Abstract
As face masks have become more commonplace in many regions due to COVID-19, concerns have been raised about their effects on the perception of mask wearers and social cohesion more broadly, including racial profiling. In two studies we examined the effects of masks on social judgments of mask wearers, and whether masks have different effects on judgments of Black and White faces. Participants rated 20 Black and 20 White faces with and without masks on trustworthiness/approachability (Studies 1 and 2) and on dominance/competence and attractiveness (Study 2). In both studies masks increased perceived trustworthiness and reduced the effect of face race on judgments. Masks also increased perceived attractiveness, but had no effect on the perception of dominance/competence. Overall, this study found no negative effects of face masks on judgments of mask wearers, though further research is needed.
In response to the COVID-19 pandemic, face masks or coverings were widely recommended and even mandated by some authorities. While the benefits of face masks for helping to prevent the spread of airborne viruses are well-established (Leung et al., 2020), concerns have been raised about the impact masks may have on social perception and social cohesion. These concerns have been expressed across the internet, such as the following from one website: “Because basic social cues such as smiles or at least an agreeable expression cannot be read, every passerby seems more threatening.” (Social Cohesion and Public Health, 2020). In a recent study, Howard (2020) surveyed the public's views about why people often don't wear face masks and identified a factor centered on masks having negative perceptions (less trustworthy, suspicious, makes others uncomfortable, makes others uneasy). Some scholars have expressed similar concerns. For example, in discussing the implications of masks in schools, Spitzer wrote “masks may increase the perception of negative emotions and diminish the perception of positive emotions” (Spitzer, 2020, p. 5). Concerns that masks might promote racial profiling have also been reported in the media (Cineas, 2020, April 22). Indeed, one county in the US temporarily passed a law exempting people from wearing face masks if they feared racial profiling or harassment (Oregon County Rescinds Racial Profiling Mask Exception, 2020).
Clearly there is some concern that face masks have negative effects on perceptions of mask wearers, particularly in terms of making wearers appear threatening. However, there is relatively little research on the effects of face masks on social perception to address these concerns. Recent studies have found that masks interfere with face recognition (Freud et al., 2020), face matching (Carragher & Hancock, 2020; Noyes et al., 2021) and emotion recognition (Carbon, 2020; Noyes et al., 2021), but no studies to our knowledge have directly examined the effects of masks on trait judgments.
There is reason to suspect masks will have an impact on trait judgments of mask wearers. Carbon (2020) found that face masks reduced the accuracy of and confidence in emotion recognition, and led to confusions between certain emotions. Therefore masks may at least affect trait judgments that rely heavily on emotion recognition, such as trustworthiness (Oosterhof & Todorov, 2008; Vernon et al., 2014). However, the nature of the impact is unclear. There is some evidence that the mere presence of a face can increase trust (Steinbrück et al., 2002) and so, by extension, a fully visible face may be deemed more trustworthy than a partially visible face. However, a recent study found, against the authors’ expectations, that faces wearing a surgical mask were rated more trustworthy than unmasked faces (Olivera-La Rosa et al., 2020). An earlier study that blocked out specific regions of the face using a black rectangle also found that trustworthiness judgments were higher when the nose or mouth regions were obscured (Santos & Young, 2011). Thus, despite the expectation or belief that masks reduce perceived trustworthiness, the limited empirical evidence does not support this.
The concern over racial profiling of African American mask wearers raises the question of whether masks increase racial stereotyping (MacLin & Herrera, 2006; Oliver, 2003; Welch, 2007). We know that social judgments from faces draw on both specific facial cues to certain traits, as well as social categorisation and associated stereotypes (Hugenberg & Bodenhausen, 2003; MacLin & Malpass, 2001). If facial cues and categorical information combine to elicit social judgments, then masks may increase the influence of categorical information by reducing the availability of individualistic facial cues. There may also be cultural associations that give particular meanings to masks in combination with other cues. For example, when worn by African Americans, masks may carry meanings associated with criminality or aggression (MacLin & Herrera, 2006). Maclin and Herrera (2006) found American respondents associated African American criminals with wearing accessories such as bandanas.
The Current Study
Motivated by the evident concerns over the impact masks may have on the perception of mask wearers, including racial profiling, the current study was aimed at measuring the effect of masks on social judgments. We examined how face masks affect ratings of faces compared to unmasked faces, and whether the effects of masks on trait judgments were moderated by facial trustworthiness and face race. Given the potential relevance of COVID-19 to the data, we note that data collection occurred between 21st August and 30th September, 2020.
Study 1
Method
Participants
Two-hundred and fifty-one participants (54.2% male, 42.6% female, 0.4% other, 2.8% missing data) were recruited through a participant recruitment website (prolific.ac). In terms of nationalities, there were 14.3% British, 11.6% Mexican, 8.0% Polish, 6.4% North American, 4.8% Italian, 3.2% Chilean, and 2.4% Australian. The remainder were various other nationalities, each comprising less than 2% of the sample. Age ranged from 18 to 65 years (M = 30.24, SD = 10.42; 63% were between 18 and 30 years). Most participants (74%) indicated that they wore a mask when they go out in public ‘most of the time’ or ‘always’. The remainder (26%) wore a mask ‘about half the time’ or less.
Stimuli
Forty faces were selected from the Chicago Face Database (Ma et al., 2015). Models were shown front-on with a neutral expression, against a plain background and wearing an identical plain t-shirt. We selected 20 Black faces and 20 White faces, ensuring 50% male and female faces within each set. We used the norming data available for the face database to select the highest and lowest trustworthiness faces within each category. For example, the five most and five least trustworthy White male faces were selected for the White male subset, the five most and five least trustworthy White female faces were selected for the White female subset, and so on. Among the 40 chosen faces there were no significant differences in trustworthiness between male and female or Black and White faces, and no gender by race interaction (all Fs < 1). Across the 40 faces, high trustworthy faces were significantly higher in trustworthiness (M = 4.15, SD = 0.24) than low trustworthy faces (M = 2.67, SD = 0.28), t (38) = 18.30, p < .001, Cohen's d = 5.79 (using norming data rated on a 7-point Likert scale).
To create the masked images, a plain white mask was superimposed onto each face using Photoshop (see Figure 1). A third set of images was created by superimposing a branded mask onto each face. This condition was included in order to evaluate a specific branded mask on the request of the mask developer and was not intended to be published. Therefore, this condition is not analysed in the current paper.
Example of high (Left) and low (Right) trustworthiness unmasked and masked faces (Study 1).
Design
The study was a 2 (mask: no mask vs. plain mask) by 2 (face race: Black vs. White) by 2 (face trustworthiness: low vs. high) within-subjects design. The dependent variable was ratings of trustworthiness/approachability given to face images. Each of the 40 faces appeared three times, once without a mask and once with a plain and branded mask. Presentation order of all 80 images was randomized across participants.
Procedure
The study was run entirely online. After providing informed consent participants read: “On the following pages you will be shown a series of faces. Some of these will be wearing face masks and some will not. Your task is to rate each face in terms of how TRUSTWORTHY or APPROACHABLE the person appears. In this study we are interested in your first impressions, so please base your responses on your first impression or impulse about each face. You should aim to give a response to each face within one or two seconds.” We combined the terms trustworthy and approachable because both terms have been used in previous studies to capture an underlying factor and are highly correlated (Sutherland et al., 2013).
One hundred and twenty images (forty unmasked, forty with the plain mask, forty with the branded mask) were then shown in random order, one per screen, above a 9-point response scale anchored with “not at all” (1) and “very” (9). Finally, participants were asked to respond to basic demographic questions, presented with a study explanation, and thanked for their time. Most participants took between 8 and 14 min to complete the study. All participants were paid for their participation.
Results
Effect of Masks on Trustworthiness Judgments of Faces
To assess the effect of masks on trustworthiness ratings of trustworthy and untrustworthy Black and White faces, we ran a 2 (race: Black vs. White) by 2 (face trustworthiness: high vs. low) by 2 (mask: no mask vs. plain mask) repeated measures ANOVA. Ratings were collapsed across face sex because sex was not a variable of theoretical interest. We acknowledge that sex is relevant to trustworthiness judgments from faces (e.g., Mattarozzi et al., 2015). However, effects depend also on perceiver gender and are therefore complicated. As gender was not the focus of the current paper, we have collapsed across face sex to simplify the interpretation of main effects and interactions.
There were significant main effects of face trustworthiness and mask. As expected, high trustworthiness faces were rated higher in trustworthiness (M = 6.12, SD = 0.98) than low trustworthiness faces (M = 4.48, SD = 1.10), F (1, 248) = 637.49, p < .001, Eta2 = 0.720. Overall, faces with masks were rated more trustworthy (M = 5.45, SD = 0.91) than unmasked faces (M = 5.15, SD = 0.90), F (1, 248) = 182.37, p < .001, Eta2 = 0.424. There was no main effect of race, F < 1. There also were significant interactions between face trustworthiness and mask, F (1, 248) = 183.45, p < .001, Eta2 = 0.425, between face trustworthiness and race, F (1, 248) = 67.58, p < .001, Eta2 = 0.214, and between race and mask, F (1, 248) = 31.93, p < .001, Eta2 = 0.114 (see Figure 2). The three-way interaction between face trustworthiness, race and mask was not significant, F < 1. The two-way interactions are described below.
Mean trustworthiness ratings of masked and unmasked black and white faces (Study 1). Error bars are standard errors.
Face Trustworthiness and Race
The interaction between face trustworthiness and race revealed that the untrustworthy Black faces were rated less trustworthy (M = 4.41, SD = 1.17) than untrustworthy White faces (M = 4.55, SD = 1.11), F (1, 248) = 7.31, p = .007, while trustworthy Black faces were rated more trustworthy (M = 6.23, SD = 1.09) than trustworthy White faces (M = 6.02, SD = 1.02), F (1, 248) = 17.76, p < .001.
Face Trustworthiness and Mask
The significant interaction between mask and face trustworthiness showed masks increased trustworthiness for untrustworthy faces (Mdiff = 0.97), F (1, 248) = 81.28, p < .001, Eta2 = 0.247, more than for trustworthy faces (Mdiff = 0.35), F (1, 248) = 9.57, p = .002, Eta2 = .037.
Mask and Race
The interaction between mask and race showed masks increased trustworthiness of White faces (Mdiff = 0.75), F (1, 248) = 46.24, p < .001, Eta2 = 0.157, more than of Black faces (Mdiff = 0.56), F (1, 248) = 27.49, p < .001, Eta2 = 0.100. Without a mask, Black faces were rated more trustworthy (M = 5.04, SD = 1.30) than White faces (M = 4.90, SD = 1.30), F (1, 248) = 7.06, p = .008. With a mask Black (M = 5.60, SD = 1.26) and White (M = 5.66, SD = 1.28) faces were rated equally trustworthy, F (1, 248) = 1.43, p = 0.232.
To summarise the overall patterns shown in Figure 2, wearing a mask reduced the influence of face trustworthiness on trustworthiness judgments, and resulted in higher trustworthiness judgments across all face types, especially for low trustworthiness faces. Race had a comparatively minor effect on trustworthiness judgments, and masks had a similar effect on judgments of Black and White faces.
Discussion
Study 1 found that, as expected, masks had a substantial effect on trustworthiness judgments. Unsurprisingly, masks reduced the perceived difference between high and low trustworthiness faces, likely because they obscured the mouth, the primary cue to trustworthiness judgments (Oosterhof & Todorov, 2008; Vernon et al., 2014). However, counter to popular concerns, masks overall increased perceived trustworthiness ratings relative to unmasked faces, particularly for low trustworthiness faces.
This finding is consistent with Olivera-La Rosa et al. (2020) who found faces wearing surgical masks were rated more trustworthy than unmasked faces. These authors speculated that perhaps “the internalized social norm of wearing a mask is suppressing any automatic mistrust due to not seeing the whole face” (p. 5). This explanation is also valid for the current sample, since data were collected during the COVID-19 pandemic and most participants reported wearing masks themselves. However, the increase in trustworthiness could also be due to a perceptual process. It may be that perceivers are more sensitive to cues signaling low trustworthiness than to cues signaling high trustworthiness. If so, in the absence of such cues, trustworthiness judgments would increase overall. This is consistent with evolutionary perspectives on the perception of trustworthiness that emphasise threat detection (Oosterhof & Todorov, 2008), and also with Santos and Young (2011) who found that trustworthiness judgments were somewhat higher when the nose or mouth regions were obscured by a black rectangle (not a face mask). It is also more consistent with the finding that masks increased trustworthiness judgments for untrustworthy faces more than for trustworthy faces. If trustworthiness judgments are more sensitive to cues to untrustworthiness than to trustworthiness, then the effect of a mask would be expected to be larger for untrustworthy faces. The smaller effect on trustworthy faces is unlikely to be a ceiling effect since mean ratings were around 6 on a 9-point scale. It is unclear why masks tend to increase perceived trustworthiness, and both normative and perceptual processes could play a role.
This study found no evidence that masks promote racial profiling in the sense of more negative evaluations of masked Black faces. There was a significant interaction between race and mask condition on trustworthiness judgments, but Black faces were rated more trustworthy than White face in the unmasked condition, and there was no difference between the races in the masked condition. Therefore, there was no evidence that masks had more negative effects for Black than White faces.
Given that the findings of Study 1 are largely inconsistent with expressed concerns about the impact of masks on social perception and cohesion, it was deemed important to replicate the findings in a new and different sample. Furthermore, we only assessed the effect of masks on judgments of trustworthiness/approachability and it is possible that masks also affect other trait judgments. A second study was conducted to examine effects of masks on other trait judgments, and to confirm the effect on trustworthiness/approachability in a new sample. In addition to examining the effects of a plain mask, we also examined whether different mask designs could alter trait judgments.
Study 2
The aims of Study 2 were to confirm the effect of masks on trustworthiness/approachability seen in Study 1, to examine how masks affect other trait judgments, and to test the effects of different mask designs. The second study followed a similar procedure to Study 1, using the same set of faces.
Perceivers make a wide range of trait judgments from faces. However, research suggests these judgments fall along two (Oosterhof & Todorov, 2008) or three (Sutherland, et al., 2013) basic dimensions. The two primary dimensions are trustworthiness/approachability and dominance/competence, with some evidence for a third dimension relating to youthful attractiveness (Sutherland et al., 2013). These dimensions seem to be basic to person perception as well as group stereotypes (Fiske et al., 2007). Therefore, in the current study we examined the effects of masks on judgements of dominance/competence and attractiveness as well as trustworthiness/approachability.
We also examined whether different mask designs could influence trait judgments. If certain design features are found to influence trait judgments, mask designers could use this information to design masks that encourage desirable impressions. Faces are perceived holistically, so we expected that masks and their design features would influence the way the face as a whole was perceived. We created one mask design featuring an upturned line simulating a smile (see Figure 3C), expecting that this would lead to higher ratings of trustworthiness/approachability compared to a plain mask. We also created a second design by offsetting the smile to one side (see Figure 3D). The rationale behind this design was based on the composite face effect (Young et al., 1987). The composite face effect refers to a phenomenon whereby fusing the lower half of one face with the top half of another face disrupts identification of either half, because perceivers have difficulty attending to each half in isolation. However, if the two halves are misaligned so that they do not form a coherent face, recognition is restored to baseline levels. The composite face effect extends to judgments other than identity, including attractiveness (Abbas & Duchaine, 2008), emotional expressions (Calder et al., 2000) and trustworthiness (Todorov et al., 2010). The aim of the offset smile design was to simulate the misalignment of face parts in order to disrupt holistic processing and encourage social judgments to be based on the visible parts of the face.
Example stimuli showing the unmasked (A), plain (B), smile (C) and offset (D) conditions.
Method
Participants
The study was advertised to students studying at an Australian university, as well as through the researchers’ social media channels. A total of 464 responses were received. Of these, 53 were removed for having more than 10% missing data. A further 23 participants were removed for failing to correctly indicate which trait they were asked to rate the face on. Of the remaining 388 cases, 77.1% were female, 21.6% were male, 0.5% other, 0.8% did not answer. 89.4% identified as Australian. The remainder comprised 13 different nationalities, each comprising less than 2% of the sample. Age ranged from 18 to 63 years (M = 31.64, SD = 9.79; 53% were between 18 and 30). 45.6% were undergraduate students and these received course credit for their participation. The remainder were postgraduate students (4.6%), employed (44.7%), and unemployed (2.5%) individuals.
Stimuli
We used the same faces as in Study 1. In addition to the plain mask condition, we also created images of each face wearing a mask with the smile design (see Figure 3C) and another wearing the offset mask design (see Figure 3D).
Design
Study 2 was a 4 (mask: no mask vs. plain mask vs. smile mask vs. offset mask) by 2 (face race: Black vs. White) by 2 (face trustworthiness: low vs. high) by 3 (trait: trustworthiness vs. dominance vs. attractiveness) mixed design, with trait as a between-subjects factor. The dependent variable was ratings of face images on the given trait. Each of the 40 faces were shown four times, once without a mask and once with each of three masks. Presentation order of all 160 images was randomized across participants.
Procedure
The procedure was the same as for Study 1, with the following changes. Participants were randomly assigned to one of three trait conditions and asked to rate each face on that trait. The traits were ‘trustworthiness/approachability’ (n = 139), ‘dominance/competence’ (n = 123), and ‘attractiveness’ (n = 126). After receiving these instructions, participants were asked to indicate which trait they had been asked to rate the faces on. This was used to reinforce which trait they were to rate the faces on, and to screen out participants who did not recall or pay attention to the trait. As in Study 1, traits were rated on a 9-point scale from ‘not at all’ to ‘very’.
Results
Interactions Between Mask, Race and Face Trustworthiness
To examine the effects of masks on trait ratings of high and low trustworthiness Black and White faces we ran three separate 2 (mask: mask vs. no mask) by 2 (race: Black vs. White) by 2 (face trustworthiness: high vs. low) ANOVAs, one for each trait. This was chosen over one omnibus ANOVA to facilitate interpretation and because we were primarily interested in the effects on each trait separately, rather than on potential interactions involving trait type. Given that trait type was a between-subjects variable, separate ANOVAs for each were considered appropriate. To control the family-wise error rate, we adopted a conservative p value of.01 for main effects and interactions. Furthermore, we did not include mask type as a variable in this analysis in order to make it directly comparable to Study 1. Mask type was analysed separately (see below).
Trustworthiness
There were significant main effects for face trustworthiness, F (1, 138) = 490.98, p < .001, Eta2 = 0.781, race, F (1, 138) = 24.46, p < .001, Eta2 = 0.151, and mask, F (1, 138) = 13.20, p < .001, Eta2 = .087. As in Study 1, high trustworthiness faces were rated higher in trustworthiness (M = 6.22, SD = .99) than low trustworthiness faces (M = 4.37, SD = 1.24), Black faces were rated higher in trustworthiness (M = 5.42, SD = 1.05) than White faces (M = 5.16, SD = 1.06), and masked faces (M = 5.48, SD = 1.00) more than unmasked faces (M = 5.11, SD = 1.05). There also were significant 2-way interactions between mask and race, F (1, 138) = 38.35, p < .001, Eta2 = 0.217, mask and trust, F (1, 138) = 165.03, p < .001, Eta2 = 0.545, and race and trust, F (1, 138) = 32.02, p < .001, Eta2 = 0.188. The three-way interaction was not significant, F (1, 138) = 5.43, p < .021, Eta2 = .038.
Face Trustworthiness by Race
Simple main effects showed that high trustworthiness Black faces were rated more trustworthy (M = 6.44, SD = 1.04) than high trustworthiness White face (M = 6.00, SD = 1.05), F (1, 138) = 64.65, p < .001, Eta2 = 0.319, while low trustworthiness Black (M = 4.41, SD = 1.36) and White (M = 4.33, SD = 1.24) faces were rated equally trustworthy, F (1, 138) = 1.37, p = 0.243, Eta2 = .010.
Face Trustworthiness by Mask
Masks increased trustworthiness for low trustworthiness faces, F (1, 138) = 66.08, p < .001, Eta2 = 0.324, but not for high trustworthiness faces, F (1, 138) = 1.75, p = 0.188, Eta2 = .012.
Race by Mask
Masks increased trustworthiness for White faces, F (1, 138) = 30.24, p = 0.188, Eta2 = 0.180, but not for Black faces, F (1, 138) = 2.53, p = 0.114, Eta2 = .018 (see Figure 4).
Mean trustworthiness ratings for high and low trustworthiness black and white faces (Study 2). Error bars show standard errors.
Dominance
For dominance, only the main effect of race was significant, F (1, 122) = 42.18, p < .001, Eta2 = 0.230. Black faces were rated more dominant (M = 5.59, SD = 0.96) than White faces (M = 5.18, SD = 1.00). Neither face trustworthiness nor masks had any effect on dominance ratings, and none of the interactions were significant.
Attractiveness
There was a significant main effect of face trustworthiness on attractiveness ratings, F (1, 125) = 518.88, p < .001, Eta2 = 0.806. High trustworthiness faces were rated more attractive (M = 5.26, SD = 1.23) than low trustworthiness faces (M = 3.42, SD = 1.25). There also were significant interactions between race and trust, F (1, 125) = 65.88, p < .001, Eta2 = 0.345, and between trust and mask, F (1, 125) = 82.28, p < .001, Eta2 = 0.397.
Face Trustworthiness by Race
High trustworthiness Black faces were rated more attractive (M = 5.43, SD = 1.42) than high trustworthiness White faces (M = 5.09, SD = 1.20), F (1, 125) = 19.09, p < .001, Eta2 = 0.132, whereas low trustworthiness Black (M = 3.35, SD = 1.29) and White (M = 3.49, SD = 1.31) faces were rated equally attractive, F (1, 125) = 5.21, p = .024, Eta2 = .040.
Face Trustworthiness by Mask
As with trustworthiness ratings, masks increased attractiveness for low trustworthiness faces, F (1, 125) = 17.79, p < .001, Eta2 = 0.125, but not for high trustworthiness faces, F < 1.
Differences Between Mask Types
In order to assess whether the addition of a smile or an offset smile to the mask altered perceptions relative to a plain mask, planned comparisons were performed comparing each design to the plain mask for each of the three traits. The smile mask did not alter perceptions on any trait relative to the plain mask. However, compared to the plain mask the offset mask led to lower dominance/competence (Mplain = 5.41, SD = 0.96; Moffset = 4.97, SD = 1.31), F (1, 122) = 18.07, p < .001, lower attractiveness (Mplain = 4.46, SD = 1.17; Moffset = 4.29, SD = 1.28), F (1, 125) = 11.81, p = .001, and lower trustworthiness/approachability, F (1, 138) = 8.66, p = .004 (see Figure 5).
Mean ratings on trustworthiness, dominance and attractiveness for each mask condition. Response scale was 1 (not at all) to 9 (very). Error bars show standard errors.
Discussion
Study 2 found similar results to Study 1, with masks increasing perceived trustworthiness compared to unmasked faces, but only for low trustworthiness faces. We also found no evidence of racial profiling, with masks having a positive impact on social judgments for both Black and White faces. Thus, Study 2 replicated the results of Study 1 despite some differences in procedure, namely that faces were displayed four times each as opposed to three. Seeing each face four times may have increased familiarity and affected trait judgments, but the effects of masks were robust. Extending the results of Study 1 to other traits, we found masks also increased ratings of attractiveness, and had little effect on dominance/competence.
That masks increased attractiveness ratings also goes against concerns over the negative impacts of masks on social judgments. One explanation for this could have to do with the effects that different face regions have on attractiveness judgments. For example, the eyes appear to play an important role in perceived attractiveness. More than other face regions, when the eyes are obscured attractiveness is reduced relative to when they are visible (Santos & Young, 2011). Hence, it may be that in the absence of other cues that may increase or decrease attractiveness, the eyes play a major role in determining attractiveness judgments and tend to increase perceived attractiveness.
That ratings of dominance/competence were not affected by masks may be interpreted in light of evidence that suggests these judgments rely largely on structural cues such as eyes-to-eyebrows distance and width-to-height ratio (Costa et al., 2017; Vernon et al., 2014), which are not significantly altered by wearing a mask.
The effects of the two designs we explored suggest that specific mask designs, at least our designs, have relatively little effect over and above a plain mask. Our simple smile design failed to impact any trait judgment, suggesting either that perceivers effectively ignored this information when making trait judgments, or that the design itself was not a strong enough representation of a smile to influence judgments. It is possible that a more realistic smile design could have a larger effect on social judgments.
The rationale behind the offset mask was based on the composite effect (Young et al., 1987), aiming to disrupt holistic processing so that social judgments would be based more on the visible parts of the face. Ratings of trustworthiness and attractiveness were not as high in the offset condition compared to the plain mask (i.e., they were closer to ratings in the unmasked face condition), which could indicate it reduced the influence of the mask on social judgments. However, it could also have resulted from a negative evaluative response to the design. Unsolicited feedback from some participants indicated that they found the offset mask to be ‘weird’ or ‘off-putting’. Therefore, it is unclear if the offset mask functioned to disrupt holistic processing or was simply evaluated less positively than the other masks. One recent study suggests that even plain masks disrupt holistic processing (Freud et al., 2020), so the offset mask may have had no further impact beyond the other masks. Either way, our offset mask design produced less positive judgements than either a plain mask or a smile mask.
General Discussion
The studies reported here explored the effect of face masks on social judgments, examining how masks affect judgments of trustworthiness/approachability, dominance/competence, and attractiveness. We also examined the effects of different mask designs, and whether the effects of masks are moderated by face trustworthiness and face race.
Across studies we found a very clear and strong effect of face masks on trait judgments, particularly trustworthiness/approachability and attractiveness. However, the effects were not consistent with expressed concerns about face masks. To the contrary, masks increased perceived trustworthiness/approachability and attractiveness, at least for low trustworthiness faces. Olivera-La Rosa et al. (2020) speculated that the positive effect of masks on trustworthiness judgments found in their study might be due to an “internalized social norm of wearing a mask” (p. 5). This could also explain the current findings, especially since masks also increased attractiveness, suggesting a generally positive evaluation of mask wearers. However, if this were the case one would expect the benefit to extend also to high trustworthiness faces, which we did not see. The effects of masks might also be due to a perceptual process in which judgments are more sensitive to cues to low trustworthiness than to high trustworthiness, and where the eyes play an important role in attractiveness judgments of masked faces. It remains unclear why masks increased perceived trustworthiness and attractiveness and whether this is due primarily to positive social norms for mask wearing or perceptual processes.
We found no evidence that masks encourage racial profiling in the sense of more negative evaluations of Black faces. Despite the Black and White face sets being equivalent according to the norming data from the face database, the unmasked Black faces were rated more trustworthy and attractive than the unmasked White faces in both studies. It is possible that our participants perceived the Black faces as displaying more trustworthy expressions (i.e., smiling more), or our participants may have held more positive stereotypes about Blacks than Whites than those providing the norming data. Either way, there was no effect of race in the masked conditions of either study, and hence no evidence of racial profiling, at least at an explicit level. An important factor to consider is participants’ attitudes. It may be that masks do elicit more negative evaluations of Black faces among participants with pre-existing negative attitudes or ‘prejudice’ against Black people (e.g., Hugenberg & Bodenhausen, 2003). It is also possible that masks increase negative perception of Black faces at an implicit level, which was not detected with the explicit measure used in these studies. Thus, future research could include measures of implicit and explicit attitudes to examine whether attitudes moderate the effects of masks on racial profiling, and utilise implicit measures of evaluation of masked faces. Masks might also have culture-specific meanings, increasing racial profiling of certain groups within certain cultures. Indeed, most of the concerns about racial profiling have been in relation to African Americans in the US, yet we did not specifically sample US participants. Further research could examine the effects of masks specifically in US participants or contexts.
With regard to the different mask designs, we conclude that our designs had relatively little impact over and above a plain mask. The use of a simple smile design failed to significantly alter trait judgments compared to a plain mask. However, more realistic facial expressions could have a larger effect. Furthermore, our attempt to disrupt holistic processing by leveraging the face composite effect did not produce positive results. It is unclear if the design itself was inadequate to disrupt holistic processing or if the design was aversive. It may be worth pursuing the concept of disrupting holistic processing with other mask designs, but the initial results were not positive.
Limitations
There were a number of limitations in this study that warrant discussion. An inherent issue with most face perception studies is the extent to which findings are specific to the face set used. To minimise this issue we used equal numbers of male and female faces, and Black and White faces, while keeping age and expression relatively constant. Nevertheless, it is possible the effects of masks would differ with a different face set. For example, the effect of masks on race-based judgments might be different with faces of a different race. Another limitation is that we have not taken account of any perceiver variables that may moderate the effect of masks on judgments. Although our sample included a wide range of nationalities, we did not treat nationality as an independent variable. Nationality could be relevant if there are cultural differences in the meanings associated with masks. Individual differences in variables such as racial attitudes, trait trust, or pathogen disgust sensitivity (see Olivera-La Rosa et al., 2020) could also be relevant. Further research could investigate such variables.
Implications
There are a number of implications of this research. One is that we found no evidence that masks encourage racial profiling in the sense of more negative evaluations of Black faces. If anything, masks appear to increase perceived trustworthiness and attractiveness of Black as well as White faces. Therefore, fears around racial profiling of mask wearers are not supported by these studies, though further research taking account of perceivers’ attitudes as well as measuring implicit evaluations is warranted. A second implication is that the widely expressed concerns about the effects of masks on social cohesion and trust may also be unfounded. At least in terms of trait judgments, mask do not appear to pose a threat to social cohesion and trust.
Conclusion
As face masks have become more commonplace in many regions of the world due to the COVID-19 pandemic, concerns have been raised about the impact of masks on social perception and cohesion, particularly in relation to perceived trustworthiness. Concerns about racial profiling have also been expressed. The current studies do not support these concerns, suggesting instead that masks have an overall positive effect on social judgments, and that they may attenuate racial biases in face perception. Our studies were only experimental analogues of face evaluations and only captured explicit judgments, so the impact of masks could be different ‘in the wild’, and may have more subtle effects at an implicit level. Therefore, further research is needed. However, the current findings are at least encouraging that widespread mask wearing may not pose a significant threat to social perception and cohesion.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship and/or publication of this article.
