Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=vsoc20

The Journal of Social Psychology

ISSN: 0022-4545 (Print) 1940-1183 (Online) Journal homepage: http://www.tandfonline.com/loi/vsoc20

At face value: Psychological outcomes differ forreal vs. computer-generated multiracial faces

Sarah E Gaither, Jacqueline M Chen, Kristin Pauker & Samuel R Sommers

To cite this article: Sarah E Gaither, Jacqueline M Chen, Kristin Pauker & Samuel R Sommers(2018): At face value: Psychological outcomes differ for real vs. computer-generated multiracialfaces, The Journal of Social Psychology, DOI: 10.1080/00224545.2018.1538929

To link to this article: https://doi.org/10.1080/00224545.2018.1538929

Published online: 30 Oct 2018.

Submit your article to this journal

View Crossmark data

At face value: Psychological outcomes differ for real vs.computer-generated multiracial facesSarah E Gaithera, Jacqueline M Chenb,c, Kristin Paukerd, and Samuel R Sommerse

aPsychology and Neuroscience, Duke University, Durham, USA; bPsychology Department, University of Utah, SaltLake City, USA; cPsychology Department, University of California, Irvine, Honolulu, USA; dPsychology Department,University of Hawaii at Manoa, Honolulu, USA; ePsychology Department, Tufts University, Medford, USA

ABSTRACTMultiracial research emphasizes hypodescent categorizations and relies oncomputer-generated stimuli. Four experiments showed that real biracialfaces in a 2-Choice categorization task (White, Black) elicited hypodescentmore than computer-generated faces. Additionally, Experiment 2 showed a2-Choice categorization task with real biracial faces increased racial essenti-alism more than a 3-Choice categorization task. Experiment 3 showed thatmere exposure to real biracial faces did not increase essentialism. Finally,Experiments 4a and 4b replicated hypodescent outcomes when comparingreal biracial faces to computer-generated versions of those same faces. Insum, these findings initiate a discussion surrounding the methodology ofmultiracial categorizations.

ARTICLE HISTORYReceived 24 August 2017Accepted 3 October 2018

KEYWORDSComputer-generated faces;hypodescent; multiracialcategorization; racialessentialism; real biracialfaces

Since multiracial individuals are projected to be one of the fastest-growing populations in the UnitedStates across the next 40 years (Dunham & Olson, 2016; Kang & Bodenhausen, 2015; Richeson &Sommers, 2016;), it is important for research to understand the nuance surrounding how multiracialindividuals are categorized by society. It is true that not all multiracial people are racially ambiguous inappearance. However, it is clear that this emerging multiracial demographic has inspired a growing bodyof social psychological research on ambiguous categorization processes (e.g., Carpinella, Chen,Hamilton, & Johnson, 2015; Chen & Hamilton, 2012; Chen, Moons, Gaither, Hamilton, & Sherman,2014; Freeman, Pauker, & Sanchez, 2016; Gaither, 2015; Gaither, Pauker, Slepian, & Sommers, 2016;Halberstadt, Sherman, & Sherman, 2011; Ho, Kteily, & Chen, 2017; Ho, Sidanius, Levin, & Banaji, 2011;Ho, Sidanius, Cuddy, & Banaji, 2013; Krosch, Berntsen, Amodio, Jost, & Van Bavel, 2013; Peery &Bodenhausen, 2008). Moreover, other research has also focused on how exposure to this growing groupinfluences social attitudes, since multiracial individuals directly contradict U.S. society’s more fixednotions about racial group membership (e.g., Gaither, Babbitt, & Sommers, 2018a; Gaither, Toosi,Babbitt, & Sommers, 2018b; Sanchez, Good, & Chavez, 2011; Sanchez, Young, & Pauker, 2015;Pauker, Meyers, Sanchez, Gaither, & Young, 2018; Young, Sanchez, & Wilton, 2013).

However, the majority of multiracial categorization research to date has utilized computer-generated images of faces. Therefore, we do not know whether computer-generated depictions ofmultiracial individuals are ecologically valid in terms of reflecting how actual multiracial individualsare categorized. Here, we explicitly test the generalizability of these past findings by directlycomparing the use of computer-generated and real multiracial faces.

Social categorization research uses a variety of methods. Some studies utilize computer-generatedfaces and other studies use photographs of actual people. Some studies use forced-choice binary

CONTACT Sarah E Gaither sarah.gaither@duke.edu Duke University, Psychology and Neuroscience, 417 Chapel Drive, Box90086, Durham, NC 27708-0086, USAColor versions of one or more of the figures in the article can be found online at www.tandfonline.com/vsoc.© 2018 Taylor & Francis

THE JOURNAL OF SOCIAL PSYCHOLOGYhttps://doi.org/10.1080/00224545.2018.1538929

categorizations, and other studies provide more options. Although multiracial and racially ambiguousperception research has clearly used a variety of stimuli type and methods (e.g., Chen & Hamilton, 2012;Freeman, Pauker, Apfelbaum, & Ambady, 2010; Krosch et al., 2013; Pauker et al., 2009; Peery &Bodenhausen, 2008), much of this work has solely relied on computer-generated or computer-morphedstimuli. Both computer-generated and computer-morphed stimuli (faces made by morphing two photosof monoracial individuals together) attempt to represent multiracial stimuli as a 50:50 blend of the racialbackgrounds of interest. Yet real multiracial individuals may not always have features that reflect these50:50 blends, and no work has systematically compared these different types of stimuli in a single study.

We argue that computer-generated faces may be qualitatively different than actual biracial faces,since real biracial people are not perfectly controlled blends of their parents’ faces and also aresimply more realistic. Based on the Social Relevance Hypothesis, we argue that real biracial faces maybe processed differently than computer-generated faces, because the former are more realistic andshould be viewed as more socially relevant and more meaningful to the perceiver. Some work showsthat higher levels of social relevance cause that stimuli to be processed in a more purposeful andsocially meaningful manner (e.g., Bastian, Loughnan, & Koval, 2011; Mullen, Brown, & Smith, 1992;Oberman, Ramachandran, & Pineda, 2008) due to the realistic qualities of the stimuli that create apersonal connection (see also Bailenson, Yee, Merget, & Schroeder, 2006; Schroeder, 2002).

Hypodescent is one prevalent heuristic that some perceivers use during racially ambiguouscategorization, whereby ambiguously raced individuals are labeled more often as members of theirminority or socially subordinate group based on their physical appearance (Halberstadt et al., 2011;Ho et al., 2013, 2011; Krosch et al., 2013; Peery & Bodenhausen, 2008; Sanchez et al., 2011).Historically known as the one-drop rule, whereby one drop of “Black blood” identified a mixed-race individual as Black, hypodescent is consistent with either/or categorical approaches to race (e.g.,Allport, 1954; Bodenhausen & Macrae, 1998). Strictly speaking, hypodescent involves assigningindividuals to their socially subordinate group when you know their ancestry (Jordan, 2014), butindividuals could still categorize someone in a manner consistent with hypodescent by assuming atargets’ ancestry based on their outward appearance and then categorizing them according to theirsocially subordinate group. For clarity, we call this specific process of categorizing where only visualinformation is utilized—visual hypodescent. Social categorization research documenting visualhypodescent has relied primarily on computer-generated or computer-morphed faces (e.g.,Halberstadt et al., 2011; Ho et al., 2011; Krosch & Amodio, 2014; Krosch et al., 2013; Hugenberg& Bodenhausen, 2004; Peery & Bodenhausen, 2008). Therefore, we predicted that real raciallyambiguous biracial faces should also heighten social motivations and elicit higher levels of visualhypodescent compared to computer-generated faces.

To our knowledge, only one set of studies has directly compared categorizations of computer-generated and real biracial faces. Chen and Hamilton (2012; Experiment 2) found that hypodescentcategorizations were more readily elicited by real Black/White biracial faces than by morphedbiracial faces (created by morphing Black and White monoracial faces together). However, theseresearchers did not test whether they elicited different levels of visual hypodescent in a raciallydichotomous categorization task (i.e., one that only asks participants to categorize the faces as Blackor White), which is most often used in research. Moreover, these authors did not provide anexplanation for the differences in hypodescent rates by stimulus type.

Here, we aimed to answer two main questions: (1) is visual hypodescent applied differently toexperimenter controlled, computer-generated racially ambiguous biracial stimuli compared to actualracially ambiguous biracial individuals?; and (2) does visual hypodescent vary based on what racialcategories are explicitly cognitively accessible to the perceiver?

Experiment 1: 2- vs. 3-choice categorization outcomes

Experiment 1 tested whether real racially ambiguous biracial faces are categorized differently thancomputer-generated racially ambiguous biracial faces by comparing a 2-Choice versus a 3-Choice

2 S. E. GAITHER ET AL.

categorization task. We hypothesized that a 2-Choice categorization task, which asks participants todichotomously categorize racially ambiguous faces into discrete monoracial categories would lead tohigher levels of visual hypodescent (categorizing more faces as Black) for real biracial faces overcomputer-generated biracial faces because real biracial faces are likely to be seen as more sociallyrelevant (Bastian et al., 2011; Oberman et al., 2008).

Stimuli pretest

FaceGen is a frequently used program in face categorization studies that allows researchers to controlvarious aspects of faces when creating stimuli (i.e., skin color, facial features, affect). For example, tocreate a racially ambiguous face, the setting for a face would be at the midpoint between the twodesired racial categories (i.e., White and Black). To compare categorization outcomes between realbiracial faces and FaceGen computer-generated faces, we first conducted a pretest to ensure that thestimuli sets were equivalent in racial ambiguity and related characteristics.

Twenty FaceGen computer-generated racially ambiguous Black/White faces (10 female) withneutral expressions were used in this study (see Pauker et al., 2009; Gaither et al., 2014; Hinzman& Kelly, 2013; for other uses). All images were in color, cropped to an oval shape to display only theface (ears were partially visible, but no hair was visible) and were adjusted to uniform size andresolution (275 × 360 pixels; 3.8 × 5.0 inches; see Appendix A).

Thirty biracial Black/White individuals (15 female) were photographed in the lab on a white back-ground while making a neutral facial expression as part of a previous study. Twenty of these photos wereselected (13 female) based on perceived ambiguity. These images were in color and were cropped to thesame oval shape. All of these images were adjusted to uniform size and resolution (184 × 248 pixels;3.8 × 5.0 inches; see Appendix A).

Fifty participants (27 female; Mage = 24.46, SD = 11.06) were recruited using Amazon’sMechanical Turk using the keywords “short face perception study” and received 50 cents for theirwork (Buhrmester, Kwang, & Gosling, 2011). The sample included 33 White, 3 Asian, 1 Latino, 1Black, 2 Native American, 2 biracial, and 8 racially unspecified individuals. Participants first read thefollowing prompt: “You will see a series of pictures of different people. Some of these individualshave features that are more typical of African Americans, and some have features that are moretypical of Caucasian Americans in terms of skin color, hair, eyes, nose, cheeks, lips, etc. Please ratethe following photos to the best of your ability using your gut or instinctual responses.” Participantsthen were randomly assigned to rate 20 out of the 40 faces (a mix of the computer-generated and realbiracial faces) on a scale ranging from 1 (not at all) to 7 (very much) for each of the followingdimensions: prototypical White appearance, prototypical Black appearance, prototypical Whitefeatures, prototypical Black features, neutrality of facial expression, and attractiveness. Resultsshowed that the two sets of stimuli did not significantly differ on any of the ratings (all Fs < 2.91,all ps > .10; p = .10 was for attractiveness with computer-generated faces being rated as slightlyhigher). Additionally, there were no differences based on target gender (all Fs < 2.18, all ps > .15).While real biracial faces will vary considerably in the prototypicality and distinctiveness of theirfeatures in reality, the current set of faces were specifically selected to match the racial ambiguity ofthe computer-generated set to systematically test how computer-generated versus actual racialambiguity may influence race categorization.

Method

Participants

Experiment 1 was our initial investigation of this research question. One hundred and thirty-threeundergraduates (81 females; 4 gender unspecified; Mage = 19.29, SD = 1.21) participated in exchangefor partial course credit. This data was collected in 2012, and we followed the norm (at the time) of

THE JOURNAL OF SOCIAL PSYCHOLOGY 3

collecting 30 participants per between-subjects condition. Based on recruitment limitations using thedepartment participant pool, our goal was to recruit approximately 25–35 participants in each offour conditions during a one-semester recruitment period. The sample included 89 Whites, 10Asians, 4 Latinos, 2 Blacks, 23 biracials, and 5 racially unspecified.

Procedure

The categorization task was programmed using javascript in Qualtrics to allow each screen toautomatically advance to the next face after a participant entered a response. Participants firstread: “You will complete a short-face categorization task. Your job is to sort these faces intocategories by pressing the numeric key ON THE KEYBOARD (e.g., “1”, “2”, or “3”) correspondingto the appropriate category displayed on the screen as quickly as possible. The page will automati-cally advance after you type one of the numbers.”

Participants were then randomly assigned to view either the 20 pre-tested computer-generated facesor the 20 pre-tested real biracial faces in a randomized order. They were also randomly assigned to useeither a 2-Choice or a 3-Choice categorization task. Therefore, the design was a 2 (stimuli type:computer-generated, real biracial faces) × 2 (categorization task: 2-Choice, 3-Choice) between-subjectsdesign. The response options in the 2-Choice task were Black (1 key) or White (2 key), and the responseoptions in the 3-Choice task were Black (1 key), Biracial (2 key), and White (3 key).

Results

Hypodescent

Relative ratesWe computed each participant’s hypodescent tendencies by subtracting the proportion of Whitecategorizations from the proportion of Black categorizations. We hypothesized that participantswould engage in visual hypodescent significantly more when categorizing real biracial faces in the2-Choice task than when categorizing computer-generated faces and/or in the 3-Choice task. Weconducted a 2(Categorization Task) × 2(Stimuli Type) between-subjects ANOVA on hypodescent.There was a main effect of categorization task, F(1, 129) = 4.14, p = .04, ηp

2 = .03, indicating thathypodescent occurred more frequently in the 2-Choice task (M = 0.06, SD = .39) than in the3-Choice task (M = −0.06, SD = .28). There was also a main effect of stimuli type, F(1, 129) = 4.80, p = .03, ηp

2 = .04. Hypodescent was applied more to real biracial faces (M = 0.07,SD = .30) than to computer-generated faces (M = −0.06, SD = .37). The interaction was notsignificant, p = .31, ηp

2 = .01. However, we conducted follow-up pairwise comparisons given our apriori hypotheses that hypodescent would be most prevalent with biracial faces in the 2-Choice task.In the 2-Choice task, hypodescent was applied to real biracial faces (M = .15, SD = .31) more than tocomputer-generated faces (M = −.03, SD = .44), p = .02, ηp

2 = .04. There was no difference bystimulus type in the 3-choice task, p = .41, ηp

2 = .01. In addition, real faces elicited more hypodescentin the 2-Choice condition than in the 3-Choice condition (M = −.02, SD = .27), p = .03, ηp

2 = .04,whereas categorization task did not affect hypodescent levels for computer-generated faces, p = .48,ηp

2 = .004 (see Figure 1).1

Absolute ratesWe then sought to determine whether evidence for hypodescent (significantly more Black categor-izations than White categorizations) was obtained in each of the four conditions. A hypodescentscore that was positive and significantly different from zero provides evidence for hypodescent (i.e.,that participants are making consistently more Black than White categorizations, and their categor-izations are different from chance). A series of one sample t-tests showed that participants engagedin hypodescent in the 2-Choice/Real Face condition (M = .15, SD = 0.31), t(34) = 2.93, p = .01,

4 S. E. GAITHER ET AL.

Cohen’s d = 1.00, but not in the 2-Choice/Computer-generated condition (M = −.03, SD = 0.44),p = .67, or the 3-Choice/Real Face condition (M = −.02, SD = 0.27), p = .61. Participants in the3-Choice/Computer-generated condition (M = −.09, SD = 0.29) actually exhibited a marginal tendencyfor hyperdescent (more White categorizations than Black categorizations), t(31) = −1.77, p = .09,Cohen’s d = .64.

Discussion

These results provide initial evidence that hypodescent is applied more readily when categorizingreal biracial faces compared to computer-generated faces, especially when perceivers’ categorizationsare constrained to two choices. This finding suggests that experiments using computer-generatedfaces may actually underestimate the prevalence of hypodescent for real biracial individuals.

However, Experiment 1 also demonstrated that perceivers categorizing faces in the 3-Choicecondition did not engage in hypodescent regardless of face type, demonstrating that the addition of abiracial category reduced hypodescent for both types of faces. This finding clarifies the role ofcategorization task in contributing to the discrepant findings in the literature in which the same facessometimes compel hypodescent (Peery & Bodenhausen, 2008) and sometimes do not (Chen &Hamilton, 2012; Chen et al., 2014).

In sum, Experiment 1 provided preliminary evidence that perceivers respond to racially ambig-uous faces differently depending on whether they are real or computer-generated and how they areasked to categorize them. Experiment 2 sought to replicate and extend these findings by examiningperceivers’ beliefs about race.

Experiment 2: 2- vs. 3-choice task, categorization outcomes and essentialism

Some work regarding the perceptions of racially ambiguous individuals has focused on whetherincreased exposure to biracial individuals will change perceivers’ beliefs about race, specifically theirendorsement of racial essentialism (Sanchez et al., 2015; Young et al., 2013). Racial essentialism is a rigid(as opposed to malleable) view of race, in which racial groups are perceived as naturally occurringdiscrete categories (e.g., Leyens et al., 2001; Rothbart & Taylor, 1992). Biracial people represent theblurring of racial boundaries, which directly conflicts with the idea that racial categories must be

-0.2

-0.1

0

0.1

0.2

0.3

0.4

Facegen Real

Hyp

odes

cent

Type of Biracial Face

2 choice task

3 choice task

Figure 1. Hypodescent tendencies (Black categorizations – White categorizations) by type of biracial face and categorization taskin Experiment 1. Error bars represent standard error.

THE JOURNAL OF SOCIAL PSYCHOLOGY 5

discrete. Indeed, exposure to biracial individuals, especially those who identify as multiracial, has beenshown to decrease essentialism and colorblindness attitudes (Gaither, Babbitt, et al. 2018a; Gaither et al.,2018b; Sanchez et al., 2015; Young et al., 2013).

This relationship between essentialist beliefs and hypodescent appears to be bidirectional. Adultswho are high in racial essentialism believe that the boundaries between racial groups are in fact discrete(Chen & Hamilton, 2012; Haslam, Rothschild, & Ernst, 2000; Plaks, Malahy, Sedlins, & Shoda, 2012),and they tend to follow the rule of hypodescent when categorizing ambiguous others (Chao, Hong, &Chiu, 2013). Thus, essentialist beliefs influence hypodescent (e.g., Chao et al., 2013; Ho, Roberts, &Gelman, 2015). Additionally, being presented with information consistent with hypodescent increasesessentialist beliefs (Gaither et al., 2018a; Young et al., 2013). For example, when perceivers are exposedto real racially ambiguous faces that are labeled as monoracial (e.g., Black) compared to multiracial,participants actually show an increase in essentialism since the monoracial label activates a more fixed,categorical view of race (Young et al., 2013). Therefore, this suggests that if someone is asked tocategorize a face by using two monoracial categories (i.e., White, Black) versus three racial categories(one of which includes a “biracial” option), these two distinct categorization tasks may activate differinglevels of fixed views about race. In other words, a 2-Choice task could lead to higher levels ofessentialism, whereas a 3-Choice task could lead to lower levels of essentialism.

Thus, we expected that asking participants to categorize real biracial faces into monoracial categories,thereby reinforcing traditional racial boundaries, would increase perceivers’ endorsement of essentialistbeliefs, but not when the same task was completed with less socially meaningful, computer-generatedfaces. On the other hand, we predicted that a task that does not solely prime monoracial categories butalso reminds perceivers of a multiracial category option will not show these same outcomes.Consequently, we predicted that essentialist thinking would only increase when perceivers engaged inthe task that forced monoracial labels onto the more socially meaningful faces.

Method

Participants

Two hundred and forty-one undergraduates (145 females; 3 gender unspecified; Mage = 19.00,SD = 1.14) participated in exchange for partial course credit. To ensure generalizability of ourfindings from Study 1 with a larger sample size, our goal was to recruit approximately 50–65participants in each of four conditions during a one-semester recruitment period. The samplecontained 171 Whites, 35 Asians, 9 Latinos, 8 Blacks, 13 biracials, and 5 unspecified.

Procedure

The laboratory session procedure was identical to that of Experiment 1, except that we collectedparticipants’ endorsement of essentialist beliefs after the categorization task. Using the essentialismscale from No et al. (2008), participants rated eight statements about essentialist beliefs (e.g., “To alarge extent, a person’s race biologically determines his or her abilities or traits”; α = .63) using ascale of 1 = not at all to 7 = very much.

Results

Hypodescent

Relative ratesThe same analysis approach utilized in Experiment 1 was used here. We hypothesized that partici-pants would engage in more hypodescent in the 2-Choice/Real Face condition compared to the otherthree conditions. We conducted a 2(Categorization Task) × 2(Stimuli Type) between-subjectsANOVA on hypodescent. There was a main effect of categorization task, F(1, 237) = 9.54,

6 S. E. GAITHER ET AL.

p = .002, ηp2 = .04, and a main effect of stimuli type, F(1, 237) = 16.82, p < .001, ηp

2 = .07. Thesemain effects were qualified by a significant interaction, F(1, 237) = 5.89, p = .02, ηp

2 = .02. Follow-uppairwise comparisons revealed that real faces (M = .32, SD = .32) elicited more hypodescent thancomputer-generated faces (M = .05, SD = .42) in the 2-Choice condition, p < .001, ηp

2 = .08, butthere was no difference in hypodescent by stimuli type in the 3-Choice condition, p = .24.Furthermore, real faces elicited more hypodescent in the 2-Choice condition (M = .32, SD = .32)than in the 3-Choice condition (M = .09, SD = .21), p < .001, ηp

2 = .06, whereas categorization taskdid not affect hypodescent levels for computer-generated faces. It is important to note that thesepatterns replicated the follow-up comparison results in Experiment 1 (even though the two-wayinteraction was not statistically significant in that experiment) (see Figure 2).

Absolute ratesOne-tailed tests revealed participants engaged in hypodescent in the 2-Choice/Real Face condition(M = .32, SD = 0.32), t(61) = 7.86, p < .001, Cohen’s d = 2.01, and in the 3-Choice/Real Face condition(M = .09, SD = 0.21), t(56) = 3.32, p < .001, Cohen’s d = 0.89. Participants did not engage inhypodescent when categorizing computer-generated faces into two categories (M = .05, SD = 0.42), t(60) = 0.99, p = .33, Cohen’s d = .26, or three categories (M = .03, SD = 0.28), t(60) = 0.71, p = .48,Cohen’s d = .18.

Psychological essentialism

We hypothesized that categorizing the real biracial faces into two categories would lead to higherlevels of endorsing racial essentialism relative to the other three conditions. First, we conducted a 2(Categorization Task) × 2(Stimuli Type) between-subjects ANOVA on essentialism scores. Therewas a significant main effect of stimuli type, F(1, 229) = 5.21, p = .02, ηp

2 = .02. Participants whocategorized the real faces (M = 3.46, SD = .74) endorsed essentialism more strongly than those whocategorized computer-generated faces (M = 3.24, SD = .76). The categorization task by stimulus-typeinteraction was not significant, F(1, 229) = 2.36, p = .13, ηp

2 = .01. We conducted follow-up pairwise

-0.2

-0.1

0

0.1

0.2

0.3

0.4

Facegen Real

Hyp

odes

cent

Type of Biracial Face

2 choice task

3 choice task

Figure 2. Hypodescent tendencies (Black categorizations – White categorizations) by type of biracial face and categorization taskin Experiment 2. Error bars represent standard error.

THE JOURNAL OF SOCIAL PSYCHOLOGY 7

comparisons given our a priori hypotheses. They revealed that participants who categorized realbiracial faces using the 2-Choice task (M = 3.51, SD = .66) endorsed psychological essentialism morethan those who categorized computer-generated faces in the 2-Choice task (M = 3.14, SD = .68),p = .01, ηp

2 = .03. No other comparisons were significant, ps > .16, ηp2s < .008.

Discussion

In conjunction with Experiment 1, our findings indicate that stimulus type and categorization taskaffect how biracial faces are perceived and influence beliefs about race. Only when participantscategorized real biracial faces did we find visual hypodescent, and those who categorized real facesinto monoracial categories of Black or White exhibited the highest levels of essentialist beliefs. It isimportant to note that if we had only used computer-generated stimuli, we might not haveconcluded that categorizing ambiguous individuals as monoracial relative to multiracial increasesessentialism. These findings are consistent with our assertion that real biracial faces are more sociallyrelevant stimuli that elicit different psychological processes than computer-generated faces.

Experiment 2 is also consistent with past work showing that essentialist thinking is associatedwith social categorization outcomes (e.g., Chao et al., 2013; Haslam et al., 2000; Rothbart & Taylor,1992). Our findings suggest that in addition to essentialist thinking affecting social categorizationoutcomes, using traditional monoracial categories to categorize ambiguous stimuli appears toactivate more fixed thinking about race. However, an alternative explanation for our results is thatmere exposure to real racially ambiguous individuals increases essentialism and that the 2-Choicetask merely reflects how people naturally categorize those faces, whereas the 3-Choice task primespeople to think more about multiracialism and categorize more faces as multiracial than they wouldotherwise. To directly address this alternative explanation, we conducted Experiment 3 and mea-sured perceivers’ essentialist beliefs after participants either categorized real biracial faces intomonoracial categories or were simply exposed to the faces. Experiments 1 and 2 also both did notinclude reaction times during categorization—an outcome variable frequently utilized in facecategorization research to reflect ease or difficulty in categorization. Therefore, Experiment 3 alsomeasured categorization latencies during the 2-Choice categorization task.

Experiment 3: 2-choice vs. mere exposure task, categorization outcomes, andlatencies, and essentialism

Thus far, Experiments 1 and 2 have highlighted differences in outcomes for racially ambiguous faces,especially when perceivers used the 2-Choice task and saw real faces. Experiment 3 tested whetherthe essentialism findings were due to imposing monoracial categories on real biracial faces or due tomere exposure to these faces. Additionally, Experiment 3 provided another opportunity to replicatethe hypodescent effects while also examining if participants spend differing amounts of timecategorizing real versus computer-generated faces.

Here, we manipulated whether perceivers saw real or computer-generated faces and whether theycategorized those faces (into 2 categories) or viewed them without a categorization task. We predictedthat, because real biracial faces are more socially relevant stimuli, both a 2-Choice categorization taskand simply exposing participants to the same real biracial faces one-at-a-time (which still givesparticipants the opportunity to mentally categorize each face into pre-existing monoracial categories),would increase essentialist beliefs more than with computer-generated biracial faces. Moreover, sincethe 2-Choice categorization task primes categorical thinking, we hypothesized that this task would leadto higher levels of racial essentialist beliefs than mere exposure.

Lastly, since past work has shown that racially ambiguous faces take longer to categorize thanracially prototypical faces (e.g., Blascovich, Wyer, Swart, & Kibler, 1997; Castano, Yzerbyt,Bourguignon, & Seron, 2002; Chen & Hamilton, 2012), we also sought to determine whether facetype affected how long perceivers spend categorizing racially ambiguous others. We predicted that

8 S. E. GAITHER ET AL.

participants would spend more time categorizing computer-generated faces compared to real biracialfaces because participants’ increased use of hypodescent as a categorization shortcut with real faceswould speed up their categorization process.

Method

Participants

One hundred and forty-two undergraduates (102 females; Mage = 18.62, SD = 1.17) participated inthe experiment for partial course credit. Based on department recruitment limitations, our goalwas to recruit approximately 25–35 participants in each of four conditions during a one-semesterrecruitment period. The sample had 71 Whites, 42 Asians, 13 Multiracials, 9 Latinos, 4 Blacks, and3 unspecified.

Procedure

Participants were randomly assigned to a 2 (stimuli type: computer-generated, real) × 2 (task:2-Choice categorization, mere exposure) between-subjects experiment. The tasks also differed bywhether categorical thinking was primed or not when the 2-Choice categorization option was or wasnot present. Participants in the 2-Choice Categorization condition read the same prompt fromExperiments 1 and 2. Participants in the Mere Exposure condition read: “You will be a shown aseries of faces one by one. Each face will remain on the screen for three seconds and will advanceautomatically. Please pay attention to each face.” Faces were shown for three seconds (see Blascovichet al., 1997; Peery & Bodenhausen, 2008; Experiment 1).

The same stimuli, programming, and 8-item essentialism scale from Experiments 1 and 2 wereused (α = .78). A timer programmed through Qualtrics timed how long participants spent duringthe 2-Choice categorization task. All participants completed this experiment on the same labcomputers.

Results

Hypodescent

Relative ratesReplicating our previous findings, an independent samples t-test showed that participants in the RealBiracial Faces Condition engaged in more hypodescent than did participants in the Computer-generated Condition, t(68) = 2.13, p = .04, ηp

2 = .06.

Absolute ratesOne-tailed t-tests revealed that participants in the Real Biracial Faces Condition displayed a margin-ally significant tendency for hypodescent (M = .12, SD = 0.36), t(36) = 1.96, p = .06, Cohen’s d = .65.Participants in the Computer-generated Condition (M = −.08, SD = 0.42) did not engage inhypodescent, t(32) = −1.13, p = .27, Cohen’s d = −0.40.

Categorization latencies

In addition, we hypothesized that participants would take longer to categorize the computer-generated faces than real biracial faces. We computed participants’ average categorization responsetime. As expected, participants categorizing real biracial faces spent significantly less time categoriz-ing each face (M = 2.45 seconds, SD = .98) in comparison to categorizing computer-generated faces(M = 2.85 seconds, SD = 1.37; t(68) = −2.02, p = .047, ηp

2 = .03).

THE JOURNAL OF SOCIAL PSYCHOLOGY 9

Psychological essentialism

We conducted a 2(Task) × 2(Stimuli Type) between-subjects ANOVA on psychological essentialism.There was a main effect of stimuli type, F(1, 138) = 4.79, p = .03, ηp

2 = .03. As predicted, participantswho viewed real biracial faces (M = 3.76, SD = .85) endorsed essentialism more than participantswho viewed computer-generated faces (M = 3.42, SD = 1.05). This effect was qualified by asignificant stimuli type by task interaction, F(1, 138) = 4.51, p = .04, ηp

2 = .03. Follow-up pairwisecomparisons confirmed that participants endorsed essentialism more after categorizing real biracialfaces (M = 3.97, SD = .75) than after categorizing computer-generated faces (M = 3.28, SD = 1.16),p = .003, ηp

2 = .06. Additionally, categorizing real biracial faces led to marginally higher essentialismthan did simply viewing real biracial faces (M = 3.56, SD = .90), p = .07, ηp

2 = .02. There was nodifference in essentialism between participants who were merely exposed to real versus computer-generated faces, p = .96, ηp

2 = .01, nor was there a difference in essentialism between participantswho were merely exposed to versus categorized computer-generated faces, p = .25, ηp

2 = .01.

Discussion

Experiment 3 again demonstrated that, compared to computer-generated faces, real biracial faceswere categorized more often as Black than as White. Importantly, we also showed that computer-generated stimuli also took longer for participants to categorize. We believe these results provideevidence that participants may categorize real racially ambiguous biracial faces by applying thehypodescent heuristic to them more readily than computer-generated faces, resulting in a fastercategorizations. Thus, this research suggests that previous work documenting hypodescent withcomputer-generated faces may underestimate how often it is actually applied to real biracialindividuals.

Experiment 3 also found that exposure to real biracial faces actually increased perceivers’essentialist thinking when they categorized real biracial faces into fixed, monoracial categoriescompared to when they passively viewed these same faces, or compared to when they categorizedor viewed computer-generated faces.

One potential question that remains is the possibility that the real biracial face stimuli set is notmatched well enough with the computer-generated, FaceGen faces. Further, the previous studies reliedon a relatively small number of faces, and it is possible that the findings were influenced by specific,distinctive faces within a stimulus set. Therefore, Experiments 4a and 4b aimed to replicate the findingsof Experiments 1–3 by comparing real biracial faces with FaceGen versions of the same faces.

Experiments 4a & 4b: real biracial faces and their matched facegen faces

Although the faces utilized in Experiments 1–3 were equivalent on all dimensions measured by thepretest, we have not tested the possibility that the distinctiveness of the real biracial faces is reallywhat drives the differences we see between real biracial faces and computer-generated faces.Therefore, here we directly compare the same real biracial faces to FaceGen versions of thosesame faces. Study 4a compares outcomes for these faces in the 2-Choice and 3-choice categorizationtasks and Study 4b compares outcomes for these faces in the 2-Choice and mere exposure tasks. Thestimuli pretest for this set of real and computer-generated faces is presented first, followed by themethods, results, and a combined discussion section.

Stimuli pretest

All 20 real biracial faces were converted into FaceGen versions using the Photo Fit Procedure. Allimages were in color, cropped to an oval shape to display only the face, and were adjusted to uniformsize and resolution (275 × 360 pixels; 3.8 × 5.0 inches; see Appendix A).

10 S. E. GAITHER ET AL.

Forty participants (24 female; Mage = 34.98, SD = 34.18) were recruited for the pretest viaAmazon’s Mechanical Turk using the keywords “short face perception study,” and they received50 cents for their work. The sample included 29 White, 5 Black, 3 Asian, 2 biracial, and 1 Latino. Thesame prompt was used from the earlier pretest. Participants were randomly assigned to rate eitherthe 20 real biracial faces or the 20 FaceGen versions of the real biracial faces on a scale ranging from1 (not at all) to 7 (very much) for each of the following: prototypical White appearance, prototypicalBlack appearance, racial ambiguity, neutrality of facial expression, attractiveness, how real the faceappeared, and distinctiveness. Results showed that the two sets of facial stimuli did not significantlydiffer on most of the ratings (ts < 1.66, ps > .10; p = .11 was for ratings of attractiveness with the realbiracial faces being rated as slightly more attractive than the FaceGen versions). As would beexpected, however, the real biracial faces were rated as significantly more real in appearance(M = 5.45, SD = .35) compared to the FaceGen versions (M = 4.83, SD = .25), t(38) = 6.41,p < .001. Therefore, this new set of FaceGen faces created from photos from actual biracial peopleshow the same variability as actual biracial faces with regard to prototypically, racial ambiguity,emotion expression, and distinctiveness.

Experiment 4a: real biracial faces and their matched facegen faces, 2-choice vs.3-choice task, categorization outcomes and essentialism

Participants

Four hundred and eleven participants (201 females; 4 gender unspecified; Mage = 35.28, SD = 12.10)were recruited via Amazon’s Mechanical Turk using the keywords “short face perception study” andreceived 50 cents for their work. To increase generalizability of our findings from the previousstudies, we collected a larger online and non-undergraduate sample. Our goal was to recruitapproximately 80–100 participants in each of four conditions. The sample contained 309 Whites,28 Blacks, 25 Asians, 17 Latino, 19 biracials, and 2 unspecified.

Procedure

The study was identical to that of Experiment 2, except, here, participants were randomly assigned toeither the 20 pre-tested real biracial faces or the 20 pre-tested FaceGen versions of the real biracialfaces in a randomized order and either a 2-Choice or a 3-Choice categorization task. Therefore, thedesign was a 2 (stimuli type: real biracial faces, real biracial faces as FaceGen) × 2 (categorizationtask: 2-Choice, 3-Choice) between-subjects design. After the categorization task, participants com-pleted the same 8-item essentialism scale as other experiments (α = .60).

Results

Hypodescent

Relative ratesWe used the same analysis approach as in the previous experiments. We conducted a 2(Categorization Task) × 2(Stimuli Type) between-subjects ANOVA on hypodescent. There was amain effect of stimuli type, F(1, 407) = 17.72, p < .001, ηp

2 = .04, and a main effect of categorizationtask, F(1, 407) = 23.08, p < .001, ηp

2 = .05. Participants who categorized the Real Biracial Faces(M = .06, SD = .25) engaged in hypodescent more than those who categorized the Real FaceGenFaces (M = −.02, SD = .15). Participants in the 2-Choice conditions (M = .06, SD = .30) engagedin hypodescent more than participants in the 3-choice conditions (M = −.03, SD = .15). These maineffects were qualified by a significant interaction, F(1, 407) = 138.71, p < .001, ηp

2 = .25, and weconducted follow-up pairwise comparisons. Participants in the Real Biracial Faces condition engagedin hypodescent significantly more in the 2-Choice (M = .22, SD = .24) than the 3-choice condition

THE JOURNAL OF SOCIAL PSYCHOLOGY 11

(M = −.11, SD = .12), p < .001, ηp2 = .25. Participants in the Real FaceGen Faces condition engaged in

hypodescent more in the 3-choice condition (M = .04, SD = .15) than in the 2-Choice condition(M = −.10, SD = .27), p < .001, ηp

2 = .06. The 2-Choice condition elicited more hypodescent on realfaces compared to FaceGen faces, p < .001, ηp

2 = .23, whereas the 3-choice condition elicited morehypodescent on the FaceGen faces than on the real faces, p < .001, ηp

2 = .07.

Absolute ratesParticipants in the 2-Choice/Real Biracial Faces condition engaged in hypodescent (M = .22,SE = .02), t(96) = 9.15, p < .001, Cohen’s d = 1.87, as did participants in the 3-Choice/RealFacegen Faces condition (M = .04, SE = .01), t(111) = 3.02, p < .003, Cohen’s d = .57. Participantsin the 3-Choice/Real Biracial Faces condition (M = −.11, SE = .01), t(105) = −8.95, p < .001, Cohen’sd = 0.57, and the 2-Choice/Real Facegen Faces condition (M = −.10, SE = .03), t(95) = −3.50, p < .001,Cohen’s d = −0.72, exhibited hyperdescent (i.e., categorized the faces as more White than Black).

Essentialism

We conducted a 2(Categorization Task) × 2(Stimuli Type) between-subjects ANOVA on essentialism.There was a significant interaction, F(1, 398) = 12.17, p < .001, ηp

2 = .032, which we followed up on byconducting pairwise comparisons. In the 2-Choice condition, participants who categorized Real BiracialFaces (M = 3.93, SD = 1.02) had higher essentialism than participants who categorized Real FaceGenFaces (M = 3.64, SD = .84), p = .04, ηp

2 = .01. In the 3-choice conditions, participants who categorizedReal FaceGen Faces (M = 4.01, SD = .98) had higher essentialism than those who categorized RealBiracial Faces (M = 3.64, SD = .92), p = .004, ηp

2 = .02. When participants viewed the real biracial faces,they had higher essentialism after categorizing them in the 2-Choice task compared to the 3-choice task,p = .03, ηp

2 = .01. When participants viewed the real FaceGen faces, they had higher essentialism aftercategorizing them in the 3-choice task compared to the 2-Choice task, p = .01, ηp

2 = .02.

Experiment 4b: real biracial faces and their matched facegen faces, 2-choice vs. mereexposure task, categorization outcomes and essentialism

Participants

Five hundred and eleven participants (264 females; 3 gender unspecified; Mage = 33.90, SD = 11.20)were recruited via Amazon’s Mechanical Turk using the keywords “short face perception study,” andreceived 50 cents for their work. Since Experiments 1–3 all used undergraduate samples, combinedwith the failed replication during the 3-choice task for the essentialism results in Experiment 4a, ourgoal was to recruit over 100 participants in each of four conditions. The sample contained 363Whites, 47 Blacks, 27 Asians, 33 Latino, 31 biracials, and 10 unspecified.

Procedure

This study was identical to Experiment 3, except, here, participants were randomly assigned to vieweither real biracial faces or the FaceGen versions of the real biracial faces. Participants were randomlyassigned either to a 2-Choice or a Mere Exposure condition. Therefore, the design was a 2 (stimulitype: real biracial faces, real biracial faces as FaceGen) x 2 (categorization task: 2-Choice, mereexposure) between-subjects design. After the categorization task, participants completed the same8-item essentialism scale as in the other studies (α = .79).

12 S. E. GAITHER ET AL.

Results

Hypodescent

Relative ratesAn independent samples t-test confirmed that the face stimuli had a significant effect on the rate ofhypodescent, t(256) = −7.56, p < .001, Cohen’s d = −0.95. Consistent with the previous studies,participants in the 2-Choice/Real Biracial Faces condition engaged in hypodescent (M = .23,SE = .03), t(129) = 7.36, p < .001, Cohen’s d = 1.30, whereas participants in the 2-Choice/RealFaceGen Faces engaged in hyperdescent (M = −.11, SE = .03), t(127) = −3.36, p < .001, Cohen’sd = −0.60.

Absolute ratesParticipants in the 2-Choice/Real Faces condition engaged in hypodescent (M = 0.23, SD = .36), t(129) = 7.36, p < .001, categorizing faces as Black more often than as White. Participants in the2-Choice/FaceGen condition engaged in hyperdescent (M = −.11, SD = .37), t(127) = −3.36, p < .001,categorizing faces as White more often than as Black.

EssentialismWe conducted a 2(Task) × 2(Stimuli Type) between-subjects ANOVA on psychological essentialism.There was no significant interaction, F(1, 507) = .44, p = .51, ηp

2 = .001, nor was there a main effectof stimuli type, F(1, 507) = 1.67, p = .20, ηp

2 = .003, or a main effect of task, F(1, 507) = .01, p = .93,ηp

2 = .000. There were also no significant pairwise comparisons.

Experiments 4a & 4b discussion

Experiments 4a and 4b compared real biracial faces to computer-generated versions of those samefaces and replicated the previous hypodescent findings—real biracial faces were categorized as Blackmore often than computer-generated versions of the same faces.3 Together across all experiments,these findings suggest that if perceivers are asked to use monoracial categories to describe raciallyambiguous individuals, such as on the U.S. Census or in eyewitness reports, perceivers may exhibitan increase in their tendencies to categorize that ambiguous individual more often as Black relativeto as White—at least for real racially ambiguous faces.

Regarding essentialist thinking, the present results remain mixed. Experiments 2 and 3 bothshowed that categorizing a real ambiguous face compared to a computer-generated face using fixed,monoracial categories led to higher essentialism. Experiments 4a and 4b only partially replicatedthese results. There are a variety of possible explanations for these differences. Undergraduatescompleted in-lab testing sessions in Experiments 1–3 and non-student adults completed the studyonline for Experiments 4a and 4b. Although some work suggests that undergraduate and onlinesamples are comparable (e.g., Sprouse, 2011), other work shows that online crowd-sourced samplesare higher in social desirability concerns, more motivated by the payment compensation, and moreopen to new experiences (in other words potentially less fixed or essentialist in their views)compared to a college sample (Behrend, Sharek, Meade, & Wiebe, 2011). Moreover, the samplesin Experiments 1–3 were also significantly younger, suggesting perhaps the online adult samplesimply had more stable views in general that were less susceptible to change. Individual item analysesalso confirms that removing any item from the essentialism scale actually worsens the overallreliability. Given these mixed findings, our work highlights the need for additional research toexamine possible boundary effects surrounding the link between exposure to racially ambiguous,biracial faces and racial beliefs (e.g., Eberhardt, Dasgupta, & Banaszynski, 2003; Plaks et al., 2012;Williams & Eberhardt, 2008) and if the use of other essentialism scales may lead to differentoutcomes.

THE JOURNAL OF SOCIAL PSYCHOLOGY 13

General discussion

Our findings extend the literature on racially ambiguous face perception by systematically examiningsome of the possible factors that may affect the historically and socially important outcome ofhypodescent. Despite the fact that computer-generated and real biracial stimuli had the same levelsof racial ambiguity, participants consistently applied hypodescent more often toward real biracialfaces compared to computer-generated biracial faces (Experiments 1–4) and spent less time whilemaking those decisions (Experiment 3). Given how important racial categorizations are in oursociety across a variety of contexts such as healthcare, jury decisions, and job interviews (e.g.,Maddox, 2004; Sommers, 2007; van Ryn & Burke, 2000), this evidence highlights stimulus differ-ences that face perception researchers should consider in future research. Although the variabilityseen here might not apply to all computer-generated faces, this set of studies is the first to comparehow different types of stimuli and tasks may impact the categorization of racially ambiguous faces.

Hypodescent can have a number of negative consequences for targets, including stigmatization,racial profiling, and increased levels of discrimination (e.g., MacLin & Malpass, 2001; Sanchez et al.,2011). Thus, the present findings, which suggest that actual biracial individuals may face higher ratesof visual hypodescent, have important implications for the biracial demographic. Importantly, whena third, multiracial categorization option was offered, perceivers were significantly less likely toengage in hypodescent for real biracial faces. This finding emphasizes the importance of acknowl-edging the multiracial category at a societal level since minority categorizations are so stronglycorrelated with negative social outcomes. Additionally, the current studies provided mixed evidencesuggesting that exposure to real biracial faces compared to computer-generated faces may sometimesincrease racial essentialist beliefs when categorizing these faces into monoracial categories (i.e., usinga two choice categorization task). Thus, more work is needed to understand how stimuli, categor-ization task, and potential individual differences may impact essentialist beliefs. Moreover, thepresent data also cannot assess whether existing levels of essentialism endorsement also moderateracially ambiguous face categorization (see Ho et al., 2015).

The present studies also had small heterogenous samples of non-White participants. Althoughour results generally held when using race (White vs. non-White) as a covariate, we lacked the powerto systematically investigate participant race differences. Some work has shown that Black indivi-duals perceive racially ambiguous faces similarly to White individuals (Ho et al., 2017) and thatWhite, Asian, and Latino participants categorize multiracial faces in similar ways in 3-Category tasks(Chen & Hamilton, 2012; Chen et al., 2014). However, the majority of hypodescent research has yetto examine whether categorization varies systematically by participant race (Young, Sanchez, Pauker,& Gaither, 2018). A recent paper also suggests that some hypodescent categorization patterns(especially when examining visual hypodescent) may actually reflect a minority categorization bias,whereby, for example, multiracial Black/White targets are simply categorized more often as non-White than White, but not necessarily always as Black (Chen, Pauker, Hamilton, Gaither, &Sherman, 2018). Additionally, the current research (like most previous studies on hypodescentwork) used a between-subjects design, and therefore future research should consider the use ofwithin-subject designs to directly test differences based on stimuli and task.

Importantly, the present set of studies only included Black/White biracial stimuli and thesestimuli only contained 20 unique computer-generated or real biracial faces. We do not knowwhether these effects apply toward other multiracial combinations (Chen & Norman, 2016). Wedo know that perceivers often categorize Asian/White biracials in line with hypodescent (Halberstadtet al., 2011; Ho et al., 2011; cf. Chen, Kteily, & Ho, 2017), though they elicit hypodescent to a lesserextent than Black/White biracial faces (Ho et al., 2011). Based on our findings, we would expect tofind similar effects for real Asian/White biracial faces, but the differences between real biracial andcomputer-generated faces might be weaker due to skin tone differences between Asians and Blacks.Moreover, we also only utilized FaceGen computer-generated faces and thus did not measure theseresults with computer-morphed faces or faces generated by other programs. Experiments 4a and 4b,

14 S. E. GAITHER ET AL.

which created FaceGen versions of real biracial faces, help mitigate some concerns surroundingpossible differences in distinctiveness between the face stimuli sets, but replications with other typesof biracial faces is needed.

Finally, our studies only examined racial ambiguity, and there are other ambiguous groupmembers that remain to be examined within this framework (see Tskhay & Rule, 2013, for a review).Knowing that faces of actual people are more multifaceted and distinct than faces typically studied inlaboratory settings, these findings spark a needed discussion surrounding stimuli selection andstimuli generalizability to the real world.

Conclusions

Although we have shown that past findings, including computer-generated racially-ambiguousstimuli, may be limited in some contexts in their applicability to real-world perceptions of actualbiracial individuals, we do not believe that researchers should discontinue the use of thesestimuli or that there is an ideal categorization task. Depending on a researcher’s main researchquestion, it may be important to use more experimentally controlled stimuli and to haveparticipants categorize faces into dichotomous categories. Furthermore, using computer-gener-ated stimuli (in addition to being an easier form of stimuli to obtain) can also be advantageousfor examining how certain components of a face may or may not contribute toward itscategorization. However, if a researcher wants to investigate real-world exposure to multiracials,those researchers should use real biracial stimuli to increase ecological validity. On the otherhand, if a researcher wants to examine the specific social cognitive processes that underlie faceperception outcomes more broadly, using controlled computer-generated stimuli will allow thoseresearchers to isolate specific variables more easily (e.g., independent manipulation of skin toneand facial features).

As such, our objective has not been to invalidate previous findings that used either type of stimulior tasks. Rather, we feel that the current empirical findings highlight an overlooked consideration inresearch that hopefully will spark needed discussions surrounding generalizability and application ofvarious psychological findings from the lab to the real world.

Notes

1. We conducted all analyses in the manuscript using race (White vs. non-White) as a covariate, and the resultsyielded the same significance-levels with one exception: the main effect of categorization task in this studybecame marginally significant, p = .05.

2. Nine participants did not complete essentialism questions and were omitted from analysis.3. Specific stimuli effects within each condition for the 2-Choice conditions on hypodescent were tested. We

conducted a 2 (Stimuli Type) × 20 (Target Face) mixed-model ANOVA with the latter factor within-subjects oncategorizations (1 = Black, 0 = White). In all instances, the main effect of stimuli type remained significant andin the expected direction, showing that on average participants engaged in hypodescent more for real facescompared to computer-generated faces (Exps 1–3) or computer-generated versions of the same real faces (Exps4a, 4b). These results support the main effects that real biracial faces elicited higher levels of hypodescent. Study1—Stimuli Type effect, p = .04; Study 2—Stimuli Type effect, p < .001; Study 3—Stimuli Type effect, p = .04;Study 4a—Stimuli Type effect, p < .001; Study 4b—Stimuli Type effect, p < .001.

Funding

This work was supported by the Ford Foundation [Dissertation Fellowship];National Science Foundation [GraduateResearch Fellowship];Society for the Psychological Study of Social Issues [Clara Mayo Award];University of California,Davis [Chancellor’s Post-doctoral Fellowship];University of Chicago [Provost’s Postdoctoral Scholarship].

THE JOURNAL OF SOCIAL PSYCHOLOGY 15

Notes on contributors

Sarah Gaither is an Assistant Professor in the Department of Psychology and Neuroscience at Duke University. She isalso a faculty affiliate at the Cook Center on Social Equity and the Center on Health and Society. Her research focuseson the role identities and diversity exposure play in shaping our behavior and attitudes.

Jacqueline Chen is an Assistant Professor in the Department of Psychology at the University of Utah. Her researchexamines how an individual's psychology is shaped by his or her social group memberships.

Kristin Pauker is an Associate Professor in the Department of Psychology at the University of Hawaii at Manoa. Herwork explores how a person’s immediate environment and culturally-shaped theories about race impact basic socialperception, social interactions, and stereotyping in childhood and throughout development.

Samuel Sommers is an Associate Professor in the Department of Psychology at Tufts University. His researchinvestigates issues related to stereotyping, prejudice, and group diversity.

References

Allport, G. W. (1954). The nature of prejudice. Reading, MA: Addison-Wesley.Bailenson, J. N., Yee, N., Merget, D., & Schroeder, R. (2006). The effect of behavioral realism and form realism of real-

time avatar faces on verbal disclosure, nonverbal disclosure, emotion recognition, and copresence in dyadicinteraction. Presence: Teleoperators and Virtual Environments, 15, 359–372. doi:10.1162/pres.15.4.359

Bastian, B., Loughnan, S., & Koval, P. (2011). Essentialist beliefs predict automatic motor-responses to socialcategories. Group Processes & Intergroup Relations, 14, 559–567. doi:10.1177/1368430210385258

Behrend, T. S., Sharek, D. J., Meade, A. W., & Wiebe, E. N. (2011). The viability of crowdsourcing for survey research.Behavior Research Methods, 43, 800. doi:10.3758/s13428-011-0081-0

Blascovich, J., Wyer, N. A., Swart, L. A., & Kibler, J. L. (1997). Racism and racial categorization. Journal of Personalityand Social Psychology, 72, 1364–1372. doi:10.1037/0022-3514.72.6.1364

Bodenhausen, G. V., & Macrae, C. N. (1998). Stereotype activation and inhibition. In R. S. Wyer Jr. (Ed.), Stereotypeactivation and inhibition: Advances in social cognition (Vol. 11, pp. 1–52). Hillsdale, NJ: Erlbaum.

Buhrmester, M. D., Kwang, T., & Gosling, S. D. (2011). Amazon’s Mechanical Turk: A new source of inexpensive, yethigh-quality data? Perspectives on Psychological Science, 6, 3–5. doi:10.1177/1745691610393980

Carpinella, C. M., Chen, J. M., Hamilton, D. L., & Johnson, K. L. (2015). Gendered facial cues influence racecategorizations. Personality and Social Psychology Bulletin, 41, 405–419. doi:10.1177/0146167214567153

Castano, E., Yzerbyt, V., Bourguignon, D., & Seron, E. (2002). Who may enter? The impact of ingroup identificationon ingroup/outgroup categorization. Journal of Experimental Social Psychology, 38, 315–322. doi:10.1006/jesp.2001.1512

Chao, M. M., Hong, Y., & Chiu, C. (2013). Essentializing race: Its implications on racial categorization. Journal ofPersonality and Social Psychology, 104, 619–634. doi:10.1037/a0031332

Chen, J. M., & Hamilton, D. L. (2012). Natural ambiguities: Racial categorization of multiracial individuals. Journal ofExperimental Social Psychology, 48, 152–164. doi:10.1016/j.jesp.2011.10.005

Chen, J. M., & Norman, J. B. (2016). Toward a comprehensive understanding of the factors underlying multiracialperson perception. Analyses of Social Issues and Public Policy, 16(1), 417–420. doi:10.1111/asap.12122

Chen, J. M., Kteily, N. S., & Ho, A. K. (2018). Whose side are you on? Mistrust of Asian-White biracials’ allegiancesleads Asian Americans to deny them group membership. Personality and Social Psychology Bulletin, Early OnlineAccess. doi:10.1177/0146167218798032

Chen, J. M., Moons, W. G., Gaither, S. E., Hamilton, D. L., & Sherman, J. W. (2014). Motivation to control prejudicepredicts categorization of multiracials. Personality and Social Psychology Bulletin, 40, 510–603. doi:10.1177/0146167213520457

Chen, J. M., Pauker, K., Hamilton, D., Gaither, S. E., & Sherman, J. (2018). Black + White = Not White: A minoritybias in categorizations of Black-White multiracials. Journal of Experimental Social Psychology, 78, 43–54.doi:10.1016/j.jesp.2018.05.002

Dunham, Y., & Olson, K. R. (2016). Beyond discrete categories: Studying multiracial, intersex, and transgenderchildren will strengthen basic developmental science. Journal of Cognition and Development, 17, 642–665.doi:10.1080/15248372.2016.1195388

Eberhardt, J. L., Dasgupta, N., & Banaszynski, T. L. (2003). Believing is seeing: The effects of racial labels and implicitbeliefs on face perception. Personality and Social Psychology Bulletin, 29, 360–370. doi:10.1177/0146167202250215

Freeman, J. B., Pauker, K., Apfelbaum, E. P., & Ambady, N. (2010). Continuous dynamics in the real-time perceptionof race. Journal of Experimental Social Psychology, 46, 179–185. doi:10.1016/j.jesp.2009.10.002

16 S. E. GAITHER ET AL.

Gaither, S. E. (2015). “Mixed” Results multiracial research and identity explorations. Current Directions inPsychological Science, 24, 114–119. doi:10.1177/0963721414558115

Gaither, S. E., Babbitt, L. G., & Sommers, S. R. (2018a). Resolving racial ambiguity in social interactions. Journal ofExperimental Social Psychology, 76, 259–269. doi:10.1016/j.jesp.2018.03.003

Gaither, S. E., Toosi, N., Babbitt, L.G., & Sommers, S. R. (2018b). Exposure to multiracial faces decreases colorblindattitudes. Personality and Social Psychology Bulletin. Online Early Access. doi:10.1177/0146167218778012

Gaither, S. E., Pauker, K., Slepian, M. L., & Sommers, S. R. (2016). Social belonging motivates categorization of raciallyambiguous faces. Social Cognition, 34, 97–118.

Gaither, S. E., Schultz, J. R., Pauker, K., Sommers, S. R., Maddox, K. B., & Ambady, N. (2014). Essentialist thinkingpredicts decrements in children’s memory for racially ambiguous faces. Developmental Psychology, 50, 482–488.doi:10.1037/a0033493

Halberstadt, J., Sherman, S. J., & Sherman, J. W. (2011). Why Barack Obama is Black: A cognitive account ofhypodescent. Psychological Science, 22, 29–33. doi:10.1177/0956797610390383

Haslam, N., Rothschild, L., & Ernst, D. (2000). Essentialist beliefs about social categories. British Journal of SocialPsychology, 39, 113–127. doi:10.1348/01446660016436

Hinzman, L., & Kelly, S. D. (2013). Effects of emotional body language on rapid out-group judgments. Journal ofExperimental Social Psychology, 49(1), 152–155. doi:10.1016/j.jesp.2012.07.010

Ho, A. K., Kteily, N., & Chen, J. M. (2017). “You’re one of us”: Black Americans’ use of hypodescent and its associationwith egalitarianism. Journal of Personality and Social Psychology, 113, 753–768. doi:10.1037/pspi0000107

Ho, A. K., Roberts, S. O., & Gelman, S. A. (2015). Essentialism and racial bias jointly contribute to the categorizationof multiracial individuals. Psychological Science, 26(10), 1639–1645. doi:10.1177/0956797615596436

Ho, A. K., Sidanius, J., Cuddy, A. J. C., & Banaji, M. R. (2013). Status-boundary enforcement and the categorization ofBlack-White biracials. Journal of Experimental Social Psychology, 49, 940–943. doi:10.1016/j.jesp.2013.04.010

Ho, A. K., Sidanius, J., Levin, D. T., & Banaji, M. R. (2011). Evidence for hypodescent and racial hierarchy in thecategorization and perception of biracial individuals. Journal of Personality and Social Psychology, 100, 492–506.doi:10.1037/a0021562

Hugenberg, K., & Bodenhausen, G. V. (2004). Ambiguity in social categorization the role of prejudice and facial affectin race categorization. Psychological Science, 15, 342–345. doi:10.1111/j.0956-7976.2004.00680.x

Jordan, W. D. (2014). Historical origins of the one-drop racial rule in the United States. Journal of Critical Mixed RaceStudies, 1(1). Retrieved from https://escholarship.org/uc/item/91g761b3

Kang, S. K., & Bodenhausen, G. V. (2015). Multiple identities in social perception and interaction: Challenges andopportunities. Annual Review of Psychology, 66, 547–574. doi:10.1146/annurev-psych-010814-015025

Krosch, A. R., Berntsen, L., Amodio, D. M., Jost, J. T., & Van Bavel, J. J. (2013). On the ideology of hypodescent:Political conservatism predicts categorization of racially ambiguous faces as Black. Journal of Experimental SocialPsychology, 49, 1196–1203. doi:10.1016/j.jesp.2013.05.009

Krosch, A. R., & Amodio, D. M. (2014). Economic scarcity alters the perception of race. Proceedings of the NationalAcademy of Sciences, 111(25), 9079–9084. doi:10.1073/pnas.1404448111

Leyens, J. P., Rodriguez, A. P., Rodriguez, R. T., Gaunt, R., Paladino, P. M., & Vaes, J. (2001). Psychologicalessentialism and the differential attribution of uniquely human emotions to ingroups and outgroups. EuropeanJournal of Social Psychology, 31, 395–411. doi:10.1002/ejsp.50

MacLin, O. H., & Malpass, R. S. (2001). Racial categorization of faces: The ambiguous race face effect. Psychology,Public Policy, and Law, 7, 98–118. doi:10.1037/1076-8971.7.1.98

Maddox, K. B. (2004). Perspectives on racial phenotypicality bias. Personality and Social Psychology Review, 8, 383–401. doi:10.1207/s15327957pspr0804_4

Mullen, B., Brown, R., & Smith, C. (1992). Ingroup bias as a function of salience, relevance, and status: An integration.European Journal of Social Psychology, 22, 103–122. doi:10.1002/ejsp.2420220202

No, S., Hong, Y., Liao, H., Lee, K., Wood, D., & Chao, M. M. (2008). Lay theory of race moderates Asian Americans’responses toward American culture. Journal of Personality and Social Psychology, 95, 991–1004. doi:10.1037/a0012978

Oberman, L. M., Ramachandran, V. S., & Pineda, J. A. (2008). Modulation of mu suppression in children with autismspectrum disorders in response to familiar or unfamiliar stimuli: The mirror neuron hypothesis. Neuropsychologia,46, 1558–1565. doi:10.1016/j.neuropsychologia.2008.01.010

Pauker, K., Meyers, C. K., Sanchez, D. T., Gaither, S. E., & Young, D. M. (2018). A review of multiracial malleability:Identity, categorization, and shifting racial attitudes. Social and Personality Psychology Compass, 12, e12392.doi:10.1111/spc3.v12.6

Pauker, K., Weisbuch, M., Ambady, N., Sommers, S. R., Adams, R. B., Jr., & Ivcevic, Z. (2009). Not so black and white:Memory for ambiguous group members. Journal of Personality and Social Psychology, 96, 795–810. doi:10.1037/a0013265

Peery, D., & Bodenhausen, G. V. (2008). Black+ White= Black hypodescent in reflexive categorization of raciallyambiguous faces. Psychological Science, 19, 973–977. doi:10.1111/j.1467-9280.2008.02185.x

THE JOURNAL OF SOCIAL PSYCHOLOGY 17

Plaks, J. E., Malahy, L. W., Sedlins, M., & Shoda, Y. (2012). Folk beliefs about human genetic variation predict discreteversus continuous racial categorization and evaluative bias. Social Psychological and Personality Science, 3, 31–39.doi:10.1177/1948550611408118

Richeson, J. A., & Sommers, S. R. (2016). Toward a social psychology of race and race relations for the twenty-firstcentury. Annual Review of Psychology, 67, 439–463. doi:10.1146/annurev-psych-010213-115115

Rothbart, M., & Taylor, M. (1992). Category labels and social reality: Do we view social categories as natural kinds? InG. R. Semin & K. Fiedler (Eds.), Language and social cognition (pp. 11–36). Thousand Oaks, CA, US SagePublications, Inc.

Sanchez, D. T., Good, J. J., & Chavez, G. (2011). Blood quantum and perceptions of Black-White biracial targets: Theblack ancestry prototype model of affirmative action. Personality & Social Psychology Bulletin, 37, 3–14. doi:10.1177/0146167210389473

Schroeder, R. (2002). Social interaction in virtual environments: Key issues, common themes, and a framework forresearch. In R. Schroeder (Ed.), The social life of avatars: Presence and interaction in shared virtual environments(pp. 1–18). London, UK: Springer.

Sommers, S. R. (2007). Race and the decision-making of juries. Legal and Criminological Psychology, 12, 171–187.doi:10.1348/135532507X189687

Sprouse, J. (2011). A validation of Amazon Mechanical Turk for the collection of acceptability judgments in linguistictheory. Behavior Research Methods, 43, 155–167. doi:10.3758/s13428-010-0039-7

Tskhay, K. O., & Rule, N. O. (2013). Accuracy in categorizing perceptually ambiguous groups a review and meta-analysis. Personality and Social Psychology Review, 17, 72–86. doi:10.1177/1088868312461308

van Ryn, M., & Burke, J. (2000). The effect of patient race and socio-economic status on physicans’ perceptions ofpatients. Social Science and Medicine, 50, 813–828. doi:10.1016/S0277-9536(99)00338-X

Williams, M. J., & Eberhardt, J. L. (2008). Biological conceptions of race and the motivation to cross racial boundries.Journal of Personality and Social Psychology, 94, 1033–1047. doi:10.1037/0022-3514.94.6.1033

Young, D., Sanchez, D., Pauker, K., & Gaither, S. E. (2018, under review). Testing the predictions of hypodescent and theingroup over-exclusion effect: Meta-analytic and systematic review of multiracial and racially ambiguouscategorization.

Young, D., Sanchez, D. T., & Wilton, L. S. (2013). At the crossroads of race: Racial ambiguity and biracial identifica-tion influence psychological essentialist thinking. Cultural Diversity and Ethnic Minority Psychology, 19, 461–467.doi:10.1037/a0032565

18 S. E. GAITHER ET AL.

Appendix A

FaceGen Computer-Generated Black/White Face Samples

Real Biracial Black/White Face Samples

Real Biracial Black/White Faces as FaceGen Samples Of the Above Real Faces

Note: Stimuli were shown in color for all experiments.

THE JOURNAL OF SOCIAL PSYCHOLOGY 19