Social Functioning and the Presentation of Anxiety in Children on the Autism Spectrum: A Multi-Method, Multi-Informant Analysis

Abstract

Co-occurring anxiety in children on the autism spectrum is associated with greater social challenges, including poorer social skills and relationships, which may influence the severity and presentation of anxiety symptoms, particularly social anxiety. The current study used Bayesian network analytics (Williams & Mulder, 2020) and a multi-method approach to examine (1) how different facets of social functioning relate to one another and to anxiety severity and comorbidity; (2) which facet(s) are most influential and thus may represent optimal targets for intervention; and (3) how social functioning relates to the presentation of social fears in a large treatment-seeking sample of autistic children with anxiety disorders (n = 191, 7-13 years). Results indicated strong associations among measures of social ability (i.e., theory of mind [ToM], social motivation, friendship attainment) and among measures of social integration (i.e., bullying, interpersonal and peer difficulties), with only bullying demonstrating a significant association with anxiety. ToM was the most interconnected variable in the network, and social motivation demonstrated the strongest individual connections with other variables, particularly with other facets of social ability. Socially anxious children with impaired ToM were less likely to express fears of negative evaluation, resulting in a distinct diagnostic presentation of social fears. Findings suggest that social motivation and ToM may represent important targets for intervention for autistic children with co-occurring anxiety. Further, social-cognitive difficulties associated with autism, like ToM, may play a role in distinct manifestations of anxiety in these youth.

General Scientific Summary

Autistic children with co-occurring anxiety disorders tend to experience significant difficulties with social skills (e.g., perspective-taking, social communication) and interpersonal relationships (e.g., friendships, bullying), and those social difficulties may in turn influence how anxiety, particularly social anxiety, is expressed in these youth. This study demonstrates that autistic children with difficulties in one area of social well-being are likely to have difficulties across many other, related areas of social well-being; in particular, children with more limited perspective-taking ability and/or less interest in social interaction are likely to have fewer reciprocal friendships and greater difficulty with social communication. Findings also suggest that differences in perspective-taking ability might explain a unique presentation of social anxiety in autistic individuals, whereby socially anxious children who have more difficulty understanding the thoughts, feelings and intentions of others might become anxious due to confusion and uncertainty in social situations rather than fear of being judged (the key criteria for a diagnosis of Social Anxiety Disorder).

Autistic children tend to experience widespread challenges in social functioning, including difficulties with social communication and understanding (e.g., social reciprocity, perspective-taking) (Kimhi, 2014; Chevallier, Kohls, Troiani, Brodkin, & Schultz, 2012), that often lead to friendship difficulties and peer rejection (Rowley et al., 2012; Schroeder, Cappadocia, & Weiss, 2014). These difficulties are often compounded by co-occurring anxiety disorders, which occur at above-population rates in children on the autism spectrum (Kerns, Rast & Shattuck, 2021; Simonoff et al., 2008).

Anxiety symptoms in autistic youth are associated with less support and intimacy in peer relationships, greater peer victimization and rejection, and reduced social reciprocity and social motivation (Chou et al., 2020; Duvekot, der Ende, Verhulst, & Greaves-Lord, 2018; Eussen et al., 2013; McVey et al., 2018), though findings are mixed. Chang et al. (2012) found that social anxiety disorder, but not other forms of anxiety, was associated with poorer parent-reported social skills in autistic children. Similarly, Kerns et al. (2015) did not find an association between anxiety disorders and parent-reported ability to initiate social interactions and develop relationships in autistic youth. Taken together, these findings suggest a potentially complex association between anxiety and social functioning in these youth.

Theory of Mind (ToM), or the ability to infer the thoughts, emotions, and intentions of others (Baron-Cohen, Leslie, & Frith, 1985; Happé & Frith, 1994), may play an influential role in the link between social functioning and anxiety. A wealth of evidence suggests that autistic individuals tend to have greater difficulty understanding the emotional expressions of others, inferring their beliefs and perspectives, and detecting and understanding “higher-order” social cues (e.g., deception) (Kimhi, 2014; Uljarevic & Hamilton, 2013). Poorer ToM skills have been linked to greater difficulties with social skills and reduced peer acceptance (Liu et al., 2018; Peterson, Slaughter, Moore, & Wellman, 2016) and may mediate the links between these difficulties and anxiety (Lei & Ventola, 2018; van Roekel, Scholte, & Didden, 2010) in autistic youth. Thus, poorer ToM and the associated social differences may lead to greater peer rejection, or to a sense of unpredictability and confusion in social interactions, which may then contribute to the development of social fears. Alternatively, intense anxiety may impede accurate perspective-taking and therein diminish social ability and social relationships. Either way, ToM may be central to the social ability and relational functioning of autistic youth with co-occurring anxiety, and thus may represent an optimal treatment target for social intervention programs. Two recent studies found that ToM training interventions led to significant improvements in parent-reported bullying experiences as well as teacher-reported social skills and peer acceptance (Liu et al., 2018; Szumski, Smogorzewska, Grygiel, & Orlando, 2019).

ToM may also be related to the manner in which anxiety, particularly social anxiety, is expressed. Autistic children can present with anxiety symptoms that are qualitatively distinct from DSM-defined categories and related to autism symptoms (Kerns et al., 2020; White, Bray & Ollendick, 2012). For example, some autistic individuals present with social fears not due to fear of negative evaluation (a key criterion of Social Anxiety Disorder; American Psychological Association, 2013), but instead possibly reflecting fears related to confusion or uncertainty about social behavior (Halim et al., 2018; Kerns et al., 2020). Clarifying if different presentations of anxiety, particularly traditional and distinct forms of social anxiety, are associated with unique social functioning profiles, particularly varied ToM ability, may help to better understand and treat these disorders and associated psychosocial sequelae.

The extant literature on this topic has several methodological limitations. First, studies have primarily assessed social functioning via single, broad measures that represent multiple facets of social ability and relationships in a single score (e.g., Chang et al., 2012; Duvekot et al., 2018; McVey et al., 2018), limiting the ability to investigate how specific social skills and experiences may differentially relate to anxiety symptoms and to one another. Including multiple, specific measures would allow for more nuanced investigation into the facets of social functioning that are most strongly related to one another and to anxiety. Second, most studies use a single-informant approach, primarily self- or parent-report, which are often discrepant and may not fully represent a child’s social ability across contexts (De Los Reyes & Kazdin, 2005). A multimethod approach could clarify associations among different facets of social functioning and anxiety in different contexts. Third, though studies suggest that social skills programs may indirectly lead to improvements in anxiety symptoms (Ke, Whalon, & Yun, 2018; Laugeson, Ellingsen, Sanderson, Tucci, & Bates, 2014), they have not identified the specific aspects of social functioning most strongly associated with anxiety severity, and it is likely that targeting aspects of social functioning that are most strongly related to anxiety will result in the most effective interventions. Finally, prior studies have failed to consider how the social profile of autistic children may relate to the form or focus of the anxiety they express.

The current study employed a multi-method, multi-informant battery of standard and specific measures of social ability and experiences, as well as careful characterization of anxiety, in a treatment seeking sample of cognitively-able autistic children with co-occurring anxiety. Aims were to: (1) examine how different facets of social functioning relate to anxiety severity and comorbidity and identify which aspects of social functioning are most strongly associated with anxiety; (2) examine which facets of social functioning are most strongly related to other forms of social functioning when controlling for anxiety severity and comorbidity; and (3) to explore whether traditional and distinct expressions of social anxiety are associated with unique patterns of social difficulty.

Network analysis, a statistical technique that identifies unique associations among specific elements of a broad construct (e.g., social functioning), was used to address Aims 1 and 2. Until recently, popular methods to estimate psychological networks have had a major limitation; namely, that the partial correlation networks from which these networks are generated capture only the non-zero effects among constructs (i.e., direct, independent associations among variables when controlling for the variance of all other variables in the model) without the ability to test hypotheses regarding the strength and centrality of those associations relative to other associations in the model. To address these limitations, this study conducted network analyses using a novel Bayesian statistical framework that allows researchers to test the hypothesis that a given variable is more strongly associated with other variables in the model (i.e., more central), and to conduct confirmatory testing of a priori expectations that are not typically possible using current methods (Kuismin & Sillanpää, 2017). Using this approach (Williams & Mulder, 2020; Williams, 2020), we tested the hypothesis that ToM would emerge as a central component of the network, connecting most other facets of social functioning. We also hypothesized that poor ToM and limited social motivation would be uniquely associated with a distinct presentation of social anxiety without fears of negative evaluation (referred to herein as other social fears), while poor friendship quality and severe and frequent and/or severe experiences of bullying would be uniquely associated with traditional social anxiety disorder. These latter hypotheses were tested using regression models.

Methods

Participants

Participants were part of the Treatment for Anxiety in Autism Spectrum Disorders trial (Wood et al., 2020), a 3-site (University of Los Angeles: California, University of South Florida, Temple University) randomized clinical trial comparing modular (Behavioral Interventions for Anxiety in Children with Autism; Wood et al., 2009) to standard CBT (Coping Cat; Kendall & Hedtke, 2006) for anxiety in autistic children. Families who met the following inclusion criteria were invited to complete diagnostic intake interviews: (a) child aged 7–13 years; (b) community diagnosis of autism; (c) verbal proficiency in English; (d) IQ ≥ 70 (if known by caregiver); (e) if on medication, able to maintain a stable dose during the trial; and (f) not receiving, or willing to stop, other psychosocial services during the trial. Of the 214 families who participated in intake interviews, families were excluded from the current study if: (a) child did not meet clinical cutoff scores on both the Autism Diagnostic Observation Schedule-2^nd Edition (ADOS-2) and Childhood Autism Rating Scale-2^nd Edition (CARS-2HF) (Lord et al., 2012; Schopler et al., 2010) (n = 9); (b) child did not meet a primary diagnosis of anxiety on the Anxiety Diagnostic Interview Schedule-Autism Spectrum Addendum (ADIS/ASA) (n = 13) (Kerns et al., 2017); or (c) parent did not complete the ADIS/ASA (n = 1). Participants also completed the Wechsler Intelligence Scales for Children-4^th edition (WISC-IV). Those who received an FSIQ below 70 (n = 5) were included in the current study to capture a more representative range of cognitive functioning. The final sample included 191 autistic children with anxiety disorders and their caregivers.

Procedure

This study was approved by the Institutional Review Board and conducted at research clinics at each site. Participants completed a phone screen and then those who met eligibility criteria completed diagnostic intake interviews. Caregivers provided written informed consent, and children provided written or verbal informed assent. Intake interviews included assessments of anxiety (ADIS/ASA, Pediatric Anxiety Rating Scale [PARS]; RUPP Anxiety Study Group, 2002), autism (ADOS-2, CARS-2HF), cognitive functioning (WISC-IV), and other socioemotional difficulties (Child Behaviour Checklist [CBCL]; Achenbach, 1999), as well as basic demographic questionnaires. Assessments were administered by doctoral-level clinicians and doctoral-level psychology students trained to research-reliable standards (see Wood et al., 2020 for details). Inter-rater reliability was excellent (100% diagnostic agreement by an independent rater, unaware of the original diagnostic decision on the ADOS-2 and CARS-2HF; PARS r = .88; ADIS/ASA ICC = 0.82-0.98) (Wood et al., 2020; Kerns et al., 2017). ADIS/ASA interviews were audio-recorded with the consent of participating families. Families received $25 for participating.

Measures

Anxiety and Other Psychological Disorders

Anxiety disorders as well as major depressive disorder, oppositional defiant disorder (ODD), and attention-deficit hyperactivity disorder (ADHD) were assessed via the ADIS-IV-P (Silverman & Albano, 1996), a semi-structured interview conducted with parents to assess DSMIV disorders in children (APA, 1994) with demonstrated inter-rater reliability (.91) and convergent validity (.51) in autistic youth (Ung et al., 2014). The current study also used the Autism Spectrum Addendum (ASA), a set of additional items intended to help differentiate autism vs. anxiety symptoms (see “ASA Social Functioning Items” below) and to capture expressions of anxiety experienced by autistic individuals that do not fit traditional diagnostic categories. In particular, the ASA includes modules to assess fears of change, social fears unrelated to fears of negative evaluation (i.e., “other social fears”), unusual phobias, and fears related to special interests. The interview has shown favorable psychometric properties including inter-rater reliability, convergent and discriminant validity (Kerns et al., 2017). The Social Anxiety Disorder (SAD)/Other Social Fears (OSF) module includes probes designed to distinguish social anxiety concerns that do (SAD) or do not (OSF) relate to fears about social judgment. For this and all other modules assessing anxiety and other co-occurring psychopathology, clinicians assigned a Clinical Severity Rating (CSR) to capture the degree of interference across domains (school, home/family, social) from 0 (no interference) to 8 (severe interference). A CSR of 4 indicates clinically significant interference and, when all criteria are met, diagnosis. A CSR of 3 indicates subclinical symptoms. The current study used the ADIS/ASA anxiety CSR ratings as a measure of the presence and severity of both traditional and distinct anxiety presentations. The 7-Item PARS Total score (RUPP Anxiety Study Group, 2002) was used to characterize the overall severity and interference of anxiety symptoms in the current sample. The PARS is a clinician-rated scale of anxiety symptoms and severity in children ages 7–17 years with demonstrated construct validity and inter-rater and test-retest reliability in autistic youth (Storch et al., 2012).

Social Functioning

Different facets of social functioning were assessed via (1) specific items related to social abilities (e.g., ToM, social motivation) and experiences (e.g., friendships, bullying) on the ADIS/ASA; (2) the Social Affect composite of the ADOS-2; and (3) parent questionnaires assessing broad social challenges (CBCL Social Problem Subscale) and autism-related social challenges (SRS-2).

ASA Social Functioning Items.

The ASA contains items designed to assess aspects of children’s social ability and experience pertinent to the assessment of anxiety. Parents are asked a set of questions about their child’s Theory of Mind (i.e., awareness of other’s thoughts and emotions), Social Motivation (i.e., interest in social interactions and relationships), Friendships (i.e., ability to establish and maintain a reciprocal friendship), and Bullying (i.e., history and severity of peer victimization and exclusion). Based on parent report (including concrete behavioural examples), clinicians assign items a Likert rating from 0 (no difficulties) to 3 (severe). For example, a ToM score of 0 indicates awareness and sensitivity to the thoughts, feelings, and opinions of others, while 3 indicates very poor to no awareness that has caused difficulties in the child’s functioning. These items have demonstrated inter-rater reliability (ICC = 0.82-0.89) (Kerns et al., 2017).

ADOS-2 Social Affect Composite (ADOS:SA)

The ADOS-2 (Lord et al., 2012) is a semistructured observational assessment of social, communicative, and restricted or repetitive behaviours associated with autism with strong diagnostic sensitivity (.91) and specificity (.84; Gotham, Risi, Pickles, & Lord, 2007). The Social Affect composite (ADOS:SA) captures children’s social communication ability, social interest, and social awareness and has demonstrated construct validity and diagnostic reliability (Hus & Lord, 2014; Kuhfeld & Sturm, 2018).

Social Responsiveness Scale-2 (SRS-2).

The SRS-2 (Constantino & Gruber, 2012) is a parent-report measure of social difficulties associated with autism as they occur on a continuum in the general population, including social motivation, social cognition, and social communication as well as repetitive and restricted behaviors and interests. The SRS-2 has demonstrated sound psychometric properties, including good internal (.72–.93), inter-rater (.8), and test-retest reliability (.83; Frazier et al., 2014).

Child Behavior Checklist: Social Problems subscale (CBCL:SP).

The Child Behavior Checklist: Social Problems subscale (CBCL:SP) (Achenbach, 1999) is a broad measure of peer difficulties and immaturity. The CBCL syndrome scales have strong psychometric properties and reliability (r_c = 0.69-0.94), including the Social Problems subscale (r_c = 0.84), in autistic children (Pandolfi, Magyar, & Dill, 2012).

Data Analytic Plan

Network estimation.

All network analyses were conducted using the R package BGGM, which performs network estimations within a Bayesian statistical framework (Williams & Mulder, 2020). First, we calculated the independent associations between each variable and the rest of the variables (i.e., nodes) in the network, controlling for all other linear associations in the model. These unique node-to-node connections (i.e., edges) reflect partial correlation coefficients. Missing data were handled using multiple imputation (20 imputations, 10 iterations) and estimates were based on pooled results. This model was then used to plot the network using the R package qgraph (Epskamp et al., 2012). Node placement was determined by the Fruchterman-Reingold (1991) algorithm, which places the most strongly connected nodes centrally and weakly connected nodes peripherally while also avoiding overlap among nodes and edges. Finally, network communities (i.e., strongly connected clusters of nodes) were estimated using the Spinglass algorithm from the R package igraph (Csardi & Nepusz, 2006) to describe how variables cluster together. The Spinglass algorithm tests for communities whereby the number and strength of edges within a cluster exceeds the number and strength of edges between clusters (Reichardt & Bornholdt, 2006).

Aim 1: Pairwise comparisons of edges between anxiety and social nodes.

To address Aim 1, network analyses were conducted to identify which facets of social functioning were most strongly related to anxiety symptom severity and comorbidity (number of anxiety disorders). To test this, we conducted pairwise comparisons of the association between each social node and anxiety (severity and comorbidity, respectively) using posterior probability (PP) and 95% credible intervals (CIs) for differences. 95% CIs reflect a range of values, given the present results, that have a 95% probability of including the true population value, and PP reflects the probability of an outcome given the current data. Very high (i.e., ≥ 0.975) or very low (i.e., ≤ 0.025) PPs were interpreted as evidence in favor of a true difference.

Aim 2: Pairwise comparisons of predictability of social nodes.

To address Aim 2, we conducted pairwise comparisons of the predictability (i.e., the strength of all associations that a node has with other nodes) of each social node, controlling for anxiety severity and comorbidity, to determine whether any social node(s) in the network are most strongly related to difficulties in other areas of social functioning. To test this, we computed PPs and 95% CIs of pairwise comparisons between the predictability estimates of each social node in the network. For example, we calculated PPs and 95% CIs to determine whether ToM is more strongly related to other nodes in the network than friendship quality.

Aim 3: Binary logistic regressions.

For Aim 3, a subsample (n = 118) of participants with clinical or subclinical social fears (i.e., CSR of ≥ 3) were examined to investigate how social functioning relates to the presentation of social fears. Participants who met diagnostic criteria for SAD, including fear of negative evaluation, were placed in the SAD group, whereas those who met diagnostic criteria without fear of negative evaluation were placed in the OSF group. Four binary logistic regressions conducted in SPSS 23 examined whether scores indicating significant social difficulties on four ADIS/ASA Social Functioning items (ToM, Social Motivation, Bullying, Friendships) differentiated the SAD versus OSF groups. Items were dummy-coded as 1 “present” (original scores of 2 “moderate” or 3 “severe”) and 0 “absent or minimal” (original scores of 0 or 1, respectively). Covariates in Step 1 of each regression included research site, anxiety severity (highest CSR across all anxiety diagnoses), anxiety comorbidity (number of co-occurring anxiety diagnoses), and diagnosis per the ADIS/ASA of co-occurring ADHD, depression, and ODD (1 “present” and 0 “absent”). Step 2 included anxiety group (SAD, OSF), with SAD as the reference group.

Results

Descriptive Statistics of Demographic and Clinical Variables

Table 1 presents demographic and clinical characteristics of the current sample. Participants were predominantly male (79.1%) and White (76.7%), and presented with varied autism and anxiety symptom severity and mildly impaired to superior intellectual functioning. Five participants (2.6%) received an FSIQ below 70, with four (2.1%) above 130. Thirty-nine percent were using psychiatric medication, primarily stimulants (20.4%) and selective serotonin reuptake inhibitors (16.8%). Visual inspection of box plots revealed no extreme outliers (i.e., values > 3 times the interquartile range away from the median) on any of the variables of interest. See Supplemental Table 1 for descriptive statistics of the variables of interest.

Table 1.

Descriptive statistics of sample demographic and clinical characteristics.

	N	n (%) or M (SD)
Demographic Characteristics
Age (years)	191	9.96 (1.80)
Sex	191
Male		151 (79.1%)
Female		40 (20.9%)
Ethnicity	189
Black or African American		11 (5.8%)
Asian		15 (7.9%)
White		145 (76.7%)
Aboriginal		4 (2.1%)
Other		14 (7.4%)
Living Arrangements	189
Both biological parents		131 (69.3%)
Both biological parents (joint custody)		7 (3.7%)
Biological parent and stepparent/partner		14 (7.3%)
Single biological parent		24 (12.7%)
Adoptive parent(s)		7 (3.7%)
Other relative(s)/friend(s)		2 (1.0%)
Other		4 (2.1%)
Site	191
University of California Los Angeles		74 (38.7%)
University of South Florida		69 (36.1%)
Temple University		48 (25.1%)
Clinical Characteristics
Full-Scale IQ (WISC-IV)	185	100.77 (16.16)
ADOS-2 Total Severity Score	181	7.40 (2.02)
PARS 7-Item Anxiety Severity Score	187	23.86 (3.61)
Co-Occurring Psychopathology	191
Attention Deficit Hyperactivity Disorder		125 (65.4%)
Oppositional Defiant Disorder		34 (17.8%)
Major Depressive Disorder or Dysthymia		13 (6.8%)
Obsessive-Compulsive Disorder		24 (12.6%)
Post-Traumatic Stress Disorder		1 (0.5%)
Traditional Anxiety	191
Social Anxiety Disorder		93 (48.7%)
Separation Anxiety Disorder		42 (22.0%)
Specific Phobia		86 (45.0%)
Generalized Anxiety Disorder		136 (71.2%)
Special Interest Fear		3 (1.6%)
Fear of Change		40 (20.9%)
Distinct Anxiety	191
Other Social Fear		25 (13.1%)
Unusual Phobia		16 (8.4%)
Atypical OCD		3 (1.6%)
Medication (% taking)	191	74 (38.7%)
Selective Serotonin Reuptake Inhibitors		32 (16.8%)
Antipsychotics		13 (6.8%)
Stimulants		39 (20.4%)
α-agonists		21 (11.0%)
Other		17 (8.9%)

Note. N: Number of responding participants, n(%): Number (percentage) of participants, M: Mean, SD: Standard deviation

Network Estimation & Pairwise Comparisons of Edges to Anxiety Nodes

Two distinct communities of strongly connected nodes were detected by the Spinglass algorithm that seem to reflect strong associations among measures of social ability and interpersonal relatedness, respectively, with few associations between these communities (see Figure 1). The first community consisted of nodes related to social communication and reciprocity: the ADOS-2 social affect composite (i.e., ADOS:SA) and ASA friendship, social motivation, and ToM items. The second community consisted of nodes related to relational difficulties and general social competence, including the CBCL Social Problems subscale (i.e., CBCL:SP), SRS-2 total, and ASA Bullying item, as well as the anxiety severity and comorbidity nodes, which notably were only connected via bullying. Using PPs and 95% CIs of pairwise comparisons between anxiety severity and comorbidity and each of the social nodes, we found evidence of a stronger association between anxiety comorbidity and bullying than between anxiety comorbidity and ToM (PPs = 0.01). For the remaining social variables, no statistically significant differences emerged in the strength of associations with anxiety severity or comorbidity. Given that bullying was the only social node that demonstrated a unique association with either of the anxiety nodes, post-hoc pairwise comparisons were conducted to examine whether any other social nodes were significantly associated with bullying, and thus may be indirectly associated with anxiety comorbidity. Post-hoc analyses revealed that bullying demonstrated significantly stronger associations with social problems (i.e., CBCL-SP), social affect (i.e., ADOS:SA), and ToM compared to the other social nodes (i.e., Friends, Social Motivation, SRS-2) (PPs = 0.98-1.00). Full results are presented in Table 2 and Table 3.

Figure 1. Network structure of social and anxiety variables of interest.

Notes. Green edges depict a positive association; red edges (i.e., the edge between bullying and ADOS) depict a negative association. Only edges in which the 95% credible interval for the edge weight did not cross zero are included. Communities are indicated by the blue (light gray) and orange (dark grey) nodes.

Table 2.

Within-sample posterior probabilities (PPs) and 95% credible intervals (within parentheses) for select social nodes being more strongly linked to anxiety severity (upper panel) and anxiety comorbidity (lower panel) than other social nodes. Comparisons for which the CI does not include zero are highlighted in bold.

	Partial r to anxiety sev.	SRS-2-- anxiety sev.	CBCL:SP-- anxiety sev.	ADOS:SA-- anxiety sev.	ToM--anxiety sev.	Soc Mot-- anxiety sev.	Friends-- anxiety sev.	Bullying-- anxiety sev.
SRS-2--anxiety sev.	.068	-
CBCL:SP--anxiety sev.	−.007	0.58 (−0.253, 0.291)	-
ADOS:SA--anxiety sev.	−.081	0.86 (−0.103, 0.382)	0.84 (−0.109, 0.353)	-
ToM--anxiety sev.	.105	0.38 (−0.320, 0.214)	0.31 (−0.338, 0.206)	0.09 (−0.463, 0.086)	-
Soc Mot--anxiety sev.	.006	0.63 (−0.209, 0.281)	0.55 (−0.211, 0.248)	0.23 (−0.373, 0.176)	0.70 (−0.220, −0.387)	-
Friends--anxiety sev.	−.113	0.87 (−0.108, 0.407)	0.84 (−0.123, 0.382)	0.53 (−0.239, 0.256)	0.90 (−0.095, 0.502)	0.75 (−0.216, −0.434)	-
Bullying--anxiety sev.	.120	0.28 (−0.310, 0.161)	0.25 (−0.366, 0.187)	0.03 (−0.426, 0.008)	0.42 (−0.303, 0.248)	0.18 (0.355, 0.124)	0.05 (−0.489, 0.048)	-
	Partial r to anxiety com.	SRS-2-- anxiety com.	CBCL:SP-- anxiety com.	ADOS:SA-- anxiety com.	ToM--anxiety com.	Soc Mot-- anxiety com.	Friends-- anxiety com.	Bullying-- anxiety com.
SRS-2--anxiety com.	.049	-
CBCL:SP--anxiety com.	−.006	0.69 (−0.195, 0.324)	-
ADOS:SA--anxiety com.	.031	0.53 (−0.233, 0.244)	0.31 (−0.284, 0.159)	-
ToM--anxiety com.	−.103	0.89 (−0.096, 0.391)	0.75 (−0.177, 0.330)	0.75 (−0.118, 0.392)	-
Soc Mot--anxiety com.	.001	0.68 (−0.181, −0.287)	0.45 (−0.226, 0.204)	0.45 (−0.213, −0.318)	0.24 (−0.381, 0.199)	-
Friends--anxiety com.	.027	0.66 (−0.197, 0.305)	0.46 (−0.254, 0.241)	0.46 (−0.201, 0.288)	0.25 (−0.376, 0.179)	0.49 (−0.302, 0.315)	-
Bullying--anxiety com.	.201	0.09 (−0.377, 0.067)	0.05 (−0.479, 0.042)	0.05 (−0.373, 0.041)	0.01 (−0.563, −0.043)	0.03 (−0.431, 0.013)	0.05 (−0.471, 0.044)	-

Table 3.

Post-hoc analyses of the strength of associations between the bullying node and each other social node in the network using withinsample posterior probabilities of predictability and 95% credible intervals (within parentheses) for differences.

	r to Bullying	SRS-2	CBCL:SP	ADOS:SA	ToM	Soc Mot	Friends
Full sample
SRS-2	−.046	-
CBCL:SP	.360	0 (−0.636, −0.096)	-
ADOS:SA	−.202	0.85 (−0.124, 0.384)	1.00 (0.285, 0.725)	-
ToM	.227	0.032 (−0.511, 0.013)	0.84 (−0.123, 0.360)	0 (−0.632, −0.117)	-
Soc Mot	−.112	0.71 (−0.180, −0.323)	1.00 (0.213, 0.666)	0.32 (−0.332, 0.201)	0.98 (0.011, 0.606)	-
Friends	.066	0.18 (−0.402, 0.147)	0.97 (−0.001, 0.491)	0.02 (−0.499, −0.018)	0.80 (−0.156, 0.411)	0.11 (−0.504, 0.120)	-

Notes. PPs are based on the assumption that the association with the bullying node for the variable in the first column is larger than that of the variables in columns three to eight, respectively. The difference scores are computed by subtracting the partial correlation between the node in the first column and bullying from the partial correlation between bullying and each of the nodes in columns three to eight. A larger PP indicates a stronger association with bullying for the node in the first column compared to the nodes in columns three to eight, while a smaller PP indicates a weaker association with bullying for the node in the first column. CIs indicate where the difference between the two partial correlations are expected to fall 95% of the times and CIs that do not include zero are considered statistically significant.

Pairwise Comparisons of Predictability of Social Nodes

Using PPs and 95% CIs of pairwise comparisons of predictability for the social nodes, we found robust evidence that social motivation demonstrated the strongest associations in the network (PPs ranging from 0.93-0.99), resulting primarily from its strong connections to the ToM, friendship quality and social affect nodes. However, ToM demonstrated the greatest number of edges with other nodes in the network. Interestingly, ADOS:SA scores were negatively associated with bullying when considered in concert with ToM. Full results are presented in Table 4.

Table 4.

Within-sample posterior probabilities of predictability of social nodes and 95% credible intervals (within parentheses) for differences.

	Mean predictability	SRS-2	CBCL:SP	ADOS:SA	ToM	Soc Mot	Friends	Bullying
Full sample
SRS-2	.276	-
CBCL:SP	.330	0.17 (−0.174, 0.060)
ADOS:SA	.224	0.79 (−0.081, 0.181)	0.94 (−0.029, 0.240)	-
ToM	.267	0.56 (−0.113, 0.133)	0.84 (−0.063, 0.194)	0.23 (−0.157, 0.072)	-
Soc Mot	.440	0.01 (−0.305, −0.026)	0.07 (−0.249, 0.033)	0 (−0.359, −0.070)	0.01 (−0.310, −0.038)	-
Friends	.310	0.31 (−0.174, 0.103)	0.62 (−0.118, 0.162)	0.11 (−0.224, 0.052)	0.24 (−0.165, 0.079)	0.95 (−0.024, 0.283)	-
Bullying	.322	0.25 (−0.179, 0.083)	0.56 (−0.115, 0.134)	0.06 (−0.220, 0.026)	0.18 (−0.178, 0.065)	0.95 (−0.023, 0.260)	0.43 (−0.151, 0.130)	-

Notes. PPs are based on the assumption that the predictability value for the variable of interest is larger than the predictability value for the other social nodes. The difference scores are computed by subtracting predictability estimates for the other social nodes (e.g., CBCL:SP) from predictability estimates for the social node of interest (e.g., SRS-2). A larger PP and positive CIs indicate greater predictability for the social node presented at the top of the column, while a smaller PP and negative CIs indicate lower predictability (e.g., the SRS-2 has significantly lower predictability than Social Motivation).

Binary Logistic Regressions of ADIS/ASA Social Functioning Items and Traditional vs. Distinct Social Anxiety

Only ToM demonstrated significant differences across the two social anxiety groups (β = 1.06, p = .032, OR = 2.90 [1.10, 7.65]): Children with OSF were significantly more likely to experience difficulties in ToM than children with SAD. The finding was retained at trend-level after accounting for covariates (i.e., co-occurring psychopathology, severity and comorbidity of anxiety, site) (β = 1.00, p = .051, OR = 2.71 [1.00, 7.35]). No significant differences were observed between groups on the Social Motivation, Friends, or Bullying items. However, several significant covariates emerged: greater anxiety severity was associated with better ability to establish and maintain reciprocal friendships (β = −0.77, p = .016, OR = 0.46 [0.25, 0.87]), and greater anxiety comorbidity was associated with more severe bullying (β = 0.46, p = .008, OR = 1.59 [1.13, 2.25]). Co-occurring depression was also associated with more severe bullying at trend-level β = 1.30, p = .054, OR = 3.67 [0.97, 13.80]). See Supplemental Table 2 for full regression results.

Discussion

Psychosocial interventions for autistic children could be optimized by an enhanced understanding of the ways in which social functioning and co-occurring anxiety relate to one another. The current study employed multi-method, multi-informant reports as well as network analytic techniques in a Bayesian statistical framework and hierarchical regressions to examine how different aspects of social functioning are related to one another and to co-occurring anxiety, particularly the presentation of social anxiety, in a large treatment-seeking sample of autistic children.

Two distinct communities emerged in the current network. Both included measures from multiple unique sources (i.e., clinical observation, clinical interview with parent, parent-report questionnaire), extending prior work which has primarily relied on parent-report measures (e.g., Duvekot et al., 2018; McVey et al., 2018). The first, which included measures of ToM, social motivation, friendship quality, and social affect, seems to reflect social difficulties characteristic of autism (i.e., difficulties in social communication and understanding that affect social behaviour and relationships; APA, 2013). That is, in the current treatment-seeking sample of autistic children with clinical anxiety, it seems that the ability to engage in reciprocal social communication, and the social awareness, interest and engagement needed to develop positive relationships with peers, may be a highly interconnected set of social skills. The second community included measures of broader difficulties with interpersonal relatedness (SRS-2, CBCL social problems), bullying, and anxiety severity and comorbidity, possibly indicating that more severe anxiety may be associated with greater peer rejection and reduced social competence. Notably, the SRS-2 total score captures differences in social communication as well as restricted and repetitive behaviours and interests, aligning with prior work identifying both social communication difficulties and restricted behaviours and interests as risk factors for bullying among autistic youth (Schroeder et al., 2014). Additionally, though the SRS-2 aims to provide a continuous measure of autistic traits, prior research suggests the measure may also be elevated by other forms of psychopathology, like anxiety, which can also have negative social effects (South et al., 2017).

Anxiety severity and comorbidity were highly peripheral in the network with minimal direct links to social skills and peer relationships, in contrast to some prior findings (Duvekot et al., 2018; Eussen et al., 2013). It is possible that the range of anxiety severity was too narrow in this treatment-seeking sample to detect significant associations, in contrast to prior studies which primarily involved samples with non-clinical anxiety (e.g., Eussen et al., 2013; McVey et al., 2018). Network analysis may also obscure spurious connections between anxiety and different facets of social functioning that are better explained by a third variable (e.g., bullying). Nonetheless, bullying experiences were independently associated with the likelihood of having multiple anxiety disorders, which is consistent with prior work (Chou et al., 2020). Bullying experiences were also strongly associated with general difficulties getting along with peers, poorer social communication ability, and poorer ToM, which may suggest an indirect link between difficulties understanding and connecting socially with others and increased levels of anxiety. Perhaps this is due to the increased social adversity that is often experienced by children with greater social challenges (Schroeder et al., 2014), though further testing is needed to confirm this exploratory finding. Overall, these findings highlight a need for early anxiety intervention and prevention strategies to help autistic children better cope with negative peer experiences.

Among the different facets of social functioning captured, social motivation emerged as the most influential in the network, demonstrating strong associations with other facets of social ability (i.e., ToM, friendship quality, social affect) and playing a markedly stronger role in the overall social well-being of children in the current sample than ToM, ADOS-2 social affect scores or the SRS-2. To develop positive relationships with peers and to foster the social skills necessary to maintain those relationships, it may be necessary to first have an internal drive or desire for social interaction and relationships, and to seek opportunities to practice those social skills. Studies of both autistic and non-autistic youth indicate associations between social motivation and the development of social skills and positive peer relationships (Chevallier et al., 2012; Sedgewick, Hill, Yates, Pickering, & Pellicano, 2015). Socially disinterested children, while not distressed by their relative social disconnection, may have less opportunity or motivation to develop the social skills that would allow them to foster positive peer relationships, and tend to be less well-liked and more excluded by their peers (Coplan et al., 2013). Notably, programs that involve learning and practicing social skills in a naturalistic setting are associated with significant improvements in social skills and peer relationship quality, as well as reductions in anxiety symptoms (Laugeson et al., 2014; McVey et al., 2016). Perhaps the opportunity to socialize with peers with similar social difficulties in a controlled and supportive environment, with social skills training to bolster self-efficacy and increase positive social experiences, may also lead to increased social interest. However, this possibility requires further evaluation.

Contrary to our hypotheses, ToM was not the most central or strongly connected node in the network, though it was the most highly interconnected, demonstrating significant associations with four of the six social nodes. The development of ToM thus may also play an important role in the social well-being of autistic children with clinical anxiety. Though we are unable to determine directionality given the cross-sectional nature of this work, it seems that the ability to understand others’ thoughts and feelings is associated with both seeking out and experiencing positive peer relationships, as well as reducing risk of negative social experiences. Indeed, prior work has shown that ToM is related to better social well-being (e.g., peer acceptance, friendships) and less bullying among autistic and non-autistic youth (Liu et al., 2018; Peterson et al., 2016). Interestingly, when controlling for this relationship between ToM and bullying, ADOS:SA scores were negatively associated with bullying. This finding may underscore the critical link between ToM ability and bullying in youth on the autism spectrum, and is consistent with the Rowley et al. (2012) finding that autistic children in a mainstream school setting who were less socially impaired experienced higher levels of victimization than more socially impaired children. ToM skills training programs involving autistic youth have demonstrated improvements in social skills and peer acceptance, as well as reductions in peer victimization (Liu et al., 2018; Szumski et al., 2019). Taken together, these findings suggest that ToM training may offer particular benefits to the social well-being of autistic children with cooccurring anxiety.

Notably, there is much debate about whether seeking to improve “neurotypical” social skills among autistic youth may do more harm than good by increasing the emotionally taxing task of camouflaging social difficulties, and thus potentially increasing risk of negative mental health outcomes such as anxiety, depression and suicide (Beck et al., 2020; South, Costa, & McMorris, 2021). Alternatively, providing anxious autistic youth with opportunities to develop their social skills via social skills training may reduce the uncertainty and confusion of social interactions and provide the option to exercise neurotypical social skills when needed, the pros (e.g., greater social acceptance) and cons (e.g., compromised sense of self) of which must be carefully weighed. Another key consideration for the efficacy of ToM and social skill interventions is the possible role of language ability; though verbal IQ (assessed via the WISCIV) was not significantly correlated with any of the variables of interest in the current study, it may be valuable to explore the role of language in ToM and social skill acquisition using a more specific tool in future work.

Hypotheses regarding the relationship of social functioning to different presentations of social anxiety were partially supported. Autistic children who demonstrated an impairing fear of social situations, but not necessarily concerns about negative social evaluation, had more difficulty understanding the thoughts and feelings of others; however, they were not more or less socially motivated, nor more or less likely to have poorer quality friendships or experience peer victimization or rejection, than autistic children with classic SAD. Thus, differences in the presentation of social anxiety in autistic children, particularly the presence or absence of fear of negative evaluation, may emerge from differences in social cognition. As reported in qualitative studies of social fears in autistic adults (e.g., Halim et al., 2018), these children may fear and avoid social situations as a result of the confusion and uncertainty that undoubtedly results from struggling to understand others’ intentions and emotions and feeling unsure of the rules of social interactions and relationships. Alternatively, it could be that these children are less likely to communicate their mental states and/or the source of their worries, such that their parent underestimate both their ToM abilities and their social evaluative concerns. Notably, children with OSF in this study demonstrated poorer social communication ability (per the ADOS:SA) compared to children with SAD, replicating prior research that has found a link between direct observation of social communication difficulties and distinct presentations of anxiety in autistic youth (Kerns et al., 2017). Our findings affirm the distinct presentation of social anxiety in youth on the autism spectrum and suggest the inclusion of social skills and ToM training into interventions for autistic children with co-occurring social anxiety, particularly distinct expressions of social anxiety.

The current study had several limitations that suggest important future directions. The data is cross-sectional, prohibiting an understanding of developmental changes in the network or the direction of effects between variables. Future research is needed to clarify the directionality of the associations detected here. For example, though this study suggests ToM and social motivation play a key role in the social well-being of autistic children with anxiety, it is unclear if more developed ToM skills result in greater social motivation and better social communication, or if children who are more socially motivated and have more positive social experiences then develop better ToM skills by learning from their peers (McAlister & Peterson, 2013). Longitudinal analyses investigating the development of social skills, peer relationships, and anxiety symptoms across childhood and adolescence could better illuminate these trajectories and provide insight into early risk factors for social difficulties and anxiety symptoms.

Second, although the clinical severity ratings of the ADIS/ASA were made by trained clinicians, the ratings were based on parent report; thus, most measures in the study ultimately relied on parent perspectives, albeit interpreted through a clinical lens. Parent report may not be fully representative of children’s social skills and relationships, given parents’ limited access to their child’s social world (Matsunaga, 2009). Additionally, though parents may have a good sense of their child’s functional ToM skills and social motivation in daily life, experimental tasks or naturalistic observations could add to our understanding of these constructs and the potential for parental bias in the future. Third, the current sample was treatment-seeking and consisted of children with relatively higher IQ, which may have resulted in a different profile of social functioning than a community sample with a wider range of anxiety severity and cognitive ability. The current sample was also predominantly male and White, which may limit the generalizability of findings to all individuals on the spectrum. Finally, the OSF group was relatively small (n = 26); thus, we may have been underpowered to detect differences between social anxiety groups. Further work is needed to replicate and extend these findings in larger, more representative samples.

In conclusion, the current study offers a nuanced view into how difficulties across different aspects of social functioning may relate to one another and influence the presentation of anxiety disorders among autistic children. Our findings suggest that problems in one social domain have a domino effect on other, related forms of social functioning. In particular, social motivation may be especially influential in the overall social well-being of these children, highlighting the importance of intervention programs that provide social skills training and offer opportunities for anxious, autistic youth to develop positive social connections. ToM may also be related to a broad range of social abilities and experiences, and may affect the expression of social worries among autistic children. Our findings provide further support for the notion that a qualitatively distinct presentation of social worries, characterized by impaired ToM abilities, may occur among a subset of autistic children. Tailoring intervention to address this unique presentation of social anxiety, potentially by incorporating ToM skill training into traditional anxiety-focused strategies, may be critical to enhance outcomes for these youth.