black, White–Hispanic, and White–Asian) or only in particular contrasts (e.g., in the White–Black contrast but not White–Asian).

In addition to documenting the comparative strength of stereotypes across contrasts (e.g., White–Black versus White–Asian), the use of large samples allowed for subgroup analyses that are typically incomputable. Perhaps most importantly, we tested whether this Human=Own Group association was expressed by all participants or only by members of advantaged racial/ethnic groups (e.g., White Americans). Further, robust samples permitted tests of the role of other demographic characteristics, besides race/ethnicity, on human–animal stereotypes. For example, DeLuca-McLean and Castano (30) observed higher levels of infrahumanization (i.e., the denial of secondary emotions; ref. 5) among self-identified political conservatives, and Kteily et al. (6) observed a positive relationship between social dominance orientation (SDO) and the tendency to characterize non-White groups as less evolved. Such variation, if also observed on implicit measures, may reveal features that predict the emergence of implicit human–animal stereotypes. As such, potential variability across age, gender, political ideology, education, and religion was additionally assessed.

Finally, we explored two questions to probe the mechanisms underlying this effect. First, do implicit human–animal stereotypes measure anything beyond valence? That is, are “Human” and “Animal” merely proxies for “Good” and “Bad”, respectively, given the valence asymmetry of these terms? Or do any observed associations reflect genuine engagement with the semantic meaning of these terms? Directly testing these competing interpretations has important implications for the construct validity of the test. Second, which pole of the IAT drives human–animal stereotypes? The relative nature of the IAT makes it challenging to determine whether Human=White associations, Animal=non-White associations, or both drive these stereotypes. However, we created operationalizations designed to disentangle these possibilities. Ultimately, an individual’s racial or ethnic membership has no biologically defensible bearing on their degree of humanity. Accordingly, if an implicit stereotype tying Human more to one racial/ethnic group exists, then it is critical to document this deviation from accuracy, as denials of humanity have been associated with reduced altruism (31, 32) and empathy (33), increased aggression (34, 35), support for punitive policies (6), and greater acceptance of violence (23, 36).

Experiments 1-2

As an initial point of departure, experiments 1-2 examined whether White and Black groups were equally and automatically associated with the attributes Human and Animal. These two groups were selected because of their unique position in US history and contemporary society. Black Americans have been depicted as less than human in media (e.g., refs. 26–29), and many were explicitly designated as such by the 1787 Three-Fifths compromise. Importantly, extending work by Goff et al. (37), the present experiments tested the association of White and Black with the more basic attributes Human and Animal.

Although a plenary representation of Human can be created with clear exemplars (e.g., man, woman, human, and person), nonhuman animals vary greatly across dimensions (e.g., valence and traits) that may influence the default mental representation of the term Animal. To represent multiple meanings of the term and therefore allow maximal generalization, four conceptualizations of Animal were created: pets, farm animals, wild animals, and vermin. On the one hand, associations between Animal and the social groups White and Black may vary according to the instantiation of Animal, suggesting that only specific representations of Animal discriminate between social groups. On the other hand, if all instantiations of the attribute Animal (e.g., pets and vermin) produce similar effects, then the conclusion that a core concept of Animal is involved can be more confidently drawn.

Moreover, given that the data were collected in the months following George Floyd’s murder, we administered an exploratory set of items about police violence and engagement in antiracism activities. These items are reported in SI Appendix, Appendix 3.

Results and Discussion

Explicit Attitudes.

On a 1 to 10 scale, participants reported feeling “somewhat” warm toward both White people (M = 6.95, SD = 2.10) and Black people (M = 7.56, SD = 1.95), with greater warmth reported toward Black than White people, t(846) = 8.93, P < 0.0001, Cohen’s d = 0.31, 95% CI [0.24, 0.38]. This was true among both White participants (t(591) = 4.90, P < 0.0001, Cohen’s d = 0.20, 95% CI [0.12, 0.28]) and non-White participants (t(240) = 8.16, P < 0.0001, Cohen’s d = 0.53, 95% CI [0.39, 0.66]).^*

Explicit Stereotypes.

Participants clearly affirmed the humanity of both groups, associating Black (M = 1.63, SD = 1.15) and White (M = 1.78; SD = 1.29) targets with Human more than Animal.^† In fact, less than 4% of participants reported that they considered either group to be more Animal than Human. Overall, the data suggest that, at least explicitly, participants affirmed the humanity of both groups.

Implicit Stereotypes.

The critical result from experiments 1–2 is that implicit associations diverged both from biological fact and from participants’ self-reported beliefs. On all four IATs, Human was more associated with White, and Animal was more associated with Black (all IAT Ds > 0.17).

Despite considerable heterogeneity that was intentionally introduced to create diverse representations of the Animal attribute (e.g., varying in valence and trait attributions; ref. 38), one-sample t-tests indicated that this White–Human/Black–Animal (hereafter “Human=White”) association differed significantly from zero in all four animal conditions (all ps < 0.0001, all Cohen’s d > 0.40). The strength of association was statistically indistinguishable when the attribute Animal was represented by highly negative “vermin” or by more positive “pets,” t(22360) = 0.02, P = 0.988, Cohen’s d = 0.00, 95% CI [−0.04, 0.06]. In addition, a supplemental experiment (N = 2,912) demonstrated that this effect persisted when exemplars such as beast and brute served as the attribute stimuli (SI Appendix, Appendix 2, experiment S1). This evidence of a generalized association between Black with Animal (and White with Human) is not in conflict with Goff et al.’s (37) finding that Black Americans are relatively more associated with Apes and White Americans are relatively more associated with Big Cats. The same group can simultaneously be more associated with the high-level representation of Animal, relative to Human, and more associated with a specific class of animals (e.g., nonhuman primates), relative to another (e.g., big cats).

Effect of participant racial group on implicit stereotypes.

Participants’ self-reported race/ethnicity reliably predicted implicit human–animal stereotypes (see Table 1 for condition-specific means across experiments). As visualized in Fig. 1, White participants displayed an implicit Human=White association in all four animal conditions, IAT D_avg = 0.22, t(7451) = 45.42, P < 0.0001, Cohen’s d = 0.53, 95% CI [0.50, 0.55]. By contrast, Black participants did not display a symmetric Human=Black effect. Instead, and aligning with their self-reported beliefs, Black participants more equally associated Human and Animal with their own group and White, IAT D_avg = –0.06, t(590) = –3.59, P < 0.0001, Cohen’s d = –0.15, 95% CI [–0.23, –0.07].

Table 1.

Overview of IAT Conditions and Effects

Experiment [condition]	Category labels (Category stimuli)	Attribute labels (Attribute stimuli)	Overall IAT D-score	IAT D-score across demographics
Experiments 1–2 [Pets]	Black (faces: set 1) White (faces: set 1)	Human (human, person, man, woman) Animal (dogcat, hamster, gerbil)	0.17 95% CI [0.157, 0.189]	White: 0.21 (n = 1793) Third-Party: 0.15 (n = 776) Black: −0.07(n = 155)
Experiments 1–2 [Farm]	Black (see above) White (see above)	Human (see above) Animal (pigcow, horse, goat)	0.19 [95% CI: 0.170, 0.201]	White: 0.22 (n = 1938) Third-Party: 0.16 (n = 764) Black: −0.07 (n = 145)
Experiments 1–2 [Vermin]	Black (see above) White (see above)	Human (see above) Animal (snakerat, cockroach, lizard)	0.18 [95% CI: 0.159, 0.191]	White: 0.19 (n = 1887) Third-Party: 0.19 (n = 702) Black: −0.08 (n = 140)
Experiments 1–2 [Wild]	Black (see above) White (see above)	Human (see above) Animal (tigerhippo, rhino, bear)	0.24 [95% CI: 0.220, 0.251]	White: 0.27 (n = 1834) Third-Party: 0.21 (n = 780) Black: −0.03 (n = 151)
Experiment 3 [Black/White]	Black (faces: set 1) White (faces: set 1)	Human (human, person, man, woman) Animal (bear, hippo, cow, horse)	0.23 [95% CI: 0.218, 0.236]	White: 0.27 (n = 4114) Third-Party: 0.21 (n = 2564) Black: −0.02 (n = 500)
Experiment 3 [Hispanic/White]	Hispanic (names: set1) White (names: set 1)	Human (see above) Animal (see above)	0.27 [95% CI: 0.264, 0.282]	White: 0.33 (n = 3888) Third-Party: 0.26 (n = 1904) Latinx: 0.07 (n = 849)
Experiment 3 [Asian/White]	Asian (faces: set 1) White (faces: set 1)	Human (see above) Animal (see above)	0.26 [95% CI: 0.254, 0.273]	White: 0.32 (n = 3721) Third-Party: 0.21 (n = 2564) Asian: 0.04 (n = 326)
Experiment 4 [Black/White]	Black (faces: set 2) White (faces: set 2)	Human (see above) Animal (see above)	0.19 [95% CI: 0.160, 0.212]	White: 0.22 (n = 628) Third-Party: 0.17 (n = 218) Black: 0.01 (n = 86)
Experiment 4 [Hispanic/White]	Hispanic (names: set1) White (names: set 1)	Human (see above) Animal (see above)	0.27 [95% CI: 0.241, 0.293]	White: 0.32 (n = 519) Third-Party: 0.22 (n = 167) Latinx: 0.05 (n = 79)
Experiment 4 [Asian/White]	Asian (faces: set 2) White (faces: set 2)	Human (see above) Animal (see above)	0.26 [95% CI: 0.236, 0.284]	White: 0.28 (n = 654) Third-Party: 0.22 (n = 268) Asian: insufficient n (n = 17)
Experiment 5a [Human–God]	Black (faces: set 2) White (faces: set 2)	Human (human, people, person, man, woman) God (god, almighty, divine, deity, creator)	0.39 [95% CI: 0.343, 0.439]	White: 0.39 (n = 204)
Experiment 5a [Human–Gods]	Black (see above) White (see above)	Human (see above) Gods (gods, zeus, apollo, aphrodite, venus)	0.49 [95% CI: 0.439, 0.539]	White: 0.49 (n = 190)
Experiment 5a [Human–Flower]	Black (see above) White (see above)	Human (see above) Flower (flower, rose, daffodil, tulip, daisy)	0.13 [95% CI: 0.075, 0.190]	White: 0.13 (n = 203)
Experiment 5a [Human–Dessert]	Black (see above) White (see above)	Human (see above) Dessert (dessert, pie, cookie, cake, macaroon)	0.19 [95% CI: 0.140, 0.246]	White: 0.19 (n = 211)
Experiment 5a [Human–Clothing]	Black (see above) White (see above)	Human (see above) Clothing (clothing, shoe, hat, sweater, jacket)	0.12 [95% CI: 0.057, 0.179]	White: 0.12 (n = 196)
Experiment 5a [Human–Furniture]	Black (see above) White (see above)	Human (see above) Furniture (furniture, chair, table, sofa, desk)	0.17 [95% CI: 0.114, 0.226]	White: 0.17 (n = 190)
Experiment 5a [Human–Appliance]	Black (see above) White (see above)	Human (see above) Appliance (appliance, stove, fridge, toaster, freezer)	0.19 [95% CI: 0.134, 0.244]	White: 0.19 (n = 199)
Experiment 5a [Human–Robot]	Black (see above) White (see above)	Human (see above) Robot (robot, bot, machine, automaton, droid)	0.39 [95% CI: 0.336, 0.451]	White: 0.39 (n = 183)
Experiment 5b [Animal–Death]	Black (see above) White (see above)	Animal (animal, beast, brute, animalistic, creature) Death (dead, dying, casket, graveyard, tombstone)	−0.27 [95% CI: −0.318, −0.229]	White: -0.27 (n = 220)
Experiment 5b [Animal–Ghost]	Black (see above) White (see above)	Animal (see above) Ghost (ghost, ghoul, spirit, phantom, haunted)	−0.03 [95% CI: −0.090, 0.029]	White: -0.03 (n = 187)
Experiment 5b [Animal–Toxin]	Black (see above) White (see above)	Animal (see above) Toxin (toxin, chemical, pesticide, pollution, acid)	0.00 [95% CI: −0.049, 0.059]	White: 0.00 (n = 191)
Experiment 5b [Animal–Disaster]	Black (see above) White (see above)	Animal (see above) Disaster (disaster, flood, fire, hurricane, storm)	−0.04 [95% CI: −0.094, 0.014]	White: -0.04 (n = 207)
Experiment 5b [Animal–Robot]	Black (see above) White (see above)	Animal (see above) Robot (robot, bot, machine, automaton, droid)	−0.14 [95% CI: −0.196, −0.087]	White: -0.14 (n = 179)
Experiment 5b [Animal–Appliance]	Black (see above) White (see above)	Animal (see above) Appliance (appliance, stove, fridge, toaster, freezer)	0.09 [95% CI: 0.037, 0.136]	White: 0.09 (n = 211)
Experiment 5b [Animal–Furniture]	Black (see above) White (see above)	Animal (see above) Furniture (furniture, chair, table, sofa, desk)	0.10 [95% CI: 0.041, 0.149]	White: 0.10 (n = 192)
Experiment 5b [Animal–Clothing]	Black (see above) White (see above)	Animal (see above) Clothing (clothing, shoe, hat, sweater, jacket)	0.10 [95% CI: 0.053, 0.154]	White: 0.10 (n = 207)

Positive IAT D-scores indicate either implicit White–Human/Black–Animal (experiments 1–4), White–Human/Black–[OTHER] associations (experiment 5a), or Animal–Black/White–[OTHER] (experiment 5b) associations.

Fig. 1.

Implicit White-Black/Human-Animal Stereotypes Across Group Membership, by Experiment. Positive IAT D-scores indicate an implicit White-Human/Black-Animal association, and negative IAT D-scores indicate an implicit Black-Human/White-Animal association. Condition-specific estimates with fewer than 20 participants were not visualized. Error bars represent 95% confidence intervals around the mean estimate. Third-party participants are participants from racial/ethnic groups that are not invoked on the IAT. On the White-Black/Human-Animal IAT, third-party participants include participants who identify as neither White nor Black (e.g., Asian participants).

Interestingly, when the IAT captured stereotypes toward two outgroups (e.g., Asian participants taking a White–Black/Human–Animal IAT), participants (hereafter “third-party participants”) did not associate Human with both outgroups equally. Instead, third-party participants assigned the attribute Human more to the socially advantaged group (White) and Animal more to the socially disadvantaged group (Black) in every condition (Table 1). In fact, the Human=White association was only “slightly” stronger among White participants than among third-party participants, IAT D_avg = 0.17, t(3021) = 22.2, P < 0.0001, Cohen’s d = 0.41, 95% CI [0.37, 0.44].

Taken together, three results from experiments 1-2 are noteworthy. First, a sharp dissociation between explicit and implicit measures emerged. Despite overwhelming recognition of the humanity of Black and White social groups, White participants displayed a pervasive implicit Human=White association. Indeed, this factually inaccurate stereotype emerged when Animal was represented in multiple ways, from amiable pets to vermin. Second, only White participants displayed the Human=Own Group effect. Black participants’ associations more closely aligned with the biological and moral imperative that all racial/ethnic groups are equally human. Third and finally, third-party participants (e.g., Asian and Latinx participants taking a White–Black/Human–Animal IAT) demonstrated greater automatic ascription of humanity to White than to Black. This finding demonstrates the pervasive power of group status in shaping the minds of members of more and less advantaged groups alike.

Experiment 3

Despite recognizing the humanity of both Black and White groups on self-report measures, White and third-party participants consistently displayed an implicit Human=White association (experiments 1–2). Are these results specific to the White–Black comparison? Or do they reflect a general tendency to ascribe Human more to White and Animal more to non-White groups? To test this, experiment 3 compared White not only to Black (providing a replication of experiments 1–2) but also to Latinx and Asian. These groups were selected because they are all prominent racial/ethnic groups in the United States but vary in their cultural histories and trait attributions. For instance, Black Americans are stereotyped as threatening, lazy, and poor (e.g., refs. 39 and 40). By contrast, Asian Americans are considered a model minority and are stereotyped as intelligent but shy (e.g., refs. 41 and 42). Thus, even if implicit human–animal stereotypes exist, they may emerge only in specific contrasts (e.g., White–Black). Further, to assess the extent to which this stereotype was widely shared, variability across six demographic characteristics—race/ethnicity, gender, political ideology, age, education, and religion—was newly examined.

Results and Discussion

Explicit Attitudes and Stereotypes.

Explicit attitudes.

Replicating experiment 1, participants reported feeling at least slightly warm (all means > 6) toward White, Black, Hispanic, and Asian groups. However, on the new measure of explicit preference, both White and non-White participants reported a relative preference for their own group, with group means ranging from 3.81 (White participants’ preference for White over Black) to 5.20 (Black participants’ preference for Black over White). Additional details are reported in SI Appendix, Appendix 3.

Explicit stereotypes.

Participants overwhelmingly affirmed the equal humanity of both groups. In fact, over 88% of non-White participants and over 96% of White participants endorsed the equal humanity of both groups.^‡ In the minority (<7%) of cases where equal humanity was not affirmed, participants asserted a social correction by rating Black, Asian, and Hispanic groups as more human than White groups.

Implicit Stereotypes.

The primary aim of experiment 3 was to explore whether the implicit Human=White associations observed in experiments 1-2 were a) specific to the White–Black contrast or b) reflect a more general tendency to associate Human more with White and Animal more with non-White groups.

Replicating the results of experiments 1-2, White (IAT D = 0.27, t(4114) = 43.4, P < 0.0001, Cohen’s d = 0.68, 95% CI [0.64, 0.71]) and third-party test takers (IAT D = 0.20, t(2563) = 26.0, P < 0.0001, Cohen’s d = 0.51, 95% CI [0.47, 0. 55]) displayed an implicit Human=White association in the White–Black IAT condition, while Black participants displayed relative accuracy (IAT D = −0.02, t(500) = -2.23, P = 0.253, Cohen’s d = −0.05, 95% CI [−0.14, 0.04]).

Newly, and providing evidence of a general tendency to associate Human more with White, White participants displayed an implicit Human=White association in both the White–Hispanic contrast (IAT D = 0.33, t(3888) = 55.4, P < 0.0001, Cohen’s d = 0.89, 95% CI [0.78, 1.03]) and the White–Asian contrast (IAT D = 0.32, t(3720) = 51.6, P < 0.0001, Cohen’s d = 0.85, 95% CI [0.81, 0.88]). Similarly, third-party participants displayed an implicit Human=White association in both the White–Hispanic contrast (IAT D = 0.26, t(1903) = 30.4, P < 0.0001, Cohen’s d = 0.70, 95% CI [0.65, 0.75]) and the White–Asian contrast (IAT D = 0.21, t(2563) = 27.8, P < 0.0001, Cohen’s d = 0.55, 95% CI [0.52, 0.59]), but to a lesser degree than White participants (all ps < 0.005).

Interestingly, unlike Black participants, when their own group served as the contrast to White on the IAT (e.g., Asian participants taking a White–Asian/Human–Animal IAT), Latinx and Asian participants did not assign Human to both groups equally. Instead, Latinx participants (IAT D = 0.07, t(848) = 5.23, P < 0.0001, Cohen’s d = 0.18, 95% CI [0.11, 0.25]) and Asian participants (IAT D = 0.04, t(325) = 2.02, P = 0.044, Cohen’s d = 0.11, 95% CI [0.003, 0.22]) displayed weak implicit Human=White associations. This pattern was replicated in the meta-analytic estimates (SI Appendix, Appendix 4). Overall, these results provide evidence for the robustness of the Human=White association across racial/ethnic comparisons and the disconnect between stated beliefs and implicit associations.

In SI Appendix, Appendixes 2 and 3, we probed the generality of human–animal stereotypes in two additional ways. First, we examined whether White participants’ Human=Own Group associations similarly emerged when Asian was represented as specific national/cultural (rather than racial/ethnic) entities. Providing evidence for the pervasiveness of human–animal stereotypes, White participants displayed implicit Human=Own Group associations when White was compared to groups identified as Chinese and Japanese (N_S5 = 10,045; Cohen’s d > 0.82; SI Appendix, Appendix 2). Similarly, White Americans implicitly associated Human more with (White) Americans than (White) Canadians (N_S6 = 875; Cohen’s d = 0.68; SI Appendix, Appendix 2); in fact, these Human=Own Group associations were just as strong as the effect in the White–Black contrast. These findings indicate that the results observed in experiments 1–3 were not specific to White versus non-White contrasts. Instead, they suggest a more general tendency for advantaged groups to associate their own group more with Human, regardless of whether the ingroup is demarcated by race, ethnicity, or nationality.

Second, we assessed whether this Human=White effect is unique to White Americans or emerges in White participants globally. Providing evidence for the generality of this stereotype across individuals who identify as White, non-US White participants consistently displayed Human=White associations, irrespective of country or geographic region of residence (SI Appendix, Appendix 3). Nevertheless, a related question remains: Is this Human=Own Group effect a signature of White participants, specifically, or of dominant groups in any society? We conducted a preliminary analysis of a dominant non-White group and find initial evidence that East Asians living in East Asia display an implicit Human=Own Group effect (In SI Appendix, Appendix 3). In other words, this effect may not be unique to White Americans or even White participants globally; instead, the effect may reflect the automatic association of Human to Own Group in any socially dominant group in a society, culture, or region. However, this test was an exploratory analysis that requires further confirmation in planned experiments.

Taken together, these data indicate that White participants display a pervasive, factually incorrect, implicit Human=Own Group stereotype. Moreover, they suggest that this stereotype a) extends beyond the White–Black comparison to other racial/ethnic and even national group comparisons and b) is not limited to Americans who identify as White. Indeed, this initial evidence suggests that a Human=White association emerges globally in those who identify as White.

Demographic moderators.

This consistency across target category contrasts is noteworthy, but is the effect similarly invariant across participant gender, age, level of education, political ideology, and religion? In other words, are these stereotypes observed across demographic subgroups or driven by particular demographic subgroups (e.g., older participants)?

Analyses revealed that education and religion did not moderate the strength of implicit Human=White associations in any of the three racial/ethnic contrasts (White–Black, White–Hispanic, White–Asian). Given the statistical power of these analyses, we can confidently assume that these null effects were not type II errors. Participant age was a significant predictor of stereotype strength only in certain contrasts.^§ Specifically, in the White–Hispanic and White–Asian contrasts (but not in the White–Black contrast), older participants (>40 y old at the time of testing) displayed stronger implicit Human=White associations than younger participants (<20 y old at the time of testing). By contrast, gender and political ideology emerged as significant moderators of the effect,^** with men and self-identified conservatives expressing slightly stronger implicit Human=White associations than women and self-identified liberals, respectively (Table 2).

Table 2.

Demographic Moderators of Implicit and Explicit Human–Animal Stereotypes, by IAT Target Category Comparison

		Explicit stereotypes			Implicit stereotypes
Moderating variable		White–Black	White–Hispanic	White–Asian	White–Black	White–Hispanic	White–Asian
Racial ingroup	White	4.00 (0.30) a	4.00 (0.30) a	3.99 (0.29) a	0.27 (0.40) a	0.33 (0.37) a	0.32 (0.40) a
	Third party	4.07 (0.56)b	4.09 (0.54)b	4.08 (0.56)b	0.20 (0.40)b	0.26 (0.37)b	0.21 (0.38)b
	Non-White Ingroup	4.30 (0.30)c	4.17 (0.68)c	4.11 (0.65)b	0.02 (0.40) c	0.07 (0.37)c	0.04 (0.40)c
Gender	Male	3.99 (0.47)a	4.00 (0.40)a	4.00 (0.45)a	0.28 (0.41)a	0.32 (0.38)a	0.30 (0.40)a
	Female	4.06 (0.45)b	4.06 (0.45)b	4.04 (0.41)b	0.20 (0.40)b	0.25 (0.38)b	0.25 (0.39)b
Political ideology	Liberal	4.07 (0.47)a	4.07 (0.45)a	4.05 (0.44)a	0.20 (0.41)a	0.26 (0.39)a	0.25 (0.39)a
	Neutral	4.04 (0.46)a	4.03 (0.45)a	4.01 (0.44)a	0.23 (0.41)a	0.27 (0.37)a	0.26 (0.39)a
	Conservative	3.95 (0.39)b	3.96 (0.35)b	3.97 (0.41)b	0.32 (0.40)b	0.35 (0.35)b	0.33 (0.37)b
Age	Over 40	4.01 (0.32)a	4.01 (0.31)a†	4.01 (0.32)a†	0.25 (0.44)	0.34 (0.39)a	0.36 (0.40)a
	20 and under	4.06 (0.54)b	4.07 (0.48)b†	4.04 (0.52)b†	0.22 (0.38)	0.23 (0.36)b	0.23 (0.36)b
Education	No College Degree	4.06 (0.52)a	4.06 (0.49)a†	4.03 (0.46)	0.23 (0.39)	0.25 (0.37)a†	0.24 (0.37)a†
	College Degree	4.02 (0.40)b	4.02 (0.37)b†	4.03 (0.39)	0.22 (0.42)	0.30 (0.38)b†	0.29 (0.41)b†
Religion	Christian	4.03 (0.43)a†	4.04 (0.45)	4.01 (0.38)	0.24 (0.41)	0.27 (0.37)	0.30 (0.38)a†
	Other-Religious	4.07 (0.50)b†	4.06 (0.46)	4.04 (0.47)	0.21 (0.41)	0.28 (0.37)	0.25 (0.40)b†
	Not Religious	4.02 (0.44)a†	4.04 (0.41)	4.05 (0.47)	0.22 (0.41)	0.28 (0.39)	0.25 (0.40)b†

Explicit stereotypes were measured on a scale from 1 (White much more human) to 7 ([Black][Hispanic][Asian] much more human), with the midpoint (4) indicating that both groups are “equally human.” Implicit stereotypes are reported in terms of IAT D-scores; positive IAT D-scores Human=White associations.

Different superscripts within each moderator–measure–IAT combination denote significant differences at P < 0.005. ^†The Omnibus test indicates no significant differences at an alpha of 0.05 in the multivariate model.

A similar pattern was observed for explicit stereotypes. Specifically, age, religion, and education did not moderate the strength of explicit stereotypes in the White–Hispanic or White–Asian contrasts. Age and education were related to explicit stereotypes only in the White–Black contrast, with younger and college-educated participants recording higher rates of explicit Human=Black stereotypes. As with implicit stereotypes, gender and political ideology emerged as significant moderators of the effect in every contrast (Table 2).

Experiment 4

Having demonstrated the robustness and generality of the Human=White effect, experiment 4 served to clarify the mechanisms underlying the effect. Specifically, experiment 4 examined whether implicit human–animal stereotypes capture anything beyond generalized valence or attitudes. That is, given the evaluative asymmetry of the attributes, with Human being more evaluatively positive in this context than Animal (N_S7 = 491; Cohen’s d = 1.02; SI Appendix, Appendix 2), an association between Human and White may largely reflect valence consistency rather than a semantic association.

Past work has attempted to eliminate a valence confound by selecting attribute stimuli that have been equated for valence (43) or by demonstrating that these associations are not moderated by participants’ implicit preferences (37). Here, to provide a direct test of the latter, we implemented a valence decomposition technique introduced by Kurdi et al. (44). This technique requires each participant to complete two measures: an implicit stereotype IAT (e.g., White–Black/Human–Animal IAT) and an implicit attitude IAT (e.g., White–Black/Good–Bad IAT). Crucially, the analysis quantifies the variance left unexplained by a) implicit attitudes and b) measurement error. If the variance left unexplained is statistically distinguishable from zero, then the implicit Human=White stereotype effect can be said to be conceptually unique from an implicit attitude.

Results and Discussion

Replicating the results of experiment 3, implicit Human=White associations were again observed in every race contrast (all IAT Ds > 0.18). Further, and in line with the well-established race attitude effect (e.g., refs. (45–47)), participants overall displayed an implicit preference for White, relative to Black (IAT D = 0.26), Hispanic (IAT D = 0.26), and Asian (IAT D = 0.20). Additionally, the correlations between implicit stereotypes (Human–Animal IATs) and implicit attitudes (Good–Bad IATs) were positive in all three racial/ethnic comparison: White–Black IAT (r = 0.43), White–Hispanic IAT (r = 0.31), and White–Asian IAT (r = 0.28). In fact, the correlations in the White–Black contrast approached the maximally expected correlations, given the IAT’s test–retest reliability of roughly r = 0.56 (48).

This initial result suggests that, as expected, implicit human–animal stereotypes detect an evaluative component, making a variance decomposition analysis necessary. In all cases, the 95% bootstrap confidence interval for residual true variance (RTV) did not include zero (see Table 3 for the full variance decomposition output). These data suggest that implicit Human=White associations are not fully reducible to valence alone. Instead, the Human–Animal IATs do detect a semantic association between Human with White (and Animal with non-White), independently of valence.

Table 3.

Variance Decomposition Results

IAT categories	Subjects	rmin	Attitude IAT mean D (SD)	Stereotype IAT mean D (SD)	Attitude–stereotype IAT r	Error variance	Variance accounted for by attitude	Residual true variance (RTV)
White–Black	All (N = 942)	0.09	0.26 (0.44)	0.19 (0.41)	0.43 [0.38, 0.48]	36.06 [30.50, 41.75]	41.25 [30.77, 53.86]	22.69 [7.61, 35.17]
	White (n = 628)	0.11	0.31 (0.43)	0.22 (0.41)	0.46 [0.40, 0.52]	35.43 [29.38, 42.04]	47.58 [34.15, 61.91]	16.99 [1.02, 32.45]
	Non-White (n = 304)	0.16	0.14 (0.44)	0.13 (0.41)	0.37 [0.27, 0.46]	36.93[28.38, 46.15]	30.38 [15.90, 47.22]	32.69 [13.06, 49.42]
White–Hispanic	All (N = 769)	0.10	0.26 (0.39)	0.27 (0.37)	0.31 [0.24, 0.37]	39.22[32.93, 46.79]	25.75 [15.61, 38.16]	35.03 [19.15, 47.79]
	White (n = 519)	0.12	0.31 (0.38)	0.31 (0.36)	0.27 [0.19, 0.35]	41.14 [33.48, 49.67]	20.45 [9.20, 34.38]	38.41 [20.38, 52.78]
	Non-White (n = 246)	0.18	0.14 (0.40)	0.16 (0.36)	0.30 [0.18, 0.41]	36.96 [30.48, 52.52]	25.60 [7.86, 47.02]	37.45 [11.60, 60.95]
White–Asian	All (N = 954)	0.09	0.20 (0.40)	0.26 (0.37)	0.28 [0.22, 0.34]	39.83 [34.71, 45.77]	21.40 [12.85, 31.41]	38.76 [26.44, 50.03]
	White (n = 654)	0.11	0.23 (0.40)	0.28 (0.37)	0.26 [0.18, 0.33]	41.62 [35.30, 49.21]	18.61 [9.45, 30.44]	39.78 [25.10, 51.72]
	Non-White (n = 285)	0.17	0.12 (0.39)	0.21 (0.39)	0.33 [0.22, 0.43]	37.05 [28.36, 47.58]	30.46 [14.25, 49.75]	32.49 [9.73, 50.67]

The “r_min” column reports the smallest correlation for which the sample size provides 80% power. Positive D scores indicate an implicit preference for White (over non-White), or a Human=White association for the attitude and stereotype IATs, respectively. The “Attitude–Stereotype IAT r” column reports the correlation between IATs. 95-percent confidence intervals are reported in square brackets. “RTV” represents the variance left unexplained by measurement error or implicit attitudes.

Experiments 5a-b

The IAT is an inherently relative measure. As such, the results from experiments 1–4 alone cannot determine which component(s)—the Human=White, the Animal=non-White, or both associations equally—is driving the effect. Experiments 5a-b were designed to disentangle these possibilities. Specifically, in experiment 5a, a series of eight IATs were created where Human was contrasted to non-Human attributes (e.g., Human–God, Human–Dessert, and Human–Robot) that varied in both semantic meaning and valence. Similarly, in experiment 5b, a series of eight IATs were created where Animal was contrasted to various non-Human attributes (e.g., Animal–Ghost, Animal–Toxin, Animal–Robot). If White–Human/Black–Animal associations were driven by a tendency to associate Human more with White (versus Animal more with Black), then the Human=White effect should emerge even when Human is contrasted to new attributes such as God, Robot, and Dessert. If, however, the Human=White effect emerged only because Animal was associated more with Black, then the Human=White effect should disappear or attenuate when the contrast to Animal is removed and replaced with these other nonhuman attributes.

Moreover, the heterogeneous valence of the non-Human attributes provided an additional test of the role of evaluations initiated in experiment 4. If Human=White associations or Animal=Black associations merely reflect valence consistency, then Human=White associations should attenuate when Human is contrasted with a positive attribute (e.g., Human–God), and Animal=Black associations should attenuate when Animal is contrasted with a negative attribute (e.g., Animal–Toxin).

Results and Discussion

Explicit Stereotypes.

As in previous experiments, participants overwhelmingly affirmed the humanity of White and Black groups. Across conditions, 92.8–99.2% reported that both groups are equally human. Other self-repot items are reported in SI Appendix, Appendix 3.

Implicit Stereotypes.

The IATs developed for experiments 5a–b served to clarify whether White–Human/Black–Animal associations can be isolated to one or both poles of the association. Providing evidence for a semantic association between Human with White, Human=White associations emerged in 8/8 IATs (all Cohen’s d > 0.27; see Table 1), with IAT D-scores ranging from 0.12 (Clothing) to 0.49 (Gods). In fact, in 6/8 IATs, the effect was just as strong (Dessert, Appliance, and Furniture conditions) or even stronger (Robot, Gods, and God conditions) than the White–Human/Black–Animal association observed in experiments 1–4.

Notably, this tendency among White participants to associate Human more with White than other groups remained even when the contrastive attribute to Human was selected to be extremely positive (e.g., God, Gods, and Dessert). In fact, the Human=White association was stronger when God and Gods (rather than Animal) served as the comparison attribute to Human. These data further affirm the variance decomposition analysis of experiment 4 and provide an independent, conceptual replication: Human=White associations reflect semantics (stereotypes), not merely valence (attitudes).

The relative nature of the IAT makes it difficult to disambiguate whether associations such as White–Human/Black–Animal were driven by the Human=White effect, Animal=Black effect, or both effects equally. However, experiment 5b provided evidence of a critical nature. An implicit Animal=Black association emerged weakly in only 3/8 conditions—when Animal was compared to Appliance, Furniture, and Clothing—and even in these conditions, the effect was as half as strong as the original effect (all IAT Ds < 0.11). These data indicate that Black may be associated more with the concept of Animal, but only in contrast to Human. In other words, Black Americans may be associated with specific animals due to historical narratives (e.g., ape; ref. 37), but the Human=White effect observed in experiments 1–4 more likely reflects a tendency to associate Human with White and non-Human—whether nonhuman is represented by Animal, Robot, or other attributes—with Black.

Taken together, these findings suggest that the Human versus Animal dichotomy embraced in many empirical reports (e.g., refs. (22) and 49) may be better conceptualized as primarily detecting the association of Human with advantaged social groups (e.g., White in the US context) rather than detecting the association of Animal with other groups (Black, Hispanic, and Asian). Put differently, the distinction between Human and non-Human appears to be fundamental, at least when Animal is represented by terms like beast and brute.^†† This distinction is important as it supports the interpretation that the term Human is imbued with a unique meaning that allows it to be transduced into an essential quality of advantaged groups, especially by members of the advantaged group itself. It also demonstrates the inability of members of advantaged groups to extend the quality Human to other groups.

General Discussion

By definition, all human groups are equally human. Indeed, modern genetics has supported this fundamental oneness of all humans by documenting the presence of more shared features across racial/ethnic groups than within groups (e.g., refs. 50–52). Indeed, although blatant denials of humanity still exist ((53, 54); for a review, see ref. 55), this basic fact was ratified by the vast majority of participants who reported that all racial/ethnic groups are equally human. Nevertheless, both fact and self-reported beliefs sit in contradiction to the evidence obtained from measures of implicit cognition across 13 experiments.

In the present work, 42 IATs embedded in six experiments and seven supplemental experiments with large samples (N = 61,377) were conducted in pursuit of three questions: a) Are US-based racial/ethnic groups equally associated with the concept of Human on a measure of implicit cognition? Do differences emerge even in the absence of explicit endorsement of this belief?; b) How pervasive is the tendency to associate some groups more with the concept of Human? Does it emerge broadly or only in specific demographic subgroups (e.g., older participants), and does it emerge invariably or only in contrast to particular non-Human attributes (e.g., Human vs. Animal)?; and c) Is the effect merely reflective of a valence match between the concepts of Human and White or can the effects be interpreted as revealing a semantic Human=Own Group association? Further, if a semantic association does exist, can the effect be isolated to the Human=White pole of the association, or does it also equally capture an Animal=Black effect?

Are Racial/Ethnic Groups Equally Associated With the Concept of Human?

On self-report measures, participants overwhelmingly affirmed the equal humanity of White and non-White groups. In fact, as many as 96.9% of White participants and 89.8% of non-White participants reported that White and Black, Asian, and Latinx groups were equally human (experiment 3). Despite this result, White participants consistently and inaccurately associated Human more strongly with White (relative to Black, Latinx, and Asian) on a measure of implicit cognition (experiments 1-5a). This result reflects a special category of bias. Most tests of implicit bias tap into dimensions that are untethered to facts (e.g., good–bad, warmth–competence) and therefore are inherently subjective. However, the present results, like work showing that European Americans are more associated with “American” (versus “Foreign”) than Asian Americans (e.g., refs. 56–58), are striking because they demonstrate a failure of human minds to hew with factual evidence and explicitly exposed beliefs.

In addition to the stark dissociation between implicit and explicit beliefs, the consistency of this effect was noteworthy. White participants displayed a Human=White association across six different conceptualizations of the attribute Animal, and in all comparisons to White: Black (experiments 1–5a; Supplemental experiments 1–2; 4), Latinx (experiments 3–4), Asian (experiments 3–4; Supplemental experiments 1–2; 4), East Asian (Supplemental experiment 3), Japanese (Supplemental experiment 5), and Chinese (Supplemental experiment 5). In fact, White Americans associated Human more with their own group even in comparison to Canadians, indicating that White participants’ tendency to ascribe Human more with their own group is not restricted to White versus non-White comparisons (Supplemental experiment 6). Indeed, this finding converges with other demonstrations of implicit human–animal stereotypes in nonracial comparisons (20, 22, 49, 59–61).

Crucially, however, this Human=White effect was not confined to White Americans. Supplemental analyses revealed that White participants, regardless of their country of citizenship or residence, consistently displayed implicit Human=White associations. Accordingly, it may be time to look outward and view such results as not limited to the United States or a specific country but as representing a pervasive effect of White solidarity and colonialism’s long shadow.

The role of historical and contemporary social hierarchies is also highlighted in the associations held by non-White groups. When testing their own group’s association to Human relative to White (e.g., Black participants taking a White–Black/Human–Animal IAT), non-White participants displayed no bias (Black participants; experiments 1–4) or a weak Human=White association (Latinx and Asian participants; experiment 3). In other words, rather than displaying an equal and opposite Human=Own Group association, non-White US participants showed a remarkable hewing to neutrality—a result that may be indicative of system-justifying tendencies (62).

Further demonstrating the power of sociopolitical status in driving automatic social perception, third-party participants (e.g., Black participants taking a White–Asian/Human–Animal IAT) displayed a Human=White association, although to a somewhat lesser degree than observed in White participants (experiments 1–3). This result is an intriguing indicator of the difficulty of obtaining neutrality in implicit social cognition; even though third-party participants should have no stake in the ascription of Human to an outgroup, their data reveal a Human=White effect.

How Pervasive Is this Tendency to Associate Some Groups More with the Concept of Human?

As previously discussed, the Human=White effect was observed in a large sample of non-US White participants, suggesting a broader rather than America-specific effect. But a related question can also be posed: Is the effect restricted to only White individuals or is it a feature of socially advantaged groups? The present data do not permit a strong test of whether there are non-White groups in other countries that display a similar Human=Own Group effect. However, an exploratory analysis of a small sample of East Asian participants living in East Asia (a socially dominant group in that region) provides preliminary evidence that this finding extends beyond the United States: East Asian participants living in East Asia displayed a slight Human=Asian association, whereas East Asian participants living in North America displayed an equal and opposite Human=White association (SI Appendix, Appendix 3). If this result persists in future experiments, it may be possible to conclude that Human=Own group associations may emerge in any dominant group, and that the Human=White effect observed in the United States reflects the behavior of any socially dominant group.

In addition to the emergence of stereotypes across cultures, we examined variability across sources of demography. In general, age, education, and religion did not moderate the strength of implicit or explicit Human=White associations. In other words, regardless of whether a participant was young or old, more or less educated, or religious or secular, they displayed the Human=White effect to a similar degree. However, two variables consistently predicted the strength of the Human=White effect in every target contrast (White–Black, White–Hispanic, White–Asian): gender and political ideology. Men showed a slightly stronger implicit Human=White effect than women, and participants who self-identified being political conservatives showed a slightly stronger implicit Human=White effect than participants who self-identified as liberals. Gender and political ideology also emerged as significant moderators of explicit stereotypes. Future work is well positioned to explore why political ideology moderated the strength of implicit human–animal stereotypes but not implicit dehumanization assessed via a conceptually related method (reverse correlation; ref. 16).

Beyond demographic variation, existing work has documented a relationship between SDO and denials of equal humanity. For example, Kteily and colleagues (6) showed that the subdimension of SDO related to more overt group-based dominance (SDO-Dominance) was more strongly correlated with blatant forms of dehumanization than the subdimension of SDO related to more subtle forms of hierarchy-enhancing ideologies (SDO-Egalitarianism). By contrast, the subdimensions were equally correlated with a subtle form of dehumanization: the denial of secondary emotions. As reported in SI Appendix, Appendix 3, structural equation models (SEMs) indicated that SDO-Dominance (but not SDO-Egalitarianism) was related to the strength of implicit stereotypes in the White–Black and White–Asian contrasts (other contrasts were not sufficiently powered). However, multigroup SEMs indicated that this pattern was true only among White participants; for non-White participants, neither subdimension related to the strength of implicit human–animal stereotypes.

Can the Human=Own Group Effect be Reduced to One of Valence?

In the study of implicit social cognition, it is often necessary to disambiguate attitude (valence) effects from belief (stereotype) effects. These experiments were focused on a specific stereotype involving the attributes Human and Animal/Other. To more confidently conclude that the obtained effect was not merely a reflection of the valence of the terms Human and Animal, we conducted three tests.

First, if human–animal stereotypes were primarily driven by valence rather than the specific semantic meaning of Human, then the Human=White effect should have been strongest when Animal was represented by negative animal types. However, this was not the case; the effect was just as strong when Animal was represented by pets or vermin (experiments 1–2). Second, a variance decomposition approach (44) indicated that Human=White associations could not be fully explained by evaluations; the residual true variance of the stereotype IAT remained significantly different from zero after accounting for measurement error and the residual variance accounted for by the attitude IAT (experiment 4). Third and finally, the Human=White effect persisted even when Human was compared to positive attributes (e.g., God, Dessert; experiment 5a). Ultimately, these data suggest that, while valence surely contributes to the effect (valence is, after all, a fundamental component of word meaning ref. 63), implicit human–animal stereotypes are not fully reducible to the good–bad dimension.

Can the Effect be Isolated to the Human=White Pole of the Association?

Experiments 5a–b also examined a methodological feature of the IAT that shapes the interpretation of the basic effect. As a relative measure, the IAT cannot discriminate whether White–Human/non-White-Animal associations emerge from a) Human=White associations; b) Animal=non-White associations; or c) both poles equally. In these experiments, designed to arbitrate between these interpretations, the data provided evidence for an independent implicit Human=White association but not for an independent Animal=Black association. When Human was contrasted with other non-Human attributes (e.g., Robot, God), Human=White associations persisted and were even stronger than the original effect in some cases (i.e., God, Gods, Robot). Conversely, Animal=Black associations largely disappeared when Animal was contrasted with non-Human attributes (e.g., Robot, Toxin). In other words, it is the distinction between Human and non-Human that appears to be fundamental; despite being factually incorrect, the Human=White effect is indeed a reflection of White participants’ representation of their own group as uniquely human.

Limitations and Future Work.

Despite its advances, two features limit the generalizability of the present results. First, all data were collected from Project Implicit. Although previous work has replicated the findings derived from Project Implicit in a nationally representative sample of the United States (64), the status of Project Implicit as a convenience sample poses an inherent limitation to generalizability. Second, the experiments highlighting the importance of Human versus nonhuman comparisons (rather than Human versus Animal comparisons) tested only White–Black comparisons. Future work is well-positioned to explore whether this result generalizes to other comparisons (e.g., White–Asian, Young–Old, and Fat–Thin) and to explore the variability observed in the strength of Human–White/[Other]–Black associations (experiment 5a). Finally, and perhaps most pressing, we hope that future work will examine the consequences of implicit human–animal stereotypes. Initial work suggests that explicit denials of humanity have broad consequences (e.g., reduced altruism (31) and increased aggression (34, 35)). However, whether implicit human–animal stereotypes predict similar or different behavioral outcomes is not well understood (see ref. 37; but also see ref. 55). Given the strength and consistency of the implicit Human=White association, whether these nonfactual stereotypes predict unequal treatment or discrimination is an important empirical question.

Conclusion

Throughout history, from Greek mythology to the language found in contemporary policing (65, 66), humans of European descent have considered themselves more civilized and therefore more human than other racial/ethnic groups. However, this false belief has been refuted by advances in genetics and challenged by moral values that have evolved as the temporal distance from colonialism has increased. Experimental psychology has demonstrated that attitudes and beliefs linger in a less conscious form, and the present work offers a comprehensive investigation of the automatic association of Human to primarily US-based racial/ethnic groups. The signature result from over 61,000 participants is that White and third-party test takers consistently associated Human more with White than other racial/ethnic groups. In fact, these associations emerged despite participants' explicit rejection of the sentiment captured by the Orwellian adage that some groups are more human than others. This mental failure to associate Human with all human groups equally was robust and emerged a) in thirteen experiments that provided exact and conceptual replications in a combined sample of over 61,000 participants; b) across multiple group comparisons (e.g., White–Black, White–Latinx, and White–Asian); and c) equally across variations in participant age, religion, and education but more strongly in self-identified conservatives and men. Additionally, multiple experiments served to clarify the mechanism and provided evidence that the effect was driven by the semantic relationship (rather than valence matching) between Human and White. In other words, the distinction between Human and non-Human, not Human and Animal, was paramount. Taken together, this work suggests that contemporary humans who belong to a socially dominant group are not freed from the belief that their group is more human than others.

Materials and Methods

Institutional Approval and Informed Consent.

All experiments, including those in SI Appendix, were granted ethical approval by the Committee on the Use of Human Subjects at Harvard University. Participants provided informed consent at the beginning of each experiment.

Experiments 1–2

Method.

Participants.

In experiment 1, N = 1,001 volunteer participants were recruited from Project Implicit’s research website (https://implicit.harvard.edu/implicit/research/). In experiment 2, a near replication of experiment 1 using a more robust sample, N = 10,969 volunteer participants were recruited from Project Implicit’s demonstration website (https://implicit.harvard.edu/implicit/takeatest.html). As the demonstration website is publicly available, sample sizes could not be specified a priori.

Following standard procedures for preparing and analyzing IAT data (67), participants who failed to complete the study (n_E1 = 33; n_E2 = 696) or recorded response latencies below 300 ms on more than 10% of all IAT trials (n_E1 = 2; n_E2 = 58), an indicator of insufficient attention (67), were excluded from all analyses. Exclusions for subsequent studies are provided on OSF: https://osf.io/wxhct?view_only=7e1ab8b096754cc08dd21a4bc838c801. The main dependent variable, IAT D-scores, did not vary across experiments. As such, the data are reported across both experiments, yielding a total sample of N = 11,181 (M_age = 36.57, 64% female, 67% White, 8% Latinx, 7% Asian, and 5% Black or African American, and 65% American residents).

Measures and Materials.

Open practices statement.

All materials and measures used in this project are reported in SI Appendix, Appendix 1. Statistical analyses were performed using the R statistical computing environment (68), and although the analyses were not preregistered, the data objects and R code required to reproduce all analyses are available from the Open Science Framework (OSF; https://osf.io/h7xst/?view_only=4d7e64fec9a2491dbb79057d65ff7ee5).

Implicit association tests.

In experiments 1-2, participants completed a standard seven-block IAT where they categorized photos of White and Black faces alongside words representing the attributes “Human” and “Animal” (Table 1). In both experiments, participants received one of four versions of the IAT, where Animal was represented by pets, farm animals, wild animals, or vermin stimuli. All other aspects of the IAT were identical across conditions.

Greenwald, Nosek, and Banaji’s (67) improved scoring algorithm was used to estimate the degree of association between the target categories (Black and White) and attributes (Human and Animal). Positive IAT D-scores indicate an implicit White–Human/Black–Animal association, whereas negative IAT D-scores indicate an implicit Black–Human/White–Animal association.

Demographic items.

Participant’s self-reported race, ethnicity, gender, age, religiosity, political identity, citizenship, occupation, and geographic measures (e.g., zip code of current residence) were collected.

Self-Report Items.

Explicit attitudes.

Two thermometers measured participants’ feelings toward White and Black people on a scale from 0 (“Extremely cold”) to 10 (“Extremely warm”).

Explicit stereotypes.

Two items probed participants’ recognition that White and Black human groups are indeed human (rather than animal). Specifically, these items asked participants to report how strongly they associated White and Black people with “human” versus “animal” on a scale from 1 (“I associate [White] [Black] people much more with human than with animal”) to 7 (“I associate [White] [Black] people much more with animal than with human”).

Police violence.

An exploratory set of items was administered to measure participants beliefs toward recent instances of police violence (e.g., the killing of George Floyd) and engagement in antiracism activities. These items are outside the purview of the current project but are reported in SI Appendix, Appendix 3.

Procedure.

In experiment 1, participants were randomly assigned to one of four White–Black/Human–Animal IATs. The only difference across IATs was whether Animal was represented by pets, farm animal, wild animal, or vermin. Then, participants completed the IAT and self-report items in a randomized order; demographic information was obtained by Project Implicit prior to participation. In experiment 2, participants completed one of the four IATs and demographic items in a randomized order.

Experiment 3

Method.

Participants.

In experiment 3, N = 29,285 volunteer participants were recruited via Project Implicit’s demonstration site. Exclusions produced a final sample of N = 20,587 participants (M_age = 30.85, 66% female, 57% White, 13% Latinx, 9% Asian, 7% Black or African American, and 67% American residents).

Measures and Materials.

Implicit association tests.

Participants received a seven-block IAT with one of three target comparisons: White–Black, White–Hispanic, or White–Asian. The stimuli used to represent Human and Animal were identical across IATs (Table 1).

Demographics.

The demographic questionnaire was identical to experiment 2.

Self-Report Items.

Explicit attitudes and stereotypes.

The items were identical to experiment 1 except for two new items. First, participants reported their relative preference on a scale from 1 (“I strongly prefer White people to [Black] [Hispanic] [Asian] people”) to 7 (“I strongly prefer [Black] [Hispanic] [Asian] people to White people”). Second, to ascertain whether participants recognized the equal humanity of White and non-White groups, participants were asked to report whether they considered White or [Black] [Hispanic] [Asian] people to be “more human” on a scale from 1 (“I consider White people to be much more human than [Black] [Hispanic] [Asian] people”) to 7 (“I consider [Black] [Hispanic] [Asian] people to be much more human than White people”).

Procedure.

Participants were randomly assigned to receive one of three IATs where the target categories varied: White–Black, White–Hispanic, or White–Asian. Participants completed the IAT, demographic questionnaire, and self-report items in a randomized order.

Experiment 4

Method.

Participants.

A total of 3,168 volunteer participants were recruited via Project Implicit’s research site. Exclusions produced a final sample of N = 2,665 participants (M_age = 37.84, 68% female, 68% White, 8% Black or African American, 8% Asian, and 8% Latinx).

Measures and Materials

Implicit association tests.

Stereotype IATs.

Human–animal stereotypes were assessed using the same IATs as experiment 3 except that a standardized set of stimuli was used to represent the targets in the White–Black and White–Asian conditions. These stimuli were validated in a supplemental experiment (N_S2 = 6,621; SI Appendix, Appendix 2).

Attitude IATs.

Implicit attitudes were measured using a standard seven-block IAT. “Good” and “Bad” served as the attribute labels (for stimuli, SI Appendix, Appendix 1), and the stimuli used to represent the target categories were identical to the stereotype IATs.

Demographics.

Demographic characteristics of each subject were obtained directly from Project Implicit.

Self-report items.

Explicit attitudes and explicit stereotypes were assessed using the measures described in experiment 3. As a purely exploratory test of construct validation, SDO was additionally assessed using the 8-item SDO_7(s) scale (69). Individuals higher on SDO more strongly favored statements such as “Some groups of people are simply inferior to other groups.” Full details are provided in SI Appendix, Appendix 1, and the results from SEMs and multigroup SEMs that relied on data from experiment 4 and supplemental experiments are reported in SI Appendix, Appendix 3.

Analytic Strategy.

The aim of experiment 4 was to determine whether implicit human–animal stereotypes are related to or fully redundant with implicit attitudes (valence). To do so, a bootstrapping approach introduced by Kurdi et al. (44) was employed to decompose implicit human–animal stereotypes into three components: a) error variance, as indexed by the mean split-half correlations of 1,000 bootstrapped samples; b) variance accounted for by attitudes, as indexed by the disattenuated correlations between stereotype and attitude IATs; and c) residual true variance (RTV), or the variance left unexplained by error or implicit attitudes. Crucially, if 95% bootstrap confidence intervals for RTV did not contain zero, then we inferred that implicit human–animal stereotypes are not fully redundant with implicit attitudes and that the semantic meaning that underlies “Human” and “Animal” contributes to the observed effects, beyond the effect of valence. Given the variability in implicit stereotypes across participant race, RTV was estimated for a) the overall sample; b) White participants; and c) non-White participants.

Procedure.

Participants were randomly assigned to one of three group IATs: White–Black, White–Hispanic, or White–Asian comparisons. Each participant completed two IATs—the stereotype IAT (e.g., White–Black/Human–Animal IAT) and attitude IAT (e.g., White–Black/Good–Bad IAT)—in a counterbalanced order. Additionally, participants completed self-report measures. The order of the self-report measures and IATs was counterbalanced.

Experiments 5a–b

Method.

Participants.

In experiment 5a (N = 1,648) and experiment 5b (N = 1,678), White volunteer participants were recruited via Project Implicit’s research site. After exclusions, N = 1,577 participants were retained in experiment 5a (M_age = 36.70, 69% female, 100% White, 59% American residents), and N = 1,594 participants were retained in experiment 5b (M_age = 37.00, 70% female, 100% White, 59% American residents).

Measures and Materials.

White–black/human–other IATs.

Participants in experiment 5a received a seven-block IAT in which the attribute “Human” was contrasted to one of eight non-Human attributes that varied in their valence: God, Gods, Flowers, Dessert, Clothing, Furniture, Appliance, or Robot. In every IAT, the target categories “White” and “Black” were represented by the standardized set of faces introduced in experiment 4. Positive IAT D-scores indicate White–Human/Black–[OTHER] associations (Human=White effect), whereas negative IAT D-scores indicate Black–Human/White–[OTHER] associations.

White–black/animal–other IATs.

Participants in experiment 5b also received a seven-block IAT in which the attribute “Animal” was contrasted to one of eight non-Human attributes that varied in their valence: Clothing, Furniture, Appliance, Robot, Disaster, Toxin, Ghost, or Death. As in experiment 5a, the target categories were represented by the standardized set of faces introduced in experiment 4. Positive IAT D-scores indicate Black–Animal/White-[OTHER] associations (Animal=Black effect), whereas negative IAT D-scores indicate White–Animal/Black–[OTHER] associations.

Demographics.

Demographic characteristics of each subject were obtained directly from Project Implicit.

Self-report items.

In addition to the items administered in experiments 3–4, participants were asked to report their relative feelings toward both target attributes (e.g., Human, Robot) and how strongly they associated White and Black people with each attribute.

Procedure.

In experiments 5a–b, participants were randomly assigned to one of eight IAT conditions where the comparative attribute to Human (experiment 5a) or Animal (experiment 5b) varied. Afterward, participants completed the IAT, followed by the self-report items.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Anonymized Numerical data have been deposited in Center for Open Science - OSF (https://osf.io/h7xst/?view_only=7e1ab8b096754cc08dd21a4bc838c801) (70). Following the guidelines proposed by Morehouse and Nosek (forthcoming), clean study data is available; researchers interested in obtaining the raw datasets (including redacted birthdate and zip code information) are directed to contact kirsten_morehouse@g.harvard.edu for access.