Sexual Selection,Agonistic Signaling,and the Effect of Beards on Recognition of Men’s Anger Displays

Abstract

The beard is arguably one of the most obvious signals of masculinity in humans. Almost 150 years ago, Darwin suggested that beards evolved to communicate formidability to other males, but no studies have investigated whether beards enhance recognition of threatening expressions, such as anger. We found that the presence of a beard increased the speed and accuracy with which participants recognized displays of anger but not happiness (Experiment 1, N = 219). This effect was not due to negative evaluations shared by beardedness and anger or to negative stereotypes associated with beardedness, as beards did not facilitate recognition of another negative expression, sadness (Experiment 2, N = 90), and beards increased the rated prosociality of happy faces in addition to the rated masculinity and aggressiveness of angry faces (Experiment 3, N = 445). A computer-based emotion classifier reproduced the influence of beards on emotion recognition (Experiment 4). The results suggest that beards may alter perceived facial structure, facilitating rapid judgments of anger in ways that conform to evolutionary theory.

Keywords

facial hair emotion recognition face processing intrasexual selection open data

Agonistic interactions between males during competition over resources, status, and mating opportunities occur across the mammalian class and have shaped the evolution of weaponry and threat displays (Darwin, 1871; Emlen, 2008; Kokko, Jennions, & Brooks, 2006). In humans, these displays are manifest in a variety of bodily and facial dimorphisms, of which beardedness is one of the most visually salient (B. J. Dixson & Vasey, 2012; B. J. W. Dixson, Lee, Sherlock, & Talamas, 2017). Beards provide an accurate indication of male sexual maturity, and bearded faces are rated as more masculine, dominant, and aggressive than clean-shaven faces (B. J. Dixson & Brooks, 2013; Muscarella & Cunningham, 1996; Neave & Shields, 2008). These effects stem from the fact that beards grow around the jaw and mouth, and thus emphasize jaw size and masculine facial structure (B. J. W. Dixson et al., 2017; B. J. W. Dixson, Sulikowski, Gouda-Vossos, Rantala, & Brooks, 2016; Sherlock, Tegg, Sulikowski, & Dixson, 2017). Beardedness has a greater influence on ratings of masculinity and dominance than does craniofacial shape or jaw size (B. J. W. Dixson et al., 2017; Sherlock et al., 2017).

The enhancing effects of facial hair on judgments of men’s facial masculinity, dominance, and aggressiveness by framing components of masculine facial shape have been measured using stimuli depicting neutral facial expressions. However, faces carry multiple sources of social information, including emotional facial expressions that can convey internal states and intentions. Facial expressions such as displays of anger can be enacted in agonistic interactions to signal interpersonal threat (Blair, 2003; Schmidt & Cohn, 2001; Sell, Cosmides, & Tooby, 2014; Tay, 2015). It has been hypothesized that beardedness facilitates recognition of threatening displays, including displays of anger, by enhancing masculine facial features related to dominance (particularly jaw size; Blanchard, 2009; Goodhart, 1960; Guthrie, 1970), but to date, there have been no behavioral studies detailing whether beards influence recognition of angry expressions.

Although the influence of facial hair on recognition of expressions of anger has not been directly tested, previous findings suggest that it is plausible that beards facilitate the recognition of anger. Previous research has demonstrated that people are faster to recognize anger when it is displayed on male faces than when it is displayed on female faces (Becker, Kenrick, Neuberg, Blackwell, & Smith, 2007). This influence of masculinity on recognition of anger has been partly attributed to overlap between structural cues of anger and masculinity. It has been suggested that angry facial expressions emphasize masculine facial structures, such as the prominence of the jaw (Becker et al., 2007; Hess, Adams, Grammer, & Kleck, 2009; Sacco & Hugenberg, 2009). Facial hair grows around the areas involved in expressing a range of emotions, including anger, and also enhances masculine craniofacial structure and the prominence of the jaw (B. J. W. Dixson et al., 2017; Sherlock et al., 2017). These observations suggest that the presence of a beard may facilitate recognition of angry expressions.

To test whether beards amplify displays of anger, we presented participants with photographs of standardized expressions of anger and happiness posed by the same men when bearded and clean-shaven. Participants categorized the emotion displayed in each face, and we examined how facial hair affected their speed and accuracy. If participants were faster to recognize anger but not happiness on bearded than on clean-shaven faces, this would indicate that beardedness facilitates recognition of a threatening emotional expression specifically and not emotional expressions more generally. After we found such a specific effect, we explored possible underlying mechanisms in two behavioral experiments and a final experiment with a computer-based emotion classifier.

The protocols for the behavioral experiments were approved by the human-ethics boards at Curtin University and The University of Queensland. All participants gave informed consent and were free to withdraw from the study without prejudice at any point. Online participants received monetary compensation. We report all measures, manipulations, and exclusions. No further data were collected after analyses had been conducted. Follow-up comparisons in all four experiments were conducted using two-tailed paired-samples t tests, and we report Cohen’s d_z (standardized mean difference) as our measure of effect size.

Experiment 1

Method

Participants

A total of 227 participants (97 males, 129 females, 1 who did not respond; mean age = 30.88, SD = 10.22) provided some data for Experiment 1. This experiment was conducted both in the laboratory and online. The in-lab participants were 78 undergraduate students at Curtin University (8 males, 69 females, 1 who did not respond; mean age = 24.85, SD = 9.68). They received partial course credit in return for their participation. The online participants were 149 Amazon Mechanical Turk (MTurk) workers (89 males, 60 females; mean age = 34.10, SD = 8.99).¹

Stimuli

Photographs of 10 young adult Caucasian males (mean age = 23.50, SD = 3.57, range = 20–30) were taken from a stimulus set developed and used in previous research (B. J. Dixson & Vasey, 2012). The men were photographed with happy and angry expressions when clean-shaven and again with a full beard (at least 8 weeks of untrimmed facial-hair growth). The emotional expressions were generated using instructions based on the Facial Action Coding System (Ekman, Friesen, & Hager, 2002). The total stimulus set consisted of 40 photographs (10 bearded angry faces, 10 bearded happy faces, 10 clean-shaven angry faces, and 10 clean-shaven happy faces). The use of this stimulus set is a particular strength of this experiment, as the same individuals were presented posing the same expressions with and without facial hair. This eliminated the influence of possible systematic facial-structure or expression differences between men who choose to be bearded and those who choose to be clean-shaven, and thus is an advancement over previous work using different individuals for the stimuli in the bearded and clean-shaven conditions. These images were edited to remove clothing and backgrounds, and the faces were placed on a uniform gray background 485 × 709 pixels in size (Fig. 1).

Fig. 1.

Examples of the stimuli used in Studies 1 through 4. These images show the same man posing happy, sad, and angry expressions when bearded (upper row) and clean-shaven (lower row).

Procedure

Laboratory experiment

Participants entered a small computer laboratory with four testing terminals and were seated in front of a 24-in. LED monitor with a screen resolution of 1,920 × 1,080 pixels and a refresh rate of 120 Hz. The task was executed using DMDX software (Forster & Forster, 2003). Participants were instructed that faces would appear on the screen one at a time. They were asked to categorize each face as either “happy” or “angry” as quickly and accurately as possible by pressing the right or left shift key. Response mapping was counterbalanced across participants. On each trial, a fixation cross was presented for 1,000 ms on a gray background and was followed by 1 of the 40 photographs, which remained on the screen until a response was made, or for a maximum of 3,000 ms. A blank screen was presented for 500 ms between trials. Each of the 40 images was presented three times, for a total of 120 trials.² Participants provided their age, sex, and ethnicity at the end of the testing session.

Online experiment

The online experiment proceeded similarly to the laboratory experiment, but with a few adaptations to the online testing environment. Inquisit 4 Web software ran the experiment, which participants completed on their personal desktop or laptop computer in a location of their choice. Participants made their responses using the “A” and “L” keys rather than the shift keys. Throughout the task, written reminders of the response mapping were presented in the bottom left and right of the screen (e.g., “Press ‘A’ for angry,” “Press ‘L’ for happy”). Participants completed a short (10-trial) practice task prior to the main task. Feedback was provided after an incorrect response only during the practice task, and at the end of the practice task, participants were given their mean response time and accuracy. Because the screen size and resolution varied across participants, the image was set to take up 35% of the height of the screen with a fixed aspect ratio. A black background (rather than a gray background) was used. All participants completed 120 trials and then responded to a demographic survey asking their age, sex, and ethnicity and whether they had a beard. They were also asked to describe their environment and were given the chance to provide any final comments.

Data processing and analysis

Trials with incorrect responses and trials with response times faster than 100 ms or more than 3 SD faster or slower than the participant’s mean were removed as invalid. Participant sex has been found to moderate the influence of beardedness on person perception in some previous studies (B. J. Dixson & Vasey, 2012). In the current study, we considered it plausible that male and female observers may also differ in their responses. For example, anger cues may be seen as more threatening by women owing to sex differences in physical strength (Puts, 2010). On the other hand, male-male competition has played an important role in the evolution of men’s secondary sexual traits (Puts, 2010), and therefore men might be more sensitive to displays of anger and beardedness than women. However, participant’s sex did not consistently influence performance in the current investigation. Therefore, results are reported collapsed across this variable, but for completeness, we report analyses including participant’s sex in the Supplemental Material available online. Response times and error rates were averaged within each condition, and the condition means for each dependent variable were submitted to a separate 2 (facial hair: bearded, clean-shaven) × 2 (emotional expression: happy, angry) repeated measures analysis of variance (ANOVA).

Additionally, to examine whether a systematic tendency toward responding “angry” or “happy” to a particular face type could account for any response time differences observed, we calculated a measure of response bias (c) separately for bearded and clean-shaven faces using a signal-detection-theory calculator created by Gaetano (2017). Response bias takes into account hits (here conceptualized as the proportion of correct “angry” responses) as well as false alarms (conceptualized as the proportion of incorrect “angry” responses to happy faces). In the present context, c provides an index of whether the participants had a bias toward making more “angry” responses (negative values), no bias (c values around zero), or a bias toward making more “happy” responses (positive values). Corrections for hit and false alarm rates of 0 or 1 were made in line with the recommendations of Stanislaw and Todorov (1999). With this correction, 0 is replaced with the value of 0.5/n, and 1 is replaced with the value (n – 0.5)/n, where n is the number of signal trials (angry expressions) for hit rates and the number of noise trials (happy expressions) for false alarm rates. Because hit and false alarm rates of 0 and 1 correspond to z scores of –∞ and ∞, correction is needed so that finite outputs are provided when response bias is calculated.

Two participants who completed the laboratory-based task could not be included in analyses because they made no valid responses in at least one condition. Five participants in the online task were not included in the analyses because they had error rates approaching chance (> 40% of responses were incorrect or invalid). The participant who did not report his or her sex was not included in analyses so that the data set for the main and supplementary analyses would be consistent, but this exclusion did not change the significance or direction of results.

Results

Response times

The main effect of facial hair was not significant, F(1, 218) = 0.40, p = .529, η_p² < .01, 90% confidence interval (CI)= [.00, .02], and the main effect of emotional expression was also not significant, F(1, 218) = 0.71, p = .402, η_p² < .01, 90% CI = [.00, .03]. However, facial hair and emotional expression interacted to influence response speed, F(1, 218) = 41.42, p < .001, η_p² = .16, 90% CI = [.09, .23]. Participants were significantly faster to categorize anger on bearded than on clean-shaven faces, t(218) = 5.06, p < .001, d_z = 0.34, 95% CI = [0.20, 0.48], but were significantly faster to categorize happiness on clean-shaven than on bearded faces, t(218) = 4.73, p < .001, d_z = 0.32, 95% CI = [0.18, 0.45] (see Fig. 2a).

Fig. 2.

Mean speed of emotion categorization and mean percentage of incorrect responses in (a, b) Experiment 1 (angry vs. happy expressions) and (c, d) Experiment 2 (sad vs. happy expressions). Results are shown separately for bearded and clean-shaven faces. Error bars represent ±1 SEM. For figures displaying the distributions of response times and error rates, see the Supplemental Material available online.

Accuracy

A similar pattern was observed in the error rates. Participants’ accuracy did not differ overall between bearded and clean-shaven faces, F(1, 218) = 1.55, p = .214, η_p² = .01, 90% CI = [.00, .04], but accuracy was significantly better for happy than for angry expressions, F(1, 218) = 3.99, p = .047, η_p² = .02, 90% CI = [.00, .06]. This effect was moderated by a significant Facial Hair × Emotional Expression interaction, F(1, 218) = 30.66, p < .001, η_p² = .12, 90% CI = [.06, .19]. Participants were more accurate in recognizing anger on bearded than on clean-shaven faces, t(218) = 5.12, p < .001, d_z = 0.35, 95% CI = [0.21, 0.48], but were more accurate in recognizing happiness on clean-shaven than on bearded faces, t(218) = 3.45, p = .001, d_z = 0.23, 95% CI = [0.10, 0.37] (see Fig. 2b).

Response bias

There was a significant difference between the response bias on trials with bearded faces (M = −0.03, SD = 0.25) and on trials with clean-shaven faces (M = 0.08, SD = 0.24), t(218) = 4.84, p < .001, d_z = 0.33, 95% CI = [0.19, 0.46]; however, although the bias to respond “happy” when faces were clean-shaven was significantly different from zero (no bias), t(218) = 4.90, p < .001, d_z = 0.33, 95% CI = [0.19, 0.47], the response bias to respond “angry” when faces were bearded was not significantly different from zero, t(218) = 1.53, p = .127, d_z = 0.10, 95% CI = [−0.03, 0.24].

Discussion

In Experiment 1, facial hair facilitated recognition of anger, and the advantage in response times cannot be attributed to a shift toward responding “angry.” Recognition of facial expressions of happiness, which are positive and nonthreatening, was slowed by the presence of a beard in this task. The quicker recognition of happiness in clean-shaven faces than in bearded faces may have been partly due to a small bias to respond “happy” on trials with clean-shaven faces. The quicker recognition of anger in bearded faces than in clean-shaven faces could have been due to beards enhancing visual cues of anger, but it also could have been due to implicit negative evaluations of beardedness being activated and priming negative categorization. This latter mechanism explains the influence of race and age cues on emotion classification in similar tasks (Craig, Koch, & Lipp, 2017; Craig & Lipp, 2018; Hugenberg, 2005). To test this possibility, in Experiment 2 we asked participants to categorize expressions of sadness and happiness on the same bearded and clean-shaven faces. Like anger, sadness is negatively valenced, but it is nonthreatening and not structurally related to beardedness or masculinity (Hess et al., 2009; Hugenberg & Sczesny, 2006). If congruence between negative evaluations of beardedness and negative facial expressions explains the results of Experiment 1, then participants would be expected to be faster to recognize sadness on bearded than on clean-shaven faces. If facial hair specifically enhances recognition of threat, beards would not be expected to facilitate recognition of sadness.

Experiment 2

Method

Participants

We requested 90 participants on the MTurk platform. After the experiment was complete, there were data available from 92 people (45 males, 46 females, 1 who did not respond; mean age = 35.51, SD = 9.96).³

Stimuli, procedure, and data analysis

The same happy faces used in Experiment 1 were presented in Experiment 2, but the angry faces were replaced with photographs of the same individuals expressing sadness (see Fig. 1). The photographs of sad expressions were taken under the same conditions and were edited in the same manner as described for the stimuli in Experiment 1.The procedure for Experiment 2 was the same as the procedure for the online version of Experiment 1 except for the change in the stimuli presented. The data were processed and analyzed as described for Experiment 1. Response time and accuracy data were submitted to separate 2 (facial hair: bearded, clean-shaven) × 2 (emotional expression: happy, angry) repeated measures ANOVAs. Response bias was again calculated and analyzed. Data from 1 participant were not included in analyses because 85% of this person’s responses were missing or invalid. One person who did not report his or her sex was also excluded so that the data sets for the main and supplementary analyses would be consistent, but this exclusion did not influence the pattern of results.

Results

Response times

As in Experiment 1, the main effect of facial hair was not significant, F(1, 89) = 1.57, p = .213, η_p² = .02, 90% CI = [.00, .09], and the main effect of emotional expression was also not significant, F(1, 89) = 2.76, p = .100, η_p² = .03, 90% CI = [.00, .11], but there was a significant Facial Hair × Emotional Expression interaction, F(1, 89) = 6.34, p = .014, η_p² = .07, 90% CI = [.01, .16]. Participants were slower to categorize sadness when it was displayed on bearded rather than clean-shaven faces, t(89) = 3.16, p = .002, d_z = 0.33, 95% CI = [0.12, 0.54], although response times for categorizing faces as happy did not differ between clean-shaven and bearded faces, t(89) = 0.68, p = .496, d_z = 0.07, 95% CI = [−0.14, 0.28] (see Fig. 2c).

Accuracy

Overall, accuracy did not differ between happy and sad expressions, F(1, 89) = 0.05, p = .826, η_p² < .01, 90% CI = [.00, .02], but participants were significantly more accurate categorizing expressions when they were displayed on clean-shaven rather than bearded faces, F(1, 89) = 4.60, p = .035, η_p² = .05, 90% CI = [.00, .14]. This effect was moderated by a significant Facial Hair × Emotional Expression interaction, F(1, 89) = 4.82, p = .031, η_p² = .05, 90% CI = [.00, .14] (see Fig. 2d). Participants were less accurate categorizing sadness, t(89) = 3.04, p = .003, d_z = 0.32, 95% CI = [0.11, 0.52], but not happiness, t(89) = 0.06, p = .951, d_z = 0.01, 95% CI = [−0.20, 0.21], when it was displayed on bearded faces rather than clean-shaven faces.

Response bias

There was a significant difference between response bias on trials with bearded faces (M = 0.05, SD = 0.23) and response bias on trials with clean-shaven faces (M = −0.04, SD = 0.20), t(89) = 2.54, p = .013, d_z = 0.27, 95% CI = [0.05, 0.48]. Unlike in Experiment 1, there was a bias to categorize bearded faces as happy; however, neither the response bias for bearded faces, t(89) = 1.95, p = .054, d_z = 0.21, 95% CI = [0.00, 0.41], nor the response bias for clean-shaven faces, t(89) = 1.68, p = .096, d_z = 0.18, 95% CI = [−0.03, 0.38], differed significantly from zero.

Discussion

Participants were slower to recognize sad expressions on bearded faces than on clean-shaven faces, which indicates that the recognition advantage for bearded faces observed in Experiment 1 does not generalize to all negative expressions. A response bias is unlikely to explain the response time differences in Experiment 2 as there was no significant response bias in either direction for bearded or clean-shaven faces. The results of this experiment suggest that the recognition advantage for angry expressions on bearded faces in Experiment 1 was not driven by congruence between negative evaluations of beardedness and anger.

Another possible explanation for the recognition advantage for angry expressions on bearded faces observed in Experiment 1 is that implicit stereotypes (information-based associations) about bearded men were activated and influenced emotion recognition. For example, bearded men may be seen as more aggressive and less sociable than clean-shaven men, and this could facilitate recognition of anger but not happiness or sadness (Guthrie, 1970). Such an effect seems possible given that racial and gender stereotypes influence emotion recognition under some circumstances (Bijlstra, Holland, & Wigboldus, 2010).

To investigate whether stereotypes might explain our results in Experiment 1, in Experiment 3 we measured how beards influence explicit ratings of aggressiveness, prosociality, and masculinity. If emotionally expressive bearded men are perceived as more aggressive and masculine but less prosocial than their clean-shaven counterparts, congruence between stereotypes about bearded men and emotional expressions could account for the results of Experiment 1.

Experiment 3

Method

Participants

We requested 450 participants on MTurk. After the experiment was complete, there were some data available from 455 MTurk workers (231 males, 214 females, 3 other, 7 who did not respond; mean age =35.67, SD = 10.91).⁴

Measures and procedure

Participants were randomly assigned to complete one of three rating tasks on the Qualtrics platform. In each task, each of the 40 images used in Experiment 1 (i.e., happy and angry bearded and clean-shaven faces) was presented once. Participants rated the aggressiveness, masculinity, or prosociality of each face on a scale from 0, not at all, to 100, extremely, by clicking on and dragging a slider. Because participants may not have been very familiar with the concept of prosociality, we defined a prosocial person as someone who appears “positive, helpful, and friendly, or someone who would act in a way that benefits others.” After rating each of the 40 faces, participants completed a brief demographic questionnaire, indicating their age, sex, ethnicity, and sexual orientation (on the Kinsey scale, from 0, exclusively heterosexual, to 6, exclusively homosexual, or 7, asexual). They also provided a description of their environment while completing the survey and could leave additional comments.

Data processing and analysis

Participants’ ratings were averaged to create condition means for each rating task, and the means for each task were submitted to a separate 2 (facial hair: bearded, clean-shaven) × 2 (emotional expression: happy, angry) repeated measures ANOVA. Data from 10 participants could not be included in analyses because they did not identify their sex as male or female or they provided ratings for fewer than 20% of the faces. These exclusions left 151 participants (81 males) in the aggressiveness rating task, 146 participants (78 males) in the prosociality rating task, and 148 participants (72 males) in the masculinity rating task.

Results

Aggressiveness ratings

Overall, bearded faces were rated as more aggressive than clean-shaven faces, F(1, 150) = 69.96, p < .001, η_p² = .32, 90% CI = [.22, .41], and angry faces were rated as more aggressive than happy faces, F(1, 150) = 519.53, p < .001, η_p² = .78, 90% CI = [.73, .81], but facial hair and emotional expression interacted to influence aggressiveness ratings, F(1, 150) = 52.65, p < .001, η_p² = .26, 90% CI = [.16, .35]. Angry faces were rated as more aggressive when bearded than when clean-shaven, t(150) = 9.43, p < .001, d_z = 0.77, 95% CI = [0.58, 0.95], whereas happy faces were rated similarly irrespective of beardedness, t(150) = 1.53, p = .129, d_z = 0.12, 95% CI = [−0.04, 0.28] (see Fig. 3a).

Fig. 3.

Results from Experiment 3: mean ratings of the (a) aggressiveness, (b) prosociality, and (c) masculinity of the angry and happy faces. Results are shown separately for bearded and clean-shaven faces. Error bars represent ±1 SEM. For figures displaying the distributions of these ratings, see the Supplemental Material available online.

Prosociality ratings

Overall, bearded faces were rated as more prosocial than clean-shaven faces, F(1, 145) = 7.31, p = .008, η_p² = .05, 90% CI = [.01, .11], and happy faces were rated as more prosocial than angry faces, F(1, 145) = 1,443.29, p < .001, η_p² = .91, 90% CI = [.89, .92], but facial hair and emotional expression interacted to influence prosociality ratings, F(1, 145) = 38.89, p < .001, η_p² = .21, 90% CI = [.12, .30]. Angry faces were rated as more prosocial when clean-shaven than when bearded, t(145) = 2.44, p = .016, d_z = 0.20, 95% CI = [0.04, 0.37], but happy faces were rated as more prosocial when bearded than when clean-shaven, t(145) = 5.04, p < .001, d_z = 0.42, 95% CI = [0.25, 0.58] (see Fig. 3b).

Masculinity ratings

Overall, bearded faces were rated as more masculine than clean-shaven faces, F(1, 147) = 281.67, p < .001, η_p² = .66, 90% CI = [.58, .71], and angry faces were rated as more masculine than happy faces, F(1, 147) = 71.65, p < .001, η_p² = .33, 90% CI = [.23, .42], but facial hair and emotional expression interacted to influence masculinity ratings, F(1, 147) = 4.17, p = .043, η_p² = .03, 90% CI = [.00, .08]. Bearded faces were rated as more masculine than clean-shaven faces both when expressing happiness, t(147) = 17.09, p < .001, dz = 1.40, 95% CI = [1.17, 1.62], and when expressing anger, t(147) = 14.88, p < .001, dz = 1.22, 95% CI = [1.00, 1.43], but this difference was smaller for angry expressions. Interestingly, bearded happy faces were judged as more masculine than clean-shaven angry faces, t(147) = 6.74, p < .001, d_z = 0.53, 95% CI = [0.38, 0.72], which suggests that beardedness influences masculinity judgments more than emotional expressions do (see Fig. 3c).

Discussion

Beardedness increased ratings of aggressiveness for angry faces, but also increased ratings of prosociality for happy faces. These ratings suggest that bearded men are not uniformly stereotyped as negative (aggressive and antisocial). If stereotypes influenced emotion recognition in Experiment 1, the presence of a beard should have facilitated the recognition of both happy and angry expressions, as beards increased perceived aggression and perceived prosociality in Experiment 3. In Experiment 4, we entered the stimuli into a computer-based emotion classifier in order to investigate the structural overlap between beardedness and anger cues as an explanation for participants’ advantage in recognizing anger when it was displayed on bearded faces. Computer classifiers are a step removed from the stereotypes and attitudes held by human observers. We predicted that the image classifier would produce higher confidence ratings for expressions of anger when they were displayed on bearded rather than on clean-shaven faces, but would produce higher confidence ratings for expressions of sadness when they were displayed on clean-shaven rather than on bearded faces.

Experiment 4

Method

Procedure

All the photographs used in Experiments 1 through 3 (men expressing happiness, anger, and sadness when bearded and when clean-shaven) were entered into the Microsoft Emotion API image classifier (ME-API; https://azure.microsoft.com/en-au/services/cognitive-services/emotion/) in the same format in which they were presented to participants. The expression-recognition algorithms in the ME-API were developed via a deep-learning neural network trained on large sets of images. The system does not recognize facial expressions on the basis of predefined features or action units, but instead learns through training and testing with large sets of images representing the different emotion categories. The ME-API and human raters have shown comparable identification accuracy (Goeleven, De Raedt, Leyman, & Verschuere, 2008; Takácˇ, Mach, & Sincˇák, 2016). For each image, the classifier provided a confidence rating between 0 (very unlikely) and 1 (very likely) for each of eight emotional expressions (angry, contemptuous, disgusted, fearful, happy, neutral, sad, and surprised). To determine whether the presence of a beard resulted in the image classifier systematically (but incorrectly) identifying a particular emotion on emotionless faces, we also entered into the ME-API photographs of the same men posing a neutral expression both when bearded and when clean-shaven. Analysis of the confidence estimates revealed that the classifier’s identification of emotion in the neutral faces was not influenced by the presence or absence of facial hair (a full analysis is provided in the Supplemental Material).

Data analysis

We report here results for the confidence ratings for the faces’ intended emotions (i.e., we analyzed the ME-API’s confidence estimates for happiness in the case of happy expressions, for anger in the case of angry expressions, and so on). These confidence estimates were submitted to a 2 (facial hair: bearded, clean-shaven) × 3 (emotional expression: angry, happy, sad) repeated measures ANOVA with Greenhouse-Geisser adjustments for violations of sphericity (adjusted degrees of freedom are reported). Each of the 10 posers acted as a single case in the analysis. For completeness, an analysis including ratings for nonintended emotions is provided in the Supplemental Material.

Results

Confidence estimates varied by emotional expression, F(1.3, 11.3) = 19.70, p = .001, η_p² = .69, 90% CI = [.40, .78] (see Fig. 4). Confidence estimates for the intended expression were higher for happy faces than for angry faces, t(9) = 4.72, p = .001, d_z =1.49, 95% CI = [0.63, 2.35], and were higher for angry faces than for sad faces, t(9) = 3.35, p = .009, d_z = 1.06, 95% CI = [0.31, 1.81]. This effect was moderated by a significant Facial Hair × Emotional Expression interaction, F(1.6, 14.7) = 7.58, p = .008, η_p² = .46, 90% CI = [.12, .61]. Confidence estimates were higher for bearded than for clean-shaven angry faces, t(9) = 2.00, p = .077, d_z = 0.63, 95% CI = [−0.03, 1.30], but were lower for bearded than for clean-shaven sad faces, t(9) = 2.43, p = .038, d_z = 0.77, 95% CI = [0.08, 1.46]. Estimates for bearded and clean-shaven happy faces were almost at ceiling (> .99 out of 1) and did not differ, t(9) = 1.27, p = .236, d_z = 0.40, 95% CI = [−0.24, 1.04].

Fig. 4.

Results from Experiment 4: the image classifier’s mean confidence rating for the intended emotion, shown separately for angry, happy, and sad faces when bearded and clean-shaven faces. Error bars represent ±1 SEM. For a figure displaying the distributions of the confidence ratings, see the Supplemental Material available online.

General Discussion

Evolution by sexual selection has shaped secondary sexual characteristics employed during male-male agonistic signaling (Darwin, 1871; Emlen, 2008) and may have acted on sexually dimorphic traits in men (A. Dixson, Dixson, & Anderson, 2005). Here we have shown that beardedness, a highly sexually dimorphic and visually salient masculine characteristic, enhances recognition of angry facial expressions (Experiment 1). Participants were faster and more accurate to categorize anger than happiness on bearded but not on clean-shaven faces. Experiment 2 ruled out the possibility that beards facilitated recognition of anger displays because of congruence between negative evaluations of beardedness and anger, as beards slowed recognition of a negative expression, sadness. Explicit ratings of the faces in Experiment 3 demonstrated that beardedness enhanced perceived aggressiveness of angry faces but also enhanced perceived prosociality of happy faces. These ratings are inconsistent with a stereotype-based explanation of the categorization advantage for angry expressions displayed on bearded faces. Finally, in Experiment 4, a computer-based image classifier was more accurate classifying anger on bearded faces but was more accurate classifying sadness on clean-shaven faces.

Considering these results together with previous research, we propose that the advantage for classifying angry bearded faces may have emerged because beards increase perceived masculinity by enhancing the prominence of the jaw (B. J. W. Dixson et al., 2017). The mouth and jaw are important in anger recognition, and masculine jawlines are associated with facilitated recognition of anger (Becker et al., 2007). Enhanced prominence and angularity of the jawline due to the presence of facial hair potentially facilitated anger classification in our study. This visual structural explanation is also consistent with the results of Experiment 2, in which sad expressions were employed. Sadness is partly conveyed through drooping the lip corners (Reed & DeScioli, 2017). Facial hair masks areas of the lips and chin that convey sadness while augmenting the size of the lower face and jaw, and this may have reduced the visibility of sadness cues. Experiment 4 suggests that the advantage for recognizing anger on bearded faces may be carried in visual information to some extent, as a computer-based image classifier produced the same biases as human observers even though computers do not directly hold the attitudes and stereotypes held by human observers. However, we note that human involvement in training the image classifier could have led to the bias observed if, for example, more bearded than clean-shaven faces were selected as training images for anger. We did not develop the image classifier, so we cannot rule out this possibility. Further research into the structural overlap between facial hair and emotional expression would be beneficial.

The finding that beards facilitate recognition of anger has wide-reaching implications that intersect with theory and practice. Our data suggest that beards may be an evolved signal of masculinity and formidability, which is consistent with studies finding that facial hair is not functionally protective (B. J. W. Dixson, Sherlock, Cornwell, & Kasumovic, 2018). Beards may have evolved via intrasexual selection; that is, facial hair may have increased the perceived formidability of opponents, thereby curtailing bearded individuals’ conflicts and leading them to enjoy increased social status and access to mates (Darwin, 1871). Alternatively, beards may serve no function for survival or reproduction. They may simply mask the lower part of the face and draw more attention to the eyes, an area more important for recognizing anger than for recognizing happiness (Calvo & Fernandez-Martin, 2013). Regardless of the origin of beards or how they facilitate recognition of angry facial expressions, understanding how facial hair influences recognition of emotions will be important for the development of person-perception research.

In the current study, we used only emotionally expressive faces in the experiments with human observers. Although the presence or absence of facial hair did not influence the image classifier’s emotion estimates, future research may investigate whether beards influence human observers to perceive emotion in neutral faces. Future research may also investigate whether beards influence emotion perception because they alter the contrast between teeth and the rest of the face. Teeth were visible in all the happy and angry expressions used in our study, so the presence of teeth in only some images cannot explain our results. Given that exposed teeth have an important influence on the perception of emotion (Becker & Srinivasan, 2014), it could be valuable to investigate whether beards influence the recognition of facial displays of emotion differently depending on whether the expressions are displayed with teeth exposed or covered. More generally, our results demonstrate the importance of considering the influence of social information, including facial hair, in emotion-perception research, as these influences are often overlooked (Schmidt & Cohn, 2001).

We have demonstrated that the presence of a beard biases how a facial expression of emotion is processed. This initial bias could affect how bearded men are perceived and suggests that a man might be treated differently when bearded than when clean-shaven. In the current study, beards facilitated anger recognition, which suggests that professionals responsible for detecting and responding to threat (e.g., police officers) could more readily perceive bearded men than clean-shaven men as threatening. This is consistent with previous research showing that beards influence perceivers’ behavior, including their voting decisions (Herrick, Mendez, & Pryor, 2015) and judgments of criminality (Conti & Conti, 2004). Together, this research demonstrates that beardedness affects nonverbal communication and suggests new lines of inquiry into how masculine characteristics influence social interactions beyond the lab and into men’s daily lives.

Supplemental Material

CraigOpenPracticesDisclosure_rev – Supplemental material for Sexual Selection, Agonistic Signaling, and the Effect of Beards on Recognition of Men’s Anger Displays

Supplemental material, CraigOpenPracticesDisclosure_rev for Sexual Selection, Agonistic Signaling, and the Effect of Beards on Recognition of Men’s Anger Displays by Belinda M. Craig, Nicole L. Nelson and Barnaby J. W. Dixson in Psychological Science

Supplemental Material

CraigSupplementalMaterial_rev – Supplemental material for Sexual Selection, Agonistic Signaling, and the Effect of Beards on Recognition of Men’s Anger Displays

Supplemental material, CraigSupplementalMaterial_rev for Sexual Selection, Agonistic Signaling, and the Effect of Beards on Recognition of Men’s Anger Displays by Belinda M. Craig, Nicole L. Nelson and Barnaby J. W. Dixson in Psychological Science

Footnotes

Action Editor

Alice J. O’Toole served as action editor for this article.

Author Contributions

B. M. Craig and B. J. W. Dixson conceptualized the initial idea for this study. B. J. W. Dixson and N. L. Nelson acquired funding to support the project. All the authors contributed to the development of the methods. B. J. W. Dixson developed the stimuli, and B. M. Craig prepared the protocols. B. M. Craig and N. L. Nelson contributed to data acquisition. B. M. Craig analyzed the data. All the authors wrote the original draft and reviewed and proofread the manuscript.

ORCID iDs

Belinda M. Craig

Nicole L. Nelson

Declaration of Conflicting Interests

The author(s) declared that there were no conflicts of interest with respect to the authorship or the publication of this article.

Funding

This study was supported by a University of Queensland Postdoctoral Research Fellowship awarded to B. J. W. Dixson.

Supplemental Material

Additional supporting information can be found at

Open Practices

All data have been made publicly available via the Open Science Framework and can be accessed at https://osf.io/zuwxt/. The stimuli are not publicly available because the individuals depicted in them did not agree for their likeness to be shared in this manner. Requests for the materials can be sent via e-mail to the Corresponding Author. The design and analysis plans for the experiments were not preregistered. The complete Open Practices Disclosure for this article can be found at https://journals-sagepub-com.web.bisu.edu.cn/doi/suppl/10.1177/0956797619834876. This article has received the badge for Open Data. More information about the Open Practices badges can be found at .

Notes

References

Becker

D. V.

Kenrick

D. T.

Neuberg

S. L.

Blackwell

K. C.

Smith

D. M.

(2007). The confounded nature of angry men and happy women. Journal of Personality and Social Psychology, 92, 179–190. doi:10.1037/0022-3514.92.2.179

Becker

D. V.

Srinivasan

(2014). The vividness of the happy face. Current Directions in Psychological Science, 23, 189–194. doi:10.1177/0963721414533702

Bijlstra

Holland

R. W.

Wigboldus

D. H. J.

(2010). The social face of emotion recognition: Evaluations versus stereotypes. Journal of Experimental Social Psychology, 46, 657–663. doi:10.1016/j.jesp.2010.03.006

Blair

R. J. R.

(2003). Facial expressions, their communicatory functions and neuro-cognitive substrates. Philosophical Transactions of the Royal Society B: Biological Sciences, 358, 561–572. doi:10.1098/rstb.2002.1220

Blanchard

D. C.

(2009). Of lion manes and human beards: Some unusual effects of the interaction between aggression and sociality. Frontiers in Behavioral Neuroscience, 3, Article 45. doi:10.3389/neuro.08.045.2009

Calvo

M. G.

Fernández-Martín

(2013). Can the eyes reveal a person’s emotions? Biasing role of the mouth expression. Motivation and Emotion, 37, 202–211. doi:10.1007/s11031-012-9298-1

Conti

R. P.

Conti

M. A.

(2004). Mock jurors’ perceptions of facial hair on criminal offenders. Perceptual and Motor Skills, 98, 1356–1358. doi:10.2466/pms.98.3c.1356-1358

Craig

B. M.

Koch

Lipp

O. V.

(2017). The influence of social category cues on the happy categorisation advantage depends on expression valence. Cognition & Emotion, 31, 1493–1501. doi:10.1080/02699931.2016.1215293

Craig

B. M.

Lipp

O. V.

(2018). Facial age cues and emotional expression interact asymmetrically: Age cues moderate emotion recognition. Cognition & Emotion, 32, 350–362. doi:10.1080/02699931.2017.1310087

10.

Darwin

(1871). The descent of man, and selection in relation to sex. London, England: Murray.

11.

Dixson

Anderson

(2005). Sexual selection and the evolution of visually conspicuous sexually dimorphic traits in male monkeys, apes, and human beings. Annual Review of Sex Research, 16, 1–19.

12.

Dixson

B. J.

Brooks

R. C.

(2013). The role of facial hair in women’s perceptions of men’s attractiveness, health, masculinity and parenting abilities. Evolution & Human Behavior, 34, 236–241. doi:10.1016/j.evolhumbehav.2013.02.003

13.

Dixson

B. J.

Vasey

P. L.

(2012). Beards augment perceptions of men’s age, social status, and aggressiveness, but not attractiveness. Behavioral Ecology, 23, 481–490. doi:10.1093/beheco/arr214

14.

Dixson

B. J. W.

Lee

A. J.

Sherlock

J. M.

Talamas

S. N.

(2017). Beneath the beard: Do facial morphometrics influence the strength of judgments of men’s beardedness? Evolution & Human Behavior, 38, 164–174. doi:10.1016/j.evolhumbehav.2016.08.004

15.

Dixson

B. J. W.

Sherlock

J. M.

Cornwell

W. K.

Kasumovic

M. M.

(2018). Contest competition and men’s facial hair: Beards may not provide advantages in combat. Evolution & Human Behavior, 39, 147–153. doi:10.1016/j.evolhumbehav.2017.11.004

16.

Dixson

B. J. W.

Sulikowski

Gouda-Vossos

Rantala

M. J.

Brooks

R. C.

(2016). The masculinity paradox: Facial masculinity and beardedness interact to determine women’s ratings of men’s facial attractiveness. Journal of Evolutionary Biology, 29, 2311–2320. doi:10.1111/jeb.12958

17.

Ekman

Friesen

W. V.

Hager

J. C.

(2002). Facial Action Coding System (FACS): The manual. Salt Lake City, UT: A Human Face.

18.

Emlen

D. J.

(2008). The evolution of animal weapons. Annual Review of Ecology, Evolution, and Systematics, 39, 387–413. doi:10.1146/annurev.ecolsys.39.110707.173502

19.

Faul

Erdfelder

Lang

A.-G.

Buchner

(2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39, 175–191. doi:10.3758/BF03193146

20.

Forster

K. I.

Forster

J. C.

(2003). DMDX: A Windows display program with millisecond accuracy. Behavior Research Methods, Instruments, & Computers, 35, 116–124. doi:10.3758/bf03195503

21.

Gaetano

J. M.

(2017). Signal detection theory calculator 1.2 [Excel workbook]. Retrieved from https://www.researchgate.net/publication/316642315_Signal_detection_theory_calculator_12

22.

Goeleven

De Raedt

Leyman

Verschuere

(2008). The Karolinska Directed Emotional Faces: A validation study. Cognition & Emotion, 22, 1094–1118. doi:10.1080/02699930701626582

23.

Goodhart

C. B.

(1960). The evolutionary significance of human hair patterns and skin colouring. In Advancement of Science (Vol. 17). London, England: British Association for the Advancement of Science.

24.

Guthrie

R. D.

(1970). Evolution of human threat display organs. In Dobhansky

Hecht

Steers

(Eds.), Evolutionary biology (pp. 257–302). New York, NY: Appleton-Century-Crofts.

25.

Herrick

Mendez

J. M.

Pryor

(2015). Razor’s edge: The politics of facial hair. Social Science Quarterly, 96, 1301–1313. doi:10.1111/ssqu.12183

26.

Hess

Adams

R. B.

Jr. Grammer

Kleck

R. E.

(2009). Face gender and emotion expression: Are angry women more like men? Journal of Vision, 9(12), Article 19. doi:10.1167/9.12.19

27.

Hugenberg

(2005). Social categorization and the perception of facial affect: Target race moderates the response latency advantage for happy faces. Emotion, 5, 267–276. doi:10.1037/1528-3542.5.3.267

28.

Hugenberg

Sczesny

(2006). On wonderful women and seeing smiles: Social categorization moderates the happy face response latency advantage. Social Cognition, 24, 516–539. doi:10.1521/soco.2006.24.5.516

29.

Kokko

Jennions

M. D. B.

Brooks

(2006). Unifying and testing models of sexual selection. Annual Review of Ecology, Evolution, and Systematics, 37, 43–66. doi:10.1146/annurev.ecolsys.37.091305.110259

30.

Muscarella

Cunningham

M. R.

(1996). The evolutionary significance and social perception of male pattern baldness and facial hair. Ethology and Sociobiology, 17, 99–117. doi:10.1016/0162-3095(95)00130-1

31.

Neave

Shields

(2008). The effects of facial hair manipulation on female perceptions of attractiveness, masculinity, and dominance in male faces. Personality and Individual Differences, 45, 373–377. doi:10.1016/j.paid.2008.05.007

32.

Puts

D. A.

(2010). Beauty and the beast: Mechanisms of sexual selection in humans. Evolution & Human Behavior, 31, 157–175. doi:10.1016/j.evolhumbehav.2010.02.005

33.

Reed

L. I.

DeScioli

(2017). The communicative function of sad facial expressions. Evolutionary Psychology, 15(1). doi:10.1177/1474704917700418

34.

Sacco

D. F.

Hugenberg

(2009). The look of fear and anger: Facial maturity modulates recognition of fearful and angry expressions. Emotion, 9, 39–49. doi:10.1037/a0014081

35.

Schmidt

K. L.

Cohn

J. F.

(2001). Human facial expressions as adaptations: Evolutionary questions in facial expression research. American Journal of Physical Anthropology, 116(s33), 3–24. doi:10.1002/ajpa.20001

36.

Sell

Cosmides

Tooby

(2014). The human anger face evolved to enhance cues of strength. Evolution & Human Behavior, 35, 425–429. doi:10.1016/j.evolhumbehav.2014.05.008

37.

Sherlock

J. M.

Tegg

Sulikowski

Dixson

B. J. W.

(2017). Facial masculinity and beardedness determine men’s explicit, but not their implicit, responses to male dominance. Adaptive Human Behavior and Physiology, 3, 14–29. doi:10.1007/s40750-016-0047-7

38.

Stanislaw

Todorov

(1999). Calculation of signal detection theory measures. Behavior Research Methods, Instruments, & Computers, 31, 137–149. doi:10.3758/BF03207704

39.

Takácˇ

Mach

Sincˇák

(2016, October 12). Cloud-based facial emotion recognition for real-time emotional atmosphere assessment during a lecture. Paper presented at the 2016 IEEE International Conference on Systems, Man, and Cybernetics (SMC), Budapest, Hungary.

40.

Tay

P. K.

(2015). The adaptive value associated with expressing and perceiving angry-male and happy-female faces. Frontiers in Psychology, 6, Article 851. doi:10.3389/fpsyg.2015.00851

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.58 MB

1.60 MB