Frequent experience with face coverings for 10 months improves emotion perception among individuals with high autistic traits: A repeated cross-sectional study

Face coverings pose difficulties for emotion recognition, but it is unclear whether improvement in recognising emotions from the eyes is possible with experience and whether this might be dependent on one’s autistic traits, given the associations between high autistic traits and poorer emotion perception and reduced gaze to the eye region. In this preregistered study, participants completed a forced-choice emotion recognition task with photographs of eyes and demographic questionnaires that measure their autistic traits and their interaction frequency with others wearing face coverings at two time points: once at the start of the face covering mandate and again 10 months later. We found that after 10 months, individuals with high autistic traits as a cohort recognised emotions from just the eyes better as a function of their experience with others wearing face coverings, suggesting that emotion perception is malleable even for those who have difficulties with emotion perception.

2 Section S1: Data Analysis using ANOVA Following the study preregistration protocol, we analysed the data using an ANOVA, given that it is commonly used in emotion perception research though we recognise the limitations of doing so given the binary nature of the dependent variable (Jaeger, 2008). Similar to the mixed effects model, we entered autistic traits (AQ), Wave (Wave 1 vs Wave 2), Emotion (Basic vs. Complex), Experience with Others Wearing Face Coverings (hereafter 'Face Covering': Rarely vs. Sometimes vs. Often) and all the possible interactions between them as fixed effects using the aov() function. In our case, we obtained similar findings to that found in the mixed effects model, which are described below.
There was also a main effect of Emotion (F(1, 554) = 236.98, p < .001), which was qualified by an Emotion × Wave interaction (F(1, 554) = 9.02, p < .003). Post-hoc comparisons revealed that the difference in performance across Waves was not significant for Basic emotions (t(944) = 0.58, p = .564) but performance was better for Wave 2 than Wave 1 for Complex emotions (t(944) = 2.48, p = .013). Importantly, we also found a significant Face Covering × AQ × Wave interaction (F(2, 554) = 4.60, p = .011). Examining the effect of AQ at each level of Face Covering by Wave revealed a significant negative AQ effect for Rarely Wave 2 (t(554) = 2.96, p = .003) and Often Wave 1 (t(554) = 4.07, p < .001). Direct comparison of the estimated AQ slope between Waves for each Face Covering level revealed a significant difference for the Often condition (t(554) = 2.33, p = .020), with the slope being more negative for Wave 1 than for Wave 2, and a marginally significant difference for the Rarely condition (t(554) = 1.97, p = .050), with the slope being more negative for Wave 2 than for Wave 1.

Section S2: Analysis on Wave 1 Data
As stated in the study preregistration protocol, we analysed data from Wave 1 to replicate expected findings (i.e., the negative relationship between autistic traits and emotion recognition accuracy) using both binomial mixed effects model and ANOVA. For the binomial mixed effects model, we fitted a similar model as the main analysis minus the Wave predictor and all its interactions. The output is shown below in Table S2.1. There was a negative effect of AQ ( 2 (1) = 8.89, p = .003; B = -0.01, SE = 0.00) as expected, and a Face Covering × AQ interaction ( 2 (2) = 7.89, p = .019). Follow up tests revealed that the AQ effect was only significantly negative for the Often condition (z = 3.94, p < .001), and direct comparisons of the AQ effect between Face Covering levels showed a significant difference only between Often and Rarely (z = 2.73, p = .018), with the AQ slope being more negative for the former (see Figure S2.1).

Section S3: Data Analysis using Autism Diagnosis
To examine whether our findings would be replicated using autism diagnosis rather than autistic traits, we ran a similar mixed effects model as reported in the manuscript, with the exception that we replaced AQ (a continuous predictor) with participants' self-reported autism diagnosis (Diagnosis; a categorical predictor). In both waves, approximately 10% of the participants identified as autistic (Wave 1 n = 29; Wave 2 n = 27) whereas the rest were classified as neurotypical (Wave 1 n = 279; Wave 2 n = 231). The results of the mixed effects model can be seen below in Table S3.1. Generally, the same findings were observed. The Diagnosis predictor ( 2 (1) = 7.06, p = .008) and the interaction between Face Covering × Wave ( 2 (2) = 9.77, p = .008) were significant, both of which were qualified by a Face Covering × Diagnosis × Wave interaction 6 ( 2 (2) = 8.65, p = .013) akin to the three-way interaction of Face Covering × AQ × Wave seen in the main analysis. Pairwise comparisons revealed that improvement across Waves was only observed among autistic participants in the Often condition (z = 3.41, p = .001; see Figure S3.1).

Section S4: Analysis on Wave 2 Data using Autistic Traits and Alexithymia Scores
To examine the relative contribution of autistic traits and alexithymia on emotion recognition ability, we fitted a binomial mixed effects model on Wave 2 data (during which we collected both autistic traits and alexithymia scores) using the following as fixed effects: Emotion (Basic vs. Complex), Experience with Others Wearing Face Coverings (Face Covering; Rarely vs. Sometimes vs. Often), autistic traits (AQ), alexithymia scores (TAS), and all the possible interactions between them. As random effects, we entered random intercept for participant and item, and random by-participant slope for Emotion. The model results are displayed in Table S4.1.  Figure S4.1). Additionally, there was also a three-way Emotion × Face Covering × AQ interaction ( 2 (2) = 6.16, p = .046). Subsequent tests for the effect of AQ by each level of Emotion and Face Covering revealed no significant effect of AQ, but the slope for AQ was significantly more negative for the Complex emotions relative to the Basic emotions only in the Often condition (z = 2.31, p = .021; see Figure S4.2).