SHORT REPORT

Exploring Social Subtypes in Autism Spectrum Disorder:

A Preliminary Study
Mirko Uljarevic , Jennifer M. Phillips, Rachel K. Schuck, Salena Schapp, Elizabeth M. Solomon,
Emma Salzman, Lauren Allerhand, Robin A. Libove, Thomas W. Frazier , and Antonio Y. Hardan

Impairments in social functioning are considered a hallmark diagnostic feature of autism spectrum disorder (ASD). Yet,
individuals diagnosed with ASD vary widely with respect to specific presentation, severity, and course across different
dimensions of this complex symptom domain. The aim of this investigation was to utilize the Stanford Social Dimen-
sions Scale (SSDS), a newly developed quantitative measure providing parental perspective on their child’s social abilities,
in order to explore the existence of homogeneous subgroups of ASD individuals who share unique profiles across specific
dimensions of the social domain. Parents of 164 individuals with ASD (35 females, 129 males; meanage = 7.54 years,
SD = 3.85) completed the SSDS, the Social Responsiveness Scale (SRS-2) and the Child Behavior Checklist (CBCL). Data
on children’s verbal and nonverbal intellectual functioning (FSIQ) were also collected. The Latent Profile Analysis was
used to classify participants according to the pattern of SSDS subscale scores (Social Motivation, Social Affiliation, Expres-
sive Social Communication, Social Recognition, and Unusual Approach). Five profiles were identified. Profiles did not dif-
fer in terms of chronological age nor gender distribution but showed distinct patterns of strengths and weaknesses across
different social components rather than simply reflecting a severity gradient. Profiles were further differentiated in terms
of cognitive ability, as well as ASD and internalizing symptom severity. The implications of current findings and the nec-
essary further steps toward identifying subgroups of individuals with ASD who share particular constellation of strengths
and weaknesses across key social domains as a way of informing personalized interventions are discussed. Autism Res
2020, 00: 1–8. © 2020 International Society for Autism Research, Wiley Periodicals, Inc.

Lay Summary: People with autism spectrum disorder (ASD) vary greatly in terms of their social abilities and social motiva-
tion. However, researchers lack measures that can fully assess different components of social functioning. This paper pro-
vides initial evidence for capturing subgroups of individuals with ASD with specific strengths and weakness across
different aspects of social functioning.

Keywords: autism spectrum disorder; social processing; social motivation; individual differences

Introduction treatment [Beglinger & Smith, 2005; Vivanti, Prior,

Williams, & Dissanayake, 2014]. Reduced ability to initi-
Impairments in social functioning are an early emerging ate, regulate, and maintain social relationships across dif-
and stable cardinal feature of autism spectrum disorder ferent social contexts can result from impairments across
(ASD; American Psychiatric Association, 2013). Yet, indi- distinct, basic social processes. Therefore, the ability to
viduals diagnosed with ASD vary widely with respect to identify subgroups of individuals with unique constella-
specific presentation, severity, and course across different tions of strengths and weakness across distinct compo-
dimensions of social functioning [Bauminger et al., 2008; nents of social processing is an essential step in informing
Leekam, 2016; Uljarevic & Hamilton, 2013; Vivanti & personalized intervention and case management plans.
Nuske, 2016]. Wide phenotypic variability likely reflects Social functioning is a complex, multifaceted construct
diverse aetiological mechanisms, and phenotypically encompassing a range of distinct yet related processes.
diverse individuals will differ not only in terms of Although a universally agreed upon definition and con-
treatment needs but also in their response to a given tent for this construct is yet to be reached, the following

From the Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, California (M.U., J.M.P., R.K.S., L.A., R.A.L., A.Y.H.); Depart-
ment of Psychiatry, Kaiser Permanente, Redwood City, California (S.S.); Department of Psychiatry and Behavioral Sciences, Davis Medical Center, Univer-
sity of California, Sacramento, California (E.M.S.); Department of Psychiatry, University of California San Francisco, San Francisco, California (E.S.);
Autism Speaks, New York (T.W.F.); Melbourne School of Psychological Sciences Faculty of Medicine, Dentistry, and Health Sciences, The University of
Melbourne, Victoria (M.U.); Department of Psychology, John Carroll University, University Heights, Ohio (T.W.F.)
Mirko Uljarevic and Jennifer M. Phillips are cofirst authors since they contributed equally to the work.
Received February 14, 2020; accepted for publication March 3, 2020
Address for correspondence and reprints: Mirko Uljarevic, Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA. E-
mail: [email protected]
Published online 00 Month 2020 in Wiley Online Library (wileyonlinelibrary.com)
DOI: 10.1002/aur.2294
© 2020 International Society for Autism Research, Wiley Periodicals, Inc.

INSAR Autism Research 000: 1–8, 2020 1

dimensions of social processing consistently emerge 2009; Lerner, De Los Reyes, Drabick, Gerber, & Gadow,
across the literature examining different neurodevelop- 2017; Livingston et al., 2018; Kang, Gadow, & Lerner,
mental and neuropsychiatric disorders including ASD: 2018]. For example, Livingston et al. [2018] estimated the
(a) the ability to perceive and interpret social signals; ability of 136 adolescents with ASD to compensate for
(b) motivation to engage in social interactions; and social deficits by contrasting their social impairments
(c) skills necessary for initiating, maintaining, and ending (measured by ADOS Social Affiliation score) and theory of
social interactions [Green, Horan, & Lee, 2015; Happe & mind (TOM) ability (measured by the computerized
Frith, 2014; Happe, Cook, & Bird, 2017; Huber, Plötner, & Frith-Happe Animations [Abell, Frith, & Happe, 2000]).
Schmitz, 2019; Kennedy & Adolphs, 2012; Pallathra Participants classified as high compensators (TOM and
et al., 2018]. These constructs are also in line with RDoC ADOS scores above the median for the group) showed
constructs of affiliation and attachment, social communi- higher IQ scores and better executive functioning perfor-
cation, and perception and understanding of others, mance but higher anxiety when compared to participants
which are described under the social processes domain characterized as low compensators (TOM and ADOS
(Insel et al., 2010; National Advisory Mental Health scores below the median). A recent study by Kang et al.
Council Workgroup on Tasks and Measures for Research (2018) utilized 13 atypical communication characteristics
Domain Criteria, 2016; Social Processes Workshop, (e.g., speech delay, babbling, pronoun reversal, pragmatic
2012). Importantly, noted processes can be assessed difficulty) from the Parent Questionnaire (Gadow et al.,
through other (parental, clinician) and self-reports and 2008) and derived four subgroups of individuals with
are applicable across a wide range of development, and ASD sharing similar symptom profiles that authors
relatedly, across the spectrum of intellectual functioning. labeled as (a) Speech Delay + Pragmatic Difficulty
Given the complexity and interrelatedness of the noted + Fixated Language; (b) Pervasive Atypical Communica-
processes that underpin social functioning, it is clear that tion; (c) Little Professors; and (d) Moderate Pragmatic Dif-
relative strengths and weaknesses across these domains ficulty Only. Identified subgroups showed differences in
can lead to highly variable social phenotype seen in ASD. terms of the severity of ASD symptoms and the severity
However, there have been surprisingly few attempts to of psychiatric symptoms as well as in terms of functional
identify subgroups of individuals who share distinct pat- outcomes.
terns of individual variation along different social In this study, we aimed to utilize the Stanford Social
domains. Dimensions Scale (SSDS; Phillips et al., 2019), a new scale
In their pioneering work on the typology of social developed to enable a detailed depiction of individual
impairments in ASD, Wing and Gould [1979] utilized variation in different components of social motivation—
individual variability in social interaction style to identify (a) the drive or desire to interact socially and affiliate with
the following three subgroups: Aloof, characterized by others, independent of the quality of the interaction;
social indifference and unresponsiveness to others’ social (b) expressive social communication—the ability to con-
approaches; Passive, characterized by lack of spontaneous vey social and emotional information when interacting
social initiations but adequate responsiveness to social with others (e.g., eye contact, facial expression, gestures,
approach; and Active-but-Odd, characterized by active body postures, prosody); and (c) social recognition and
seeking of social contact, but in a manner inappropriate understanding—the ability to perceive and interpret
to a given context and perceived as unusual by others. social signals ranging from basic emotion signals to com-
Subsequent work, mainly utilizing the Wing Subgroups plex mental states, communicated through the face,
Questionnaire [Castelloe & Dawson, 1993], has relatively body, and tone of voice. The initial set of items was con-
consistently shown that while the aloof and active-but- ceptually developed through systematic search of the lit-
odd subtypes lie on opposite ends of the severity spec- erature and in consultation with clinical and research
trum in terms of cognitive ability and autism symptom- experts in order to tap into the noted social constructs
atology, individuals classified as active-but-odd show and to reflect the full range of behaviors seen across ASD
significantly more impairments in attention, cognitive, and normative social development. Exploratory structural
and emotional regulation, as well as more severe internal- equation modeling was used to examine the SSDS for
izing symptoms (Bonde, 2000; Scheeran et al., 2012). latent components in a sample of children and adoles-
Although noted evidence provides some support for cents with ASD and largely confirmed social motivation
the utility and validity of the Wing and Gould subtypes, (consisting of separate motivation/drive and affiliation
this taxonomy was based only on a broad conceptualiza- components), expressive social communication and
tion of social interaction style without considering other social recognition factors and yielded an additional
social dimensions. In addition to the summarized work “unusual approach” factor (capturing the approach/initi-
by Wing and Gould, several relatively recent investiga- ation of social interactions in a manner that is unusual in
tions have focused on identifying informative subgroups terms of content and/or intensity). Noted factor structure
of individuals with ASD [Anderson, Oti, Lord, & Welch, showed good to excellent fit (comparative fit index

2 Uljarevic et al./Social subtypes in autism spectrum disorder INSAR

[CFI] = 0.940, Tucker–Lewis index [TLI] = 0.919, root Stanford University Institutional Review Board. Parents
mean square error of approximation [RMSEA] = 0.048, and/or legal guardians participating in the study provided
standardized root mean square residual [SRMR] = 0.038) consent and completed all study questionnaires through
and good internal consistency as evidenced by Compos- a secure online portal. See Table 1 for the descriptive sta-
ite Reliability scores of ≥0.72. Our primary aim was to use tistics of the sample.
the SSDS in order to explore the existence of subgroups
of individuals with ASD who share particular patterns of Procedures and Measures
peaks and troughs across distinct dimensions of the social
The SSDS. The SSDS is a 58-item dimensional measure
domain. We also aimed to further characterize subgroups
designed to provide parental perspective on their child’s
by examining differences in subgroups’ cognitive ability
social abilities. Initial factorization returned a five-factor
and severity of ASD symptoms and co-occurring internal-
solution with factors interpreted as Social Motivation
izing and externalizing problems.
(SM), Social affiliation (SA), Expressive Social Communi-
cation (ESC), Social Recognition (SR), and Unusual
Approach (UA) [Phillips et al., 2019]. The scale has
Methods
Participants been shown to have a good construct, divergent, and
convergent validity and good to excellent reliability as
One hundred sixty-four individuals with ASD and their indicated by Composite Reliability Index scores of 0.90,
parents took part in the study (35 females, 129 males; 0.80, 0.74, 0.85, and 0.72 for SM, SA, ESC, SR, and UA
meanage = 7.54 years, SD = 3.85, range: 3–17 years). Par- factors.
ticipants were either recruited through ongoing research
projects conducted in the Stanford Autism and Develop- The Social Responsiveness Scale-Second Edition
mental Disorders Research Program or by advertisements (SRS-2). The SRS-2 is a 65-item parent-report measure
posted in the Stanford clinic and online (e.g., emails to designed to index autism severity (Constantino &
the lab research registry and parent listservs). Participants Gruber, 2012). Here we focused on social communica-
in ongoing studies (N = 76) had cognitive testing (using tion/interaction (SCI) and restricted/repetitive behavior
the Stanford Binet [5th edition, SB5, Roid, 2003] or (RRB) factors derived by Frazier et al. [2014].
Mullen Scales of Early Learning [Mullen, 1995]) and diag-
nostic assessment using the Autism Diagnostic Interview- The Child Behavior Checklist, Ages 11/2–5 and 6–18
Revised (ADI-R; Lord, Rutter, & Le Couteur, 1994; Rutter, (CBCL). The CBCL is a parent-report instrument
Le Couteur, & Lord, 2003; available for N = 44 partici- designed to assess behavior and emotion related prob-
pants) and/or the Autism Diagnostic Observation Schedule, lems in children. Here we focused on the internalizing
Second Edition (ADOS-2; Lord, Luyster, Gotham, & Guthrie, and externalizing problems scales [Achenbach &
2012; available for N = 73 participants). Diagnosis was Rescorla, 2000].
established by the multidisciplinary team. For partici-
pants recruited online (N = 88), inclusion criteria was a Analysis Plan
reported clinical diagnosis of ASD (clinical report was not
required) and SRS-2 total T-score of ≥60 [Constantino & Latent profile analysis (LPA) was conducted in Mplus Ver-
Gruber, 2005, 2012]. The two subsamples did not differ sion 8 [Muthén & Muthén, 1998-2015] using the robust
on any of the SSDS subscales, nor in terms of SRS-2 and maximum likelihood estimator in order to classify partici-
CBCL scores. This research study was approved by the pants according to the pattern of SM, SA, ESC, SR, and

Table 1. Participant Characteristics

Mean (SD) Range Ethnicity % Income %

Age (years) 7.54 (3.85) 3–17 Caucasian 42.2 >150,000 50.7

FSIQa 75.21 (27.84) 14–122 Asian 31.1 125–150,000 12.7
SRS-2 total T-score 76.50 (10.75) 52–102b Mixed race 14.9 100–125,000 6.0
CBCL internalizing 62.59 (10.77) 34–82 Hispanic 7.5 75–100,000 11.2
CBCL externalizing 56.70 (10.66) 28–80 Middle Eastern 1.2 50–75,000 9.0
Native American 1.2 35–50,000 4.5
Pacific Islander 1.2 25–35,000 1.5
African American 0.6 <25,000 4.5

CBCL, Child Behavior Checklist; FSIQ, Full-Scale IQ; SCI, social communication/interaction; SRS-2, Social Responsiveness Scale.
a
Data available for N = 76 participants.
b
Participants who had SRS-2 T scores < 60 (N = 9) met the clinical cutoff on ADOS and/or ADI-R.

INSAR Uljarevic et al./Social subtypes in autism spectrum disorder 3

UA SSDS subscale scores. Models with 1–8 profiles were and BIC values continued to increase for seventh and
estimated. The decision on an optimal number of profiles eighth profile. Considering that BIC performs best in
was guided by the statistical solution adequacy indexed smaller samples such as ours [Nylund, Asparouhov, &
by: (a) the Akaïke Information Criterion (AIC); (b) the Muthén, 2007], 5-profile solution was selected as the best
Consistent AIC (CAIC); (c) the Bayesian Information Cri- fitting solution. Examination of the adjacent 4- and
terion (BIC); (d) the sample-size Adjusted BIC (ABIC); and 6-profile solutions confirmed added value of 5 when com-
(e) the Bootstrap Likelihood Ratio Test (BLRT). The profile pared to the 4-profile solution and that the addition of
selection was assisted by (a) the higher entropy values sixth profile did not provide additional meaningful infor-
which indicate fewer classification errors and (b) the BIC mation. There were significant differences between the
(lower values indicate better fit) and the BLRT (significant profiles across all SSDS subscales: SM (F = 39.64,
values indicate that additional profile leads to fit P < 0.001, Partial η2 = 0.51), SA (F = 90.77, P < 0.001, Par-
improvement) that have been demonstrated as the most tial η2 = 0.70), ESC (F = 57.03, P < 0.001, Partial η2 = 0.60),
reliable information criteria and likelihood-based statis- SR (F = 75.52, P < 0.001, partial η2 = 0.66) and UA
tics, respectively [Nylund, Asparouhov, & Muthén, 2007; (F = 3.58, P = 0.008, Partial η2 = 0.09).
Lubke & Tueller, 2010; Masyn, Henderson, & Green- Table 3 shows SM, SA, ESC, SR, and UA SSDS scores
baum, 2010. Model selection was also guided by parsi- and comparison statistics across profiles. Table 4 shows
mony and interpretability [Bauer & Curran, 2003]. Once CA, FSIQ, SRS-2, and CBCL scores and comparisons across
the profiles were extracted, subgroup-level differences in profiles. Profiles did not differ in terms of CA (F = 2.16,
mean SSDS subscale scores were determined via one-way P = 0.076, Partial η2 = 0.054) nor gender distribution
analysis of variance (ANOVA) with Bonferroni-correction (χ 2 = 6.03, P = 0.20). As can be seen from Table 3, there
applied to post hoc tests. Pearson’s chi-square tests and were both severity and shape differences among identi-
one-way ANOVAs were then performed to explore differ- fied profiles. Profiles 4 and 5 had the highest scores across
ences in individual characteristics and chronological age most of the SSDS subscale scores (best performance/low-
(CA), Full-Scale IQ (FSIQ), SRS-2, and CBCL scores as a est impairments) with Profile 5 showing significantly
function of social dimensions subgroup. Partial eta higher SM, SA, ESC, and SR scores when compared to Pro-
squared (η2) and Cohen’s d were computed as a measure file 4 (all P < 0.001, Cohen’s d range: 0.97–2.27) These
of effect size for each comparison. To ensure the conver- two profiles were labeled as socially adaptive and mild,
gence across subtyping methods, a supplementary k- respectively. In addition, socially adaptive and mild pro-
means cluster analysis was also used to classify partici- files had the highest FSIQ and lowest SRS-2 SCI and RRB
pants based on their pattern of SSDS subscale scores. scores as well as CBCL Internalizing and Externalizing
scores when compared to other three profiles. Profile
2 had the lowest scores across all SSDS subscales (apart
Results from UA) as well as lowest FSIQ scores and highest SRS-2
and CBCL scores and was labeled as socially severe. Pro-
Information criteria and likelihood-based statistics for the files 1 and 3 showed distinctive patterns of strengths and
1–8 profile solutions are presented in Table 2. As can be weaknesses across SSDS subscales. More specifically, these
seen, all models had high entropy values (>0.9) and each two profiles had comparable SM and UA scores, however,
additional profile resulted in significant fit improvement while Profile 3 showed relative weakness in terms of SR
in information criteria for 1–5 profile solutions. The addi- scale (differences were significant when compared to Pro-
tion of sixth profile resulted in the increase of BIC value, file 1, adaptive and mild profiles and comparable to

Table 2. Fit Indices from Latent Profile Analysis Models

Model AIC BIC ABIC Entropy BLRT

1 Profile 19,387.724 19,650.562 19,378.338 — —

2 Profiles 18,031.452 18,428.764 18,017.262 0.973 1444.273**
3 Profiles 17,652.154 18,183.941 17,633.162 0.955 467.298**
4 Profiles 17,471.215 18,137.476 17,447.420 0.964 268.939**
5 Profiles 17,333.286 18,134.022 17,304.688 0.970 225.929**
6 Profiles 17,246.910 18,182.121 17,213.510 0.976 180.942**
7 Profiles 17,223.189 18,292.875 17,184.986 0.976 139.694**
8 Profiles 17,230.064 18,434.225 17,187.059 0.977 68.113

ABIC, sample-size adjusted BIC; AIC, Akaïke Information Criterion; BIC, Bayesian Information Criterion; BLRT, Bootstrap Likelihood Ratio Test.
Note: ** p < .001.

4 Uljarevic et al./Social subtypes in autism spectrum disorder INSAR

Table 3. Profile Comparisons Across SSDS Subscales
Profile 1 Profile 2 Profile 3 Profile 4 Profile 5
Mean (SD) Mean (SD) Mean (SD) Mean (SD) Mean (SD) Statistics Post hoc

SM 33.81 (6.27) 27.54 (6.69) 28.37 (6.21) 42.29 (8.04) 49.85 (7.58) F = 39.64, P < 0.001, Profile 5 > 1–4; Profile 4 > 1–3;
η2 = 0.51 Profile 1 > 2
SA 20.90 (3.35) 19.18 (4.77) 30.87 (6.22) 30.15 (3.37) 36.69 (2.29) F = 90.77, P < 0.001, Profile 5 > 1–4; Profile 4 > 1,2;
η2 = 0.70 Profiles 1 and 2 < 3
ESC 33.87 (3.46) 28.46 (5.24) 42.12 (5.03) 38.39 (4.19) 45.85 (3.36) F = 57.03, P < 0.001, Profile 5 > 1, 2, 4; Profile 4 > 1,2;
η2 = 0.60 Profile 3 > 1, 2
SR 19.93 (3.15) 11.90 (2.76) 13.12 (2.88) 20.98 (3.14) 24.15 (3.41) F = 75.52, P < 0.001, Profile 5 > 1–4; Profile 4 > 2, 3;
η2 = 0.66 Profile 1 > 2, 3
UA 9.61 (2.96) 9.73 (3.56) 9.87 (3.31) 10.43 (2.48) 12.31 (2.56) F = 3.58, P = 0.008, Profiles 4 > 1
η2 = 0.09

ESC, expressive social communication; SA, social affiliation; SM, social motivation; SR, social recognition; UA, unusual approach.

Table 4. Profile Comparisons in Terms of CA, FSIQ, SRS-2, and CBCL Scores
Profile 1 Profile 2 Profile 3 Profile 4 Profile 5
Mean (SD) Mean (SD) Mean (SD) Mean (SD) Mean (SD) Statistics Post hoc

CA 7.48 (4.08) 5.79 (3.6) 9.63 (5.42) 7.50 (3.89) 7.0 (3.44) F = 2.16, P = 0.076, NS
η2 = 0.054
FSIQa 69.59 (33.06) 50.84 (25.5) 67.0 (24.17) 82.70 (22.11) 82.92 (18.52) F = 4.67, P = 0.002, Profiles 4 and 5 > 2
η2 = 0.22
SRS-2 87.93 (17.51) 98.59 (16.1) 90.25 (12.02) 73.90 (16.59) 59.38 (14.78) F = 21.65, P < 0.001, Profile 5 < 1–4; Profile
SCI η2 = 0.37 4 < 1, 2
SRS-2 19.23 (5.97) 21.69 (7.2) 17.0 (5.71) 17.29 (6.74) 13.62 (6.34) F = 4.73, P = 0.001, Profiles 4 and 5 < 2
RRB η2 = 0.11
CBCL Int 65.59 (10.09) 65.54 (8.6) 60.50 (13.19) 60.21 (10.34) 55.90 (14.54) F = 3.23, P = 0.01, Profiles 1 and 2 > 4, 5
η2 = 0.09
CBCL 57.38 (9.86) 58.46 (8.7) 53.17 (7.55) 56.96 (12.11) 51.10 (9.43) F = 1.17, P = 0.336, NS
Ext η2 = 0.03

CBCL, Child Behavior Checklist; Ext, externalizing; FSIQ, Full-Scale IQ; Int, internalizing; RRB, restricted/repetitive behavior; SCI, social communica-
tion/interaction; SRS-2, Social Responsiveness Scale.
a
Data available for N = 76 participants.

severe profile) and strengths in terms of ESC scores comparable ESC scores with Cluster 4) but significantly
(higher than Profile 1 and severe profile and comparable lower SR score, was comparable to moderate, SR impaired
to mild and adaptive profiles), Profile 1 showed the oppo- profile, derived through the LPA. Finally, Cluster 2 which
site pattern (e.g., strengths in terms of SR but weakness in had higher SR scores (better performance) than Cluster
terms of ESC scale). In addition, Profile 1 (labeled as mod- 3 (and comparable SR scores with Cluster 4) but signifi-
erate but expressive communication and affiliation- cantly lower ESC scores, was comparable to moderate,
impaired) showed weakness in SA scale relative to Profile expressive communication, and affiliation-impaired pro-
3 (labeled as moderate but SR impaired). These two pro- file derived through the LPA.
files did not differ in terms of FSIQ nor SRS-2 scores; how-
ever, moderate, expressive communication, and
affiliation-impaired profile had higher internalizing prob- Discussion
lems scores. The distribution of the participants from two
subsamples did not differ across the profiles. There is pronounced heterogeneity in the social pheno-
Supplementary k-means cluster analysis indicated four type across the autism spectrum, with individuals pre-
clusters as an optimal solution. Clusters 1 and 4 showed senting with varying profiles of strengths and weakness
the most and least impairments across the SSDS scales, across different social processes. Our study utilized the
respectively, and were therefore comparable to severe SSDS [Phillips et al., 2019], a newly developed measure of
(Cluster 1) and mild and adaptive (Cluster 4) profiles social processes, to explore whether differences across
derived through the LPA. Cluster 3, which demonstrated particular social domains can be used to identify distinct
higher ESC scores when compared to Cluster 2 (and ASD subgroups.

INSAR Uljarevic et al./Social subtypes in autism spectrum disorder 5

 Five profiles were identified. Profiles showed distinct and depression [Bellini, 2004, 2006; Pickard, Rijsdijk,
patterns of peaks and troughs across different social com- Happe, & Mandy, 2017; Pickard, Happe, & Mandy, 2018;
ponents rather than simply reflecting a severity gradient. Spain, Sin, Linder, McMahon, & Happé, 2018]. These
Profiles 2, 4, and 5 were on opposite ends of the spectrum observations in relation to Profile 1 are preliminary and
in terms of the distribution of SM, SA, ESC, and SR scores, need to be further explored and tested by incorporating a
with Profiles 4 and 5 (labeled as mild and adaptive) show- dedicated measure of self-regulation and utilizing longitu-
ing the fewest and Profile 2 the greatest impairments dinal design. Finally, Profile 3 would particularly benefit
(labeled as severe). Profiles 1 and 3 showed distinctive from the theory of mind focused interventions.
patterns of strengths and weaknesses across SSDS sub- As noted in the introduction, the social typology pro-
scales. Profile 3, labeled as moderate, SR impaired profile posed by Wing and Gould [1979] and more recent work
showed strengths in terms of ESC scores but weaknesses by Lerner et al. [2017], Livingston et al. [2018], and Kang
in terms of SR. Profile 1, labeled as moderate, ESC, and et al. [2018] represent rare attempts aimed at identifying
affiliation-impaired profile showed strengths in terms of subgroups of individuals with ASD based on social phe-
SR, but weaknesses in terms of SA and ESC scales. Identi- notype. However, it is difficult to draw direct parallels
fied clusters did not differ in terms of age or gender, but between profiles derived here and subtypes identified in
provided explanatory value in terms of their relations to previous studies. Firstly, we utilized a data-driven cluster
cognitive ability, as well as the severity of ASD and inter- approach rather than relying on prescribed cutoff scores
nalizing symptoms. More specifically, severe profile had used by Wing and Gould and median scores used by Liv-
the highest SCI and repetitive behaviors symptom sever- ingston et al. [2018]. Secondly, rather than considering a
ity (as measured by the SRS-2) and highest CBCL internal- more comprehensive constellation of peaks and troughs
izing scores and lowest FSIQ scores, while adaptive and across a range of social dimensions, Wing’s typology only
mild profiles showed the opposite patterns. Importantly, focused on social interaction style and work by Lerner
moderate, ESC, and affiliation-impaired profile had com- et al. [2017] and Kang et al. [2018] focused on different
parable internalizing symptoms with severe profile and aspect of social processing as well as a range of communi-
significantly higher severity when compared to other cation items.
three profiles. Several study limitations are important to note. First,
As noted in the introduction, adoption of a fine- for participants who only took the survey online, it was
grained approach to identifying individual differences in not possible to administer the cognitive assessment nor
social phenotype is important for identifying groups with verify the diagnosis of the ASD beyond the parent report;
distinct patterns of strengths and weakness that might however, all participants included scored above the rec-
necessitate distinct treatment approaches. Although pro- ommended SRS-2 threshold. Although missing IQ data
files derived here are preliminary and warrant further rep- reduced the statistical power to some extent (for FSIQ
lication (please see a detailed overview of the limitations comparisons), all comparisons were supplemented with
below), they show a potential clinical utility. More specif- effect sizes for the majority of comparisons. Although
ically, Profile 1 was characterized by relative strengths in recruitment of participants through two different sources
SR, moderate SM but weaknesses in ESC and affiliation. can introduce a significant degree of variability, it is
This profile would particularly benefit from interventions important to highlight that the two subsamples did not
aimed at increasing social and communication skills. differ on any of the SSDS subscales, SRS-2, and CBCL
Importantly, this profile was also characterized by high scores nor in terms of the distribution of the participants
levels of internalizing problems. The presence of high across the clusters. Our sample included a relatively wide
levels of internalizing problems would warrant adequate age range and, although identified profiles did not differ
support and interventions and could be partially in terms of CA, it will be important for future research to
explained by poorer ESC skills, increased awareness/ explore social subtypes in more constrained developmen-
insight into own difficulties and potential impairments tal periods, as well as to utilize longitudinal designs to
in self-regulation. It has been suggested that ASD individ- explore subtype stability and potential mechanisms
uals who show a desire for social engagement (Profile mediating and moderating developmental continuities
1 has moderate levels of SM) might be at particularly high and discontinuities of the social subtypes. Although mul-
risk for developing anxiety and depression [Bellini, 2004; tidimensional, SSDS does not capture full spectrum of
Hedley, Uljarevic, Wilmot, Richdale, & Dissanayake, social processes. Therefore, in addition to the dimensions
2018; White, Oswald, Ollendick, & Scahill, 2009; Wing, assessed in this study, a range of other relevant domains
1992]. Despite preserved SM in some individuals with such as biological motion processing, eye gaze processing,
ASD, issues in self-regulation as well as limited social and and social hierarchy mapping warrant further attention
communication skills may lead to repeated social failures, in future studies. Importantly, SSDS is a new scale, and
leading to increased emotional pain and isolation, which, although it has demonstrated initial strong psychometric
in turn, would contribute to the emergence of anxiety properties, it will be important to explore and confirm

6 Uljarevic et al./Social subtypes in autism spectrum disorder INSAR

the structure of the scale and its invariance across a range in children with Autism: A preliminary investigation. Journal
of different relevant factors such as age, gender, and clini- of Autism and Developmental Disorders, 35(3), 295–303.
cal status in the future investigations. Finally, SSDS is a Bellini, S. (2004). Social skill deficits and anxiety in high func-
parent-report measure providing parental perspective on tioning adolescents with autism spectrum disorders. Focus on
Autism and Other Developmental Disabilities, 19(2), 78–86.
their child’s social processing domains, and it will be cru-
Bellini, S. (2006). The development of social anxiety in adoles-
cial for future research to further refine and expand sub-
cents with autism spectrum disorders. Focus on Autism and
typing work by including experimental and objective Other Developmental Disabilities, 21(3), 138–145.
indexes of different social domains. Bonde, E. (2000). Comorbidity and subgroups in childhood
Despite the limitations, our study provides a significant autism. European Child and Adolescent Psychiatry, 9(1), 7–10.
contribution by identifying subgroups of individuals with Castelloe, P., & Dawson, G. (1993). Subclassification of children
ASD who shared distinct social domain profiles. Impor- with autism and pervasive developmental disorder: A ques-
tantly, these subgroups reflected differential individual tionnaire based on Wing’s subgrouping scheme. Journal of
variability in terms of cognitive ability and severity Autism and Developmental Disorders, 23(2), 229–241.
of both ASD and internalizing symptoms. These findings Constantino, J. N., & Gruber, C. P. (2005). The social responsive-
represent an initial step toward reducing phenotypical ness scale. Los Angeles, CA: Western Psychological Services.
Constantino, J. N., & Gruber, C. P. (2012). Social responsive scale
heterogeneity in ASD, which ultimately may lead to more
manual (2nd edition, SRS-2). Los Angeles, CA: Western Psy-
personalized interventions.
chological Services.
Frazier T. W., Ratliff K. R., Gruber C., Zhang Y., Law P. A.,
Constantino J. N. (2014). Confirmatory factor analytic structure
Acknowledgments
and measurement invariance of quantitative autistic traits mea-
We would like to thank the study participants and their sured by the Social Responsiveness Scale-2. Autism, 18(1), 31–44.
parents. The research reported in this paper was Gadow, K. D., DeVincent, C., & Schneider, J. (2008). Predictors
supported by the Department of Psychiatry and Behav- of psychiatric symptoms in children with an autism spectrum
disorder. Journal of Autism and Developmental Disorders, 38,
ioral Science small research grant awarded to J.M.P.
1710–1720.
Green, M. F., Horan, W. P., & Lee, J. (2015). Social cognition in
Conflict of Interest schizophrenia. Nature Review Neuroscience, 16(10), 620–631.
Happe, F., Cook, J. L., & Bird, G. (2017). The structure of social
cognition: In(ter)dependence of sociocognitive processes.
Authors declare no conflict of interest.
Annual Review of Psychology, 68, 243–267.
Happe, F., & Frith, U. (2014). Annual research review: Towards a
developmental neuroscience of atypical social cognition.
References
Journal of Child Psychology and Psychiatry, 55(6), 553–577.
Abell, F., Happe, F., & Frith, U. (2000). Do triangles play tricks? Hedley, D., Uljarevic, M., Wilmot, M., Richdale, A., &
Attribution of mental states to animated shapes in normal Dissanayake, C. (2018). Understanding depression and
and abnormal development. Cognitive Development, 15(1), thoughts of self-harm in autism: A potential mechanism Involv-
1–16. ing loneliness. Research in Autism Spectrum Disorders, 46, 1–7.
Achenbach, T. M., & Rescorla, L. A. (2000). Manual for the Huber, L., Plötner, M., & Schmitz, J. (2019). Social competence
ASEBA preschool forms & profiles. Burlington, VT: University and psychopathology in early childhood: A systematic review.
of Vermont, Research Center for Children, Youth and European Child and Adolescent Psychiatry, 28(4), 443–459.
Families. Insel, T. R. (2010). The challenge of translation in
American Psychiatric Association. (2013). Diagnostic and statistical socialneuroscience: a review of oxytocin, vasopressin, and
manual of mental disorders (5th ed.). Arlington, VA: Author. affiliative behavior. Neuron, 65, 768–779.
Anderson, D. K., Oti, R. S., Lord, C., & Welch, K. (2009). Patterns Kang, E., Gadow, K. D., & Lerner, M. D. (2018). Atypical commu-
of growth in adaptive social abilities among children with nication characteristics, differential diagnosis, and the autism
autism spectrum disorders. Journal of Abnormal Child Psy- spectrum disorder phenotype in youth. Journal of Clinical
chology, 37(7), 1019–1034. Child & Adolescent Psychology, 49(2), 251–263.
Bauminger, N., Solomon, M., Aviezer, A., Heung, K., Gazit, L., Kennedy, D. P., & Adolphs, R. (2012). The social brain in psychi-
Brown, J., & Rogers, S. (2008). Friendship manifestations, atric and neurological disorders. Trends in Cognitive Sciences,
dyadic qualities of friendship and friendship perception in 16(11), 559–572.
high-functioning preadolescents with autism spectrum disor- Leekam, S. R. (2016). Social cognitive impairment and autism:
der. Journal of Abnormal Child Psychology, 36, 135–150. What are we trying to explain? Philosophical Transactions of
Bauer, D. J., & Curran, P. J. (2003). Distributional assumptions of the Royal Society of London Series B: Biological Sciences, 371
growth mixture models: implications for overextraction of (1686), 20150082–20150088.
latent trajectory classes. Psychological Methods, 8(3), 338–363. Lerner, M. D., De Los Reyes, A., Drabick, D. A. G.,
Beglinger, L., & Smith, T. (2005). Concurrent validity of social Gerber, A. H., & Gadow, K. D. (2017). Informant discrepancy
subtype and IQ after early intensive behavioral intervention defines discrete, clinically useful autism spectrum disorder

INSAR Uljarevic et al./Social subtypes in autism spectrum disorder 7

 subgroups. Journal of Child Psychology and Psychiatry, 58 victimisation and friendship quality in the development of
(7), 829–839. social anxiety in childhood. Journal of Anxiety Disorders,
Livingston, L. A., Colvert, E., Social Relationships Study Team, & 60, 1–10.
Bolton, P. (2018). Good social skills despite poor theory of Pickard, H., Rijsdijk, F., Happe, F., & Mandy, W. (2017). Are
mind: Exploring compensation in autism spectrum disorder. social and communication difficulties a risk factor for the
Journal of Child Psychology and Psychiatry, 60(1), 102–110. development of social anxiety? Journal of American Academy
Lord, C., Luyster, R. J., Gotham, K., & Guthrie, W. (2012). of Child and Adolescent Psychiatry, 56(4), 344–351.
Autism diagnostic observation schedule, 2nd edition (ADOS- Roid, G. H. (2003). Stanford Binet’s intelligence scales. Itasca, IL:
2) manual. Torrance, CA: Western Psychological Services. Riverside.
Lord, C., Rutter, M., & Le Couteur, A. (1994). Autism diagnostic Rutter, M., Le Couteur, A., & Lord, C. (2003). Autism diagnostic
interview-revised: A revised version of a diagnostic interview interview–revised (ADI–R) manual. Western Psychological
for caregivers of individuals with possible pervasive develop- Services: Los Angeles, CA.
mental disorders. Journal of Autism and Developmental Dis- Scheeren, A. M., Koot, H. M., & Begeer, S. (2012). Social interac-
orders, 24(5), 659–685. tion style of children and adolescents with high-functioning
Masyn, K. E., Henderson, C. E., & Greenbaum, P. E. (2010). autism spectrum disorder. Journal of Autism and Develop-
Exploring the latent structures of psychological constructs in mental Disorders, 42(10), 2046–2055.
social development using the dimensional-categorical spec- Spain, D., Sin, J., Linder, K. B., McMahon, J., & Happé, F.
trum. Social Development, 19(3), 470–493. (2018). Social anxiety in autism spectrum disorder: A sys-
Mullen, E. M. (1995). Mullen scales of early learning (AGS ed.). tematic review. Research in Autism Spectrum Disorders, 52,
Circle Pines, MN: American Guidance Service. 51–68.
Muthén, L. K., Muthén, B. O. (1998-2015). Mplus User’s Guide. Tueller, S., & Lubke, G. (2010). Evaluation of structural equation
Seventh Edition. Los Angeles, CA: Muthén & Muthén. mixture models: Parameter estimates and correct class assign-
National Advisory Mental Health Council Workgroup on Tasks ment. Structural Equation Modeling, 17(2), 165–192.
and Measures for Research Domain Criteria. (2016). Behav- Uljarevic, M., & Hamilton, A. (2013). Recognition of emotions in
ioral assessment methods for RDoC constructs. Bethesda, autism: A formal meta-analysis. Journal of Autism and Devel-
MD: National Institute of Mental Health. opmental Disorders, 43(7), 1517–1526.
National Institute of Mental Health (2012). Social Processes Vivanti, G., & Nuske, H. J. (2016). Autism, attachment, and
Workshop. social learning: Three challenges and a way forward. Behav-
Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding ioural Brain Research, 325, 251–259.
on the number of classes in latent class analysis and growth Vivanti, G., Prior, M., Williams, K., & Dissanayake, C. (2014).
mixture modeling: A monte carlo simulation study. Structural Predictors of outcomes in Autism Intervention: Why don’t
Equation Modeling,14(4), 535–569. we know more? Frontiers in Pediatrics, 2, 58.
Pallathra, A. A., Calkins, M. E., Parish-Morris, J., Maddox, B. B., White, S. W., Oswald, D., Ollendick, T., & Scahill, L. (2009). Anx-
Perez, L. S., Miller, J., et al. (2018). Defining behavioral compo- iety in children and adolescents with autism spectrum disor-
nents of social functioning in adults with autism spectrum dis- ders. Clinical Psychology Review, 29, 216–229.
order as targets for treatment. Autism Research, 11(3), 488–502. Wing, L. (1992). Manifestations of social problems in high func-
Phillips, J. M., Uljarevic, M., Schuck, R. K., Frazier, T. W., tioning autistic people. In E. Schopler & G. Mesibov (Eds.).
Schapp, S. S., Solomon, E. M., … Hardan, A. Y. (2019). Devel- High functioning individuals with autism, New York, NY: Ple-
opment of the Stanford Social Dimensions Scale (SSDS): Ini- num Press.
tial validation in autism spectrum disorder (ASD). Molecular Wing, L., & Gould, J. (1979). Severe impairments of social inter-
Autism, 10(1), 1–16. action and associated abnormalities in children: Epidemiol-
Pickard, H., Happe, F., & Mandy, W. (2018). Navigating the ogy and classification. Journal of Autism and Developmental
social world: The role of social competence, peer Disorders, 9(1), 11–29.

8 Uljarevic et al./Social subtypes in autism spectrum disorder INSAR