facial appearance affects science communication...(e.g., likeability and friendliness), and morality...
TRANSCRIPT
PSYC
HO
LOG
ICA
LA
ND
COG
NIT
IVE
SCIE
NCE
S
Facial appearance affects science communicationAna I. Gheorghiua, Mitchell J. Callana, and William J. Skylarkb,1
aDepartment of Psychology, University of Essex, Colchester CO4 3SQ, United Kingdom; and bDepartment of Psychology, University of Cambridge,Cambridge CB2 3SQ, United Kingdom
Edited by Alexander Todorov, Princeton University, Princeton, NJ, and accepted by Editorial Board Member Michael S. Gazzaniga April 19, 2017 (receivedfor review December 16, 2016)
First impressions based on facial appearance predict many impor-tant social outcomes. We investigated whether such impressionsalso influence the communication of scientific findings to lay audi-ences, a process that shapes public beliefs, opinion, and policy.First, we investigated the traits that engender interest in a scien-tist’s work, and those that create the impression of a “good scien-tist” who does high-quality research. Apparent competence andmorality were positively related to both interest and quality judg-ments, whereas attractiveness boosted interest but decreasedperceived quality. Next, we had members of the public choose realscience news stories to read or watch and found that people weremore likely to choose items that were paired with “interesting-looking” scientists, especially when selecting video-based commu-nications. Finally, we had people read real science news items andfound that the research was judged to be of higher quality whenpaired with researchers who look like “good scientists.” Our find-ings offer insights into the social psychology of science, and indi-cate a source of bias in the dissemination of scientific findings tobroader society.
science communication | impression formation | social cognition
Public discourse and policy are increasingly shaped by sci-entific research, and scientists are increasingly encouraged
to communicate directly with the public (1, 2). Newspaper andtelevision interviews, science festivals, dedicated websites, andonline videos are just some of the channels by which researchersdescribe their work to nonexpert audiences (3). These commu-nications shape people’s beliefs about the physical and socialworld, and correspondingly influence personal decision-makingand government action (4, 5).
However, contrary to traditional conceptions of the scien-tific process as a dispassionate sifting of evidence (6), extrane-ous variables can influence whether a given piece of research iswidely discussed and believed or ignored and discredited. Peo-ple’s selection and evaluation of science communications areswayed by the use of imagery (7), clarity of expression (8), andinclusion of jargon (9). These stylistic features interact with therecipient’s preconceptions and social context to influence thespread and impact of a scientist’s work (10, 11).
We investigated whether science communication is also affect-ed by the scientist’s facial appearance. People form an impres-sion of an individual’s personality, character, and abilitieswithin a few hundred milliseconds of viewing their face (12).These impressions predict important social outcomes in domainsincluding law (13), finance (14), and politics (15). Different traitsare important in different domains (16), but there is good agree-ment between individuals and cultures about the extent to whicha face signals core social traits such as trustworthiness, compe-tence, and sociability (17, 18). However, these inferences gen-erally have poor validity, meaning that facial appearance is animportant source of bias even when more diagnostic informationabout a person is available (19, 20).
Given the potency of face-based impressions and the suscep-tibility of science communication to extraneous presentationalfactors, we hypothesized that a scientist’s face will influence twokey components of the science communication process: selection
(which research the public chooses to find out about) and evalua-tion (the opinions they form about that research). There is a longtradition of research into scientist stereotypes (21–23), includingevidence that people have a sense of what a scientist “looks like”(24), but the facial features that shape the public’s selection andevaluation of science communications have not previously beenexamined.
We focused on three core sociocognitive traits: competence(encompassing, for example, intelligence and skill), sociability(e.g., likeability and friendliness), and morality (e.g., trustworthi-ness and honesty). These factors capture the basic dimensionson which people evaluate groups and individuals (25–28), andall three are germane to science communication. Facial com-petence predicts positive outcomes in many domains (29), and,although some depictions of scientists emphasize elements ofincompetence (e.g., absent-mindedness; ref. 21), intelligence andskill are central to both competence (27) and scientist stereotypes(22), suggesting a positive effect of apparent competence on suc-cessful science communication. Trust is important both to effec-tive communication and to the scientific process (6, 30, 31), andtrustworthy-looking scientists may enjoy greater research success(32). However, face-based inferences about morality have surpris-ingly weak effects in other domains where trust is important, suchas politics (15, 33, 34), so their impact on science communicationis an open question. Finally, although science is a social enterprise(6, 31), scientists are often perceived as solitary and socially awk-ward (22, 23). Thus, although apparent sociability may be desir-able in a communicator/educator (35), it might also weaken theperception that a researcher is a “good scientist” and hence dimin-ish the public’s regard for their work (cf. ref. 33). A similar logicapplies to facial attractiveness, whose influence we also exam-ined: Attractiveness is valued in communicators (35) but doesnot predict research success (32), and may even be detrimentalto having one’s work taken seriously by the public (cf. ref. 34).
Significance
The dissemination of scientific findings to the wider public isincreasingly important to public opinion and policy. We showthat this process is influenced by the facial appearance of thescientist. We identify the traits that engender interest in ascientist’s work and the perception that they do high-qualitywork, and show that these face-based impressions influenceboth the selection and evaluation of science news. Thesefindings inform theories of person perception and illuminatea potential source of bias in the public’s understanding ofscience.
Author contributions: A.I.G., M.J.C., and W.J.S. designed research; A.I.G. performedresearch; A.I.G. contributed new reagents/analytic tools; A.I.G. and W.J.S. analyzed data;and A.I.G., M.J.C., and W.J.S. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission. A.T. is a guest editor invited by the EditorialBoard.
Freely available online through the PNAS open access option.
1To whom correspondence should be addressed. Email: [email protected].
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1620542114/-/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.1620542114 PNAS Early Edition | 1 of 6
Facial competence, morality, sociability, and attractiveness aretherefore plausible influences on both the selection and evalu-ation stages of science communication, but the existence, loci,direction, and magnitude of their effects are open questions thatthe current work seeks to address.
ResultsStudies 1 and 2: Which Facial Traits Are Important to ScienceCommunication? In study 1, we randomly sampled the faces ofscientists from physics (N =108) and genetics/human genetics(N =108) departments of 200 US universities. One group ofparticipants rated these faces on a variety of social traits (e.g.,“How intelligent is this person?”) as well as attractiveness andperceived age. Two other groups of participants indicated howinterested they would be in finding out more about each sci-entist’s research (“interest” judgments) or how much the per-son looked like someone who conducts accurate and importantresearch (“good scientist” judgments). Study 2 was a replicationof study 1, using larger samples of faces and participants andmore social traits. The faces were a representative sample fromthe biological sciences (N =200) and physics (N =200) depart-ments of UK universities.
Confirmatory Factor Analysis established that the trait ratingscomprised three factors: competence (αStudy1 =0.92, αStudy2 =0.91), sociability (αStudy1 =0.95, αStudy2 =0.95), and moral-ity (αStudy1 =0.95, αStudy2 =0.92) (SI Appendix). Interest judg-ments and good scientist judgments were reliable and cor-related, but were distinct constructs (study 1: αInt =0.72,αGood =0.89, correlation between mean judgments for eachface r =0.182, P =0.008; study 2: αInt =0.75, αGood =0.89,r =0.279, P =0.001).
Separate mixed-effects regression analyses predicted interestjudgments and good scientist judgments from facial traits (com-petence, morality, sociability, and attractiveness), scientist demo-graphics [gender, age, discipline, and ethnicity (white vs. non-white, ref. 36)], and participant-level variables (age, gender, and
−0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8
Interest Judgments
Regression Coefficient
P_Science
P_Age
P_Female
Physics
Age
Female
Non−white
Attractiveness
Competence
Sociability
Morality Study 1Study 2Pooled
−0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8
"Good Scientist" Judgments
Regression Coefficient
Fig. 1. Regression coefficients for studies 1 and 2, and pooled across studies. All predictors were standardized. Error bars show 95% confidence inter-vals; coefficients with CIs that exclude zero are highlighted in black. P Age, participant age; P Female, participant gender; and P Sci, participant scienceengagement.
level of science engagement), with all predictors entered simulta-neously. Science engagement was measured with a custom ques-tionnaire and is a potentially important source of variation inpeople’s overall interest in scientists’ communications that mightmodulate the strength of superficial, appearance-based cues (9).We analyzed the two studies separately, and pooled the data toget an overall estimate of effect size. (None of the effects weremodulated by study; see SI Appendix.)
Interest in a scientist’s work was more pronounced among par-ticipants with higher science engagement (Fig. 1, Left). Moreimportantly, interest was related to the facial traits of the scien-tist: People were more interested in learning about the work ofscientists who were physically attractive and who appeared com-petent and moral, with only a weak positive effect of apparentsociability. In addition, interest was somewhat stronger for olderscientists and slightly lower for females than for males, with littledifference between white and nonwhite scientists and no consis-tent effects of participant gender or age.
Judgments of whether a scientist does high-quality work werepositively associated with his or her apparent competence andmorality, but negatively related to both attractiveness and per-ceived sociability (Fig. 1, Right). In addition, older scientists andnonwhite scientists were judged more likely to do good-qualitywork, but there was little overall effect of the scientists’ genderor of participant-level predictors.
In sum, scientists who appear competent, moral, and attrac-tive are more likely to garner interest in their work; those whoappear competent and moral but who are relatively unattractiveand apparently unsociable create a stronger impression of doinghigh-quality research. We found similar results in an additionalstudy that used a standardized face database rather than scien-tists (see SI Appendix).
Studies 3 and 4: Interest in a Scientist’s Work. We next investi-gated whether facial appearance affects people’s choices aboutwhich science to engage with, by pairing the titles of real science
2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1620542114 Gheorghiu et al.
PSYC
HO
LOG
ICA
LA
ND
COG
NIT
IVE
SCIE
NCE
S
news stories with faces that had received low or high Interestjudgments in studies 1 and 2. By counterbalancing the assign-ment of faces to articles, we tested whether facial appearancebiases people’s selection of science news stories. Study 3 exam-ined whether the effects of face-based impressions were moder-ated by the scientist’s gender, academic discipline, and commu-nication format (text versus video); study 4 explored the distinctcontributions of facial competence and attractiveness, and themoderating influence of participant demographics.
In study 3, members of the public were told that they wouldread an article or watch a video in which a scientist describes hisor her work. On each trial, participants chose which one of fouritems they would like to read/watch. Two of the titles were pairedwith “uninteresting” faces, and two were paired with “interest-ing” scientists, selected from those with the lowest and highestinterest judgments in study 1. The article titles were taken fromreal news items published on ScienceDaily.com and prerated tobe of similar, moderate interest to the public (see SI Appendix).The page layout mimicked the selection of science news itemsor blogs on popular websites. All participants made four choices,one for each combination of the scientists’ gender and researchdiscipline (biology vs. physics), on the understanding that theywould subsequently watch/read their chosen items.
Choices were coded according to whether the participantselected an article paired with a “low” face (coded 0) or a “high”face (coded 1). A mixed-effects logistic regression predictedchoices from format (text vs. video), discipline, scientist gender,and their interactions, as well as participant age, gender, and sci-ence confidence. (The complexity of the design meant we did notinclude interactions between experimental and participant-levelvariables for this study.) The choice proportions and regressioncoefficients are plotted in Fig. 2.
Participants were more likely to choose research that waspaired with a photo of an interesting-looking scientist, as indi-
0.00
0.25
0.50
0.75
FemBio
FemPhys
MaleBio
MalePhys
P(C
hose
high
face
)
Text Video
Study 3
123456
LowComp
HighComp
Inte
rest
Rat
ing Low Attract High Attract
Study 4
P_SciP_FemP_Age
Vid*Phys*FemPhys*Fem
Vid*FemVid*Phys
VidFemPhys
Int
−0.25 0.00 0.25 0.50Coefficient
Com*P_AgeCom*P_Fem
Com*P_SciAtt*P_AgeAtt*P_Fem
Att*P_SciP_AgeP_Fem
P_SciCom*Att
ComAtt
0.0 0.2 0.4Coefficient
Fig. 2. (Top) (Left) The choice data from study 3 and (Right) the interest ratings from study 4. (Bottom) The corresponding regression coefficients. Allpredictors were standardized (prior to computing interaction terms). Error bars show 95% confidence intervals; coefficients with CIs that exclude zero arehighlighted in black. Fem, female scientist; Int, intercept; Phys, physics news item; Vid, video format.
cated by the significant intercept term. This bias was presentfor both male and female scientists, physics and biologynews stories, and video and text formats (all Ps< 0.05). Theeffect was more pronounced for videos than written articles,and was stronger for biology than for physics, although theeffect of discipline depended on the scientist’s gender (formales, BDisc =−0.338, P < 0.001; for females, BDisc =0.014,P =0.893). Finally, female participants were more swayedby the scientist’s appearance than were male participants,and the effect of facial appearance diminished with partici-pant age.
Study 4 built on the finding that competence and attractivenesswere two key predictors of interest judgments in studies 1 and 2by varying the attractiveness and competence of the scientists ina 2× 2 within-subject design. Participants were asked to imag-ine that they were browsing a website hosting videos of scien-tists describing their research. Each trial presented one putativevideo, comprising a biology article title taken from study 3 pairedwith a male scientist’s photo taken from those scoring in the bot-tom or top octile on competence and attractiveness in study 2.(The ecological stimulus sample meant that the resulting manip-ulation of attractiveness was weaker than that of competence;see SI Appendix.) Participants rated how likely they would beto watch the video, completing one trial per cell of the design.A mixed-effects regression predicted interest ratings from com-petence, attractiveness, and their interaction, along with partic-ipant age, gender, science engagement, and their interactionswith the facial traits.
Interest judgments were higher for participants with highscience engagement and for older participants (Fig. 2). Moreimportantly, interest was positively related to the facial compe-tence of the scientist. There was also some indication that par-ticipants were more likely to select articles that were paired withattractive faces, but the effect was small, most likely because the
Gheorghiu et al. PNAS Early Edition | 3 of 6
manipulation was weaker. None of the participant-level variablesmoderated the effects of facial traits.
Taken together, these studies show that facial appearanceaffects the public’s selection of science news stories.
Studies 5 and 6: Evaluation of a Scientist’s Work. Finally, wetested the consequences of face-based impressions for the pub-lic’s appraisal of a scientist’s work. We paired articles from newswebsites with faces that did or did not look like good scientists.Study 5 examined the moderating effects of the scientist’s disci-pline and gender; study 6 dissected the contributions of apparentcompetence and physical attractiveness, and examined the mod-erating influence of participant demography.
In study 5, participants were told that they would read articlesfrom a new magazine section comprising profiles of people dis-cussing their interests and work. The articles were adapted fromnews websites (e.g., newser.com) so as to be of similar length andclarity and to be expressed in the first person, such that a scientistis describing his or her own work to a general audience. Partici-pants read two articles, each presented with a photo of its puta-tive author—one with a high good scientist rating in study 1 andone with a low rating. The scientists’ gender and discipline (biol-ogy vs. physics) were varied between subjects. After two fillerarticles that profiled athletes, participants rated the quality ofthe two pieces of research. A mixed-effects regression predictedquality judgments from face type, discipline, scientist gender, andtheir interactions, as well as participant age, gender, and scienceengagement.
Research that was paired with the photo of a good scientistwas judged to be higher quality, and this effect was unaffected bythe scientist’s gender and discipline (Fig. 3). In addition, qualityjudgments were higher for physics articles than for biology arti-cles, and higher among participants who were more engaged withscience.
123456
FemBio
FemPhys
MaleBio
MalePhys
Qua
lity
Rat
ing High Face Low Face
Study 5
123456
LowComp
HighComp
Qua
lity
Rat
ing Low Attract High Attract
Study 6
P_SciP_AgeP_Fem
HiFace*Phys*FemPhys*Fem
HiFace*FemHiFace*Phys
FemPhys
HiFace
−0.2 −0.1 0.0 0.1 0.2Coefficient
Com*P_AgeCom*P_Fem
Com*P_SciAtt*P_AgeAtt*P_Fem
Att*P_SciP_AgeP_Fem
P_SciCom*Att
ComAtt
−0.1 0.0 0.1 0.2Coefficient
Fig. 3. (Top) The mean quality ratings from (Left) study 5 and (Right) study 6. (Bottom) The corresponding regression coefficients. All predictors werestandardized (prior to computing interaction terms). Error bars show 95% Wald confidence intervals; coefficients with CIs that exclude zero are highlightedin black. HiFace, researcher looks like a good scientist.
Study 6 used the same 2× 2 factorial manipulation of compe-tence and attractiveness as study 4. Participants read four physicsnews stories, each paired with a male face from one cell of thedesign. They were subsequently shown the face–article pairingsone at a time and asked to imagine that they had been selected tojudge how much each piece of research deserved to win a prizefor excellence in science. The data were analyzed in a mixed-effects regression with the same predictors as study 4.
More-competent-looking scientists were judged more deserv-ing of the prize (Fig. 3). There was only a very weak nega-tive effect of attractiveness, and no competence× attractivenessinteraction. (As in study 4, the weak effect of attractiveness maybe due to the relative weakness of the manipulation due to stim-ulus constraints; see SI Appendix.) In addition, older participantsand female participants judged the scientists’ work to be moreprize-worthy than did younger/male participants, but participantvariables did not modulate the effects of facial traits.
DiscussionThe traits that engender initial engagement with a scientist’swork are distinct from, and sometimes opposite to, those thatencourage the belief that the scientist does high-quality research.People reported more interest in the research of scientists whoappear competent, moral, and attractive; when judging whethera researcher does “good science,” people again preferred scien-tists who look competent and moral, but also favored less socia-ble and more physically unattractive individuals. Notably, thesesociocognitive traits “trumped” the influence of age, gender, andethnicity—variables that are the primary of focus of much workon stereotypes and bias (37, 38)—implying an underlying sourceof influence that has received little attention in public discourseor academic studies of scientist stereotypes.
Our results further demonstrate the centrality of apparentcompetence and morality to social outcomes (29, 39), and
4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1620542114 Gheorghiu et al.
PSYC
HO
LOG
ICA
LA
ND
COG
NIT
IVE
SCIE
NCE
S
support the idea that sociability and morality are distinct compo-nents of social warmth (25, 40). The conflicting effects of attrac-tiveness on interest and good scientist judgments indicate that,although the stereotypical scientist may be an impartial truthseeker with limited personal appeal (23, 31), people partly treatscience communication as a form of entertainment, where emo-tional impact and aesthetic appeal are desirable qualities (41).Presumably, it is pleasant to look at attractive researchers evenif they do not fit one’s conception of a top-notch scientist, asuggestion that is consistent with evidence that good-lookingacademics receive higher teacher evaluations but do not enjoygreater research success (32).
These face-based impressions affected both the selection andevaluation of science news: People preferentially chose commu-nications that were paired with scientists who looked interesting,and judged real science news stories more favorably when theywere paired with faces that looked like good scientists. Theseresults held for male and female researchers, for biology andphysics news stories, and for text- and video-based communi-cations, a breadth that implies that real-world metrics of com-munication success (e.g., web page views or social media feed-back) will be positively correlated with the apparent competenceof practicing academics.
Although appearance can be an accurate signal of a person’sdisposition or abilities (42), this is limited to specific circum-stances and traits (19), and the same face can produce radicallydifferent impressions (43). Thus, the fact that the same pieceof research is evaluated differently when arbitrarily paired withdifferent faces means that facial cues are a potential source ofbias in science communication. This bias was not always large,but it is practically significant given the current scale of web-based media production and dissemination, where the 60% pref-erence for “interesting-looking” scientists found in the Videocondition of study 3 would amount to tens or hundreds of thou-sands of extra views. Indeed, the effect was particularly strong forvideo communications, and the rising use of video media such asTED talks means that face-based judgments are likely to playan increasing role in shaping the public’s engagement with sci-entific research. Moreover, although people with greater scienceengagement reported more interest in scientists’ work, engage-ment did little to moderate the effects of facial appearance onthe selection and evaluation of science communications, indicat-ing a pervasive bias that may not readily be rectified by improvingmotivation or education.
Our results show that science is a social activity whose out-comes depend on facial appearance in ways that may bias publicattitudes and government actions regarding key scientific issuessuch as climate change and biotechnology. Moreover, becauseeffective communication is increasingly important to scientists’career progression (44), face-based biases may influence not justwhich scientists’ work gains popularity or acceptance among thepublic but also which scientific research is actually conducted,and by whom.
Materials and MethodsEthical approval was granted by the University of Essex Faculty of ScienceEthics Sub-committee. Participants gave informed consent and were givenlinks to the original sources of the science news stories. The data are avail-able via the University of Cambridge Data Repository. Studies 4 and 6 werepreregistered on the Open Science Framework (osf.io/ev794; osf.io/fterb).Additional information about participants, stimuli, procedures, and resultsis provided in SI Appendix.
Participants. Participants in study 1 who provided trait ratings for the sci-entist face set were members of the University of Essex (United King-dom) participant panel and participated in the laboratory; all otherparticipants were members of the US population recruited via an onlineplatform (45). At the end of all studies, participants provided demographicinformation and completed a questionnaire to measure their engagement
with science (e.g., “I am knowledgeable about science,”“I find scientificideas fascinating”).
Design and Procedure. Trial order, block order, stimulus locations, andassignment of participants to conditions were randomized. Assignments ofnews items to faces and conditions were counterbalanced. Unless otherwisenoted, all studies presented stimuli sequentially.Studies 1 and 2. The study 1 faces were a random sample of profile picturesfrom the websites of the physics and genetics/human genetics departmentsof the top 200 ranked US universities (46), cropped and edited to have agray background and uniform height (130 pixels). Study 2 used 400 facesrandomly sampled from the biological sciences and physics departments ofUK universities in proportion to the number of scientists from each insti-tution submitted to the United Kingdom’s 2014 Research Excellent Frame-work, cropped and standardized to 150-pixel height and presented againsttheir original background (47).
Participants made judgments on a nine-point scale (1 = “not at all,”9 = “extremely”). In study 1, 54 participants each rated the faces on traitsrelated to competence (competence, intelligence), sociability (likability,kindness), and morality (trustworthiness, honesty) (48), as well as judgingthe attractiveness of the faces and estimating the face’s age in years (val-ues below 16 and above 100 were discarded). Each dimension was judgedin a separate block. The face set was divided into two subsets (54 biologistsand 54 physicists per subset); 27 participants judged one subset; 26 judgedthe other. Two separate groups of participants indicated for all 216 pho-tos “How interested would you be in finding out more about this person’sresearch?” (N = 27) or “How likely is it that this person is a good scientist?”(N = 27), with the latter defined as “someone who conducts accurate scien-tific research which yields valid and important conclusions.”
In study 2, 762 participants rated all faces on 1 of 12 social traits related tocompetence (competent, intelligent, capable, effective), morality (trustwor-thy, honest, moral, fair), and sociability (likable, friendly, warm and socia-ble), or judged attractiveness; a further 68 judged age. Participants couldskip a face if they recognized it. Two separate groups provided Interestjudgments (N = 103) and good scientist judgments (N = 103); each partici-pant judged one of six sets of 200 faces.
In both studies, two independent judges rated the ethnicity (white vs.nonwhite) of the photos, with a third judge resolving discrepancies.Studies 3 and 4. Study 3 (N = 849) used the titles of eight biology and eightphysics news stories selected from a prerated pool. For each scientist gender,the four lowest- and four highest-scoring faces on the interest dimensionwere selected from the study 1 stimuli. To boost the plausibility of the coverstory, participants in the video condition completed an audio check at thestart of the session. Study 4 (N = 408) used the four biology titles from study3 with the least-extreme interest preratings. On each trial, one of two facesinstantiating the relevant attractiveness–competence combination was ran-domly presented. Ratings were on a seven-point scale.Studies 5 and 6. Study 5 (N = 558) used four biology and four physics newsstories selected from a prerated set for being of similar, moderate quality, ofhigh clarity, and very seldom recognized. The faces were those with the twolowest and two highest good scientist scores for each gender from study 1(after excluding the lowest-scoring male because of conspicuous headwear).Study 6 (N = 824) used the four physics news stories from study 3 with theleast-extreme quality preratings, and the face stimuli from study 4.
After reading all their articles, participants were shown the title andphoto for each science article; they rated the rigor, importance, validity, andoverall quality of the work on a seven-point scale and indicated whetherthey had seen the scientist (study 6) or read about the research (studies 5and 6) before the experiment (recognized trials were excluded). The fourjudgments were averaged (αStudy5 = 0.882; αStudy6 = 0.875).
Data Analysis. All analyses used mixed-effects regression (49) with max-imal but uncorrelated random effects, i.e., by-participant random inter-cepts and random slopes for all effects that are nested within partici-pants (studies 1 to 6) and by-face random intercepts and random slopesfor participant-level predictors (studies 1 and 2). Categorical predictorswere coded as: gender (male = 0, female = 1); ethnicity (white = 0, non-white = 1); discipline (biology = 0, physics = 1); format (text = 0, video = 1); andface type (low on dimension of interest = 0; high = 1). All predictors werestandardized (before computing interaction terms). To test simple maineffects in study 3, we refit the model using dummy coding of the relevantpredictor.
ACKNOWLEDGMENTS. This work was supported by Economic and SocialResearch Council studentship ES/J500045/1.
Gheorghiu et al. PNAS Early Edition | 5 of 6
1. Lok C (2010) Science funding: Science for the masses. Nature 465:416–418.2. Peters HP, et al. (2008) Interactions with the mass media. Science 321:204–205.3. Castell S, et al. (2014) Public Attitudes to Science 2014 (Ipsos MORI Social Res Inst,
London).4. Scheufele DA (2014) Science communication as political communication. Proc Natl
Acad Sci USA 111:13585–13592.5. Satterfield JM, et al. (2009) Toward a transdisciplinary model of evidence-based prac-
tice. Milbank Q 87:368–390.6. Godfrey-Smith P (2003) Theory and Reality (Univ Chicago Press, Chicago).7. O’Neill SJ, Smith N (2014) Climate change and visual imagery. Wiley Interdiscip Rev
Clim Change 5:73–87.8. Hartley J (2003) Improving the clarity of journal abstracts in psychology: The case for
structure. Sci Commun 24:366–379.9. Weisberg DS, Taylor JCV, Hopkins EJ (2015) Deconstructing the seductive allure of
neuroscience explanations. Judgment Decis Making 10:429–441.10. Kahan D (2010) Fixing the communications failure. Nature 463:296–297.11. Knobloch-Westerwick S, Johnson BK, Silver NA, Westerwick A (2015) Science exem-
plars in the eye of the beholder: How exposure to online science information affectsattitudes. Sci Commun 37:575–601.
12. Ballew CC, Todorov A (2007) Predicting political elections from rapid and unreflectiveface judgments. Proc Natl Acad Sci USA 104:17948–17953.
13. Zebrowitz LA, McDonald SM (1991) The impact of litigants’ baby-facednessand attractiveness on adjtions in small claims courts. Law Hum Behav 15:603–623.
14. Rezlescu C, Duchaine B, Olivola CY, Chater N (2012) Unfakeable facial configurationsaffect strategic choices in trust games with or without information about past behav-ior. PLoS One 7:e34293.
15. Todorov A, Mandisodza AN, Goren A, Hall CC (2005) Inferences of competence fromfaces predict election outcomes. Science 308:1623–1626.
16. Olivola CY, Eubanks DL, Lovelace JB (2014) The many (distinctive) faces of lead-ership: Inferring leadership domain from facial appearance. Leadersh Q 25:817–834.
17. Rule NO, et al. (2010) Polling the face: Prediction and consensus across cultures. J PersSoc Psychol 98:1–15.
18. Sussman AB, Petkova K, Todorov A (2013) Competence ratings in US predict presiden-tial election outcomes in Bulgaria. J Exp Soc Psychol 49:771–775.
19. Olivola CY, Funk F, Todorov A (2014) Social attributions from faces bias human choices.Trends Cognit Sci 18:566–570.
20. Olivola CY, Todorov A (2010) Fooled by first impressions? Reexamining thediagnostic value of appearance-based inferences? J Exp Soc Psychol 46:315–324.
21. Haynes R (2010) From alchemy to artificial intelligence: Stereotypes of the scientist inwestern literature. Public Understanding Sci 12:243–253.
22. Mead M, Metraux R (1957) Image of the scientist among high-school students. Science126:384–390.
23. Schinske J, Cardenas M, Kaliangara J (2015) Uncovering scientist stereotypes and theirrelationships with student race and student success in a diverse, community collegesetting. CBE Life Sci Educ 14:1–16.
24. Litterer OF (1933) Stereotypes. J Soc Psychol 4:59–69.25. Leach CW, Ellemers N, Barreto M (2007) Group virtue: The importance of morality (vs.
competence and sociability) in the positive evaluation of in-groups. J Pers Soc Psychol93:234–249.
26. Brambilla M, Rusconi P, Sacchi S, Cherubini P (2011) Looking for honesty: The primaryrole of morality (vs. sociability and competence) in information gathering. Eur J SocPsychol 41:135–143.
27. Fiske ST, Cuddy AJ, Glick P (2007) Universal dimensions of social cognition: Warmthand competence. Trends Cognit Sci 11:77–83.
28. Wojciszke B (2005) Morality and competence in person- and self-perception. Eur RevSoc Psychol 16:155–188.
29. Todorov A, Olivola CY, Dotsch R, Mende-Siedlecki P (2015) Social attributions fromfaces: Determinants, consequences, accuracy, and functional significance. Annu RevPsychol 66:519–545.
30. Fiske ST, Dupree C (2014) Gaining trust as well as respect in communicating to moti-vated audiences about science topics. Proc Natl Acad Sci USA 111:13593–13597.
31. Shapin S (1996) The Scientific Revolution (Univ Chicago Press, Chicago).32. Dilger A, Lutkenhoner L, Muller H (2015) Scholars’ physical appearance, research per-
formance, and feelings of happiness. Scientometrics 104:555–573.33. Castelli L, Carraro L, Ghitti C, Pastore M (2009) The effects of perceived competence
and sociability on electoral outcomes. J Exp Soc Psychol 45:1152–1155.34. Mattes K, et al. (2010) Predicting election outcomes from positive and negative trait
assessments of candidate images. Polit Psychol 31:41–58.35. Mendez JM, Mendez JP (2016) Student inferences based on facial appearance. High
Educ 71:1–19.36. Banducci SA, Karp JA, Thrasher M, Rallings C (2008) Ballot photographs as cues in
low-information elections. Polit Psychol 29:903–917.37. Moss-Racusin CA, Dovidio JF, Brescoll VL, Graham MJ, Handelsman J (2012) Science
faculty’s subtle gender biases favor male students. Proc Natl Acad Sci USA 109:16474–16479.
38. Williams WM, Ceci SJ (2015) National hiring experiments reveal 2:1 faculty preferencefor women on STEM tenure track. Proc Natl Acad Sci USA 112:5360–5365.
39. Goodwin GP (2015) Moral character in person perception. Curr Dir Psychol Sci 24:38–44.
40. Goodwin GP, Piazza J, Rozin P (2014) Moral character predominates in person percep-tion and evaluation. J Pers Soc Psychol 106:148–168.
41. Milkman KL, Berger J (2014) The science of sharing and the sharing of science. ProcNatl Acad Sci USA 111:13642–13649.
42. Bonnefon JF, Hopfensitz A, De Neys W (2015) Face-ism and kernels of truth in facialinferences. Trends Cognit Sci 19:421–422.
43. Todorov A, Porter JM (2014) Misleading first impressions: Different for different facialimages of the same person. Psychol Sci 25:1404–1417.
44. Jensen P, Rouquier JB, Kreimer P, Croissant Y (2008) Scientists who engage with societyperform better academically. Sci Public Policy 35:527–541.
45. Buhrmester M, Kwang T, Gosling SD (2011) Amazon’s mechanical turk: A new sourceof inexpensive, yet high-quality, data? Perspect Psychol Sci 6:3–5.
46. US News & World Report (2014) National university rankings. Available at colleges.usnews.rankingsandreviews.com/best-colleges/rankings/national-universities/data/spp%2B50. Accessed September 5, 2014.
47. Sutherland CA, et al. (2015) Personality judgments from everyday images of faces.Front Psychol 6:1616.
48. Landy J, Piazza JR, Goodwin G (2016) When it’s bad to be friendly and smart: Thedesirability of sociability and competence depends on morality. Pers Soc Psychol Bull42:1272–1290.
49. Bates D, Machler M, Bolker B, Walker S (2015) Fitting linear mixed-effects modelsusing lme4. J Stat Softw 67:1–48.
6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1620542114 Gheorghiu et al.
1
APPENDIX: SUPPORTING INFORMATION
Participants
For all the online studies, a participant was defined as a row in the data file that had completed the
task, was over 18, and was the first occurrence of that IP and MTurk ID (both within and across
studies). The vagaries of participant recruitment meant that our pre-chosen sample sizes were often
specified as minima, with some overshooting. Power calculations for mixed-effects models can be
difficult, so we primarily based our estimates on simpler tests (e.g., a t-test to contrast two
conditions) that are typically less sensitive than our actual analyses. Power calculations used G*Power
[1].
Study 1. The sample size was based on previous impression-formation studies, which have found that
20-25 participants per stimulus gives reliable measures (e.g., [3]). One participant was removed for
having zero variance in their responses, demonstrating little engagement with the task. The final
sample of participants rating the faces on the predictor variables comprised 9 males and 44 females,
ages 18-50 (M = 20.0, SD = 4.6); 50.9% reporting English as their first language. The participants who
rated the faces on the criterion variables (“Interest” or “Good Scientist”) comprised 16 males and 38
females, ages 18-40 (M= 21.4, SD = 4.9), 94% first-language English.
Study 2. Akin to study 1, sample size was based on obtaining 25-30 judgments per each dimension, so
we sought to recruit at least 780 participants for the predictor variables and 120 for the criterion
variables. The initial ratings of the criterion variables showed rather low reliability, so the criterion
sample was boosted by 100 people. Four participants were removed for having zero variance in their
responses, another was removed because of a computer error. The final sample who rated faces on
the predictor dimensions comprised 450 males and 380 females, ages 18-72 (M = 35.3, SD = 10.8),
98% first-language English; the sample rating the faces on the criterion variables comprised 107 males
and 99 females, ages 20-75 (M = 34.3, SD = 10.3), 97% first-language English. The numbers of
participants rating each predictor trait/criterion variable are listed in Table S1.
Variable N
Age 68
Capable 56
Competent 66
Effective 58
Fair 53
Friendly 58
Honest 59
Intelligent 62
Likeable 55
Moral 59
Physically Attractive 58
Sociable 62
Trustworthy 60
Warm 56
Interest judgments 103
“Good Scientist” judgments 103
Table S1. Number of participants rating each dimension in Study 2.
2
Study 3. The sample size was based on 80% power to detect a small effect (w=0.1) in a chi-square test
of whether face type influences the article that is chosen, resulting in a desired sample size of at least
785 participants. (This study was based on a pilot study that indicated a small effect size; full details
are available from the authors.) No participants were excluded. The final sample comprised 526
males, 323 females, ages 18-73 (M = 32.4, SD = 10.6), 93% first-language English, with 427
participants assigned to the Text condition and 422 to the Video condition.
Study 4. The sample size was based on 95% power to detect a small-to-medium effect size (d=0.2,
based on the modest effect found in Study 3) in a within-subject t-test for an effect of face-type,
resulting in a minimum required sample of 330 participants. Two participants were excluded for
reporting technical issues (i.e., the photos did not load properly). The final sample comprised 192
males, 216 females, ages 18-74 (M = 35.9, SD = 11.1), 98% first-language English.
Study 5. The sample size was based on obtaining 95% power to detect a small-to-medium effect (d =
0.15) in a within-subjects t-test comparing the two face types. Seventy participants were excluded for
recognizing either of the two articles they read. The final sample comprised 261 males, 297 females,
ages 18-81 (M = 36.4, SD = 12.5), 97% first-language English, with the Male-Biology, Male-Physics,
Female-Biology, and Female-Physics conditions having 150, 144, 129, and 135 participants,
respectively.”
Study 6. The minimum sample size of 800 participants was calculated based on 80% power to detect a
small effect (d = 0.1, estimated from the effect size from Study 5). Participants were asked a simple
memory/attention check question after reading the science stories; those who failed were redirected
away from the survey and counted as “non-completers”. Of those who completed the task, 3 were
excluded for reporting technical problems, and one was excluded for recognizing all of the articles.
This study only excluded people who recognized all articles, whereas Study 5 excluded people who
recognized any articles; this discrepancy arose because of an error when we submitted our pre-
registration for Study 6, which was intended to have the same policy as Study 5. We decided it was
best to keep to the publicly pre-registered plan for this study. The final sample comprised 369 males,
455 females, ages 19-73 (M = 37.5, SD= 12.0), 98% first-language English.
3
Stimuli
Study 1. We randomly sampled photos of scientists from the departmental websites of the top-200 US
Universities (National University Rankings, 2014). We randomly selected a university; if it had a
genetics/human genetics department, then we randomly selected 10 photos of scientists from their
departmental web pages; only photos of main faculty were selected. If the university did not have a
genetics/human genetics department, then we randomly selected another university and randomly
sampled 10 photographs of scientists from their web pages, and so on. Sampling continued until we
had acquired at least 250 photos. This procedure was repeated for physics departments. We edited the
photos (254 geneticists and 271 physicists) to have a grey background and cropped them to start at the
top of the head and finish immediately below the chin, and to be reasonably centred. Images that were
below 130 pixels in height were removed, and the remaining images were resized to have a height of
130 pixels. Poor-quality images were excluded, resulting in a final stimuli set of 108 photos of geneticists
and 108 photos of physicists.
Study 2. We randomly sampled photos of biologists and physicists who had been submitted to the
UK’s 2014 Research Excellence Framework (REF), a nationwide audit of university research. The
power calculation was based on one of the smallest effects of interest in Study 1, that of sociability of
“Good scientist” judgments (partial R2 = .024); 85% power to detect this effect in a simple multiple
regression requires at least 368 participants, so we sampled a total of 400 photos (200 from each
discipline). We drew up a list of all scientists submitted to the relevant “unit of assessment”
(Biological Science or Physics). After excluding 3 universities for which no photos were available, we
randomly sampled from this list in proportion to the number of individuals from each university (e.g.,
the University of Cambridge constituted roughly 8% of the total researchers evaluated within the
Biological Sciences unit of assessment, so Cambridge contributed roughly 8% of the biologists within
our set of 200 photos). If the scientist selected did not have a suitable photo on the university
webpages then we randomly selected another scientist within that university; if were unable to reach
the desired number of photos for a given university then we randomly sampled from the whole list of
scientists. The photos were cropped around the top of the head and the shoulders, and standardised
to 150 pixels in height. Any photos that were too blurry were replaced using the original sampling
procedure.
Studies 3 and 4. The titles of 60 science news stories were collected from ScienceDaily.com, 30 from
the “Health and Medicine” category, and 30 from the “Physics” category. In the pre-rating task, 105
participants were presented with either the biology or the physics titles, in a random order, and rated
them on how interested they would be in reading the full article (0 – not at all interested, 10 –
extremely interested). Mean interest ratings were computed for each title, averaging across
participants. The titles selected for later studies had average ratings close to the mid-point of the
scale, and similar ratings to each other (Table S2).
4
Biology Article Titles Mean Rating
Opinions on vaccinations heavily influenced by online comments 5.12
Confidence in government linked to willingness to vaccinate 5.17
Texting may be more suitable than apps in treatment of mental illness* 5.19
Cow immune system inspires potential new therapies* 5.27
Reasons why winter gives flu a leg up could be key to prevention* 5.35
Stress balls, DVDs and conversation ease pain, anxiety during surgery* 5.37
Risk for autism increases for abandoned children placed in institutions 5.38
Elementary teachers' depression symptoms related to students' learning 5.52
Physics Article Titles Mean Rating
Laser pulse turns glass into a metal: New effect could be used for ultra-fast logical switches
5.13
Doing more with less: Steering a quantum path to improved internet security 5.17
A 'Star Wars' laser bullet -- this is what it really looks like 5.23
'Solid' light could compute previously unsolvable problems 5.25
How to make mobile batteries last longer by controlling energy flows at nano-level 5.26
Universe may face a darker future: Is dark matter being swallowed up by dark energy?
5.32
Hunt for Big Bang particles offering clues to the origin of the universe 5.45
Electronics that need very little energy? Nanotechnology used to help cool electrons with no external sources
5.45
Table S2. Mean interest ratings for article titles used in Studies 3 and 4. Titles marked with an asterisk
were used in Study 4.
The faces used in Study 4 were selected to score low or high on competence and attractiveness. Table
S3 lists the mean ratings that the chosen faces had received in Study 2. As noted in the main text, the
Competence manipulation was stronger than the Attractiveness manipulation: the mean Interest
rating for the low-attractiveness faces is 4.89; that for the high-attractiveness faces is 5.38, giving a
difference of only 0.49, compared with the difference of 0.99 between the low-competence and high-
competence faces. Likewise, the low-attractiveness faces received “Good scientist” ratings that were
only are only 0.86 above those of the high-attractiveness faces, as compared with a difference of 1.96
between the high- and low-competence faces.
5
Low Competence High Competence
Low
attractiveness High
attractiveness Low
attractiveness High
attractiveness
Attractiveness 2.65 5.60 2.81 5.12
Competence 4.62 5.02 6.65 6.69
Interest 4.23 5.05 5.55 5.71
“Good Scientist” 4.96 4.34 7.16 6.06
Age 42.38 26.07 52.62 42.02
Sociability 5.80 4.61 5.64 4.91
Morality 5.16 5.23 6.14 5.74
Warmth 5.48 4.92 5.89 5.32
Table S3. Mean ratings for the face stimuli used in Studies 4 and 6.
Studies 5 and 6. Twenty scientific articles (10 biology and 10 physics) were selected from news
websites (e.g., newser.com) and re-written in first person, and in an accessible fashion, simulating
“scientist profiles” found in magazines. In the pre-rating task, 128 participants saw 5 biology and 5
physics articles (randomly selected and displayed), and rated them on questions related to the quality
of research presented in the article, as well as their comprehension and recognition of the work, using
7-point scales (1=Not at all, 7=Extremely). Trials where the participant recognized the research were
excluded, and mean quality ratings were computed by averaging across participants and questions.
The articles selected for use had ratings close to the mid-point of the scale, good scores on
comprehension and less than 10% recognition rate (Table S4).
Biology Articles Mean
Quality
Mean
Comprehension
Recognition
Study Suggests Earth Life Began on Mars 3.98 4.76 6.35%
Slime Mould Is Smarter Than You Think 4.39 4.97 4.76%
Beneath Pacific Lies Ancient, Barely Alive Bacteria 4.52 5.18 1.49%
Earth Holds 8.7M Species, and Most of Them are Still
Undiscovered
4.52 5.17 4.76%
Physics Articles Mean
Quality
Mean
Comprehension
Recognition
Dark Matter Particles Detected Deep in Mine 4.01 4.84 9.52%
Bloodhound Diary: It's rocket science 4.53 5.29 4.84%
World's Next Timekeeper: Quantum Superclock? 4.67 4.63 0%
Final chapter to be published, in decades-long Gravity
Probe B project
4.69 3.91 1.56%
Table S4. Titles of articles use in studies 5 and 6; Study 6 only used the Physics stories.
6
Counterbalancing
Study 3. We constructed three 8x8 Latin-squares that equally allocated articles to faces for each
discipline, and four counterbalancing tables that equally allocated face-types to disciplines. Combining
these gave 24 versions of the task, with participants randomly assigned to a version.
Study 4. The four article titles were paired with the four cells of the design (low/high attractiveness
and low/high competence) using a 4x4 Latin Square; participants were randomly allocated to a
version. One of the two photos with the appropriate attractiveness-competence combination was
randomly selected on each trial.
Study 5. For each of the four gender-discipline combinations, we constructed a 4x4 Latin Square that
ensured that each article was assigned to each face type equally often. Participants were randomly
allocated to one of the resulting 16 versions of the task. On each trial, one of the two faces of the
relevant type was randomly selected to be displayed alongside the article.
Study 6. The four articles were assigned to the four cells of the design using a 4x4 Latin square,
creating four versions of the task with random allocation to version. On each trial, one of the two
faces with the relevant competence-attractiveness combination was selected to be displayed
alongside the allocated article.
7
Consistency of measures
Table S5 shows Cronbach’s alphas for the measures in Studies 1 and 2, and indicated good
consistency. For Study 1, Cronbach’s Alpha was calculated separately for each face-set (participants
saw one of two face-sets) and the average is reported. For Study 2 there were 3 pairs of face-sets,
where one member of each pair comprised 50% of the faces and the other member comprised the
complimentary 50%. Cronbach’s alpha was calculated for each combination of 3 non-complimentary
sets (i.e., with no members of a pair in the analysis) and an average taken.
Type Measure Study 1 Study 2
Predictor variables
Capable - 0.74 Competent 0.85 0.78 Effective - 0.72 Intelligent 0.88 0.78 Friendly - 0.93 Likeable 0.91 0.84 Sociable - 0.91 Warm - 0.88 Kind 0.92 - Fair - 0.75 Honest 0.89 0.81 Moral - 0.79 Trustworthy 0.88 0.79 Age 0.99 0.99 Physically Attractive 0.95 0.91
Composite measures
Competence 0.92 0.91 Sociability 0.95 0.95 Morality 0.95 0.92
Criterion “Good Scientist” 0.89 0.89 variables Interest 0.72 0.75
Table S5. Cronbach’s Alpha values for the individual traits, outcomes and composite measures in
Studies 1 and 2.
8
Confirmatory Factor Analysis
The correlations between the trait ratings are shown in Table S6 and S7, and are consistent with the
three-factor structure we were expecting.
Intelligent Likeable Kind Trustworthy Honest
Competent 0.860* 0.385* 0.404* 0.543* 0.522*
Intelligent 0.310* 0.376* 0.515* 0.511*
Likeable 0.914* 0.799* 0.850*
Kind 0.850* 0.904*
Trustworthy 0.903*
Table S6. Correlations among the items forming each trait for Study 1 (* indicates p <.05).
Table S7. Correlations between the items forming each trait, for Study 2 (* indicates p <.05).
For both studies, we ran a CFA on the three-factor model: competence (comprising Competent and
Intelligent in Study 1, and Competent, Intelligent, Capable, and Effective in Study 2), sociability (Study
1: Likeable and Kind; Study 2: Likeable, Sociable, Friendly, and Warm) and morality (Study 1:
Trustworthy and Honest; Study 2: Trustworthy, Honest, Moral, and Fair). The models had an
acceptable fit for both Study 1 (SRMR = .018, RMSEA = .102, CFI = .991, TLI = .978, BIC = 2056.26) and
Study 2 (SRMR = .064, RMSEA = .128, CFI = .933, TLI = .913, BIC = 5579.77), supporting the three-
factor model of social judgement.
For both studies, the three-factor model was a better fit than an alternative “competence and
warmth” two-factor model in which “warmth” combines morality and sociability. For Study 1, the
two-factor model had SRMR = .051, RMSEA = .247, CFI = .931, TLI = .870, BIC = 2139.70, and a chi-
square test for the difference in model fit gave χ2diff(2) = 94.19, p<.001. For Study 2, SRMR = .107,
RMSEA = .190, CFI = .845, TLI = .808, BIC = 6004.34, χ2diff(2) = 436.56, p<.001). The three-factor
model also fit better than a single-factor model; Study 1: SRMR = .142, RMSEA = .419, CFI = .776, TLI =
.627, BIC = 2370.55; χ2diff(3) = 330.41, p<.001; Study 2: SRMR = .204, RMSEA = .284, CFI = .650, TLI =
.572, BIC = 6969.18, χ2diff(3) = 1407.4, p<.001).
Intel. Capab. Effect. Lik. Soc. Friend. Warm Trust. Hon. Mor. Fair
Competent 0.678* 0.715* 0.727* 0.295* 0.207* 0.155* 0.244* 0.454* 0.405* 0.459* 0.338*
Intelligent 0.737* 0.675* 0.123* 0.054 0.069 0.073 0.278* 0.319* 0.315* 0.135*
Capable 0.733* 0.210* 0.099* 0.097 0.132* 0.39* 0.404* 0.387* 0.192*
Effective 0.231* 0.144* 0.093 0.132* 0.361* 0.400* 0.364* 0.241*
Likeable 0.819* 0.806* 0.825* 0.742* 0.727* 0.729* 0.799*
Sociable 0.890* 0.867* 0.597* 0.612* 0.576* 0.737*
Friendly 0.912* 0.64* 0.692* 0.624* 0.744*
Warm 0.677* 0.673* 0.645* 0.753*
Trustworthy 0.786* 0.808* 0.709*
Honest 0.803* 0.702*
Moral 0.704*
9
Correlations between the composite traits and the criterion variables in Studies 1 and 2 are shown in
Table S8.
Competence Sociability Morality “Good Scientist”
Interest
Competence 0.39* 0.555* 0.778* 0.505*
Sociability 0.168* 0.893* 0.098 0.632*
Morality 0.424* 0.798* 0.304* 0.624*
“Good Scientist” 0.689* -0.069 0.163* 0.182*
Interest 0.585* 0.422* 0.534* 0.279*
Table S8. Correlations between the composite traits and criterion variables. The top-half of the table
(above the diagonal) represents Study 1, the bottom half represents Study 2 (*indicates p <.05).
10
Regression coefficients
Tables S9-S13 give the numeric values of the regression coefficients plotted in the main text, along
with 95% CIs and p-values based on Satterthwaite-adjusted degrees of freedom. In all tables, P =
participant, Sci = science engagement. Unless otherwise noted, all predictors were standardized (z-
scored), with standardization prior to computing any interaction terms. Standardization was across
participants for participant-level variables and across faces for face-level variables.
INTEREST JUDGMENTS
Study 1 Study 2
B CIlow CIhigh p B CIlow CIhigh p
Age 0.047 -0.072 0.166 0.446 0.074 0.012 0.137 0.021
Female -0.115 -0.242 0.011 0.084 -0.051 -0.141 0.039 0.268
Non-white -0.009 -0.079 0.060 0.792 0.032 -0.014 0.078 0.176
Physics 0.013 -0.036 0.062 0.613 -0.013 -0.044 0.018 0.406
Attractiveness 0.374 0.233 0.516 <.001 0.213 0.142 0.284 <.001
Competence 0.136 0.022 0.251 0.026 0.200 0.122 0.277 <.001
Sociability 0.059 -0.109 0.226 0.496 0.049 -0.032 0.131 0.236
Morality 0.124 -0.007 0.255 0.068 0.132 0.039 0.225 0.006
P_Age 0.030 -0.332 0.393 0.872 0.020 -0.226 0.265 0.876
P_Female -0.169 -0.564 0.226 0.409 0.273 0.024 0.523 0.034
P_Sci 0.382 -0.020 0.785 0.073 0.232 -0.017 0.482 0.071
“GOOD SCIENTIST” JUDGMENTS
Study 1 Study 2
B CIlow CIhigh p B CIlow CIhigh p
Age 0.177 0.056 0.298 0.007 0.059 -0.019 0.137 0.140
Female -0.068 -0.158 0.022 0.143 0.023 -0.072 0.119 0.633
Non-white 0.079 -0.034 0.192 0.179 0.040 -0.014 0.094 0.146
Physics 0.039 -0.015 0.094 0.160 0.024 -0.019 0.067 0.283
Attractiveness -0.252 -0.382 -0.122 <.001 -0.325 -0.415 -0.235 <.001
Competence 0.698 0.578 0.819 <.001 0.516 0.429 0.604 <.001
Sociability -0.152 -0.282 -0.022 0.023 -0.123 -0.203 -0.043 0.003
Morality 0.204 0.046 0.362 0.012 0.111 0.003 0.219 0.045
P_Age -0.247 -0.565 0.070 0.138 -0.054 -0.275 0.167 0.635
P_Female -0.099 -0.418 0.220 0.548 0.152 -0.072 0.376 0.187
P_Sci 0.026 -0.294 0.345 0.877 0.128 -0.084 0.340 0.239
Table S9. Regression coefficients for Studies 1 and 2.
11
To obtain a better estimate of the overall effect of each predictor, we pooled the data from Studies 1
and 2. The main text reports the results of simple pooling (with all predictors standardized across the
pooled sample); Tables S10 and S11 show the regression coefficients, and those obtained when Study
and its interactions are included as fixed-effects (with Study 1 coded -1, Study 2 coded +1; we did not
standardize this variable). There is no indication that study modulated the other effects.
INTEREST JUDGMENTS
Studies 1 and 2 Pooled Data Pooled Data with Effect of Study
B CIlow CIhigh p B CIlow CIhigh p
Age 0.073 0.017 0.130 0.012 0.061 -0.005 0.128 0.070
Female -0.063 -0.138 0.012 0.101 -0.084 -0.173 0.006 0.069
Non-white 0.026 -0.017 0.069 0.232 0.015 -0.032 0.062 0.531
Physics -0.007 -0.034 0.019 0.586 -0.000 -0.031 0.030 0.979
Attractiveness 0.266 0.197 0.336 <.001 0.285 0.204 0.365 <.001
Competence 0.215 0.139 0.292 <.001 0.177 0.090 0.263 <.001
Sociability 0.057 -0.018 0.132 0.140 0.051 -0.041 0.143 0.276
Morality 0.149 0.058 0.240 0.002 0.139 0.044 0.234 0.005
P_Age -0.024 -0.235 0.186 0.820 0.028 -0.620 0.677 0.932
P_Female 0.185 -0.032 0.402 0.097 0.048 -0.254 0.350 0.756
P_Sci 0.265 0.048 0.481 0.018 0.319 0.018 0.620 0.040
Study -0.195 -0.876 0.486 0.576
Study*Age 0.015 -0.051 0.081 0.650
Study*Female 0.032 -0.058 0.122 0.485
Study*Non-white 0.023 -0.024 0.069 0.347
Study*Physics -0.013 -0.044 0.017 0.391
Study*Att -0.037 -0.118 0.043 0.365
Study*Comp 0.070 -0.016 0.156 0.114
Study*Soc 0.001 -0.091 0.093 0.979
Study*Mor 0.034 -0.061 0.129 0.483
Study*P_Age 0.000 -0.648 0.649 0.999
Study*P_Female 0.204 -0.098 0.506 0.187
Study*P_Sci -0.095 -0.397 0.206 0.537
Table S10. Regression coefficients for Interest Judgments when data from Studies 1 and 2 are pooled,
either with or without including Study and its interactions with other variables.
12
Table S11. Regression coefficients for “Good Scientist” judgments when data from Studies 1 and 2 are
pooled, either with or without including Study and its interactions with other variables.
“GOOD SCIENTIST” JUDGMENTS
Studies 1 and 2 Pooled Data Pooled Data with Effect of Study
B CIlow CIhigh p B CIlow CIhigh p
Age 0.094 0.027 0.160 0.006 0.115 0.040 0.191 0.003
Female -0.004 -0.077 0.070 0.921 -0.022 -0.109 0.064 0.613
Non-white 0.056 0.005 0.107 0.034 0.056 0.002 0.110 0.044
Physics 0.029 -0.006 0.063 0.102 0.031 -0.007 0.069 0.109
Attractiveness -0.331 -0.414 -0.247 <.001 -0.297 -0.392 -0.202 <.001
Competence 0.600 0.518 0.681 <.001 0.592 0.503 0.682 <.001
Sociability -0.139 -0.206 -0.072 <.001 -0.135 -0.215 -0.056 0.001
Morality 0.167 0.073 0.262 0.001 0.158 0.058 0.258 0.002
P_Age -0.009 -0.193 0.175 0.927 -0.258 -0.654 0.138 0.204
P_Female 0.072 -0.112 0.257 0.444 0.024 -0.218 0.266 0.846
P_Sci 0.100 -0.083 0.283 0.284 0.087 -0.151 0.325 0.475
Study 0.290 -0.147 0.726 0.195
Study*Age -0.054 -0.130 0.021 0.160
Study*Female 0.046 -0.041 0.132 0.301
Study*Non-white -0.009 -0.063 0.046 0.759
Study*Physics -0.008 -0.046 0.031 0.694
Study*Att -0.081 -0.176 0.014 0.098
Study*Comp 0.044 -0.045 0.134 0.333
Study*Soc 0.003 -0.077 0.083 0.945
Study*Mor -0.013 -0.112 0.087 0.803
Study*P_Age 0.206 -0.190 0.602 0.310
Study*P_Female 0.126 -0.116 0.367 0.310
Study*P_Sci 0.029 -0.209 0.267 0.812
13
Testing for modulation by participant gender
Study 1 had more female participants than males (Study 2 showed a slight preponderance of males). To test whether participant gender modulates our results we re-ran our regression analyses including participant gender and its interaction with all other variables as additional predictors. (The analysis was as before, including the standardization of variables prior to computing interaction terms. We did not include by-face random slopes for any of these interaction terms because of convergence problems.) The regression coefficients are shown in Tables S12 and S13 and indicate that our results are consistent across male and female participants (only one of 40 interaction terms has p <.05). Likewise, comparison of the models with and without the interaction terms indicated that the simpler
models are to be preferred: Interest judgments for Study 1, 2 (10) = 5.50, p = .955, BICno_int = 22137,
BICint = 22218; for Study 2, 2 (10) = 11.90, p = .292, BICno_int = 82123, BICint = 82210; “Good Scientist”
judgments for Study 1 2 (10) = 11.89, p = .292, BICno_int = 21846, BICint = 21921; for Study 2, 2 (10)
= 8.13, p = .616, BICno_int = 81706, BICint = 81797.
INTEREST JUDGMENTS
Study 1 Study 2
B CIlow CIhigh p B CIlow CIhigh p
Intercept 4.912 4.503 5.320 0.000 4.831 4.584 5.079 0.000
Age 0.047 -0.069 0.162 0.433 0.074 0.012 0.137 0.021
Female -0.115 -0.241 0.011 0.082 -0.051 -0.141 0.039 0.266
Non-white -0.009 -0.079 0.060 0.792 0.032 -0.014 0.077 0.176
Physics 0.013 -0.036 0.062 0.613 -0.013 -0.044 0.018 0.408
Attractiveness 0.374 0.235 0.514 0.000 0.213 0.143 0.284 0.000
Competence 0.136 0.022 0.251 0.026 0.200 0.123 0.277 0.000
Sociability 0.059 -0.107 0.225 0.492 0.049 -0.032 0.131 0.235
Morality 0.124 -0.008 0.255 0.069 0.132 0.039 0.224 0.006
P_Age -0.071 -0.718 0.577 0.832 0.036 -0.214 0.286 0.776
P_Female -0.218 -0.700 0.264 0.383 0.275 0.027 0.524 0.032
P_Sci 0.388 -0.007 0.782 0.064 0.247 -0.005 0.499 0.057
P_Female*Age -0.072 -0.185 0.041 0.222 -0.037 -0.096 0.022 0.220
P_Female*Female -0.021 -0.146 0.104 0.744 0.037 -0.050 0.124 0.401
P_Female*Non-white 0.006 -0.060 0.073 0.852 -0.025 -0.068 0.018 0.251
P_Female*Physics -0.018 -0.062 0.026 0.416 -0.020 -0.047 0.007 0.147
P_Female*Att 0.027 -0.110 0.164 0.701 0.024 -0.043 0.091 0.491
P_Female*Comp 0.004 -0.109 0.117 0.943 0.060 -0.014 0.133 0.115
P_Female*Soc 0.084 -0.075 0.243 0.306 0.008 -0.068 0.085 0.831
P_Female*Mor -0.032 -0.149 0.085 0.592 0.036 -0.047 0.120 0.396
P_Female*P_Age 0.235 -0.715 1.184 0.632 0.015 -0.237 0.266 0.910
P_Female*P_Sci 0.237 -0.173 0.647 0.268 -0.131 -0.384 0.121 0.311
Table S12. Regression coefficients for Interest judgments when interactions between Participant
Gender and all other predictors are included.
14
“GOOD SCIENTIST” JUDGMENTS
Study 1 Study 2
B CIlow CIhigh p B CIlow CIhigh p
Intercept 5.664 5.360 5.967 0.000 5.752 5.526 5.977 0.000
Age 0.177 0.057 0.297 0.006 0.059 -0.019 0.136 0.141
Female -0.068 -0.156 0.019 0.132 0.023 -0.072 0.119 0.635
Non-white 0.079 -0.034 0.192 0.178 0.040 -0.014 0.093 0.147
Physics 0.039 -0.015 0.094 0.160 0.024 -0.019 0.067 0.282
Attractiveness -0.252 -0.381 -0.123 0.000 -0.325 -0.415 -0.235 0.000
Competence 0.698 0.579 0.818 0.000 0.516 0.429 0.604 0.000
Sociability -0.152 -0.282 -0.022 0.023 -0.123 -0.203 -0.043 0.003
Morality 0.204 0.046 0.362 0.012 0.112 0.004 0.220 0.043
P_Age -0.205 -0.578 0.167 0.289 -0.113 -0.354 0.128 0.360
P_Female -0.137 -0.448 0.175 0.397 0.167 -0.057 0.390 0.148
P_Sci 0.023 -0.298 0.344 0.888 0.133 -0.078 0.344 0.218
P_Female*Age 0.058 -0.054 0.169 0.320 0.019 -0.042 0.080 0.544
P_Female*Female -0.047 -0.124 0.029 0.236 0.063 -0.018 0.144 0.131
P_Female*Non-
white
0.024 -0.084 0.133 0.662 0.014 -0.025 0.054 0.469
P_Female*Physics 0.019 -0.023 0.061 0.385 0.002 -0.024 0.027 0.907
P_Female*Att 0.032 -0.085 0.149 0.597 0.019 -0.056 0.095 0.617
P_Female*Comp 0.046 -0.067 0.158 0.434 0.030 -0.042 0.102 0.422
P_Female*Soc 0.113 0.013 0.212 0.028 -0.014 -0.062 0.034 0.569
P_Female*Mor -0.100 -0.228 0.028 0.128 -0.020 -0.084 0.045 0.550
P_Female*P_Age -0.151 -0.637 0.335 0.548 0.145 -0.091 0.380 0.232
P_Female*P_Sci 0.181 -0.179 0.540 0.334 0.025 -0.186 0.236 0.818
Table S13. Regression coefficients for “Good Scientist” judgments when interactions between
Participant Gender and all other predictors are included.
15
Results for each gender and ethnicity
For completeness (and as requested by a reviewer), Tables S14 and S15 present the results of
analysing the pooled data from Studies 1 and 2 split by face-gender and face-ethnicity. (The sample
sizes are small for the female- and non-white groups, resulting in low power.)
INTEREST JUDGMENTS
Male Faces Female Faces
B CIlow CIhigh p B CIlow CIhigh p
Intercept 4.748 4.534 4.963 0.000 5.103 4.851 5.355 0.000
Age 0.099 0.038 0.161 0.002 -0.061 -0.146 0.024 0.160
Non-white 0.034 -0.011 0.080 0.137 -0.007 -0.080 0.065 0.843
Physics -0.005 -0.034 0.024 0.738 -0.020 -0.072 0.031 0.442
Attractiveness 0.237 0.169 0.305 0.000 0.230 0.119 0.341 0.000
Competence 0.212 0.129 0.296 0.000 0.198 0.107 0.289 0.000
Sociability 0.065 -0.010 0.141 0.091 0.001 -0.109 0.111 0.985
Morality 0.129 0.045 0.214 0.003 0.118 -0.014 0.250 0.083
P_Age -0.015 -0.230 0.200 0.891 -0.070 -0.323 0.184 0.590
P_Female 0.155 -0.066 0.376 0.172 0.288 0.028 0.548 0.032
P_Sci 0.303 0.082 0.524 0.008 0.107 -0.151 0.366 0.417
White Faces Non-white Faces
B
B
CIlow CIhigh p B CIlow CIhigh p
Intercept 4.816 4.606 5.026 0.000 4.873 4.615 5.131 0.000
Age 0.074 0.014 0.133 0.016 0.017 -0.076 0.110 0.722
Female -0.057 -0.132 0.019 0.145 -0.095 -0.224 0.033 0.153
Physics -0.013 -0.041 0.015 0.375 0.058 -0.021 0.137 0.155
Attractiveness 0.259 0.187 0.330 0.000 0.299 0.184 0.415 0.000
Competence 0.217 0.140 0.295 0.000 0.125 0.014 0.237 0.034
Sociability 0.063 -0.015 0.141 0.115 0.035 -0.129 0.199 0.680
Morality 0.143 0.049 0.237 0.003 0.149 -0.061 0.358 0.171
P_Age -0.025 -0.236 0.186 0.816 -0.046 -0.301 0.210 0.727
P_Female 0.193 -0.024 0.410 0.084 0.101 -0.166 0.369 0.458
P_Sci 0.276 0.060 0.492 0.014 0.147 -0.114 0.408 0.273
Table S14. Regression coefficients for Interest Judgments when data from Studies 1 and 2 are pooled,
with separate analyses for each gender and ethnicity.
16
Table S15. Regression coefficients for “Good Scientist” Judgments when data from Studies 1 and 2 are
pooled, with separate analyses for each gender and ethnicity.
“GOOD SCIENTIST” JUDGMENTS
Male Faces Female Faces
B CIlow CIhigh p B CIlow CIhigh p
Intercept 5.808 5.625 5.992 0.000 5.564 5.335 5.792 0.000
Age 0.141 0.070 0.212 0.000 -0.155 -0.261 -0.049 0.005
Non-white 0.063 0.008 0.117 0.024 0.035 -0.058 0.128 0.462
Physics 0.026 -0.011 0.064 0.174 0.026 -0.044 0.096 0.462
Attractiveness -0.271 -0.350 -0.193 0.000 -0.450 -0.579 -0.321 0.000
Competence 0.596 0.511 0.681 0.000 0.511 0.402 0.620 0.000
Sociability -0.134 -0.204 -0.065 0.000 -0.161 -0.296 -0.026 0.021
Morality 0.161 0.071 0.252 0.001 0.132 -0.027 0.291 0.107
P_Age -0.014 -0.199 0.170 0.879 -0.013 -0.242 0.215 0.911
P_Female 0.049 -0.137 0.234 0.608 0.186 -0.044 0.415 0.115
P_Sci 0.028 -0.156 0.211 0.767 0.409 0.184 0.634 0.001
White Faces Non-white Faces
B CIlow CIhigh p B CIlow CIhigh p
Intercept 5.701 5.517 5.885 0.000 6.260 6.030 6.490 0.000
Age 0.089 0.019 0.159 0.014 0.027 -0.102 0.157 0.682
Female 0.001 -0.074 0.076 0.979 -0.058 -0.202 0.086 0.434
Physics 0.033 -0.004 0.069 0.082 0.022 -0.077 0.121 0.663
Attractiveness -0.343 -0.429 -0.257 0.000 -0.260 -0.399 -0.121 0.001
Competence 0.606 0.523 0.688 0.000 0.507 0.370 0.643 0.000
Sociability -0.150 -0.222 -0.079 0.000 -0.027 -0.231 0.178 0.800
Morality 0.172 0.073 0.271 0.001 0.136 -0.113 0.384 0.289
P_Age -0.012 -0.198 0.173 0.896 -0.006 -0.222 0.209 0.955
P_Female 0.069 -0.118 0.255 0.471 0.101 -0.114 0.315 0.360
P_Sci 0.100 -0.085 0.284 0.292 0.076 -0.135 0.288 0.480
17
Table S16. Regression coefficients for Studies 3 and 4.
Predictor B CIlow CIhigh p
Study 3 Video 0.104 0.030 0.178 .006
Physics -0.096 -0.178 -0.013 .023
Gender -0.017 -0.098 0.064 .682 Video*Physics -0.024 -0.106 0.059 .574
Video*Female 0.056 -0.025 0.136 .178 Physics*Female 0.118 0.025 0.212 .013
Video*Physics*Female 0.075 -0.018 0.168 .116 P_Age -0.089 -0.164 -0.014 .020 P_Female -0.134 -0.210 -0.058 <.001 P_Sci -0.028 -0.104 -0.048 .467 Study 4 Competence 0.083 0.007 0.158 .032 Attractiveness 0.052 -0.027 0.131 .196 Comp*Att -0.059 -0.129 0.010 .093
P_Age 0.104 -0.014 0.223 .084
P_Female 0.124 0.007 0.240 .039 P_Sci 0.380 0.261 0.499 <.001 Comp*P_Female -0.051 -0.129 0.026 .196 Comp*P_Age -0.023 -0.099 0.054 .564
Comp*P_Sci 0.029 -0.049 0.107 .471
Att*P_Female 0.009 -0.072 0.091 .821 Att*P_Age 0.060 -0.021 0.140 .148 Att*P_Sci -0.038 -0.120 0.043 .357
18
Table S17. Regression coefficients for Studies 5 and 6
Predictor B CIlow CIhigh p
Study 5 Facetype 0.161 0.083 0.238 <.001 Physics 0.119 0.028 0.210 .011 Female 0.017 -0.074 0.108 .716
Facetype*Physics 0.001 -0.076 0.078 .978 Facetype*Female 0.024 -0.054 0.101 .550
Physics*Female 0.009 -0.082 0.101 .844
Facetype*Physics*Female -0.051 -0.129 0.026 .193 P_Female 0.081 -0.014 0.175 .094
P_Age -0.068 -0.160 0.024 .150 P_Sci 0.150 0.055 0.245 .002
Study 6 Competence 0.142 0.104 0.179 <.001
Attractiveness -0.017 -0.053 0.020 .368
Comp. * Att. -0.016 -0.052 0.021 .402
P_Female 0.102 0.041 0.163 .001
P_Age -0.080 -0.140 -0.020 .009
P_Sci 0.094 0.033 0.155 .003
Comp*P_Female 0.001 -0.037 0.040 .950
Comp*P_Age 0.013 -0.024 0.050 .497
Comp*P_Sci 0.037 -0.002 0.075 .060
Att*P_Female -0.010 -0.047 0.028 .610
Att*P_Age 0.006 -0.031 0.042 .758
Att*P_Sci -0.022 -0.059 0.015 .252
19
Additional Study
We conducted an additional face-rating study using 200 neutral-expression photos from the Park
Aging Mind Face Database [2], comprising 25 men and 25 women in each of 4 age-bands (18-29; 30-
49; 50-69; 70-94), with 10-15% non-white faces per group. Using an MTurk sample, we had 80
participants rate the faces on the same traits as in Study 1; the faces were randomly divided into 2
sets and 40 participants rated all the faces in each set on all dimensions. A further 30 participants
gave Interest judgments for each face and 30 gave “Good scientist” judgments, like those in Studies 1
and 2. In addition, 31 participants were asked to indicate how far each person looked like “a scientist”
(Scientist judgments; this differs from the other judgments, which are predicated upon the person
being a scientist, and was an exploratory variable). Other aspects of the procedure and analysis were
the same as for Studies 1 and 2.
The regression coefficients are shown in Table S18. The pattern is similar to our main studies: the
confidence intervals are often somewhat wider, most likely because of the smaller samples and more
heterogeneous stimuli, but interest was again greater for attractive and competent-looking faces and
for older individuals (although there was little effect of perceived morality), and good-scientist
judgments were positively related to apparent competence and morality but negatively associated
with attractiveness and perceived sociability. “Good Scientist” ratings were also lower for non-white
than for white faces, possibly because there were more African-American faces in this face-set. Also in
contrast to our mains studies is the finding that females received lower “Good Scientist” ratings than
did males.
20
INTEREST JUDGMENTS
B CIlow CIhigh p
Age 0.198 -0.047 0.442 0.122
Female -0.048 -0.185 0.088 0.492
Non-white 0.051 -0.028 0.131 0.213
Attractiveness 0.199 0.016 0.382 0.040
Competence 0.197 0.035 0.358 0.022
Sociability 0.079 -0.059 0.217 0.263
Morality 0.003 -0.161 0.167 0.973
P_Age 0.472 -0.016 0.960 0.068
P_Female 0.279 -0.229 0.787 0.290
P_Sci 0.408 -0.082 0.899 0.113
“GOOD SCIENTIST” JUDGMENTS
B CIlow CIhigh p
Age -0.163 -0.402 0.076 0.185
Female -0.249 -0.428 -0.069 0.009
Non-white -0.116 -0.221 -0.011 0.034
Attractiveness -0.161 -0.359 0.036 0.112
Competence 0.914 0.724 1.104 0.001
Sociability -0.204 -0.434 0.027 0.085
Morality 0.233 -0.058 0.524 0.119
P_Age 0.055 -0.362 0.471 0.799 P_Female -0.243 -0.659 0.173 0.262 P_Sci 0.417 0.000 0.833 0.06
SCIENTIST JUDGMENTS
B CIlow CIhigh p
Age -0.268 -0.483 -0.053 0.016
Female -0.182 -0.346 -0.018 0.034
Non-white -0.028 -0.122 0.066 0.557
Attractiveness -0.355 -0.577 -0.132 0.002
Competence 1.074 0.866 1.282 0.000
Sociability -0.141 -0.390 0.109 0.270
Morality 0.134 -0.188 0.455 0.416
P_Age 0.073 -0.258 0.405 0.667
P_Female -0.297 -0.651 0.057 0.110
P_Sci 0.038 -0.326 0.401 0.841
Table S18. Regression coefficients from Supplementary Study.
21
References
[1] Faul, F., Erdfelder, E., Lang, A-G., & Buchner, A. (2007). G*Power 3: A flexible statistical power
analysis program for the social, behavioral, and biomedical sciences. Behavior Research
Methods, 39(2), 175-191.
[2] Minear, M., & Park, D.C. (2004). A lifespan database of adult facial stimuli. Behavior Research
Methods, Instruments, & Computers, 36(4), 630-633.
[3] Willis, J., & Todorov, A. (2006). First impressions: Making up your mind after a 100-ms exposure to
a face. Psychological Science, 17(7), 592-598.