The Impact of Face Masks on the Emotional Reading Abilities of Children—A Lesson From a Joint School–University Project

Wearing face masks has become a usual practice in acute infection events inducing the problem of misinterpreting the emotions of others. Empirical evidence about face masks mainly relies on adult data, neglecting, for example, school kids who firmly are dependent on effective nonverbal communication. Here we offer insights from a joint school–university project. Data indicate that emotional reading of 9 to 10 years old pupils (N = 57) was similarly impaired as adults on an overall performance level, but that their selective performance on specific emotions was quite different. Kids showed extreme problems in reading the emotion disgust, strong effects on fear and sadness, and only mild effects on happiness, but also even better performances for emotional states anger and neutral when faces were masked. This project did gain not only relevant data about children’s perception but also made clear how fruitful seriously conducted school projects can be to encourage the interest and commitment for Science, Technology, Engineering, and Mathematics (STEM)-relevant topics.


Introduction
impact of masks on negative emotions is less clear, but it seems that masks do not specifically 28 increase the feeling that the mask wearer shows negative emotions (Carbon, 2020;Marini, et al., 29 2021). additionally, they employed a condition where sunglasses had been graphically added. Children 44 made a specific study necessary, especially as the wearing of face masks has become a political 69 issue (Wong, 2020) and the acceptance of masks is generally under risk (Egan, et al., 2021). The 70 topic of wearing masks is particularly emotionally and politically charged when it comes to 71 children. Consequently, usage of face masks is highly debated for schools in particular (Spitzer, 72 2020). The major force behind aiming the present study was the second author (MS), a 9-year 73 old schoolboy from Florida, who contacted the first author (CCC), who is a perceptual scientist 74 with a focus on face research. After having read an article about CCC's specific research on 75 emotional reading of faces in adults (i.e., Carbon, 2020) with masks typically used in the first 76 wave of the COVID-19 Pandemic, MS was curious to know about the possibility to extend the 77 study to a sample with school kids. He planned to conduct a replication study specifically with 78 children because he identified this a valuable study as school kids strongly depend their everyday 79 communication, especially in classrooms, on non-verbal communication-and even if this 80 emotional reading might be limited in most cases to affirmations by expressing a happy face. MS 81 also aimed at submitting such a replication study to the school's STEM fair in 2020 (which he 82 won-and, fortunately, he subsequently also won the 2 nd prize in the district STEM fair in 2020 83 with this project). 84 1 We will use the term "mask" in a broad sense comprising typical accessories to cover the nose and the mouth used during the COVID-19 pandemics. These include non-professional protective items such as loop scarfs and homemade community masks as well as certified face masks such as FFP2/N95 masks. Note: Non-protective / lessprotective items like scarfs may be perceived differently than professional items of course (see Calbi, et al., 2021). important side kick of this research project later to make clear how STEM projects can attract 92 school students by active involvement of scientific advisors and supervisors. All means towards 93 the final product of a scientific paper were finely concerted with MS's mother in order to 94 optimally support MS. 95 The main aim of the present study was to analyse the performance of reading basic 96 emotions in faces that were masked vs unmasked. Here, we were particularly interested in 97 gaining knowledge about the specific confusion of expressed emotions with the perception of 98 these emotions. As the first author has conducted a similar study at the start of the first wave of 99 the COVID-19 Pandemic with adults (Carbon, 2020), we were also keen to compare both 100 datasets to gain knowledge on the specific problems children have with reading emotions from 101 masked faces. 102
to an average performance of correctly recognising the presented facial in 60% of the cases. For 111 Model #1 we furthermore assumed the effect of face masks as a slope of -2.5 which corresponds 112 to a decrease of correct recognition of emotions of 2.5% when faces with masks were presented 113 this rather small assumed slope was employed to be able to detect even small effect sizes with 114 the targeted sample size. The random intercept variance was set to 10.0 and the residual standard 115 deviation was set to 20.0; the α error level was set to 0.05. The desired test power (1-β) of 0.90 116 was approached with a minimum N = 57 (95%-CI of test power with 1,000 simulations: 0.88-117 0.92). 118

119
Material. All stimuli were based on frontal depictions of faces which were obtained from the 120 MPI FACES database (Ebner et al., 2010) by a study-specific contract. Specifically, we 121 employed frontal photos of four white European persons (previously called "Caucasian"), two 122 female and two male, who belonged to two different face age groups (young: young persons 123 #140 & #066 and medium: middle-aged persons #168 & #116-the hashtag numbers refer to the 124 MPI FACES notation)-this range of young up to middle-aged adults was used to reflect the 125 typical school setting with teachers of that age range in a school setting (but of course children 126 also interact with people of different ages, e.g., younger persons as peers, old people as

136
showing six different emotions without a mask ("no mask") and wearing a mask ("mask" Overall performance for correctly identifying facial emotions in faces without masks was 176 remarkable, M = 89.9% (chance rate: 1/6 = 16.7%), with average performance rates ranging 177 across participants between 61.1%-98.6%. As indicated by Figure 2, kids were particularly good 178 at recognising the emotional state happiness, fear and neutral, followed by recognition 179 performances, being still higher than 80%, for anger, disgust and sadness. As soon as faces were 180 covered by a typical blueish surgical mask, we detected an overall decrease of performance to 181 77.7% with average performance rates ranging across participants between 59.7%-90.3%. We 182 observed a pretty diverse pattern of performance changes from recognising faces without masks 183 to faces with masks: While we detected a dramatic drop in performance for disgust, the decreases 184 for fear and sadness were still evident but less substantial (only about 10% of performance 185 decrease). In addition, for happiness, the decline was only about 5%. Somehow unexpectedly, we 186 also registered two emotions which showed better recognition performances: anger and neutral 187 showed an increase of performance by about 4%. exprEmo as a further fixed factor following the idea that a face mask has a specific impact on the 205 recognition of certain emotions which are mainly expressed by the facial information around the 206 covered mouth-nose area. We always tested the more complex model with the preceding model, 207 for instance, Model #1 against Model #0 via likelihood ratio tests. Each model's residuals were 208 visually inspected to exclude models deviating from homoscedasticity or normality.  Table 2 shows the parameters of the finally selected Model #2 which explains 27.8% of the 218 variance of the data. From session to session, participants earned higher recognition 219 performance, indicated by significant effects of Session 2 and Session 3 tested against Session 1 220 (indicated by "Reference" in Table 2). Most importantly, we did not only find an overall effect of 221 hampered emotional reading when masked faces were shown, but face masks had specific effects emotions and face mask wearing or not. We detected a particularly large effect of face masking 224 on the reading ability of disgust, substantiated by an estimate of -56.73 for the interaction of 225 exprEmo and facemask for the emotional state of disgust (see Table 2). 226 Additionally, we followed a signal detection theory (SDT) approach to investigate whether the 230 impact of face masks was mainly about the sensitivity of reading emotions or the response bias 231 based on a different decision criterion. For conducting this additional analysis, we did not any 232 more taking the sessions and different base faces into account. Figure 3  was always below 0.60) and for happiness and sadness, the change was even less pronounced. 242 We obtained an evident change of decision criterion c for disgust only-see Figure 3.  Table 3 for statistical details. 263 264

267
We were further interested in how specifically face masks affected the ability to read emotions 268 regarding the confusion of emotions (misperceiving and expressed emotion as a different one). 269 As the confusion matrices in Figure 4  Although the overall pattern was relatively similar to the original study from May 2020 where 291 we tested adult participants (Carbon, 2020), there are also important differences to be reported. but also of elderly people, we processed the Carbon (2020) data set excluding the data for images 294 of the elderly persons in the following. As can be seen in Figure 4 (right panel), the control data 295 of adult persons showed particular confusions for the emotions of sadness and disgust when 296 masks were presented; however, for children, anger was even more often detected when a mask 297 was present vs. no mask was shown, whereas adults suffered a drop of performance in this 298 respect. These dissociations warranted a deeper look into the data, so we decided to analyze the 299 differences on basis of the signal detection theory (SDT) to be able to decide whether both 300 groups differed primarily in terms of sensitivity or the decision criterion. In order to compare 301 both data sets, we conducted two independent Linear Mixed Models (LMM), one for the 302 sensitivity measure dprime and the other for the decision criterion c. The LMM which we 303 employed (Model #C) contained Study (kids vs. adults), faceMask and exprEmo as fixed factors 304 with full interaction among these factors, and caseID (participant) as the only random factor. 305 Table 4 shows that kids showed overall lower scores for dprime, but higher scores for decision 306 criterion c which indicate more conservative responses on average. However, we have to be 307 cautious in interpreting main effects before analysing the interactive effects. We could indeed 308 find dissociate patterns of the role of face masks, depending on the respective emotion: While 309 kids showed higher sensitivity scores and more liberal response behaviour for anger, they 310 responded on disgusted faces in a more conservative response way. 311 312 313 We tested school kids aged 9-10 years on their recognition performance of facial emotions in the 320 times of the COVID-19 Pandemic where face masks were common hygienic accessories to 321 mitigate possible infections. The participants had to recognise emotional expressions displayed 322 by faces which we showed with and without masks. The recognition performance was further 323 qualified by comparing the data with a similar study that tested adult participants in May 2020 324 (Carbon, 2020). 325 First of all, the kids performed very well on a general basis. They reached nearly 90% of 326 correct responses when confronted with faces without masks. This is quite remarkable as many 327 theories claim a needed and ongoing maturation of face processing skills lasting about 12-14 328 years, particularly to develop the so-called configural processing mode (Mondloch, Le Grand, & 329 Maurer, 2002;Schwarzer, 2006). Other researchers focusing on so-called holistic processing 330 have found similar late maturation of expertise-based facial processing at an age between 11-15 331 years (Carbon, Grüter, & Grüter, 2013), while other research indicated even longer periods 332 needed to become a face expert, actually Germine, Duchaine, and Nakayama (2011) revealed in 333 an extensive online study that learning abilities on faces peak after about an age of 30 years. 334 When compared with the Carbon (2020) study employing adults with a mean age of 26.7 years 335 ranging between 18 and 87 years, we see a highly comparable level of overall performance (for detection theory approach. We revealed that face masks mainly impacted the sensitivity, but not 341 so much the decision criterion of the children. Only the emotion of disgust was very much 342 impacted by changing to a more conservative response behavior which means that children were 343 less decisive in reporting the status of that emotional expression when a face mask was present. indicating selective processes which might be at work when recognising emotions. This could be 356 interpreted by non-unitary cognitive processing of emotions, which differs from standard models 357 that assume general processing modes (Bruce & Young, 1986). While kids were nearly perfect in in faces presented with masks. In contrast, they even did better in detecting anger when faces 360 were covered with masks. This could indicate that they more pronouncedly relied on the eyes 361 region through which the emotional state of anger is mostly expressed. However, the clear drop 362 of fear recognition, which is strongly expressed by eyes wide as saucers, does not support this 363

view. 364
Previous studies identified the emotional states happiness and sadness, and to a lesser 365 degree, also anger, as being mostly expressed by the lower facial part (Bassili, 1979 Compared with these emotions, we only detected mild negative effects in recognising happiness. 370 Most unexpectedly was the finding that the emotional state of anger and a neutral emotional state 371 could even be better identified when face masks were present. These results were substantiated 372 by respective significant interactions of these emotional states with faceMask (Table 2). In 373 similar studies employing alternative means of covering the mouth area, it could be shown that 374 anger was at least affected much less by occlusions through a rectangular cardboard (Bassili, 375 1979)  Most notable in the present study is the strong negative effect of face masks on the 393 recognition of disgust. Adult participants in an earlier, very similar study (Carbon, 2020) also 394 showed a pronounced decline of performance for this specific emotion, however, adult 395 participants mainly misperceived disgust as anger-reflecting a common finding (see, e.g., 396 Cigna, Guay, & Renaud, 2015)-but they did not confuse disgust with sadness as the kids did. 397 We do not yet know the base of this effect, but it points to the relevance of such findings for 398 children's everyday life: Face masks cover a large part of the face, and only so they seem to be 399 effective in mitigating the viral load entering the respiratory passages through the mouth and the 400 nose. This comes at a price, e.g. an imposed change of the processing of faces. Changing 401 processing does not evidently mean to impede the informative value of such processes. This can were masked. Only the recognition of disgust was indeed dramatically affected, calling for 404 effective and easy to implement countermeasures: In situations where disgust is aimed to be 405 expressed, this should be accompanied by explicit and clearly pronounced verbal wording and, The present study tells us a further, important lesson: As we know, our capabilities as 417 scientists to attract young people for STEM (Science, Technology, Engineering, and 418 Mathematics) is rather limited, but explicit interest in our research expressed by children should 419 be open-heartedly taken up. For fruitful, stimulating and productive interactions between STEM 420 research and schools, it is important to take such interactions (like the present one where a 9-year 421 old schoolboy contacted a perceptual scientist) very seriously. We should support naturally 422 interested school kids with the same level of commitment as typically invested in regular 423 collaborations. Based on such a spirit, such collaborations will not be just promotionally