The Quarterly Journal of Experimental Psychology Internal ...ward-lab.bangor.ac.uk › pubs ›...

PLEASE SCROLL DOWN FOR ARTICLE

This article was downloaded by: [Bangor Universtiy]On: 5 July 2010Access details: Access Details: [subscription number 789519458]Publisher Psychology PressInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

The Quarterly Journal of Experimental PsychologyPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t716100704

Internal facial features are signals of personality and healthRobin S. S. Kramera; Robert Warda

a Bangor University, Bangor, UK

First published on: 17 May 2010

To cite this Article Kramer, Robin S. S. and Ward, Robert(2010) 'Internal facial features are signals of personality andhealth', The Quarterly Journal of Experimental Psychology,, First published on: 17 May 2010 (iFirst)To link to this Article: DOI: 10.1080/17470211003770912URL: http://dx.doi.org/10.1080/17470211003770912

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article may be used for research, teaching and private study purposes. Any substantial orsystematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply ordistribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date. The accuracy of any instructions, formulae and drug dosesshould be independently verified with primary sources. The publisher shall not be liable for any loss,actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directlyor indirectly in connection with or arising out of the use of this material.

http://www.informaworld.com/smpp/title~content=t716100704

http://dx.doi.org/10.1080/17470211003770912

http://www.informaworld.com/terms-and-conditions-of-access.pdf

Internal facial features are signals of personalityand health

Robin S. S. Kramer and Robert WardBangor University, Bangor, UK

We investigated forms of socially relevant information signalled from static images of the face. Wecreated composite images from women scoring high and low values on personality and health dimen-sions and measured the accuracy of raters in discriminating high from low trait values. We also lookedspecifically at the information content within the internal facial features, by presenting the compositeimages with an occluding mask. Four of the Big Five traits were accurately discriminated on the basisof the internal facial features alone (conscientiousness was the exception), as was physical health. Theaddition of external features in the full-face images led to improved detection for extraversion andphysical health and poorer performance on intellect/imagination (or openness). Visual appearancebased on internal facial features alone can therefore accurately predict behavioural biases in theform of personality, as well as levels of physical health.

Keywords: Face; Evolution; Personality; Health; Signal.

The face can be used to predict a person’s behav-iour. Some transient emotional states, such assurprise or fear, are indicated by motion withinthe face. And although we are frequently warnedthat appearances are deceiving, recent evidencealso suggests that static properties of the face aresimilarly expressive. Some reviewers have suggestedthat the face is a visible indicator of sex hormonelevels (e.g., Fink & Penton-Voak, 2002; Johnston,Hagel, Franklin, Fink, & Grammer, 2001). To theextent that sex hormones direct action, the facewould then be a predictor of any such hormonallydriven behaviours. For example, Swaddle andReierson (2002) note that levels of testosterone inmen are associated with both the development ofthe jaw-line and levels of aggressive behaviour

(Mazur & Booth, 1998). Shape of the jaw maytherefore be an accurate predictor of dominancebehaviours in men (Swaddle & Reierson, 2002).In women, ovulation is associated with bothvisible changes in facial attractiveness (Penton-Voak et al., 1999) and a change in sexual interestsand potential sexual behaviours (Gangestad,Thornhill, & Garver, 2002).

More recently, static properties of the face havebeen associated with enduring behavioural biases inthe form of personality. Research has found thatraters could identify certain personality traits ofstrangers (individually or in the form of compo-sites) at a level significantly above chance, basedonly on a photograph of the face with a neutralexpression (Little & Perrett, 2007; Penton-Voak,

Correspondence should be addressed to Robin S. S. Kramer, School of Psychology, Bangor University, Bangor, Gwynedd, UK.

E-mail: [email protected]

# 2010 The Experimental Psychology Society 1http://www.psypress.com/qjep DOI:10.1080/17470211003770912

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY

iFirst, 1–15

Downloaded By: [Bangor Universtiy] At: 14:09 5 July 2010

Pound, Little, & Perrett, 2006; Shevlin, Walker,Davies, Banyard, & Lewis, 2003). Using compo-sites based on the Big Five traits, extraversion,conscientiousness, and agreeableness were ident-ified accurately (Little & Perrett, 2007).Boothroyd, Jones, Burt, DeBruine, and Perrett(2008) have shown that indications of sociosexualorientation are also available from static faceimages. So not only are people quite willing tomake personality and other judgements on thebasis of appearance and other “thin slices”, butthese judgements can be accurate.

The face may also provide a visible signal ofhealth, although the picture is not yet certain. Apowerful theoretical standpoint is that preferencesin attractiveness have evolved to guide matechoice. By this perspective, attractiveness shouldbe a useful cue to traits of great adaptive impor-tance, such as fertility and health (e.g.,Grammer, Fink, Møller, & Thornhill, 2003).Significant effects of facial attractiveness andhealth have been found (e.g., Rhodes, Chan,Zebrowitz, & Simmons, 2003), but also a conceal-ing effect has been reported, in that ratings ofhealth are more accurate when effects of attractive-ness are partialled out (Kalick, Zebrowitz,Langlois, & Johnson, 1998). Further examinationof the database used by Kalick et al. suggested asmall correlation between health and averageness(an r around –.1 between health and “distinctive-ness”), but no relationship with face symmetry(Rhodes et al., 2001). There was no correlationof perceived femininity with actual health, but asmall correlation between perceived masculinityand health (Rhodes et al., 2003). An ongoingdebate is therefore the extent to which facialattractiveness indicates health—for example,there is disagreement about the importance offluctuating asymmetry as an indicator of health(see Weeden & Sabini, 2005, and a responsefrom Grammer, Fink, Møller, & Manning, 2005).

With these findings in mind, we decided toinvestigate whether faces accurately signal healthand personality. Given the importance of theirfindings, our first aim was to replicate the mainresults of Penton-Voak et al. (2006) and Littleand Perrett (2007), showing that aspects of

personality were discernible from static facialimages alone. Specifically, we looked to seewhether composite images, formed from womenwith high and low personality trait values, couldbe accurately identified. For example, when pre-sented with one composite made from the facesof extraverted women and another compositemade from introverted women, could observersidentify which is which? Second, we wished tolook again at the issue of health and appearance.Here we were specifically interested in therelationship between attractiveness and health(e.g., Grammer et al., 2003), but in some sensethe more fundamental issue of whether healthcan be accurately estimated from the face. Wetherefore looked to see whether composite faceimages from women of high and low self-reportedhealth could be accurately identified. The use ofcomposite images would effectively minimize anyinfluence of fluctuating asymmetry, so that accu-rate health identification would have to rest onother factors.

Finally, we sought to develop a method fordetermining where in the face information relatingto personality and/or health was carried.Specifically, we tested (a) whether the internal fea-tures corresponding to the area around the eyes,nose, and mouth were sufficient for trait recog-nition; and (b) for which traits did other, noninter-nal, features contribute to identification?

GENERAL METHOD

Experiments 1 and 2 each consisted of a shortrating task of about 5 to 10 minutes. Participantscompleted both experiments, presented in coun-terbalanced order between participants. For expo-sition, it is simpler and clearer to consider theresults of each task as separate experiments.

Stimuli: The composite images

For both experiments, composite face images weremade from facial photos and inventories of person-ality and health, taken from a pool of 63 Caucasianwomen undergraduates (age M ¼ 21.03, SD ¼

2 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 0000, 00 (0)

KRAMER AND WARD


1.94, age unavailable for 4 participants). Coursecredit was given for participation. Each womancompleted the Mini-IPIP (IPIP: InternationalPersonality Item Pool) personality inventory(Donnellan, Oswald, Baird, & Lucas, 2006) andthe Short-Form 12-Item Health Survey (SF-12;Ware, Kosinski, & Keller, 1996). The SF-12provides both a physical component summary(PCS) and a mental component summary (MCS).Digital photographs of each woman’s face weretaken by a professional photographer using pro-fessional-quality camera, lighting, and reflectors.Photos were constrained to reflect neutralexpression, eyes on the camera; consistent posture,lighting, and distance to the camera; no glasses;jewellery, or make-up if possible; and hair back.

The 15 highest and lowest scorers were ident-ified on each of seven traits: Big Five traits fromthe Mini-IPIP (agreeableness, extraversion, con-scientiousness, neuroticism, and intellect/imagin-ation) and physical and mental health (basedupon the PCS and MCS subscales of the SF-12).Separate composite images were made for thehigh and low scorers using Abrosoft FantaFaceMixer, based on 112 key locations within the faceand around the face outline. In addition, anaverage composite face was made for the entiregroup of 63 women. (All composite images canbe seen in Figure 1.)

Differences between traitsNot surprisingly, the participants selected for thehigh and low composites differed significantlyalong the selected trait (all ps , .001), but therewere also a few other differences. As might beexpected, there was overlap in the measures ofmental health, as assessed by the SF-12, and neur-oticism. The high mental health group had signifi-cantly lower neuroticism than the low mentalhealth group, t(28) ¼ 5.00, p , .001; and likewise,the low neuroticism group had higher mentalhealth scores than did the high neuroticismgroup, t(28) ¼ 6.77, p , .001. The overlap ofthese measures simply reflects their similardomains. In addition, the high extraversion grouphad significantly higher mental health scoresthan the low extraversion group, t(28) ¼ 2.44,

Figure 1. Composite faces based on self-reported personality (Mini-

IPIP, International Personality Item Pool; Donnellan et al., 2006)

and health (SF-12, Short-Form 12-Item Health Survey; Ware

et al., 1996). To view a colour version of this figure, please see the

online issue of the Journal.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 0000, 00 (0) 3

INTERNAL FACIAL FEATURES


p ¼ .021; and the low agreeableness group hadlower conscientiousness scores than the highagreeableness group, t(28) ¼ 2.17, p ¼ .038. Thepotential implications of these differences are con-sidered later. There were no other significantdifferences.

Internal face imagesThe composites were converted to greyscale tominimize any skin tone differences and werecropped to produce images where only the internalfeatures were visible (see Figure 2). By presentingonly this limited region of the composites, wecould explore whether the internal features of theface alone carried both health and personalityinformation.

EXPERIMENT 1: ACCURACY OFTRAIT IDENTIFICATION FROMFULL AND INTERNAL FACES

Here we measured accuracy in discriminatingcomposites made from the high and low valuescorers on each trait. The high and low value com-posite faces were presented together, along with adiscrimination question relevant to the trait (seeFigure 3). Participants judged which of the twofaces better fitted the question.

Method

DesignThe experiment was defined by two factorsdescribing the stimulus images: Trait (agreeable-ness, conscientiousness, extraversion, neuroticism,intellect/imagination, physical health, mentalhealth) × Face Type (full face or internal featuresonly). Face type was varied between participants,trait within.

ParticipantsThere were 131 participants (92 females; age M ¼

20.99, SD ¼ 2.33), including 59 of the 63 womenwho contributed to the stimulus creation pool.These women plus 31 men completed theexperiment using the full faces, for class credit.The remaining 41 participants (33 females) werenot in the class and completed the task usingthe internal face composites for printing credits.All participants were undergraduate students inthe Psychology programme at Bangor University.

ProcedureOn each of the 28 trials, the high and low compo-sites for a trait were presented to the participant(image size of 489 × 489 pixels, or about 13 ×13 cm on a 96-dpi screen), one to the left andone to the right of centre. Viewing distance wasnot fixed. The task was to judge which facebetter suited the discrimination statement appear-ing beneath the composite pair. Participantsindicated their answer using the mouse to clickon the appropriate image, and the next trial thenappeared. The experiment was self-paced, andparticipants were encouraged to make their bestanswer.

Each composite pair was presented four times,each time with a different discrimination state-ment. For the Big Five traits, the discriminationstatements were taken directly from the four rel-evant questions of the Mini-IPIP inventory usedfor scoring the women in the stimulus pool. Forphysical and mental health, the discriminationstatements were taken from four items of thePCS and MCS subscales of the SF-12. The fouritems chosen were the ones producing the largest

Figure 2. The “high physical health” composite, converted to black

and white and cropped so that only the internal features are visible.


KRAMER AND WARD


contributions to subscale scores for the women inour stimulus pool. For the PCS, we used discrimi-nation statements based on Items 1 (health isbetter), 2b (has greater difficulty climbing stairs),3a (accomplishes less due to health problems),and 5 (pain interferes more with work). For theMCS, we used discrimination statements basedon Items 4a (accomplishes less due to emotionalproblems), 4b (works less carefully due to emotion-al problems), 6a (feels more calm and peaceful),and 6c (more often feels downhearted and low).The order of face pairs and questions was random-ized for each participant. The presentation of highand low composites was balanced for field of pres-entation, both for individual participants and forthe four questions used to assess each trait.

Before beginning the rating exercise, each par-ticipant also completed a computerized version ofthe Mini-IPIP personality inventory and the SF-12.

Results and discussion

There were three main findings. First, we replicatedprevious results showing that many personalitytraits can be accurately judged from static facialfeatures (Little & Perrett, 2007; Penton-Voaket al., 2006). Second, we found that physical

health is also reflected in static facial composites.Third, we found that the internal features can bythemselves carry much of the information used forpersonality and health judgements, although therewere some elaborations and exceptions to thisgeneral rule. We now consider these points in turn.

Figure 4 shows discrimination accuracy for eachtrait. For the most part, traits clustered into twosets: one clearly at chance levels (conscientiousness,mental health), and the other set well above chance(agreeableness, extraversion, neuroticism, andphysical health, all ps , .001, in these cases withboth full and internal faces). Intellect/imaginationwas an interesting exception: Identification wassignificantly below chance levels with full faces,t(89) ¼ 2.27, p ¼ .025, yet well above chancewith internal features only, t(40) ¼ 4.93, p ,

.001. Internal features alone therefore allowed foraccurate discrimination for four of the Big Fivepersonality traits (conscientiousness was the excep-tion), as well as physical health.

Figure 5 focuses on the difference in accuracyfor full and internal faces. This difference indicatesthe benefits or costs of external features on identi-fication. As evident from the figure, there were threesignificant differences between full and internalcomposites. External features contributed to

Figure 3. An example stimulus display. Participants clicked on the face that better matched the discrimination statement. To view a colour

version of this figure, please see the online issue of the Journal.




accurate discrimination for both physical healthand extraversion, ts(129) . 3.70, ps , .001. Forphysical health, the PCS subscale includes ques-tions in which excess body weight might produce

lower scores (for example, “my health limits me inclimbing several flights of stairs”). Inspection ofFigure 1 shows that for full faces, there is evidenceof additional body weight in the outline of the low

Figure 4. Accuracy on forced-choice (two-alternative) discrimination for the Big 5 personality traits and for physical and mental health, as

measured by the appropriate subscales of the Short-Form 12-Item Health Survey (SF-12). Chance performance level is indicated by a line at

50%. Error bars indicate 95% confidence interval and can be used to compare conditions to baseline (i.e., error bars overlapping the 50% line

are not significantly different from chance). Agree ¼ agreeableness; Consci ¼ conscientiousness; Extrav ¼ extraversion; Int/Imag ¼

intellect/imagination; Neurot ¼ neuroticism; Phys Hlth ¼ physical health; Ment Hlth ¼ mental health.

Figure 5. Difference in identification accuracy for full and internal faces. Positive bars indicate greater accuracy for full faces; negative bars

indicate greater accuracy with internal features only. Error bars indicate 95% confidence intervals. Agree ¼ agreeableness; Consci ¼

conscientiousness; Extrav ¼ extraversion; Int/Imag ¼ intellect/imagination; Neurot ¼ neuroticism; Phys Hlth ¼ physical health; Ment

Hlth ¼ mental health.


KRAMER AND WARD


compared to high physical health faces. Cues tobody weight from the face and jaw outline are notavailable in the internal faces, which emphasizethe spatial relationships between facial parts.

Finally, there is the case of intellect/imagination,in which the external features actually produce sys-tematic error in identification, compared to theinternal features alone. Of the four questions onthis subscale, two related to facility with abstractideas and two to imagination. We ran a two-factoranalysis of variance (ANOVA), looking at accuracyon intellect/imagination discrimination as a func-tion of Face Type (full or internal) × QuestionType (abstract ideas or imagination related).External features produced interference on bothestimates of imagination and abstract ideas,evidenced by the main effect of face type, F(1,129) ¼ 24.32, p , .001. However, the two-wayinteraction of Face Type × Question Type wasmarginal, F(1, 129) ¼ 3.58, p ¼ .061, althoughthe form of the interaction was such that there waslittle effect of question type for full faces, and thebenefit for internal over full faces was greater withimagination than with intellect questions.

EXPERIMENT 2: PERCEIVEDATTRACTIVENESS AND HEALTH

Experiment 1 demonstrated accurate perception ofphysical health, extraversion, agreeableness, andneuroticism from the face. Internal features alonewere sufficient for better than chance recognitionof all these traits. However, external features, atleast in combination with the internal ones,improved accuracy of physical health and extraver-sion judgements. In contrast, intellect/imagin-ation, especially as tapped by imagination, wasapparent from internal features, and externalfeatures were actually misleading.

Do these results reflect accurate discriminationof specific traits? Or is it possible that our resultscould be explained by a more general effect? Theattractiveness “halo”, in which socially desirabletraits are indiscriminately applied to attractivepeople, is one such effect (Dion, Berscheid, &Walster, 1972). The problem with any such

account is that an indiscriminate halo effectcannot by itself explain the main findings ofExperiment 1—namely, the cases of accurate dis-crimination. That is, if socially desirable traitswere assigned to faces in a genuinely indiscriminateway, identification accuracy would be at chance.However, to the extent that perceived attractive-ness is correlated with actual trait measures, thenresponses based on attractiveness could producecorrect identification. Suppose, purely for illus-tration, that attractive people simply had thesocially desirable values of the traits that were accu-rately identified. That is, suppose that attractivewomen were more extraverted, more agreeable,less neurotic, and physically healthier than lessattractive women. Observers could then performwell simply by assigning the more attractive facethe more desirable trait. Other kinds of globalcharacteristics might similarly collect socially desir-able trait values. For example, given the potentialimportance of health in problems of mate choice,it might also have been the case that socially desir-able traits covaried with healthy appearance. Giventhe theoretical interest in the relationship betweenattractiveness and health outlined earlier, and theimportance of perceived health in other contexts(Kramer, Arend, & Ward, 2010), we were inter-ested in a possible health “halo”, in which sociallydesirable traits might be attributed according toperceived health.

In this experiment, we looked at how discrimi-nation performance in Experiment 1 related to theattractiveness of the different composites and tothe perceptions of their physical health. However,it may be important to reemphasize that we arenot investigating a type of halo effect in whichraters are indiscriminately applying socially desir-able traits to attractive people. Instead, we arelooking at the possibility that people who arerated as attractive (or healthy looking) actuallyhave socially desirable traits.

Method

DesignThe experiment was defined by two factors describ-ing the stimulus images: Trait (agreeableness,




conscientiousness, extraversion, neuroticism, intel-lect/imagination, physical health, mental health)× Face Type (full face or internal features only).Face type was varied between participants, traitwithin.

StimuliThe same images were used as those inExperiment 1.

ParticipantsAs described previously, all participants fromExperiment 1 took part in this experiment.

ProcedureParticipants rated single face images for physicalhealth and attractiveness, in separate blocks.Images were presented one at a time in thecentre of the screen, the same size as inExperiment 1. Under the image would appear areminder phrase indicating the task for thatblock (e.g., “How attractive is this face?”), andunder that reminder, a written 7-item scale (e.g.,very unattractive; unattractive; slightly unattrac-tive; average; slightly attractive; attractive; veryattractive). A similar scale was used for physicalhealth ratings (very unhealthy; unhealthy; slightlyunhealthy; average; slightly healthy; healthy; veryhealthy). We also included a similar block inwhich participants rated relationship preferencefor the face, but technical errors in presentationof the scale invalidated subsequent analysis.Participants clicked on the appropriate ratingwith the mouse, and the next image then appeared.

Blocks were presented in counterbalanced orderacross participants. Prior to each block of trials, aninstruction screen appeared showing an array of thefaces about to be rated and instructions on therating task to be performed (e.g., “In this sectionyou will be judging the ATTRACTIVENESS ofthe faces above. Please take a moment to considerthe range of attractiveness in these faces.”).

Results and discussion

We first wanted to confirm that ratings were equiv-alent for participants familiar and unfamiliar with

the women in the stimulus pool. We did notexpect any difference, as in a composite of 15faces, the identities of the individual faces seemedimpossible to discern. Other reports have suggestedthat individual faces are effectively disguised withincomposites of even six faces (Little & Hancock,2002). When we correlated the ratings given bythe two groups to each of the 15 face stimuli, wefound high correlations both for attractiveness,r(13) ¼ .91, p , .001, and for physical health,r(13) ¼ .75, p ¼ .001. In addition, the twogroups did not differ on attractiveness ratings,t(14) ¼ 0.27, p ¼ .793, or health ratings, t(14) ¼0.60, p ¼ .556. The ratings of the two groupswere therefore combined in further analyses.

There was general agreement in the attractive-ness ratings and, to a lesser degree, the healthratings given to the full and internal faces. Forattractiveness, the correlation between the meanratings given to each full and correspondinginternal face image was r(13) ¼ .66, p ¼ .008;the agreement on health was marginal, r(13) ¼.47, p ¼ .077.

However, our main focus is on the detailedrelationship between accuracy of trait discrimi-nation and the health and attractiveness ratings,as computed separately for full and for internalface images. Below, we present the completeresults for each rating (attractiveness and health)and then selectively highlight interesting findings.To summarize, attractiveness and perceived healthappear to be honest signals of extraversion andphysical health, even when information is limitedto internal face features. However, there arenumerous cases in which discrimination perform-ance and ratings of these signals appear to beindependent.

AttractivenessWe first consider attractiveness ratings. Figure 6shows the difference in the mean attractivenessrating given to the high and the low value compo-site for each trait. The difference is separatelyshown for full and internal face images.

To better understand the attractiveness differ-ences illustrated in Figure 6, we compared differ-ences in attractiveness with differences in


KRAMER AND WARD


discrimination accuracy in Experiment 1. Thesecomparisons are show in Table 1.

Table 1 separately summarizes the results forattractiveness and discrimination accuracy for thefull and internal face stimuli. Beginning with thefull faces (Table 1a), for three of the seven traits—agreeableness, extraversion, and physical health—

above-chance discrimination was accompanied bysignificant differences in attractiveness, such thatattractive composites possessed the more sociallydesirable trait levels. However, this relationshipbetween attractiveness and socially desirable traitsdid not hold across the board. Raters inExperiment 1 were unable to accurately discriminate

Figure 6. Difference in attractiveness for high and low trait composites. Attractiveness was rated on a 7-point scale from “very unattractive”

(1) to “very attractive” (7) for all faces. Positive bars indicate greater attractiveness for high faces, negative bars indicate greater attractiveness

for low faces. Error bars indicate 95% confidence intervals. Agree ¼ agreeableness; Consci ¼ conscientiousness; Extrav ¼ extraversion; Int/Imag ¼ intellect/imagination; Neurot ¼ neuroticism; Phys Hlth ¼ physical health; Ment Hlth ¼ mental health.

Table 1. Comparison of differences in attractiveness with differences in trait discrimination accuracy

Difference

in attractiveness?

Discriminated trait levels?

Faces Yes No

a. Full Yes Extraversion,

Physical health,

Agreeableness

Conscientiousness,

Mental health

No Neuroticism Intellect/Imagination

b. Internal Yes Extraversion,

Physical health

Conscientiousness

No Agreeableness,

Neuroticism,

Intellect/Imagination

Mental health

Note: Breakdown of traits according to whether high and low composites were accurately identified in Experiment 1 and whether there

was a difference in attractiveness such that the more socially desirable trait pair was rated higher in attractiveness in Experiment 2.




conscientiousness, even though the same raters inthe current experiment found the high conscien-tiousness face significantly more attractive thanthe low. Similarly, raters in Experiment 1 wereunable to discriminate levels of mental health,even though they found the low mental healthface more attractive than the high. The reversepattern was also found. In Experiment 1, raterswere able to discriminate levels of neuroticism inthe full faces, even though in the present experimentboth faces were rated equally attractive.

Attractiveness likewise does not provide a goodexplanation of discrimination for internal faceimages, summarized in Table 1b. As with thefull faces, extraversion and physical health wereaccurately identified in Experiment 1 and alsoshowed a significant difference in attractiveness.However, agreeableness, neuroticism, and intel-lect/imagination were also accurately identifiedfrom internal faces in Experiment 1, even thoughthe high and low values of each were ratedequally attractive. Finally, the low conscientious-ness internal face was rated significantly moreattractive than the high, but there was no accuratediscrimination of these items in Experiment 1.

A consistent result therefore with both full andinternal faces is that high levels of extraversion andphysical health are reflected in attractive faces. Inthis context, it is also interesting to note that extra-version and physical health were the two traits thatbenefited significantly from information outsidethe internal faces (Figure 4). That is, there is infor-mation present in the full, and to a lesser extent, injust the internal facial features, which both isattractive to look at and serves as an honestsignal of extraversion and physical health.However, attractiveness is not associated with alldiscriminable personality traits, and not all dis-criminable personality traits are reflected in corre-sponding attractiveness. The pattern of accurateperformance in Experiment 1 is therefore notfully explained by an association of socially desir-able traits and attractiveness.

While the above analysis investigates howattractiveness judgements at the group levelrelate to accurate trait perception, it does notaddress participants’ decisions at the individual

level. We therefore carried out regression analysesto investigate whether differences in individualparticipants’ ratings of attractiveness for the twocomposites (high minus low) predicted their accu-racy—that is, did individual ratings predict sub-sequent discrimination? Of the seven traits forthe full face judgements, only neuroticism accuracywas predicted by attractiveness ratings, b ¼ –.32,p ¼ .014 (Bonferroni corrected). For the internalfaces, attractiveness did not predict accuracy forany of the traits. Again, this highlights attractive-ness as unable to satisfactorily explain accuracy ofperception in these judgements.

HealthA similar analysis was performed for perceivedphysical health (see Figure 7) and differences inperceived health compared with discriminationaccuracy (Table 2). For the internal faces, healthand attractiveness scores were also highly corre-lated, r(13) ¼ .70, p ¼ .004. However, withthe sole exception of mental health, t(40) ¼ 1.63,p ¼ .111, all internal face pairs were perceived toreflect significantly different levels of physicalhealth, all ts(40) . 2.60, all ps , .013. In thesecase, accurate discrimination of many traits couldtherefore conceivably be explained by a “healthhalo”, such that healthy-looking people are not justperceived to have socially desirable traits, but actuallydo possess these traits. However, again, this cannotbe a complete account of our findings—the traitpairs for conscientiousness differed in perceivedhealth but could not be accurately discriminated.

Further exceptions to a perceived-health haloare found in the data for full faces. Health andattractiveness scores for the full faces were highlycorrelated, r(13) ¼ .91, p , .001, and thegeneral pattern of results is similar to that forattractiveness. Table 2a shows there were threecases in the full-face data in which socially desir-able traits were both accurately discriminated andwere also perceived as more healthy: agreeableness,extraversion, and actual physical health. But again,as with the attractiveness ratings, there were traitpairs that differ in perceived health but were notaccurately discriminated (mental health) and traitpairs that were accurately discriminated but were


KRAMER AND WARD


of equivalent apparent health (neuroticism). Insummary, it seems unlikely that attractiveness orperceived health can explain all cases of accurateidentification that we observed in Experiment 1.

As with attractiveness, we ran regression ana-lyses with individual ratings of perceived healthas a factor predicting trait accuracy. For full faces,perceived health differences predicted accuracy

Figure 7. Difference in perceived health for high and low trait composites. Health was rated on a 7-point scale from “very unhealthy” (1) to

“very healthy” (7) for all faces. Positive bars indicate greater perceived health for high faces; negative bars indicate greater perceived health for

low faces. Error bars indicate 95% confidence intervals. Agree ¼ agreeableness; Consci ¼ conscientiousness; Extrav ¼ extraversion; Int/Imag ¼ intellect/imagination; Neurot ¼ neuroticism; Phys Hlth ¼ physical health; Ment Hlth ¼ mental health.

Table 2. Comparison of differences in perceived health with differences in trait discrimination accuracy

Faces

Difference

in perceived health?

Discriminated trait levels?

Yes No

a. Full Yes Extraversion,

Physical health,

Agreeableness

Mental health

No Neuroticism Conscientiousness,


b. Internal Yes Extraversion,

Physical health,

Agreeableness,

Neuroticism,


Conscientiousness

No — Mental health

Note: Breakdown of traits according to whether high and low composites were accurately identified in Experiment 1 and whether

there was a difference in perceived health such that the more socially desirable trait pair was rated higher in perceived physical

health in Experiment 2.




for physical health only, b ¼ .38, p ¼ .001(Bonferroni corrected). For internal faces, nodifferences in ratings predicted accuracy.

These results suggest that discriminations arenot being made simply on the basis of halos relat-ing to perceived attractiveness or apparent health.For example, neither attractiveness ratings at thegroup level, nor those at the level of individualraters, can well explain performance across all thedifferent traits we have measured. Our resultsseem to dissociate raters’ perceptions of attractive-ness and healthy appearance from the ability ofthose raters to accurately judge personality traits.A related issue is to what extent having sociallydesirable traits is associated with attractiveness,independent of whether the trait can be accuratelydiscriminated. For example, our raters could notdiscriminate levels of conscientiousness, eventhough the high conscientious composite wasrated as more attractive than the low. Could it bethe case that attractive, healthy-looking peoplewill tend to have socially desirable traits, even ifthose traits are not accurately perceived by obser-vers? Again, the relationship between attractive-ness and socially desirable personality traits is notstraightforward. For example, from our results itseems that attractive faces are more likely toreflect high than low levels of conscientiousness.However, neither attractiveness nor healthyappearance were associated with low neuroticism,or high intellect/imagination. These resultssuggest a more complex picture than any simpleaccount relating a global measure such as attrac-tiveness to social desirability for a multitude ofpersonality traits.

GENERAL DISCUSSION

Previous work has shown that observers can accu-rately assess aspects of personality based on unfa-miliar, static faces with neutral expressions(Little & Perrett, 2007; Penton-Voak et al.,2006). Our main results, from Experiment 1,show further that internal features of the face,specifically the areas around the eyes, nose, andmouth, carry enough information to allow accurate

judgements relating to physical health, and to fourof the Big Five personality factors: agreeableness,extraversion, neuroticism, intellect/imagination(cf. openness). By comparing accuracy with fullfaces to internal features only, our method alsoallowed us to identify the contribution of externalfeatures (and colour) to identification. Althoughexternal features contributed to accurate identifi-cation of health and extraversion, they actuallyinterfered with accurate judgements of intellect/imagination.

Experiment 2 verified that accuracy did notresult from attractive people simply having moresocially desirable traits. That is, the traits inwhich the composite pair differed in attractivenesswere not necessarily correctly identified, and thetraits that were correctly identified did not necess-arily differ in the attractiveness of the compositepair. Likewise, our results do not seem completelyconsistent with the possibility that healthierlooking people also have simply more sociallydesirable personality traits than less healthylooking people. Analyses of individual predictorsfurther demonstrated that perceived health andattractiveness, while influencing judgements, didnot account for accurate perceptions of personality.

As noted earlier, there were a few cases in whichour composites overlapped in traits other thanthose they were created for. There was no surprisethat MCS and neuroticism dimensions coincidedfor those individuals making up the compositepairs as these scales clearly reflect similardomains. That neuroticism but not MCS was dis-criminated from the images is more surprising,though this may simply reflect that the latter tapsa more general domain that also includesdepression, anxiety, and so on. Alternatively,MCS may just be a less well validated measure ofmental health. In addition, the low agreeablenessgroup had significantly lower conscientiousnessthan the high agreeableness group. This maymean that agreeableness composites were moreeasily discriminated because they differed on twotrait dimensions. However, conscientiousness wasnot accurately discriminated, and so it seemsunlikely that this extra information would havecontributed significantly to participant accuracy.


KRAMER AND WARD


Our results provide a useful replication of thePenton-Voak et al. (2006) and Little and Perrett(2007) findings. These studies used correlationsof rated and actual traits, rather than the forced-choice identification we used. Little and Perrettused composite images, as we did, while Penton-Voak et al. also used individual images. Thesetwo studies asked observers to rate the degree ofa trait (e.g., agreeableness present in the image),whereas we asked observers the same questions asthose that were used to create the personalityratings. Despite these differences in images andobserver tasks, all three studies found accurateidentification of agreeableness and extraversion.Like Little and Perrett, we also find accurateidentification of neuroticism, although in ourinternal as well as full-face images. Little andPerrett noted several potential advantages and dis-advantages in the use of composites. One advan-tage is that traits consistently associated withspecific visual features will have increased signalto noise ratio. The fact that neuroticism is, todate, found more easily with composite thansingle images suggests that the distinguishingvisual characteristics for this trait are only weaklypresent in single images.

We have no compelling account as to why con-scientiousness was accurately identified in theLittle and Perrett (2007) images but not in ours.Similarly, there are differences in their study andours in the attractiveness differences of composites.Little and Perrett only found differences betweenthe high and low agreeableness female composites,whereas we found observed significant differencesin extraversion and conscientiousness as well.This may simply reflect the reliability of traitdifferences across different samples but at presentit is difficult to tell.

Although both Penton-Voak et al. (2006) andLittle and Perrett (2007) investigated personalitydisplays in male and female faces, our currentresearch only explored female composites. Thislimitation was due to our sampling a populationwith a low number of males, and thereforemaking it impossible to produce sufficiently separ-ate composites, and it is likely that the currentresults may differ to those expected from male

composites. Little and Perrett found that malecomposites only differed significantly for extraver-sion and suggest that males may contain fewer cuesin the face to their actual personality than dofemales. This idea finds limited support in the lit-erature, which has shown that women are believedto use more expressive and nonverbal behavioursthan men (Briton & Hall, 1995) and are betternonverbal encoders of facial expressions than men(Rosenthal, Hall, DiMatteo, Rogers, & Archer,1979). However, these relate to dynamic signals,and so further research is required to demonstratetheir applicability to static features.

As noted above, the use of composites couldpotentially lose as well as gain signals. Some pre-vious evidence suggests that fluctuating asymmetry(FA), in the face as well as the body, is a cue todevelopmental integrity and physical health(Thornhill & Møller, 1997). FA within a compo-site image will of course tend to be less than thatin any of the components. However, although FAis an unlikely cue for physical health in the compo-sites, physical health was still accurately identifiedin both full and internal face images. We alsonoted earlier that evidence of excess body weightis much reduced in the internal images. Skinblood colouration is also associated with perceivedhealth (Stephen, Coetzee, Law Smith, & Perrett,2009), with increased redness linked with higherlevels of blood oxygenation, although this cue wasalso not available in the internal face images.While we accept that skin surface properties andFA probably play a role in assessing health, ourresults show that other features also signal healthwhen these cues are minimal. At present, wesuggest that health is signalled through a varietyof cues, including FA and colour, but also thespatial arrangement of internal face features.

We close with some admittedly speculative, butperhaps intriguing, links between our results andtheories of biological signal systems. In thiscontext, we have seen that a signal, in this case,levels of socially desirable traits (such as high orlow levels of agreeableness), are expressed on theface of the “sender” and are accurately detected bythe “receiver” viewing the face. Theories of biologi-cal signal systems emphasize the perspectives of




both the sender and receiver: A signal must be suf-ficiently informative, sufficiently often, that recei-vers benefit from attending to it. That is, if asignal is uninformative or easily faked, there is noadvantage or reason for the receiver to attend toit. Conversely, in a stable-state system, attentionto a signal suggests that there is some net benefitto the receiver in attending. But this very validityopens the possibility of another selective pressure,for the sender to insert occasional deceptive mess-ages, which benefit the sender, possibly at theexpense of the receiver. That is, the receiver maybe manipulated into acting against their own bestinterests (e.g., Dawkins & Krebs, 1978). Whatthen keeps the system “honest”? For example, inthe context of mate choice, an individual whocould display false signals of exaggerated fitnessmight acquire a higher quality mate. In thiscontext, the interests of the sender and receiverare not entirely opposing, but they are divergent,producing a pressure to exaggerate fitness. Why isit then that all faces do not express a socially desir-able personality? A general conclusion from signaltheory is that in such cases of divergent self-inter-ests, a signal will generally not remain informativeunless it entails costs that impact more heavily onless fit individuals (“costly signals”; Grafen, 1990;Zahavi, 1975). An interesting issue may thereforebe identifying costs for expressing socially desirabletraits on the face.

Original manuscript received 3 August 2009

Accepted revision received 23 February 2010

First published online day month year

REFERENCES

Boothroyd, L. G., Jones, B. C., Burt, D. M., DeBruine,L. M., & Perrett, D. I. (2008). Facial correlates ofsociosexuality. Evolution and Human Behavior, 29,

211–218.Briton, N. J., & Hall, J. A. (1995). Beliefs about female

and male nonverbal communication. Sex Roles, 32,

79–90.Dawkins, R., & Krebs, J. R. (1978). Animal signals:

Information or manipulation? In J. R. Krebs &N. B. Davies (Eds.), Behavioural ecology: An

evolutionary approach (1st ed., pp. 282–309).Oxford, UK: Blackwell.

Dion, K. L., Berscheid, E., & Walster, E. (1972). Whatis beautiful is good. Journal of Personality and Social

Psychology, 24, 285–290.Donnellan, M. B., Oswald, F. L., Baird, B. M., &

Lucas, R. E. (2006). The Mini-IPIP scales: Tiny-yet-effective measures of the Big Five factors of per-sonality. Psychological Assessment, 18, 192–203.

Fink, B., & Penton-Voak, I. S. (2002). Evolutionary psy-chology of facial attractiveness. Current Directions in

Psychological Science, 11, 154–158.Gangestad, S. W., Thornhill, R., & Garver, C. E.

(2002). Changes in women’s sexual interests andtheir partners’ mate retention tactics across the men-strual cycle: Evidence for shifting conflicts of inter-est. Proceedings of the Royal Society of London B,269, 975–982.

Grafen, A. (1990). Biological signals as handicaps.Journal of Theoretical Biology, 144, 517–546.

Grammer, K., Fink, B., Møller, A. P., & Manning, J. T.(2005). Physical attractiveness and health: Commenton Weeden and Sabini (2005). Psychological Bulletin,131, 658–661.

Grammer, K., Fink, B., Møller, A. P., & Thornhill, R.(2003). Darwinian aesthetics: Sexual selection andthe biology of beauty. Biological Review, 78,

385–407.Johnston, V. S., Hagel, R., Franklin, M., Fink, B., &

Grammer, K. (2001). Male facial attractiveness:Evidence for a hormone-mediated adaptive design.Evolution and Human Behavior, 22, 251–267.

Kalick, S. M., Zebrowitz, L. A., Langlois, J. H., &Johnson, R. M. (1998). Does human facial attrac-tiveness honestly advertise health? Longitudinaldata on an evolutionary question. Psychological

Science, 9, 8–13.Kramer, R. S. S., Arend, I., & Ward, R. (2010).

Perceived health from biological motion predictsvoting behaviour. The Quarterly Journal of

Experimental Psychology, 63, 625–632.Little, A. C., & Hancock, P. J. B. (2002). The role of

masculinity and distinctiveness in judgments ofhuman male facial attractiveness. British Journal of

Psychology, 93, 451–464.Little, A. C., & Perrett, D. I. (2007). Using composite

images to assess accuracy in personality attributionto faces. British Journal of Psychology, 98, 111–126.

Mazur, A., & Booth, A. (1998). Testosterone and dom-inance in men. Behavioral and Brain Sciences, 21,

353–397.


KRAMER AND WARD


Penton-Voak, I. S., Perrett, D. I., Castles, D. L.,Kobayashi, T., Burt, D. M., Murray, L. K., et al.(1999). Menstrual cycle alters face preference.Nature, 394, 884–887.

Penton-Voak, I. S., Pound, N., Little, A. C., & Perrett,D. I. (2006). Personality judgments from natural andcomposite facial images: More evidence for a “kernelof truth” in social perception. Social Cognition, 24,

490–524.Rhodes, G., Chan, J., Zebrowitz, L. A., & Simmons,

L. W. (2003). Does sexual dimorphism in humanfaces signal health? Proceedings of the Royal Society

of London B (Suppl.), 270, S93–S95.Rhodes, G., Zebrowitz, L. A., Clark, A., Kalick, S. M.,

Hightower, A., & McKay, R. (2001). Do facial aver-ageness and symmetry signal health? Evolution and

Human Behavior, 22, 31–46.Rosenthal, R., Hall, J. A., DiMatteo, M. R., Rogers,

P. L., & Archer, D. (1979). Sensitivity to nonverbal

communication: The PONS test. Baltimore: TheJohns Hopkins University Press.

Shevlin, M., Walker, S., Davies, M. N. O., Banyard, P.,& Lewis, C. A. (2003). Can you judge a book by its

cover? Evidence of self-stranger agreement on per-sonality at zero acquaintance. Personality and

Individual Differences, 35, 1373–1383.Stephen, I. D., Coetzee, V., Law Smith, M., & Perrett,

D. I. (2009). Skin blood perfusion and oxygenationcolour affect perceived human health. PLoS ONE,4, 1–7.

Swaddle, J. P., & Reierson, G. W. (2002). Testosteroneincreases perceived dominance but not attractivenessof human males. Proceedings of the Royal Society of

London B, 269, 2285–2289.Thornhill, R., & Møller, A. P. (1997). Developmental

stability, disease and medicine. Biological Reviews,72, 497–548.

Ware, J. E., Jr., Kosinski, M., & Keller, S. D. (1996). A12 item short form health survey: Construction ofscales and preliminary tests of reliability and validity.Medical Care, 34, 220–233.

Weeden, J., & Sabini, J. (2005). Physical attractivenessand health in Western societies: A review.Psychological Bulletin, 131, 635–653.

Zahavi, A. (1975). Mate selection: A selection for ahandicap. Journal of Theoretical Biology, 53, 205–214.




The Quarterly Journal of Experimental Psychology Internal ...ward-lab.bangor.ac.uk › pubs ›...

Documents

Transcript of The Quarterly Journal of Experimental Psychology Internal ...ward-lab.bangor.ac.uk › pubs ›...