Social Anxiety Symptoms Predict Poorer Facial Emotion Recognition in Autistic Male Adolescents and Young Adults Without Intellectual Disability

Abstract

Utilizing a novel computerized task, we aimed to examine whether social anxiety symptoms would be related to individual differences in facial emotion recognition (FER) in a sample of autistic male adolescents and young adults without intellectual disability. Results indicated that social anxiety and IQ predicted poorer FER, irrespective of specific emotion type. When probing specific effects within emotion and condition types, social anxiety impacted surprise and disgust FER during a truncated viewing condition and not full viewing condition. Collectively, results suggest that social anxiety in autism may play a larger role in FER than previously thought. Future work should consider the role of social anxiety within autism as a factor that may meaningfully relate to FER assessment and intervention.

Introduction

Understanding emotion in others is fundamental to navigating the social world and requires multiple cognitive processes which support discrimination, recognition, and interpretation of facial expressions, gestures, and speech (Ekman et al., 1980; Ferretti & Papaleo, 2019). One important mechanism that facilitates understanding of emotions is facial emotion recognition (FER), or the ability to discriminate and recognize nonverbal facially-displayed emotion (Hobson et al., 1988; Kuusikko et al., 2009). In typical development, FER capabilities improve with age, such that by adulthood basic emotions (i.e., disgust, anger, fear, surprise, sadness, and happiness) are more quickly and accurately recognized, as compared to earlier developmental stages (Lawrence et al., 2015; Rump et al., 2009). Several studies have examined FER processes in autism and found diminished FER abilities to be associated with a variety of negative outcomes, including decreased social cooperation, higher incidence of internalizing symptoms, and increased social difficulties (Bal et al., 2010; Izard et al., 2001). Although several studies have examined FER in autism as compared to neurotypical individuals, this literature has yielded inconsistent results, with some studies finding that autistic samples demonstrate no FER difficulties, some indicating global FER difficulties and others indicating difficulties in select emotions (Harms et al., 2010; Lozier et al., 2014; Uljarevic & Hamilton, 2013; Weigelt et al., 2012). These findings have led to the conclusion that FER difficulties may not be universal in autism, underscoring a need to identify factors that account for the heterogeneity in findings (Nuske et al., 2013). Previous work has linked autistic traits, co-occurring social anxiety symptoms, and alexithymia (i.e., difficulties recognizing and interpreting one’s own affective states; Sifneos, 1973) to FER difficulties (Cook et al., 2013; Kleinhans et al., 2010; Yeung et al., 2020). Thus, using a novel FER task with dynamic face stimuli (Dobs et al., 2018; Richey et al., 2022), we aimed to examine the impact of these variables on FER in a sample of autistic male adolescents and adults.

In typical development, of the basic emotions studied, happiness has been reported as the easiest emotion to recognize, while fear and disgust have tended to be the most difficult to recognize (Lawrence et al., 2015; Montagne et al., 2007). FER tasks examining basic emotions have varied on dimensions such as valence, intensity, and duration of emotional stimuli, which have added to the heterogeneity in performance in autistic samples (Harms et al., 2010; Nuske et al., 2013; Yeung, 2021). In autistic adolescents and adults, some studies using static facial expression stimuli of basic emotions have found intact FER accuracy (Loveland et al., 2008; Neumann et al., 2006; Ogai et al., 2003; Rutherford & Towns, 2008; Weng et al., 2011), while other studies comparing FER of basic emotions have reported diminished accuracy particularly for negative (i.e., fear, anger) but not positive emotions (Ashwin et al., 2006; Bal et al., 2010; Corden et al., 2008; Wallace et al., 2011; Wallace et al., 2008; Yeung et al., 2020). Additionally, compared to typical adolescents and adults, FER difficulties have been found for autistic adolescents and adults in studies using tasks with complex, subtle, or morphing emotions (Baron-Cohen et al., 1997; Greimel et al., 2010; Humphreys et al., 2007; Law Smith et al., 2010; Teunisse & de Gelder, 2001). FER tasks using dynamic morphing face stimuli have allowed for the study of gradients of subtle changes in facial emotion expressions and are considered to be more ecologically valid than static images of facial expressions, as they elicit richer neural, autonomic, and behavioral responses than static images and can provide nuanced information about emotion recognition processes (Enticott et al., 2013; Kilts et al., 2003). For example, diminished FER accuracy for disgust, anger, and surprise has been found among autistic adolescents using a dynamic morphing face task with disgust not being accurately identified even at full intensity, and anger and surprise intact at full intensity, but impaired at lower levels of intensity compared to typical adolescents (Law Smith et al., 2010). Altogether, these findings indicate that tasks using subtle, ambiguous, and brief facial emotion stimuli may more finely capture FER processes in autism, and the heterogeneity of findings calls for variables that may impact performance.

Social anxiety presents as one such candidate, as prior work has highlighted it as a co-occurring pathology that contributes to social difficulties in the context of autism (Spain et al., 2018). As such, facilitating future understanding of co-occurring social anxiety symptoms and their role in FER processes within autism may elucidate novel risk pathways for social difficulties as well as mechanisms for targeted treatment (Kuusikko et al., 2009; White et al., 2014). Beginning with the observation that social anxiety is highly prevalent in autistic samples (Bejerot et al., 2014; Kent & Simonoff, 2017), a number of prior studies have further established that autistic traits are associated with increased levels of social anxiety symptomatology (Albantankis et al., 2020; Bellini, 2004; Kuusikko et al., 2008; White et al., 2009). Along similar lines, co-occurring diagnoses of social anxiety disorder (SAD) have been found to be as high as 50% in autistic adults (Maddox & White, 2015). Moreover, a systematic review of social anxiety in autism found that high social anxiety symptoms were linked to poorer social skills, social competence, and social motivation in autism (Spain et al., 2018). This work complements a related body of literature which has examined FER in socially anxious samples and found biased and impaired FER compared to non-socially anxious samples (Gutiérrez-García & Calvo, 2014; Maoz et al., 2016; Wieckowski et al., 2016). Collectively, this suggests that social anxiety may be a clinically relevant variable that may contribute to FER difficulties noted in autism.

A growing body of work has demonstrated specific alterations in facial processing and impairments as well as biased FER in social anxiety using primarily static stimuli in FER tasks. For example, compared to neurotypical adults, adults with SAD were more likely to interpret ambiguous and neutral faces as negative emotions and responded more rapidly to negative facial expressions (Gutiérrez-García & Calvo, 2014, 2017). This negative bias in adults with SAD has also been supported by slower response to happy faces and faster response to angry faces compared to non-socially anxious adults (Maoz et al., 2016). Further, in adolescents, higher social anxiety symptoms were associated with less accuracy in identifying fearful faces (Wieckowski et al., 2016). In recent work, there has been an emerging appreciation for the links between social anxiety symptoms and FER in autism. This work has primarily used FER tasks with static stimuli and has suggested mixed findings. One study found that social anxiety symptoms did not correlate with FER using static stimuli in autistic children (Wong et al., 2012). Similarly, Spain and colleagues (2016) did not find a relationship between social anxiety symptoms and social-emotional tasks (i.e., including FER tasks using static stimuli) in an autistic adult sample (Spain et al., 2016). Conversely, neural mechanisms of emotion face processing have been linked with social anxiety in autistic adults, such that activation of amygdala and temporal areas during an emotion matching task was associated with increased social anxiety, while diminished activation in the fusiform face area was associated with increased social anxiety, suggesting that autistic adults with high social anxiety may be more sensitive to emotional faces, which may contribute to differences in interpretation, avoidance, and responses to emotional face stimuli (Kleinhans et al., 2010). Further, poor fear recognition and diminished fixation of eyes has been specifically linked to social anxiety symptoms in an autistic adult sample (Corden et al., 2008). Moreover, greater socially anxious cognitions (e.g., fear of negative evaluation) in autistic adolescents has been associated with greater visual attention to static socially threatening stimuli (i.e., disgust, anger) above and beyond autistic traits (White, Maddox, & Panneton, 2015). Thus, there is some evidence to suggest altered FER processes in autistic adults with high social anxiety. Despite previous work on FER in autistic samples and socially anxious samples using dynamic face stimuli, no studies to date have examined dynamic FER performance as it relates to social anxiety in autistic samples. Such tasks may be able to further elucidate on social anxiety mechanisms that alter FER processes in autism.

Although FER tasks using dynamic images can elucidate specific mechanisms that may be impacted by social anxiety (i.e., avoidance of facial features, intolerance of emotional ambiguity, rapid interpretations of varying emotion expression intensities), literature focused on FER using dynamic images within the context of social anxiety is limited. Emerging literature has supported that FER mechanisms may not only be impacted by the threatening and complex emotions, but also in rapid interpretations of morphing images. Joormann & Gotlib (2006) found that adults with social anxiety detected angry facial expressions at lower intensity levels. Heuer and colleagues (2010), extended this work by examining FER with morphing faces from neutral to increasing emotion intensity levels in adults with high social anxiety and non-anxious adults. Participants completed two conditions, (1) a “restricted viewing condition,” in which participants indicated the earliest point they recognized the expression and the face immediately disappeared, prompting the participant to choose an emotion from a list of options, and (2) a “free viewing condition,” in which participants were allotted open-ended time to observe and manipulate the morphing expression prior to choosing an emotion. Interestingly, difficulties in FER were observed in those with high social anxiety only during the “restricted viewing condition,” such that a negative bias was present in those with high social anxiety (i.e., disgust more likely to be interpreted as contempt) and a positive bias in non-anxious adults (i.e., disgust more likely to be positively interpreted as happy). Moreover, a recent study using dynamic face stimuli found that adults with social anxiety detected negative emotions (i.e., anger, disgust) at lower intensity levels than non-anxious adults (Gutiérrez-García & Calvo, 2017). Relatedly, Stephenson and colleagues (2019) examined the role of worry traits on a FER task using dynamic face stimuli in an autistic sample and found that higher worry traits were related to faster response time of FER, and further that higher worry and lower autistic traits predicted attentional avoidance of negative stimuli, suggesting the importance of co-occurring emotional and behavioral symptoms on FER tasks. Altogether, FER differences in socially anxious individuals may be particularly prominent when rapid inferences are made about the emotion identity of low intensity expressions, and these findings parallel results from dynamic face tasks in autistic adults without intellectual disability. Thus, an important question remains, as to whether social anxiety symptoms impact dynamic FER processes in autistic individuals.

Another variable of interest that has been linked to social-emotional outcomes in autism is alexithymia, or the difficulty in recognizing and distinguishing one’s own affective states (Poquérusse et al., 2018; Sifneos, 1973; Valdespino et al., 2017). Alexithymia has been found to be pronounced in autistic samples, with a recent review identifying a prevalence of about 50% in autistic samples vs. 5% of neurotypical samples (Kinnaird et al., 2019). Mechanisms related to alexithymia in autism may also be linked to difficulties with empathy and understanding emotions in others (Bird et al., 2010; Cook et al., 2013; Keating et al., 2021; Valdespino et al., 2017; Sivathasan et al., 2020). Due to the overlapping traits between autism and alexithymia, Bird and Cook (2013) have posited the alexithymia hypothesis, which suggests that difficulties in emotion processing in autism stem from the elevated rates of alexithymia. Consistent with this line of work, Cook and colleagues (2013) assessed the degree to which alexithymia was associated with impaired FER in a sample of autistic adults and found that alexithymia, not autistic traits, predicted the precision of responses of morphing faces of disgust-anger and surprise-fear. Similarly, one study found that autistic women with high levels of alexithymia were less able to detect static images of basic emotions at lower intensity levels than autistic women with low levels of alexithymia (Ketelaars et al., 2016). Relatedly, another study using brief video clips of 14 different facial emotions found that alexithymia, not autistic traits, predicted impaired FER in autistic women (Ola & Gullon-Scott, 2020). Although there is work individually examining associations between autistic traits, alexithymia, and social anxiety on FER, no work to date has examined these variables together, especially using nuanced FER tasks. Moreover, this work is imperative as these profiles of risk can aid in our understanding of the heterogeneity of FER and ultimately inform treatments targeting social cognition. Thus, we aimed to examine the role of these variables on FER accuracy using a novel dynamic FER task.

In the current study, participants completed a novel FER task involving dynamically morphing expressions that change from a neutral to an emotional face. Similar to Heuer and colleagues (2010) our FER task used two conditions in which participants either (1) pressed a button as soon as they recognized the emotion, the morphing image immediately stopped, and they were then prompted to indicate which emotion they recognized (hereby referred to as the truncated condition), and (2) pressed a button as soon as they recognized the emotion, the trial continued with the full intensity of the emoting morph image, and they were then prompted to indicate the emotion they recognized (hereby referred to as the full viewing condition). Using dynamic morphing stimuli under these two conditions, we were able to measure dynamic FER processes that may impact FER accuracy; More specifically the truncated condition was thought to index relatively rapid FER interpretations through brief emotion recognition processing of relatively lower intensity emotion expressions vs. the full viewing condition, which indexes dynamic interpretations of emotion recognition processes of full intensity emotion expressions. The first aim of this study was to examine the condition effects of this novel FER task. A second aim of this study is understanding the role that autistic traits, alexithymia, and social anxiety have on FER accuracy. Consistent with previous work on FER in samples of autism and social anxiety, we predicted that social anxiety and alexithymia, and not autistic traits, would predict FER difficulties (Cook et al., 2013; Corden et al., 2008; Heuer et al., 2010). We used covariates of age and IQ, which have previously been linked to FER (Lozier et al., 2014; Rump et al., 2009; Trevisan & Birmingham, 2016; Sivathasan et al., 2020). Moreover, in light of prior work illustrating that social anxiety symptoms are related to dynamic FER processes, especially in tasks which require rapid interpretations (Gutiérrez-García & Calvo, 2017; Heuer et al., 2010; Joormann & Gotlib, 2006), we hypothesized that greater social anxiety would specifically predict truncated FER accuracy, and not full viewing FER accuracy. Additionally, as there is work indicating a negative bias in social anxiety (Gutiérrez-García & Calvo, 2017; Heuer et al., 2010), we planned to explore misinterpretations for any impacted emotions.

Method

Participants & Procedures

The final study sample included 19 autistic male adolescents and adults (ages 16 to 30 years), who were interested in participating in a pilot study on the development of a neurotherapy targeting FER (M age=21.82, SD=3.83). Of note, as the parent study was a feasibility pilot study on the development of a neurotechnology to target FER in autism, recruitment was specific to autistic male participants between the ages of 16–30 years of age. The rationale for this decision was to reduce heterogeneity in neural processes as previous work has suggested sex differences in FER processes in autism (e.g., Coffman et al., 2015). Moreover, this age range was chosen in order to capture FER during a relatively more stable developmental period in FER processes (e.g., Rump et al., 2009). The final sample consisted of 16/19 (84.21%) individuals that self-identified as White, 1/19 (5.26%) who self-identified as Asian, 1/19 (5.26%) who self-identified as Native American, and 1/19 (5.26%) who self-identified as Mixed Race.

In order to be eligible for the parent study, which focused on the development of neurotechnology intervention to target FER, participants were required to: (1) be male, (2) between the ages of 16 to 30 years, (3) meet criteria for autism spectrum disorder (ASD) as confirmed by Autism Diagnostic Observation Schedule, Second Edition (ADOS-2; Lord et al., 2012), (4) have a Full Scale Intelligence Quotient (FSIQ-2) ≥ 80, as confirmed by Wechsler Abbreviated Scale of Intelligence, 2^nd edition (WASI-II; Wechsler, 2011), (5) be capable of undergoing Magnetic Resonance Imaging (MRI; no contraindications such as metal in body), (6) be ambulatory with no known, uncorrected sensory deficits, (7) be free from severe psychopathology that warrants immediate clinical care (i.e., psychosis, suicidality), as confirmed by the Anxiety Disorder Interview Schedule for DSM-5 (ADIS-5; Brown & Barlow, 2014), (8) have no known genetic, medical, or neurological condition including intellectual disability, (9) no changes in medication for 4 weeks, no planned changes in immediate future and no use of anti-seizure medication, and (10) not enrolled in a treatment that directly targets facial emotion recognition.

The study was approved by the Virginia Tech Institutional Review Board. Participants were recruited through flyers in the local community (e.g., churches, schools, restaurants), existing research registry databases, university-affiliated assessment clinics, and local autism support groups. In addition to flyers about the study, the recruitment strategy included community presentations about autism research opportunities, mass emails, and individual emails or phone calls to individuals who have previously expressed interest in research (i.e., via research registry). Once an individual identified that they were interested in learning more about the study, they were contacted by a research assistant who scheduled a time to tell the participant about the study using IRB-approved language and answer any questions they may have. If interested in participating, individuals completed a verbal assent for the initial phone screening. Once preliminary eligibility was determined, an eligibility visit was scheduled. At the eligibility visit, adult participants provided verbal and written consent, and adolescent participants provided verbal and written assent with parent verbal and written consent for study participation. The eligibility visit included cognitive and diagnostic measures to determine eligibility (i.e., WASI-II, ADOS-2, ADIS-5). We excluded participants with severe psychopathology of various forms, but under the general rubric of impairment. In other words, we allowed co-occurring psychopathology, but only excluded individuals who were impaired beyond the ability to credibly participate in the larger intervention study (e.g., psychosis or suicidality). In such cases, participants were excluded from the study and began a process of monitored referral to an appropriate outlet. Once eligibility was confirmed, participants completed a pre-treatment visit, where they completed a battery of questionnaires and behavioral tasks. The Emotion Recognition Task was conducted prior to treatment. Although 23 participants were recruited in the parent study, only 19 participants had emotion recognition task data, thus, the present study only included these participants.

Measures

Liebowitz Social Anxiety Scale (LSAS; Liebowitz, 1987)

The LSAS is a 24-item self-report measure that assesses social anxiety symptoms. Using a 4-point Likert scale, participants were asked to rate their 1) fear or anxiety (0 “none” to 3 “severe”) and 2) avoidance (0 “never” to 3 “usually”) of specific social situations (e.g., “Going to a party”). Higher LSAS total scores reflect greater fear and avoidance in social situations. An LSAS total score of ≤29 indicates no social anxiety, 30–49 indicates mild social anxiety, 50–64 indicates moderate social anxiety, 65–79 indicates marked social anxiety, 80–94 indicates severe social anxiety, and ≥95 very severe social anxiety. The LSAS is considered to be a reliable and valid measure to assess social anxiety, which has demonstrated excellent internal consistency in autistic adult samples (Bejerot et al., 2014; Boulton & Guastella, 2021; Danfroth et al., 2018; McVey et al., 2016). Internal consistency of the LSAS in this sample was excellent (α = .95).

Social Responsiveness Scale, Second Edition (SRS-2; Constantino & Gruber, 2012).

The SRS-2 is a 65-item self-report measure of autistic traits, including social awareness, social cognition, social communication, social motivation, and restricted interests/repetitive behaviors. Items are rated on a 4-point scale from 1 (“Not True”) to 4 (“Almost Always True”). Higher scores indicate higher levels of autistic traits. The SRS-2 is considered to be a reliable and valid measure to assess autistic traits, with excellent internal consistencies in adolescent and adult samples (Bemmer et al., 2021; White et al., 2022). Internal consistency of the SRS-2 for this sample was excellent (α = .91).

Toronto Alexithymia Scale (TAS-20; Bagby, Parker, & Taylor, 1994).

Alexithymia was measured with the TAS-20, a 20-item self-report scale evaluating difficulties identifying and describing emotions. Items are answered on a 5-point Likert scale ranging from “strongly disagree” to “strongly agree.” TAS-20 has 3 subscales: Difficulty Describing Feelings, Difficulty Identifying Feelings, and Externally-Oriented Thinking. The total TAS-20 score is comprised of sum responses to all 20 items. A total score ≥ 61 indicates alexithymia, with scores between 52 and 60 indicating possible alexithymia. Of note, there have been concerns about reliability of the TAS-20 subscale scores, especially the Externally-Oriented Thinking scale (Kinniard et al., 2019; Kooiman et al., 2002), though the TAS-20 total score is considered to be a reliable and valid measure of alexithymia. The TAS-20 total score has frequently been used in autistic adult samples with good internal consistencies (Cook et al., 2013; Samson et al., 2012). Internal consistency of the TAS-20 total score for this sample was good (α = .83).

Emotion Recognition Task (Richey et al., 2022).

The Emotion Recognition Task (Figure 1) is a behavioral task which was completed during EEG data collection, for purposes of the parent study. The images for the Emotion Recognition Task were selected from the Cohn-Kanade image database (http://www.pitt.edu/~emotion/ck-spread.htm). All images have been Facial Action Coded and Emotion Facial Action coded, meeting the standards for portraying prototypical emotional expressions (Kanade, Cohn, & Tian, 2000). Two graduate research assistants reviewed and consensus rated the images used in the task. Given that the study sample included only male participants, only male facial stimuli were selected. Ten stimuli were selected for each of the six basic emotions (i.e., anger, disgust, fear, happiness, sadness, surprise) for a total of 60 stimuli. The number of times that an individual actor was portrayed was limited to three presentations, in order to reduce participant familiarity with the actor.

Figure 1. Depiction of the Facial Emotion Recognition Task.

a) Condition 1: Truncated Viewing depicts run 1 in which a button press (i.e., indicating emotion was recognized) at any moment triggers prompt, followed by new trial, b) Condition 2: Full Viewing depicts run 2 in which a button press does not immediately trigger prompt and the participant watches the full morphing before indicating response.

The series of still images were compiled into dynamic videos, created using custom MATLAB code that interpolated the space between the still images to make the videos appear smoother. The final 60 videos were then split into 2 sets for 2 separate tasks, which included 30 emotion videos each, with each emotion shown 5 times per task. Therefore, no videos were repeated within the Emotion Recognition Task. On average, the videos were 20.57 seconds (SD = 7.26) in length across different emotions and did not vary in length across emotional conditions. The presentation of stimuli was randomized for each participant to account for order effects.

Participants were seated in a desk chair (no wheels), approximately 50 cm from a computer monitor which displayed the task. The computer monitor was located on a desk in a quiet lab room with the room lights on. Participants were oriented to the task (i.e., identifying dominant hand for button press) and were verbally instructed: “There will be two rounds. The first round the video will stop playing when you indicate that you recognize the emotion and in the second round the video will keep playing after you indicate you recognize the emotion. For both of these rounds you will press ‘1’ as soon as you recognize the emotion. You will press ‘1’ again to scroll through the options and ‘2’ to select an emotion. Any questions?” Participants completed a practice run which was four trials to ensure they understood the task instructions. Upon starting the task, a screen was presented with written instructions which included “Please press the button as soon as you recognize the emotion.” This instruction was also presented prior to the second run of the task. The research assistant stepped out of the room while the participant completed the task and monitored for behavior and completion via a two way mirror. The emotion recognition task took approximately 20 minutes total.

As noted in the instructions, the task consisted of two conditions. During both conditions participants viewed the video and responded with a button press as soon as they recognized the emotion. After the completion of the video, participants were then asked to select the portrayed emotion from a list of six presented emotions (i.e., anger, disgust, fear, happiness, sadness, surprise), with the options always presented in the same order. As soon as the participant selected the emotion, the next trial started. In the first condition, also known as the truncated FER condition, participants viewed an actor that morphs from a neutral to emoting face and pressed the button at the earliest moment in which they recognized the emotion. For the truncated FER condition, after recognition was indicated by the participant, the morph immediately disappeared, and participants were then prompted to choose which emotion was presented. In the second condition, also known as the full viewing FER condition, trials proceed identically, with the exception that participants continued to watch the entire morph before seeing the screen in which they chose which emotion was presented. This task allowed for a precise measurement of dynamic FER processes and allowed us to examine accuracy during rapid FER interpretations vs. FER accuracy when more emoting intensity was available.

Analytic Plan

FER accuracy was measured by percentage of correct responses. All data were conducted in SPSS 27.0. A 2-by-6 (condition x emotion) within-subject repeated measures ANOVA with pairwise Bonferroni-adjusted comparisons was conducted in order to understand the main effects and interaction of the task. Hierarchical linear regressions were conducted to examine the potential impact of autistic traits, alexithymia, and social anxiety on FER accuracy, while controlling for age and IQ as covariates. The first step included covariates (age, IQ) and the following steps included predictors of autistic traits (step 2), alexithymia (step 3), and social anxiety (step 4). Significant results were probed further at the condition and emotion accuracy level with covariates included in the first step. For significant emotions impacted by an independent variable, misinterpretations were probed by examining the percentage of incorrect responses that were interpreted as another emotion.

Results

Participant Characteristics

Participant characteristics of the sample are included in Table 1. Of note, 68.42% (13/19) of the sample exceeded the cut-off (>30) for social anxiety on the LSAS. Additionally, 36.84% (7/19) of the sample surpassed the cut-off (>52) for possible alexithymia on the TAS-20.

Table 1.

Participant demographics and characteristics

Measure	M (SD)	Range
Age (years)	21.82 (3.83)	16.25 – 29.00
Age of Autism Spectrum Disorder Diagnosis (years)	9.62 (5.32)	2.5 –12
FSIQ-2	108.28 (16.47)	79 – 140
ADOS-2 Comparison Score	5.00 (1.79)	3–9
Social Responsiveness Scale-2 (SRS-2)	74.53 (24.23)	30–122
Toronto Alexithymia Scale-20 (TAS-20)	50.33 (10.78)	31 – 70
Liebowitz Social Anxiety Scale (LSAS)	42.89 (22.09)	8 – 82
	n	Percentage
Race
Asian	1/19	5.26%
Mixed Race	1/19	5.26%
Native American	1/19	5.26%
White	16/19	84.21%
Employment Status
Employed	8/19	42.11%
Not Employed	11/19	57.89%
Education Level
Some High School	3/19	15.79%
High School Diploma	4/19	21.05%
Some College	8/19	42.11%
College Diploma	4/19	21.05%
History of therapy or counseling in lifetime	16/19	84.21%
Currently in therapy or counseling	7/19	36.84%
History of psychiatric medication in lifetime	13/19	68.42%
Currently prescribed psychiatric medication	7/19	36.84%

Within-Subjects Repeated Measures ANOVA of Emotion Recognition Accuracy

In order to examine the within-subjects effects of the task a repeated measures ANOVA was conducted. The repeated measures ANOVA for the emotion recognition task revealed a main effect of Condition (F(1, 18) = 10.63, p = .004, η_p² = .37), with better accuracy in the full viewing condition (83.2%) than the truncated emotion recognition condition (70.1%). Additionally, a main effect of Emotion (F(5, 90) = 15.80, p < .001, η_p² = .47) was revealed. Pairwise Bonferroni-adjusted comparisons revealed specific significant differences in emotion types, which are indicated in Figure 2. Of note, participants were more accurate for happiness trials than all other emotions (ps<.05). Participants were more accurate for sadness trials than anger, disgust, fear (p < .05); sadness trials did not differ from accuracy for surprise trials. Accuracy for surprise was significantly higher than accuracy for disgust (p < .05). A Condition-by-Emotion effect was also revealed (F(5, 90) = 2.32, p = .05, η_p² = .12). Follow-up paired sample t-tests revealed diminished accuracy for truncated vs. full viewing conditions for fear, sadness, and surprise (p < .01; Figure 2).

Figure 2. FER Accuracy by Emotion and Condition.

Bar graph displaying facial emotion recognition accuracies by emotion and condition.

Predicting Overall FER Accuracy

Mean FER accuracy was collapsed across conditions and emotions and used as the dependent variable, overall FER accuracy. Hierarchical linear regressions were conducted to predict overall FER accuracy, including the effect of the covariates (Model 1), and additive effects of autistic traits (Model 2), alexithymia (Model 3), and social anxiety (Model 4; Table 2). Models 1–3 were not significant (ps > .33, R² values < .24). A significant ΔR² was only apparent when adding social anxiety (ΔR² = 0.34, p = .007). Model 4 significantly predicted overall FER accuracy and accounted for 57.4% of the variance in overall FER accuracy, R² = 0.57, F(5, 18) = 3.50, p = .03. Within Model 4, greater IQ predicted better overall FER accuracy, β = 0.50, t = 2.57, p < .02, and greater social anxiety predicted poorer overall FER accuracy, β = −0.67, t = −3.24, p < .01. Age, autistic traits, and alexithymia did not significantly contribute to Model 4 (ps > .45). Exploratory examination of the role of IQ was probed by using a mean split (IQ Mean = 107) with a scatter plot (Figure 3). Examination of the plot showed that higher social anxiety was related to poorer FER performance in the low IQ group (r = −.78, p < .01); the effect of social anxiety on FER in the high IQ group was apparent but attenuated (r = −.43, p = .25).

Table 2.

Hierarchical linear regression results for overall FER accuracy including covariates of age and IQ, and predictors of autistic traits, alexithymia, and social anxiety

	Model 1		Model 2		Model 3		Model 4
	β	R 2	β	R 2	β	R 2	β	R 2
Overall FER Accuracy
Age	−0.05	0.10	−0.03	0.20	−0.05	0.23	0.12	0.57*
IQ	0.31		0.29		0.32		0.50*
Autistic Traits			−0.32		−0.21		−0.004
Alexithymia					−0.23		−0.18
Social Anxiety							−0.67**

Note:

^*= p < .05

^**= p < .01

^***= p < .001

Figure 3. Relationship Between FER Accuracy and Social Anxiety.

Scatterplot demonstrating the relationship between facial emotion recognition accuracy and social anxiety in those with IQ scores between 79–106 vs IQ scores between 107–140 (mean split = 107).

Predicting Truncated and Full Viewing FER Accuracy

As social anxiety was the only predictor that impacted overall FER accuracy, we probed this effect further to test the hypothesis that social anxiety may be associated with rapid interpretations during the truncated FER condition and not by global FER difficulties as measured by the full viewing FER condition. Hierarchical linear regressions were conducted, which included covariates (age and IQ) in Model 1, and additive effect of social anxiety in Model 2. Model 2 significantly predicted truncated FER accuracy and accounted for 55.50% of the variance in truncated FER accuracy, R² = 0.56, F(3, 18) = 6.25, p = .006. Within Model 4, greater IQ predicted better overall FER accuracy, β = 0.43, t = 2.38, p = .03, and greater social anxiety predicted poorer overall FER accuracy, β = −0.74, t = −4.08, p < .001. Age did not significantly contribute to Model 4. As hypothesized, Model 2 was not significant in predicting full viewing FER accuracy (R² = 0.11, p > .62).

To assess whether the association between social anxiety and FER was restricted to specific emotion types within the truncated FER condition, hierarchical linear regressions were conducted to separately examine the role of social anxiety, above and beyond age and IQ, on accuracy for each emotion (i.e., anger, fear, disgust, happiness, sadness, surprise). Emotion accuracies for the full viewing condition were also tested, though as predicted no significant results emerged (ps > .27, R² values < .23; Table 3). Within the truncated FER condition, social anxiety significantly predicted FER accuracies for disgust and surprise, and accounted for 46.8% (R² = 0.47, F(3, 18) = 4.39, p = .02) and 62.8% (R² = 0.63, F(3, 18) = 8.44, p < .01) of the variances in each emotion accuracy, respectively. Higher IQ significantly predicted better surprise accuracy (β = 0.39, p < .05) and greater social anxiety significantly predicted poorer disgust and surprise accuracies (βs = −0.69, ps < .01). Age did not significantly add to the models. No models were significant in predicting anger, fear, happiness, and sadness FER accuracies (ps > .15).

Table 3.

Hierarchical linear regression results for accuracy by condition and emotion

	Truncated FER Condition				Full Viewing FER Condition
	Model 1		Model 2		Model 1		Model 2
	β	R 2	β	R 2	β	R 2	β	R 2
FER Accuracy
Age	−0.10	0.06	−0.01	0.47**	0.07	.07	0.10	0.11
IQ	0.24		0.43*		0.25		0.30
Social Anxiety			−0.74***				−0.21
Anger
Age	−0.13	0.02	−0.08	0.13	−0.33	0.14	−0.33	0.14
IQ	−0.04		0.05		0.22		0.22
Social Anxiety			−0.35				0.004
Fear
Age	0.10	0.08	0.15	0.20	0.35	0.15	0.36	0.16
IQ	0.25		0.34		0.15		0.16
Social Anxiety			−0.37				−0.07
Disgust
Age	−0.05	0.03	0.06	0.47*	−0.07	0.01	−0.04	0.04
IQ	0.16		0.34		−0.05		−0.002
Social Anxiety			−0.70**				−0.20
Happiness
Age	−0.14	0.04	−0.07	0.18	--	--	--	--
IQ	0.15		0.25		--		--
Social Anxiety			−0.40				--
Sadness
Age	0.06	0.10	0.13	0.29	0.14	0.23	0.14	0.23
IQ	0.30		0.41^		0.44^		0.43^
Social Anxiety			−0.47^				0.04
Surprise
Age	−0.28	0.11	−0.17	0.63**	0.26	0.07	0.31	0.19
IQ	0.20		0.39*		0.03		0.13
Social Anxiety			−0.76***				−0.38

Note:

^{^}= p < .10

^*= p < .05

^**= p < .01

^***= p < .001.

Accuracy for Happiness in the Full Viewing FER condition was 100% for all participants, thus, due to the lack of variability in the data, regressions were not computed for this variable.

Exploring Misinterpretations for Impacted Emotions within the Truncated FER Condition

At the group level, of the incorrect responses for disgust during the truncated FER condition, 85.29% (29/34) of the responses were marked as anger (14.71% (5/34) happiness). Of the incorrect responses for surprise during the truncated FER condition, 50% (11/22) of the responses were marked as fear (27.27% (6/22) anger, 13.64% (3/22) happiness, 4.55% (1/22) sadness, 4.55% (1/22) disgust).

Discussion

The present study aimed to examine the role of social anxiety, alexithymia, and autistic traits on FER using a novel FER task with neutral to emoting morphing expressions in a sample of autistic male adolescents and adults. This novel FER task revealed a main effect of condition, such that accuracy for the truncated condition was worse than accuracy in the full viewing condition, a main effect of emotion, with happiness and sadness being the easiest emotions to recognize compared to others, and a condition-by-emotion interaction effect, with worse accuracy in the truncated vs. full viewing condition for fear, sadness, and surprise. Previous work has highlighted strong relationships between poorer FER with alexithymia and autistic traits (Cook et al., 2013; Trevisan & Birmingham, 2016; Sivathasan et al., 2020), though few studies have examined links with social anxiety. We found that higher social anxiety and lower IQ predicted poorer overall FER (above and beyond age, IQ, autistic traits, and alexithymia) in a sample of autistic male adolescents and adults without intellectual disability. The effect of social anxiety on FER (above covariates of age and IQ) was prominent when participants made rapid interpretations of emotional stimuli. More specifically, when examining specific emotions, social anxiety impacted FER for truncated surprise and disgust.

Results from this study are consistent with similar effects from other FER tasks more broadly (Kessels et al., 2014; Montagne et al., 2007). Consistent with work examining intensity levels of emoting faces and emotion recognition, we found poorer FER within our truncated vs. full viewing condition (Gutiérrez-García & Calvo, 2017; Heuer et al., 2010; Joormann & Gotlib, 2006; Kätsyri et al., 2008; Law Smith et al., 2010). Moreover, the emotion effects from the task coincide with literature supporting the highest accuracy for FER of happiness and sadness trials (Lawrence et al., 2015; Montagne et al., 2007). Moreover, we found a condition-by-emotion interaction, driven by accuracies of fear, sadness, and surprise; these findings are consistent with work concluding better accuracy at high vs. low and medium emotion intensity levels, with the strongest effects for surprise and sadness (Wingenbach et al., 2018).

Although there is a mixed literature on altered FER mechanisms in autism, several studies have indicated that negative (i.e., disgust) and “cognitive” emotions (i.e., surprise) are altered in autism (Ashwin et al., 2006; Law Smith et al., 2010;. Wallace et al., 2011; Wallace et al., 2008). Consistent with calls for understanding relevant variables which may explain the FER differences seen in autistic people (Harms et al., 2010; Lozier et al., 2014; Nuske et al., 2013), we examined the roles of social anxiety, alexithymia, and autistic traits on FER. Although there is some work to suggest that alexithymia, and not autistic traits, predict FER, our results do not support a link between increased alexithymia and poorer FER. In contrast to studies that have found an effect of alexithymia on FER (Cook et al., 2013; Ketelaars et al., 2016; Ola & Gullon-Scott, 2020), our study differed in a few key ways. First, we only included autistic male adolescents and adults, previous work in this area has included female autistic individuals or all female samples. Secondly, although our mean IQ was similar to previous studies, our sample was notably younger (in the adolescent to emerging adult range), while previous studies had mean age ranges between 38.5 to 41.5 years. Mechanisms that may impact FER may change across development and more work is needed in understanding developmental trajectories of FER and social outcomes in autism. Finally, we used a novel morphing face task, which contrasts from previous literature using a task including morphing of emotion expression and identity changes (Cook et al., 2013), an emotion recognition task using static stimuli at various intensity levels (Ketelaars et al., 2016), and video clips of 14 complex and basic emotions (Ola & Gullon-Scott, 2020). Our task only measured six basic emotions and allowed us to examine the effects of a more natural procession of emotion recognition (truncated condition) vs. emotion recognition of a natural procession of full intensity emotions.

Although we did not find support for alexithymia predicting global FER difficulties, we found an effect of social anxiety on poorer FER, which was driven by diminished accuracy in the truncated condition. The association between decreased accuracy in truncated FER and increased social anxiety are consistent with previous work suggesting that individuals with social anxiety have difficulty disengaging from and interpreting ambiguous faces (Lira Yoon & Zinbarg, 2007; McKendrick et al., 2018). Moreover, this work is consistent with findings of trait worry (and not alexithymia) impacting dynamic FER performance in autistic adults (Stephenson et al,. 2019). The potential ambiguity of lower intensity emotions in the truncated condition, may have magnified FER difficulties captured by the task (e.g., truncated RT M (SD) = 11.94 (3.67) seconds vs. stimuli length in full viewing condition M (SD) = 20.57 (7.26) seconds), suggesting that on average participants only saw about half the intensity of the emotion expression in the truncated condition as compared to the full viewing condition. Moreover, these findings converge with results from morphing FER tasks in non-autistic socially anxious samples, which have found alterations in FER and interpretation of negatively biased emotions at lower intensity levels (i.e., disgust, anger; Gutiérrez-García & Calvo, 2017; Heuer et al., 2010). Altogether, social anxiety symptomatology may relate meaningfully to the noted FER differences in autism; in that autistic individuals with higher social anxiety symptoms may initially misinterpret emotions but when given more exposure may be able to accurately identify emotions.

Truncated surprise and disgust were specifically impacted by social anxiety symptoms. These findings are consistent with work by Law Smith and colleagues (2010) using a morphing face task which found that autistic adolescents demonstrated diminished FER accuracy for disgust at all intensity levels, and diminished FER accuracy for surprise at lower levels of intensity compared to typical adolescents (Law Smith et al., 2010). Both disgust and surprise have been demonstrated to be impacted by cognitive biases (Baron-Cohen et al., 1997; Clark & Wells, 1995; Rapee & Heimberg, 1997). Symptoms of social anxiety include marked fear of negative evaluation or rejection (American Psychiatric Association, 2013), and previous work has supported biased processing of facial expressions, especially negative, threatening, or ambiguous facial expressions (Mogg & Bradley, 2002; Mogg et al., 2004, Silvia et al., 2006). Disgust may be particularly salient and relevant for social anxiety as it conveys aversion and rejection and may be elicited by several types of stimuli (i.e., food, body boundary violations, poor hygiene; Rawdon et al., 2018; Rozin et al., 1994). Additionally, disgust is perceived as more negatively valenced than anger in social anxiety (Amir et al., 2010; Moser et al., 2008) and is more likely to be misinterpreted as contempt in social anxiety (Heuer et al., 2010). Altogether, important cognitive biases of social anxiety are relevant to altered FER of disgust in autism. Additionally, surprise has long been posited to be a vulnerable emotion in autism as it has been hypothesized to be a “cognitive” emotion, which requires interpretation of a belief vs. a “simple” emotion (i.e., happiness), which is due to a situation (Baron-Cohen et al., 1993; Jones et al., 2011). Further, although surprise is considered a “basic” emotion, previous work has demonstrated that it has been attributed to positive, negative, and neutral affective valences (Ortony & Turner, 1990), and the truncated condition may further exacerbate the ambiguity or potential negative valence of surprise, which may be further amplified by social anxiety symptoms in autism. Previous work has tied the social motivation hypothesis to alterations in surprise, such that cognitive biases away from social information may make surprise a particularly difficult emotion to identify for those with autism (Chevallier et al., 2012; Jones et al., 2011). Thus, cognitive biases which contribute to difficulties in recognizing surprise may be amplified in autistic adolescents and adults with high social anxiety symptoms. Although we did not find that autistic traits predicted FER, IQ impacted many aspects of FER, with the exception of disgust. It will be important to further parse the roles of IQ and social anxiety vs. autistic traits on FER, especially in the context of surprise.

Untangling the observed social-emotional differences in autism and whether they are uniquely related to autistic traits vs. other clinical correlates is important for future work, as such findings will provide insight into the specificity of mechanisms at play and development of new treatments (Antezana et al., 2015; Antezana & Yerys, 2021; Gotham et al., 2018; Stephenson et al., 2019). Additionally, the use of eye tracking will allow for insight to perceptual processes that subserve the roles of social anxiety on FER and should be considered for future studies. As a potential future direction for this type of work, we further suggest that a systematic review of autism and non-autism social anxiety research in FER may be timely in order to aggregate the available findings on this topic and perhaps point to a longer-term research agenda for the field. Specifically, it would be particularly profitable to systematically review the available data at the intersection of FER differences in autism and social anxiety for the purposes of platforming and elaborating the major outstanding questions raised by this work and related research and also to consolidate current theory in this emergent area. As our results indicated that social anxiety only impacted FER during the truncated condition, in which the participant stopped receiving information once they decided on the emotion, it may be important to examine the real-world social implications that are tied to brief emotion processing (i.e., avoidance of faces, diminished time looking at faces) and social difficulties. Moreover, as co-occurring social anxiety symptoms may impact social difficulties above and beyond autistic traits, future research should aim to disentangle what other areas of social processing (i.e., theory of mind, empathy) that may be impacted by social anxiety symptomatology in autism. Finally, the identification of social anxiety as a predictor of poorer FER in autism may aid in development of targeted interventions. For example, autistic individuals with high social anxiety may benefit from FER targeted interventions, above and beyond cognitive behavioral therapy, as social cognitive processes relay important information one receives and interprets from one’s social environment. Changes in FER may be an important mechanism that may have cascading effects on social anxiety and vice versa, and thus should also be considered in psychosocial interventions for autistic people. For example, it will be important for future work to examine whether cognitive behavioral therapy (i.e., developing fear hierarchies, and challenging cognitions through social exposures) for SAD has secondary effects on FER in autistic individuals and whether targeted FER interventions in autism have secondary effects on social anxiety symptoms (Lee et al., 2018; Wieckowski & White, 2020).

This work provides unique insights into FER processes in autistic people, such that high social anxiety in autistic adolescents and adults may specifically contribute to findings of difficulties in recognizing and interpreting brief low intensity emotions, especially potentially ambiguous and negative emotions (i.e., disgust, surprise). Further, this work provides an initial step in identifying FER processes impacted by social anxiety, which may be used as a potential target for social cognitive and cognitive behavioral interventions. It is important to note that this study was limited in several ways. First, it was a small sample of autistic male adolescents and adults without intellectual disability. Our sample was limited in heterogeneity in many ways (e.g., primarily White sample, all male participants, no intellectual disability, mean age of diagnosis in childhood) and it is important for future research to examine whether these findings are generalizable to the broader autism spectrum, diverse samples, and to autistic female adolescents and adults. Additionally, we did not have a neurotypical or SAD sample to compare to, which would have been useful in identifying autism-specific FER differences and associations. Moreover, our sample recruited adolescents and young adults, and social anxiety emerges in adolescence and young adulthood (Grant et al., 2005; Solmi et al., 2021; Stein & Stein, 2008), thus, we may have captured individuals during a critical period in the emergence of social anxiety symptomatology. It is imperative for future work to examine the relationship between FER and social anxiety developmentally in autistic people, as FER difficulties may both contribute causally to co-occurring social anxiety and may be magnified by the onset of SAD and add to vulnerabilities in FER noted in autistic people. It is likely that there are bidirectional effects between social anxiety and social difficulties that are compounded over time and amplified. Therefore, future research is needed to disentangle the developmental patterns of FER and social anxiety in autistic individuals.

Conclusion

Using a novel computerized FER task, we examined FER in a sample of autistic male adolescents and adults without intellectual disability. We found that greater social anxiety and lower IQ was associated with poorer FER accuracy, while age, autistic traits, and alexithymia did not significantly impact overall FER. When examining specific effects of social anxiety on emotion types, social anxiety impacted FER accuracy for surprise and disgust, although this was only observed in a truncated viewing condition. These results indicate that social anxiety may play an important role on FER. Future work should consider the role of social anxiety within autism, as a factor that may meaningfully relate to FER assessment and intervention.