I see how you feel: facial expressions’ recognition and distancing in the time of COVID-19

The COVID-19 pandemic has dramatically changed the nature of our social interactions. In order to understand how protective equipment and distancing measures inuence the ability to comprehend others’ emotions and, thus, to effectively interact with others, we carried out an online survey across the Italian population during the rst pandemic peak. Participants were shown static facial expressions (Angry, Happy and Neutral) covered by a sanitary mask or by a scarf. They were asked to evaluate the expressed emotions as well as to assess the degree to which one would adopt physical and social distancing measures for each stimulus. Results demonstrate that, despite the covering of the lower-face, participants correctly recognized the facial expressions of emotions with a polarizing effect on emotional valence ratings found in females. Noticeably, while females’ ratings for physical and social distancing were driven by the emotional content of the stimuli, males were inuenced by the “covered” condition. The results also show the impact of the pandemic on anxiety and fear experienced by participants. Taken together, our results offer novel insights on the impact of the COVID-19 pandemic on social interactions, providing a deeper understanding of the way people react to different kinds of protective face covering. the we


Introduction
Facial expressions are extremely important for comprehending people's emotions and intentions.
Speci cally, several studies have investigated facial features crucial to convey and correctly recognize speci c emotions. As a result, it has been demonstrated that both the upper (particularly the eyes) and lower (particularly the mouth) face are essential for conveying and decoding emotional facial expressions. Precisely, the eyes and mouth represent crucial cues for detecting angry and happy expressions, respectively [1][2][3][4][5][6][7] .
Recently, the COVID-19 pandemic (Word Health Organization, 2020) has dramatically changed the nature of our social interactions. This is true not only of the lockdown initiated in Italy on March 2020 (when the Italian government announced restrictions on all non-essential activities, with only work and healthrelated travel being authorized), but also of the present days as we still regard other people as possible sources of infection. For the foreseeable future, it will be necessary to wear sanitary masks covering the lower face (i.e., mouth and nose) and do social distancing when interacting with others. How do these unprecedented conditions in uence the way we perceive and comprehend other people's emotions?
Addressing these questions is of utmost importance to understand the consequences of the COVID-19 pandemic on our ability to effectively entertain social interactions with others.
Previous studies have investigated the effect of covered faces on emotion perception, in particular with regard to Islamic veils or headdresses [8][9][10] , showing that the interaction between contextual cues and the covering of facial features in uences emotion recognition. As expected, these results showed that when only the upper part of the face is visible (i.e., the eyes), participants perceive and recognize negative emotions (i.e., anger and fear) better than positive (i.e., happiness) ones. Furthermore, some studies also demonstrated that Islamic contextual cues bias perception toward more intense negative emotions 9,10 .
In the last months, many scholars have addressed the social and emotional consequences of COVID-19 on people's well-being [11][12][13] , but very few have considered the social negative consequences of wearing a sanitary face mask [14][15][16] . Indeed, the presence of the sanitary mask may in uence not only the recognition and comprehension of others' emotions, but, given its current status as a contextual cue of the pandemic, also the attribution of physical and social distance. The physical distance that people maintain between themselves and others, typically de ned as "interpersonal space" 17,18 , can expand or shrink depending on the situational context 17,[19][20][21] . Indeed, it is well known that intrusions into this "personal space" induce a sense of threat and discomfort in individuals 17 . For instance, previous studies have demonstrated that individuals affected by traumatic events keep larger interpersonal distance, such as physically abused children 22 and adults with post-traumatic stress disorder 23 .
On the other hand, it has been theorized and demonstrated that, after a traumatic collective event such as a pandemic, social distancing from members of one's community is modulated in the opposite direction (i.e., people reduce social distancing), resulting in a renewal of the community's sense of social bonding 24 .
In light of the aforementioned studies, the aim of the present study was to investigate how the perception of emotional facial expressions during the current pandemic is in uenced by the covering of lower-face features by a sanitary mask. We also aimed to investigate whether sanitary masks in uence the attribution of physical distance as well as perceived social distance with others. Given the novelty and the timing of the study, our third aim was to explore gender differences.
To this aim, by means of a survey available online between May 12 th and June 1 st 2020, Italian participants were shown static facial expressions (angry, happy and neutral) in two different conditions of "covering": by a sanitary mask and by a scarf. The scarf was chosen for its ecological validity, guaranteeing at the same time that the same portion of the face (from nose to neck) was covered. It has to be stressed that, while the sanitary mask is a personal protective equipment which ensures high protection, the scarf is not as effective as the sanitary mask, ensuring low protection from infection 25 .
Participants were asked to evaluate: the emotion expressed (in terms of valence and explicit categorization), the physical distance and social distance one would adopt with the person depicted in each stimulus. With regard to emotional evaluations, we expected an impairment in the recognition of happy facial expressions as the recognition of this emotion is most affected by the covering of the lower face. We also hypothesized an increase in physical and social distance attribution to negative facial expressions. For all these aspects, a signi cant difference between stimuli covered with a sanitary mask and those covered with a scarf could be expected depending on the intrinsic negative value ascribed by participants to the two kinds of protective equipment, both contextual cues of the current pandemic.

Materials And Methods
Participants Ninety-six healthy Italian volunteers took part in the study: 47 females and 49 males, mean age 36.2 years old (Standard Deviation -SD = 16; min = 19, max = 82). All participants provided written informed consent to participate in the study, which was conducted in accordance with the Declaration of Helsinki (2013). For a detailed description of the sample, please see Supplementary Table 1 online.
Power was calculated a-posteriori by means of GLIMMPSE 26 (https://v3.glimmpse.samplesizeshop.org/#) using the Hotelling-Lawley Trace to test for a repeatedmeasures design with two within-factors and the Condition by Emotion interaction. The signi cance level was set α = 0.05 resulting in an actual power > .80 with our sample size.

Procedure
We created an experimental survey to be administered online by means of Psytoolkit 27,28 . It was composed of 1) a "socio-demographic section" divided into eight different randomized parts, each  Lundqvist et al., 1998). Stimuli were manipulated using Photoshop software (cc2019); speci cally, a sanitary mask (from now on "High protective equipment", HP) or scarf (from now on "Low protective equipment", LP) was added to the faces, making sure that the same portion of the face (from nose to neck) was covered in the two conditions (see Figure  1). Each stimulus was presented four times, one for each of the following four questions: a) Valence: "How would you judge the valence of the expressed emotion?". Participants were asked to answer the questions using a Visual Analog Scale (VAS) from -50 (negative) to 50 (positive); b) Explicit categorization: "Which label would you choose to describe the person's emotion?". Participants were asked to choose the most appropriate emotional label in a group of seven (Anger, Happiness, Disgust, Fear, Neutral, Sadness, Surprise); c) Social distance: The Inclusion of Other in the Self -IOS 30 scale was used by asking: "Which picture best describes your relationship with this person (as if s/he was a member of your community)?". Participants were asked to answer selecting one of the seven images (i.e., pair of circles) in the IOS; d) Physical distance: "How (physically) distant would you like to keep this person away from you?". Participants were asked to answer by means of a VAS ranging from 0 (very close distance) to 100 (very high distance). Hence, this experimental section was composed of 288 "trials" in total, with the following design: 3 Emotions (Anger, Happiness, Neutral) * 12 Actors (Female, Male) * 2 Conditions (HP, LP) * 4 Questions (Valence, Explicit Categorization, Physical Distance, Social Distance). Participants were asked to observe each stimulus and to answer the question positioned below without any time constrains. Two different versions of the survey were created balancing the presentation order of the two sections (socio-demographic section followed by experimental section vs. experimental section followed by socio-demographic section).

Valence and Physical Distance
In order to investigate whether Valence and Physical Distance are modulated by the experimental conditions, a linear mixed effect analysis was performed. Following a hierarchical approach, we initially created a simple model using one parameter, and we progressively added others with the aim to evaluate whether their inclusion improved model t. Likelihood ratio tests, Akaike Information Criterion (AIC) and Bayesan Information Criterion (BIC) were used to establish whether the inclusion of main effects, interaction effects and random effects would signi cantly improve model t. We entered participants' For more details about selected models please see Supplementary Table 3  Oriented Thinking (EOT)) scores and with IRI subscales scores (the critical probability values for multiple comparisons were corrected with the Bonferroni method: 0.05/6 = .008).

Discussion
The aims of the present study were three-folded: 1) To investigate how the perception of emotional facial expressions is in uenced by the covering of lower-face features by different kinds of protective facecovering within the context of the COVID-19 rst pandemic peak; 2) To investigate whether protective face-covering in uenced the attribution of physical distance as well as perceived social distance with others; 3) To explore gender differences. To this purpose, by means of a survey administered online between May 12 th and June 1 st , 2020, Italian participants were asked to observe static facial expressions (Angry, Happy and Neutral) in two different conditions of "covering" (with a sanitary Mask -HP and with a Scarf -LP). Participants were asked to explicitly Categorize the emotion, rate it for Valence, and evaluate how much Physical and Social Distance they would choose to keep from the person depicted in the stimulus.
Results for Valence ratings show that both Female and Male participants rated Angry facial expressions as more negative than both Happy and Neutral ones, and they rated Happy facial expressions more positively than Neutral ones. Nonetheless, Female participants gave more negative ratings than Male ones when evaluating Angry and Neutral facial expressions, and more positive ratings when evaluating Happy facial expressions (see Figure 2). While these results show that participants rated the Valence in accord with the expressed emotion, they also reveal a polarizing effect of ratings in Females. This is consistent with previous studies demonstrating women' stronger sensibility to face stimuli (e.g., higher ratings for experienced valence and arousal, greater attention to the eyes, better decoding of emotions through facial expressions, and greater electro-cortical responsivity) [45][46][47][48][49] . Remarkably, the absence of a signi cant interaction between Emotion and Condition suggests that the presence of a HP or a LP does not modulate emotional evaluations of this kind.
Results for the explicit Categorization of expressed emotions con rm participants' ability to correctly recognize emotional expressions, as they chose emotional labels congruent with the displayed emotions. Interestingly, for Neutral facial expressions, participants also chose the Sadness label more frequently than chance, particularly in the case of Female participants (see Supplementary Table 4 online). Altogether, both the results for Valence and Categorization show that, despite the covering of the lowerface, participants were able to properly recognize the facial expressions of emotions.
Furthermore, results are also characteristic of the empirical challenges posed by operationalizing neutral emotion, and in particular of the recurrence of "negative bias" (i.e., the attribution of negative ratings to neutral stimuli) 50 . It is also possible that participants' psychological state due to the pandemic might have in uenced their responses to neutral stimuli.
With regard to Physical Distance, the expressed emotion and the condition of "covering" both in uenced the results. Indeed, participants chose to keep the least distance from positive facial expressions, followed by Neutral and Angry ones. Interestingly, Female participants chose to keep greater Physical Distance from Angry facial expressions and less Physical Distance from Happy facial expressions with respect to Male participants (see Figure 3). These results are coherent with previous studies that demonstrate an increase in interpersonal distance when participants are exposed to Angry rather than Happy facial expressions 51,52 .
With regard to the "covered" condition, less Physical Distance, only in Male participants, was attributed to facial expressions covered by HP than to those covered by LP. This latter result can be explained by the fact that the Scarf offers far less protection from the infection with respect to a sanitary Mask, facilitating a higher risk for individuals if distancing measures are not undertaken 25 . Hence, it is shown for the rst time that, in a pandemic context, the use of an appropriate protective device reduces the desire to maintain Physical Distance and thus potentially improves interpersonal social relations. Additionally, in accord with previous studies, we found that less Physical Distance was attributed to Female than to Male stimuli 19,53,54 .
Results for Social Distance revealed that, similarly to Physical Distance, participants chose to keep the least distance from Happy facial expressions, followed by Neutral and Angry ones, and to Female stimuli than to Male ones. While Female participants felt more Socially Distant than Male for stimuli covered by both HP and LP, only Male participants chose more Social Distance from faces covered with LP. Lastly, Female participants chose more Social Distance from Angry and Neutral facial expressions than Male participants. The results of the Physical and Social Distance are substantially overlapping (see Figure 4): Females show a greater tendency, compared to Males, to keep at distance the negative or ambiguous faces regardless of the type of protective equipment (HP or LP). This is very interesting, because the need to maintain greater distance from potentially dangerous stimuli in Females seems to be driven by the emotional content of the stimuli. Differently, Males' ratings of Social and Physical distances depend on the type of protective equipment. Females, in order to establish social and physical distance from other individuals, use more 'empathic qualities' of faces than Males, whose responses seem to be more dependent on contextual cues.
One could also have expected a difference between Physical and Social distance ratings (i.e., increase of interpersonal/physical distance and a reduction of social distance after a traumatic event; see [22][23][24], but it has to be considered the period during which our data was collected. Circulation of the survey began on May 12 th , 2020, two months after lockdown was initiated in Italy and the World Health Organization announced that COVID-19 would be classi ed as a global pandemic. Indeed, previous studies demonstrate that the practice of social sharing, which enhances social bonding among members of the same community, initiates immediately after a traumatic event and does not last for longer than three weeks 24,55 . Given the urgent need to mitigate the psychological consequences of the pandemic, we carried out an investigation of current attitudes toward COVID-19 by correlating the Fear of COVID-19 scale with the Health Anxiety total scores and IRI-Personal Distress subscale scores. Results showed that the higher the scores for Health Anxiety and IRI-Personal Distress, the greater the fear of COVID-19, thus pointing to a correlation between these personality aspects (see Figure 5).
This study has some potential limitations. Firstly, we should consider that future studies are needed to investigate whether the adoption of additional control conditions would have an impact on results.
Secondly, our sample of participants was composed by Italian people only, thus we cannot exclude that cultural diversity in the habit of wearing sanitary masks may in uence results. Finally, the long duration of the survey may have affected participants' attention throughout the experiment.
Taken together, the present results offer novel insights on the consequences of the COVID-19 pandemic on our ability to effectively interact with each other. This study also highlights the impact of the pandemic on anxiety and fear experienced by participants. Future studies are needed to investigate the social and emotional costs of wearing sanitary masks (e.g., for teacher-student and clinician-patient communication as well as for individuals that are hearing-impaired) 16