Unique dynamic profiles of social attention in
autistic females
Teresa Del Bianco,1 Luke Mason,1 Meng-Chuan Lai,2,3,4 Eva Loth,5 Julian Tillmann,5
Tony Charman,5 Hannah Hayward,5 Teresa Gleissl,5 Jan K. Buitelaar,6
Declan G.M. Murphy,5 Simon Baron-Cohen,3 Sven Bo¨lte,7 Mark H. Johnson,1,8
Emily J. H. Jones,1 and The EU-AIMS LEAP Group
1Centre for Brain and Cognitive Development, Birkbeck, University of London, London, UK; 2Centre for Addiction and
Mental Health and The Hospital for Sick Children, Department of Psychiatry, University of Toronto, Toronto, ON,
Canada; 3Autism Research Centre, Department of Psychiatry, University of Cambridge, Cambridge, UK;
4Department of Psychiatry, National Taiwan University Hospital and College of Medicine, Taipei, Taiwan; 5Institute of
Psychiatry, Psychology and Neuroscience, King’s College London, London, UK; 6Department of Cognitive
Neuroscience, Donders Institute for Brain, Cognition and Behaviour, Nijmegen, The Netherlands; 7Center of
Neurodevelopmental Disorders (KIND), Department of Women’s Health, Karolinska Institutet, Solna, Sweden;
8Department of Psychology, University of Cambridge, Cambridge, UK
Background: Social attention affords learning opportunities across development and may contribute to individual
differences in developmental trajectories, such as between male and female individuals, and in neurodevelopmental
conditions, such as autism. Methods: Using eye-tracking, we measured social attention in a large cohort of autistic
(n = 123) and nonautistic females (n = 107), and autistic (n = 330) and nonautistic males (n = 204), aged 6–30 years.
Using mixed Growth Curve Analysis, we modelled sex and diagnostic effects on the temporal dynamics of
proportional looking time to three types of social stimuli (lean-static, naturalistic-static, and naturalistic-dynamic)
and examined the link between individual differences and dimensional social and nonsocial autistic traits in autistic
females and males. Results: In the lean-static stimulus, average face-looking was higher in females than in males of
both autistic and nonautistic groups. Differences in the dynamic pattern of face-looking were seen in autistic vs.
nonautistic females, but not males, with face-looking peaking later in the trial in autistic females. In the naturalistic-
dynamic stimulus, average face-looking was higher in females than in males of both groups; changes in the dynamic
pattern of face looking were seen in autistic vs. nonautistic males, but not in females, with a steeper peak in
nonautistic males. Lower average face-looking was associated with higher observer-measured autistic characteristics
in autistic females, but not in males. Conclusions: Overall, we found stronger social attention in females to a similar
degree in both autistic and nonautistic groups. Nonetheless, the dynamic profiles of social attention differed in
different ways in autistic females and males compared to their nonautistic peers, and autistic traits predicted trends
of average face-looking in autistic females. These findings support the role of social attention in the emergence of sex-
related differences in autistic characteristics, suggesting an avenue to phenotypic stratification. Keywords: Autism;
social attention; eye-tracking; sex differences; male; female.
Introduction
Autism is a neurodevelopmental condition with a
prevalence rate of 1 in 89 children (Maenner et al.,
2020; Posada de la Paz, 2018) characterised by
difficulties with social communication and interac-
tion, restricted, repetitive behaviours and sensory
alterations, causing support needs (American Psy-
chiatric Association, 2013). Sex stratifies autism on
multiple levels, such as in prevalence rates (which
are 3 times more frequent in males than in females;
Loomes, Hull, & Mandy, 2017), core symptom
domains (Lai & Szatmari, 2020), and brain organi-
sation (e.g., cortical connectivity, Floris et al., 2021,
and morphometry, Hammill et al., 2021). Under-
standing the neurocognitive mechanisms that differ
between males and females may highlight mecha-
nisms contributing to differing symptoms and
presentation and inform clinical support (Con-
stantino, Charman, & Jones, 2021).
Social attention − dynamic engagement with other
people − has been a leading candidate neurocognitive
marker of autistic neurodevelopment. Several studies
have found that social attention is decreased in
autistic people (Frazier et al., 2017) and altered prior
to formal clinical diagnosis (Bedford et al., 2016;
Chawarska, Macari, Powell, DiNicola, & Shic, 2016).
However, interpretation and generalisation have been
partly limited by low female representation (Frazier
et al., 2017). Recent studies that have included a
higher number of female participants found that
clinically diagnosed autistic females showed similar
average looking time to faces (Harrop et al., 2019), and
social vs. nonsocial preferences (Harrop et al., 2018,
2020),asnonautistic females,unlikethereductionsin
social attention shown in autistic vs. nonautistic
males. In these studies, a similar pattern of sex
differences in social attention was observed in autism
and neurotypical development, and diagnostic group
differences were smaller in females than in males.
Conflict of interest statement: See Acknowledgements for full
disclosures.
The EU-AIMS LEAP Group members are present in Appendix.
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for Child and
Adolescent Mental Health.
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any
medium, provided the original work is properly cited.
Journal of Child Psychology and Psychiatry **:* (2022), pp 1–13 doi:10.1111/jcpp.13630
PFI_12mmX178mm.pdf + eps format
These observations suggest that sex differences in
social attentionmay not be explained bymodels such
as the ExtremeMale BrainHypothesis (Baron-Cohen,
2002) and the Additive Inherited Liability model
(Constantino et al., 2021), which predict that autistic
people may not show the same pattern of sex differ-
ences as neurotypical people, and that diagnostic
group differences may be bigger in females compared
to males. Other models suggest that social attention
may act as a protective (Chawarska et al., 2016) or
moderating factor (Johnson, Charman, Pickles, &
Jones, 2021) because it may maintain engagement
with others and provide opportunities for learning.
However, sexdifferences in the relationbetweensocial
attention and dimensional variation in autistic symp-
toms have not been explored.
Sex differences in social attention might be most
sensitively tested by including not only averaged
measures of social attention but using methods that
capture the temporal dynamics of social interest.
Indeed, recent studies (Del Bianco et al., 2021;
Hedger & Chakrabarti, 2021) have shown that
analytical methods that treat social attention as a
time series are sensitive to age-related and contex-
tual flexibility, aiding the elucidation of the under-
lying neurocognitive processes. In the present work,
we examined the modulation of social attention by
sex and diagnosis and its relation to dimensional
symptomatology in a large sample of the multi-site
Longitudinal European Autism Project, LEAP (Loth
et al., 2017), which included a large sample of female
participants of a wide age range. We predicted that,
consistent with the model described above, social
attention would be greater in females than males in
both autistic and nonautistic groups, and that the
pattern of diagnostic group differences would differ
in males and females, across stimuli that vary in
their complexity (Harrop et al., 2019) and ecological
validity (Chevallier et al., 2015), that have been,
respectively, found to better elicit sex and diagnostic
group differences. Furthermore, we predicted that
fewer social communication traits in autistic females
would be associated with more significant deviance
from autistic males, compared to nonautistic
females, formalised as a pattern of sex differences
exceeding diagnostic group differences.
Methods
Participants
The data come from 764 participants (453 autistic and 311
nonautistic), in four countries (the United Kingdom, Germany,
The Netherlands, Italy). See Table 1 for demographic charac-
teristics and comparisons by diagnostic group and sex.
Ethical considerations
The study was carried out upon approval of national and local
ethics review boards at each study site. Participants or their
parents signed a written consent before entering the study.
Clinical variables
For dimensional symptoms, we used the standardised T-score
of the parent-reported Social Responsiveness Scale 2, SRS-2
(Constantino & Gruber, 2005), the parent-reported Repetitive
Behaviour Scale-Revised, RBS-R (Bourreau, Roux, Gomot,
Bonnet-Brilhault, & Barthe´le´my, 2009), and Social Affect and
Restricted Repetitive Behavior Calibrated Severity Scores (SA-
CSS and RRB-CSS, respectively) of the Autism Diagnostic
Observation Schedule, ADOS (Esler et al., 2015; Lord et al.,
2000) Module 2/3/4 (see Charman et al., 2017).
Eye-trackers and Software
Sites used a Tobii T120 (3, 383 participants) or TX300 (3,381
participants) eye tracker (Tobii AB, Sweden), at a maximum
samplingrate (120and300 Hz).Thedifferenceinscreensize (17”
and23”respectively)wasuniformedbypresenting thestimulion
a 17” virtual screen with a black border on the TX300. The
freedom of head movement was similar at a standard distance
from the screen, and slightly bigger for the T120 (44*22 cm)
compared to the TX300 (37*17 cm). Stimuli were presented on
Apple Macbook Pro (Apple Inc., USA), with TaskEngine (sites.
google.com/site/taskenginedoc/). Raw gaze was recorded and
processed with Tobii Gaze Analytics SDK 3.0.
Stimuli
Three sets of stimuli were presented on a virtual screen of
33*18 visual degrees of angle on 1,280 × 1,024 monitors:

 Face pop-out (FPO): 8 static arrays of one face
(balanced by gender), a scrambled face, a car, a
bird and a mobile phone for 10 seconds (Gliga,
Elsabbagh, Andravizou, & Johnson, 2009; Gui
et al., 2020), without audio.

 Static Scenes (SS): 6 naturalistic photographs of
adults, children (nine females and four males in
foreground) and everyday objects, for 20 s (Del
Bianco et al., 2021), without audio.

 Dynamic Video (DV): a 40-seconds extract from
“50 People, One Question Brooklin” (http://
fiftypeopleonequestion.com/), presenting street
interviews with 14 people (balanced by gender),
with relaxing piano music in the background.
Procedure
The participant sat centrally at 60 cm from the screen. Five-
point calibration was performed up to 3 times before the
experimenter could skip the presentation. The presentation −
intermixed with other stimuli − proceeded automatically when
the participants fixated the interstimulus image, in 4 blocks of
7 min (Face Pop-Out and Static Scenes: 1st block; Dynamic
Video: last block), all including post-hoc calibration checks, for
an overall duration of 28 min.
Pre-processing
AOIs were manually drawn on the faces (Figure 1). Each
sample was scored according to whether the gaze coordinates
fell within the face and aggregated in time bins of half a second,
obtaining the Proportional Looking Time (PLT). For the
Dynamic Video, we defined 3-second segments from the onset
of a face, excluding scenes <3 s, and cutting exceeding time
>3 s. We excluded time bins with missing data (i.e., eyes not
detected/out of the screen) >75%.
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
2 Teresa Del Bianco et al. J Child Psychol Psychiatr 2022; 0(0): 1–13
Table 1 Essential demographic characteristics of the sample, split by diagnostic group and sex. Effect sizes of between sex
comparisons (Phi Coefficient for categorical variables, Cohen’s D for continuous variables) are provided for each group separately
Parameter
Females Phi Coefficients
(Categorical) /
Cohen’s D
(Continuous)
Males Phi Coefficients
(Categorical) /
Cohen’s D
(Continuous)Autistic Nonautistic Autistic Nonautistic
N 123 107 – 330 204 0.006
FSIQ < 75 26 (21%) 15 (14%) 0.008 50 (15%) 20 (9%) 0.005
Age Range in Year 6.08 ~ 30.28 6.89 ~ 30.78 – 6.13 ~ 30.60 6.24 ~ 30.98 –
Age Mean in Year (SD) 16.77 (6.36) 17.04 (5.92) 0.04 16.69 (5.62) 17.23 (5.96) 0.09
FSIQ Mean (SD) 95.88 (20.37) 104.39 (19.94) 0.42 97.17 (19.85) 103.48 (18.3) 0.33
SRS T-Score Mean (SD) 73.73 (12.13) 48.07 (8.94) 2.38 71.52 (11.49) 48.35 (9.78) 2.13
RBS-R Mean (SD) 15.92 (13.53) 2.34 (4.96) 1.30 17.01 (13.92) 3.1 (9.95) 1.11
ADOS SA-CSS Mean (SD) 5.45 (2.53) – – 6.42 (2.64) – –
ADOS RRB-CSS Mean (SD) 4.1 (2.58) – – 5 (2.81) – –
Parameter
NonAutistic Phi Coefficients
(Categorical) /
Cohen’s D
(Continuous)
Autistic Phi Coefficients
(Categorical) /
Cohen’s D
(Continuous)Female Male Female Male
N 107 204 – 123 330 0.006
FSIQ < 75 15 20 0.003 26 50 0.005
Age Range in Years 6.89 ~ 30.78 6.24 ~ 30.98 – 6.08 ~ 30.28 6.13 ~ 30.60 –
Age Mean in Years (SD) 17.04 (5.92) 17.23 (5.96) 0.03 16.77 (6.36) 16.69 (5.62) 0.01
FSIQ Mean (SD) 104.39 (19.94) 103.48 (18.3) 0.05 95.88 (20.37) 97.17 (19.85) 0.06
SRS T-Score Mean (SD) 48.07 (8.94) 48.35 (9.78) 0.03 73.73 (12.13) 71.52 (11.49) 0.19
RBS-R Mean (SD) 2.34 (4.96) 3.1 (9.95) 0.09 15.92 (13.53) 17.01 (13.92) 0.08
ADOS SA-CSS Mean (SD) – – – 5.45 (2.53) 6.42 (2.64) 0.37
ADOS RRB-CSS Mean (SD) – – – 4.1 (2.58) 5 (2.81) 0.33
ADOS RRB-CSS, Autism Diagnostic Observation Schedule Repetitive Restricted Behaviour Calibrated Severity Score; ADOS SA-
CSS, Autism Diagnostic Observation Schedule Social Affect Calibrated Severity Score; FSIQ, Full Scale Intelligence Quotient; N,
number; RBS-R, Repetitive Behaviour Scale Revised; SRS, Social Responsiveness Scale.
Figure 1 (A) 2 sample stimuli from the eight gendered-balanced slides of the Face Pop-Out. (B) two samples of six photographs
portraying nine females and four males of the Static Scenes (faces are pixelated for privacy reasons). (C) screenshot of three of 14
individual people (gender-balanced) interviewed on the street in the Dynamic Video
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
doi:10.1111/jcpp.13630 Dynamic Profiles of Social Attention in Females 3
Statistical analysis
Analytic models. We applied Growth Curve Analysis,
GCA (Mirman, Dixon, & Magnuson, 2008) in a series of
successive models to assess sex and diagnostic group differ-
ences (Lai et al., 2013).
We calculated the orthogonal polynomials of the time of
presentation of the stimulus, up to degree 3 of the Face Pop-
Out and the Static Scenes (3rd degree polynomials correspond
to 3 changes of focus, or Slope, Quadratic and Cubic Compo-
nents), and up to degree 2 for the Dynamic Video (2nd degree
polynomials correspond to 2 changes of focus, or Slope,
Quadratic Components). Additionally, we included random
intercepts and slopes by the participant, and random inter-
cepts by trial/scene (Equation 1).
yi,s ¼ β0 þ β1Pnxi þ β0i þ β0s þ β1iPnxi þ ϵi þ ϵs (1)
where yi,s = PLT of the -ith participant (i), for the -sth
stimulus (s), β0 = fixed intercept, β1 = fixed slope,
Pnxi = polynomial function, β0i , β0s = random inter-
cepts, β1iPnxi ¼ random polynomials, ϵi = overall
variability, ϵs = stimulus variability.
With reference to Lai et al. (2013), we fitted two sex and two
diagnostic group differences models through a process of
stepwise addition tested with Likelihood Ratio Tests: (a)
diagnosis/sex, (b) orthogonal polynomials (interaction), (c)
age (covariate/interaction), (d) proportion of missing data
(covariate; Appendix S1 Tables FPO S4, S5, S8, S9; SS S12,
S13, S16, S17; DV S20, S21, S24, S25), (e) FSIQ (covariate;
Appendix S1 section S1.4). For significant effects, we report the
Coefficients (Coef.; referenced to the female/nonautistic group,
corresponding to unstandardised effect sizes), Bootstrapped
95% Confidence Intervals (between squared brackets) and
Standard Errors (SE) of the model with the best fit.
Dimensional variation
We extracted random effects, i.e., individual coefficients, from
the sex differences model in the autistic group. Differently from
the Coefficients pertaining to the analytical models above,
random effects quantify the effect of sex on individual partic-
ipants. We used them as dependent variables in multiple linear
regressions with SRS-2, RBS-R, ADOS SA-CSS and ADOS
RRB-CSS as predictors, in interaction with sex, with stimulus
and age as covariates (Equation 2). T-adjustment based on
Monte Carlo approximations was applied to pairwise compar-
isons.
y ¼ β0 þ β1xi j i þ j i þ k1 þ k2 þ ϵ (2)
where y = random effect, β0 = fixed intercept, β1-
= fixed slope, xi = clinical variable,j i = sex, k1-
= age,k2 = stimulus, ϵ = variability.
Second, we subtracted the random effects of female partic-
ipants extracted from the sex differences model from the
corresponding random effect extracted from the diagnostic
group model, thus obtaining a different score, illustrating how
much bigger/smaller the effect of sex is for each female
compared to the effect of being autistic. We used it as
dependent variables in a multiple linear regression with ADOS
SA-CSS as a predictor, with stimulus and age as covariates
(Equation 3).
y ¼ β0 þ β1xi þ k1 þ k2 þ ϵ (3)
where y = difference score, β0 = fixed intercept, β1-
= fixed slope, xi = ADOS SA-CSS, k1 = age,k2-
= stimulus, ϵ = variability.
Results
87% of the sample provided usable data (for
autistic and nonautistic groups, respectively, final
sample sizes 388 and 275 for FPO; 417 and 245 for
SS; 388 and 271 for DV). 638 provided valid data
for all 3 stimuli (104 autistic females, 89 nonautis-
tic females, 268 autistic males, 177 nonautistic
males; missing due to tasks not presented 10.23%;
lack of acquisition 90.77%). Based on Monte Carlo
Simulations, with the current sample, there is an
estimated power of 96.38% to detect previously
observed effects (Del Bianco et al., 2021; Harrop
et al., 2020); power estimation only starts to drop
below 70–80% if assumed effects are 80% smaller
than previously found (Confidence Interval = 55.85
– 75.18% for sex differences in the autistic sub-
sample; Confidence Interval = 47.71 – 67.80% for
diagnostic group differences in the female subsam-
ple). Calibration accuracy (i.e., Euclidean distance
from the centroid of all gaze samples to the central
interstimulus image) did not vary by group (Coef. =
0.001, SE = 0.002, p-value = .44), indicating that
there were no significant changes of data quality
over the time of the session (see Appendix S1
Table S1).
The % of missing data differed between stimuli,
with a higher percentage in SS (23.42%, SD = 7.56),
and lower in FPO (17.80%, SD = 7.79) and DV (10%,
SD = 14.41), which was accounted for by inclusion
as a covariate for each model. Differences between
groups reached small effect sizes (<0.4; see
Appendix S1, Tables S2 and S3).
After controlling for missing data, FSIQ did not
contribute to any of the model fit and was thus
excluded (see Appendix S1 section S1.4).
The overall pattern of results is summarised in
Table 2. In the following paragraphs, we report the
significant coefficients/effects for each model (for the
complete list of coefficients, see Appendix S1
Table FPO S6, S7, S10, S11; SS S14, S15, S18,
S19; DV S22, S23, S26, S27).
Face Pop-Out
Sex differences models.
Average profile: In the nonautistic group, PLT
averaged to 0.20 [0.16 ~ 0.24] (SE = 0.02, Table S8)
and was significantly lower in males compared to
females (Coef. = −0.03 [−0.05~−0.01], SE = 0.01;
see Figure 2, right panel). In the autistic group
model, PLT averaged to 0.24 [0.20 ~ 0.27] (SE =
0.01, Table S9), and was significantly lower in males
than females (Coef. = −0.02 [−0.04~−0.006], SE =
0.009; see Figure 2, right panel).
Dynamic profile: 3rd degree polynomials signifi-
cantly described the change in time of PLT in both
groups, configuring a transient decline (Quadratic,
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
4 Teresa Del Bianco et al. J Child Psychol Psychiatr 2022; 0(0): 1–13
T
a
b
le
2
D
e
s
c
ri
p
ti
v
e
s
u
m
m
a
ry
o
f
a
ll
re
s
u
lt
s
S
ti
m
u
lu
s
A
v
e
ra
g
e
P
ro
fi
le
(I
n
te
rc
e
p
t)
D
y
n
a
m
ic
P
ro
fi
le
(P
o
ly
n
o
m
ia
ls
)
S
e
x
d
if
fe
re
n
c
e
s
m
o
d
e
ls
D
ia
g
n
o
s
ti
c
g
ro
u
p
d
if
fe
re
n
c
e
s
m
o
d
e
ls
S
e
x
d
if
fe
re
n
c
e
s
m
o
d
e
ls
D
ia
g
n
o
s
ti
c
g
ro
u
p
d
if
fe
re
n
c
e
s
m
o
d
e
ls
F
a
c
e
P
o
p
-O
u
t
M
a
le
s
<
F
e
m
a
le
s
(N
o
n
a
u
ti
s
ti
c
:
C
o
e
f.
=
−
0
.0
3
,
p
-v
a
lu
e
=
.0
1
;
a
u
ti
s
ti
c
:
C
o
e
f.
=
−
0
.0
2
,
p
-
v
a
lu
e
=
.0
1
)
A
u
ti
s
ti
c
fe
m
a
le
s
>
n
o
n
a
u
ti
s
ti
c
fe
m
a
le
s
(C
o
e
f
=
0
.0
8
,
p
-v
a
lu
e
=
.0
3
)
A
g
e
-d
e
p
e
n
d
e
n
t
d
e
c
re
a
s
e
in
a
u
ti
s
ti
c
fe
m
a
le
s
(C
o
e
f.
=
−
0
.0
0
5
,
p
-
v
a
lu
e
=
.0
1
)
L
a
te
p
e
a
k
in
a
u
ti
s
ti
c
fe
m
a
le
s
(C
o
e
f.
=
−
0
.1
5
,
p
-v
a
lu
e
=
.0
0
8
)
A
g
e
-d
e
p
e
n
d
e
n
t
fl
a
tt
e
n
in
g
w
it
h
a
g
e
in
fe
m
a
le
s
(C
o
e
f.
=
0
.0
1
,
p
-
v
a
lu
e
<
.0
0
1
)
M
o
re
s
o
c
ia
l
a
tt
e
n
ti
o
n
in
fe
m
a
le
s
v
e
rs
u
s
m
a
le
s
;
s
u
b
te
n
d
e
d
b
y
d
if
fe
re
n
t
p
ro
c
e
s
s
e
s
in
a
u
ti
s
ti
c
a
n
d
n
o
n
a
u
ti
s
ti
c
fe
m
a
le
s
,
w
it
h
a
g
e
-d
e
p
e
n
d
e
n
t
c
o
n
v
e
rg
e
n
c
e
a
t
th
e
a
v
e
ra
g
e
le
v
e
l,
b
u
t
th
e
d
iv
e
rg
e
n
c
e
o
f
th
e
d
y
n
a
m
ic
p
ro
fi
le
S
ta
ti
c
S
c
e
n
e
s
M
a
le
s
<
F
e
m
a
le
s
(N
o
n
a
u
ti
s
ti
c
:
C
o
e
f.
=
−
0
.0
2
,
p
-v
a
lu
e
=
.0
5
;
a
u
ti
s
ti
c
:
C
o
e
f.
=
−
0
.0
2
,
p
-
v
a
lu
e
=
.0
7
)
A
g
e
-d
e
p
e
n
d
e
n
t
in
c
re
a
s
e
b
u
t
n
o
t
s
ta
ti
s
ti
c
a
ll
y
d
if
fe
re
n
t
b
e
tw
e
e
n
g
ro
u
p
s
(M
a
le
s
:
C
o
e
f.
=
−
0
.0
0
3
,
p
-
v
a
lu
e
=
.0
7
;
F
e
m
a
le
s
:
−
0
.0
0
5
,
p
-
v
a
lu
e
=
.0
6
)
N
o
ro
b
u
s
t
s
e
x
d
if
fe
re
n
c
e
s
D
y
n
a
m
ic
V
id
e
o
M
a
le
s
<
F
e
m
a
le
s
(N
o
n
a
u
ti
s
ti
c
:
C
o
e
f.
=
−
0
.0
6
,
p
-v
a
lu
e
<
.0
0
1
;
a
u
ti
s
ti
c
:
C
o
e
f.
=
−
0
.0
4
,
p
-
v
a
lu
e
=
.0
2
)
N
o
n
a
u
ti
s
ti
c
>
a
u
ti
s
ti
c
(M
a
le
s
:
C
o
e
f.
=
−
0
.0
7
,
p
-v
a
lu
e
<
.0
0
1
;
F
e
m
a
le
s
:
−
0
.0
9
,
p
-v
a
lu
e
<
.0
0
1
)
L
o
w
e
r
m
a
x
im
u
m
a
n
d
fa
rt
h
e
r
ro
o
ts
in
a
u
ti
s
ti
c
m
a
le
s
th
a
n
fe
m
a
le
s
(C
o
e
f.
=
0
.0
5
,
p
-v
a
lu
e
=
.0
0
2
)
L
o
w
e
r
n
e
t
in
c
re
a
s
e
(C
o
e
f.
=
−
0
.0
5
,
p
-v
a
lu
e
=
.0
0
7
)
a
n
d
m
a
x
im
u
m
(C
o
e
f.
=
0
.0
5
,
p
-v
a
lu
e
<
.0
0
1
)
in
a
u
ti
s
ti
c
m
a
le
s
th
a
n
in
n
o
n
a
u
ti
s
ti
c
m
a
le
s
N
o
n
s
ig
n
ifi
c
a
n
t
lo
w
e
r
m
a
x
im
u
m
a
n
d
fa
rt
h
e
r
ro
o
ts
in
a
u
ti
s
ti
c
fe
m
a
le
s
th
a
n
n
o
n
a
u
ti
s
ti
c
fe
m
a
le
s
(C
o
e
f.
=
0
.0
4
,
p
-v
a
lu
e
=
.0
6
)
M
o
re
s
o
c
ia
l
a
tt
e
n
ti
o
n
in
fe
m
a
le
s
v
e
rs
u
s
m
a
le
s
;
s
u
b
te
n
d
e
d
b
y
a
lo
w
e
r
m
a
x
im
u
m
,
a
n
d
lo
w
e
r
n
e
t
in
c
re
a
s
e
,
in
a
u
ti
s
ti
c
m
a
le
s
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
doi:10.1111/jcpp.13630 Dynamic Profiles of Social Attention in Females 5
nonautistic: Coef. = 0.14 [0.12 ~ 0.17], SE = 0.01;
autistic: Coef. = 0.09 [0.07 ~ 0.11], SE = 0.01) fol-
lowed by a rise (Cubic, nonautistic: Coef. = −0.10
[−0.11~−0.08], SE = 0.01; autistic: Coef. = −0.07
[−0.08~−0.06], SE = 0.01), with a net decrease by
the end of the trial (Slope, nonautistic:
Coef. = −0.14 [−0.17~−0.11], SE = 0.01; autistic:
Coef. = −0.14 [−0.16~−0.11], SE = 0.01). This pat-
tern was not influenced by sex: males and females
looked away from the face similarly during the
course of the trial.
Diagnostic group differences models.
Average profile: In the males’ model, the two diag-
nostic groups scored similarly on average
(Table S12). In the females’ model, PLT was higher
on average in the autistic compared to the
nonautistic group (Coef. = 0.19 [0.12 ~ 0.24], SE =
0.03; Group Coef. = 0.08 [0.01 ~ 0.16], SE = 0.04;
Table S13) but tended to decrease more with age in
the autistic compared to the nonautistic group (Age
Coef. = 0.08 [0.01 ~ 0.16], SE = 0.04; Group*Age
Coef. = −0.005 [−0.01~−0.002], SE = 0.002; see Fig-
ure 2, left bottom panel).
Dynamic profile: The males did not show a definite
trend of PLT throughout the trial (Slope and Cubic
nonsignificant); however, a decline followed by an
asymmetrical peak emerged with age (Quadratic*Age
Coef. = 0.01 [0.005 ~ 0.02], SE = 0.003; Cubic
Coeff = −0.002 [−0.005~−0.001], SE = 0.002); this
pattern did not differ by diagnostic group. Autistic
females showed a more pronounced asymmetrical,
late peak compared to nonautistic females (Cubic
Figure 2 Bar plot of the intercepts – right side; representing the estimated average PLT by group – and comparisons of the raw (top left)
and model estimated dynamics (bottom left) for autistic and nonautistic females in the Face Pop-Out. We illustrated the comparison
between autistic and nonautistic females because the estimates significantly differ from each other. In the left plot, the colour codes of
the curves represent age, while the black lines represent the average estimate for each age group
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
6 Teresa Del Bianco et al. J Child Psychol Psychiatr 2022; 0(0): 1–13
Coef. = −0.15 [−0.25 ~ 0.02], SE = 0.06), but this
late peak flattened with age (Cubic*Group*Age
Coef. = 0.01 [0.005 ~ 0.02], SE = 0.003), while it
increased in nonautistic females (Cubic*Age
Coef. = −0.01 [−0.01~−0.002], SE = 0.002), such
that changes were in opposite directions in the two
diagnostic groups over-development (see Figure 2,
left bottom panel).
Static scenes
Sex differences models.
Average profile: In the nonautistic group, PLT
averaged to 0.35 [0.22 ~ 0.46] (SE = 0.06;
Table S16) and was marginally lower in males
(Coef. = −0.02 [−0.04~−0.004], SE = 0.01). In the
autistic group, PLT averaged to 0.39 [0.26 ~ 0.51]
(SE = 0.05; Table S17) and was only marginally
lower in males (Coef. = −0.02 [−0.04 ~ 0.003], SE =
0.01).
Dynamic profile: In the nonautistic group, 3rd
degree polynomials significantly described the
change in time of PLT, configuring a transient decline
(Quadratic Coef. = 0.14 [0.11 ~ 0.17], SE = 0.02) fol-
lowed by a rise (Cubic Coef. = −0.05 [−0.08~−0.03],
SE = 0.01), with a net decrease by the end of the trial
(Slope Coef. = −0.14 [0.18~−0.09], SE = 0.02). The
2nd degree polynomials significantly described the
change in time of PLT in the autistic group, config-
uring a progressive decline (Quadratic Coef. = 0.12
[0.10 ~ 0.15], SE = 0.01), with a net decrease by the
end of the trial (Slope Coef. = −0.19 [−0.22~−0.15],
SE = 0.01). There was no significant effect of sex on
the dynamic profiles.
Diagnostic group differences models.
Average profile: In males, the average PLT was
0.32 [0.18 ~ 0.44] (SE = 0.06; Table S20), and
increased with age (Coef. = 0.004 [0.002 ~ 0.01],
SE = 0.001), but marginally less in the autistic
group (Group*Age Coef. = −0.003 [−0.01 ~ 0.0001],
SE = 0.002). In females, the average PLT was 0.35
[0.22 ~ 0.48] (SE = 0.07; Table S21), and increased
with age (Coef. = 0.004 [0.001 ~ 0.01], SE = 0.002)
though this was marginally less in the autistic group
(Coef. = −0.005 [−0.01 ~ 0.001], SE = 0.002).
Dynamic profiles: In males, PLT marginally
decreased by the end of the trial (Slope Coef. = −0.12
[−0.25~−0.02], SE = 0.07), with an asymmetrical U-
shape (Quadratic Coef. = 0.19 [0.07 ~ 0.30], SE =
0.06; Cubic Coef. = 0.10 [0.02 ~ 19], SE = 0.04),
and a peak emerging by the end of the trial with
age (Cubic*Age, Coef. = −0.01 [−0.01~−0.004], SE =
0.002). In females, we observe a net decrease by the
end of the trial (Slope Coef. = −0.34 [−0.54~−0.10],
SE = 0.11), approximately linear. No diagnostic
group differences were significant.
Dynamic video
Sex differences models.
Average profile: In the nonautistic group, PLT
averaged to 0.75 [0.68 ~ 0.82] (SE = 0.04;
Table S24), lower in males than females
(Coef. = −0.06 [−0.09~−0.03], SE = 0.01). In the
autistic group, PLT averaged to 0.70 (SE = 0.04
[0.63 ~ 0.78]; Table S25), lower in males than in
females (Coef. = −0.04 [−0.07~−0.004], SE = 0.02;
see Figure 3, left panel).
Dynamic profile: In the nonautistic and autistic
groups, the 2nd degree polynomials significantly
described the changes of PLT, configuring a bell
shape (nonautistic: Quadratic Coef. = −0.30
[−0.32~−0.28], SE = 0.01; autistic: Coef. = −0.29
[−0.32~−0.26], SE = 0.01), with a net increase by
the end of the scene (nonautistic: Slope Coef. = 0.25
[0.23 ~ 0.27], SE = 0.01; autistic: Coef. = 0.22
[0.18 ~ 0.25], SE = 0.01). However, the bell shape
flattened more in autistic males than in females
(Quadratic*Sex, Coef. = 0.05 [0.02–0.09], SE =
0.02).
Diagnostic group differences models.
Average profile: In males, the average PLT score
was 0.70 [0.62 ~ 0.77] (SE = 0.04; Table S28) and
was lower in autistic compared to nonautistic males
(Coef. = −0.07 [−0.09~−0.04], SE = 0.01). In
females, the average PLT was 0.82 [0.77 ~ 0.89]
(SE = 0.03; Table S29) and was lower in autistic
compared to nonautistic females (Coef. = −0.09
[−0.13~−0.06], SE = 0.02; see Figure 3, left panel).
Dynamic profile: In males, we saw a net increase by
the end of the scene (Slope Coef. = 0.25 [0.22 ~ 0.27],
SE = 0.01), being lower in the autistic versus
nonautistic males (Coef. = −0.05 [−0.09~−0.02],
SE = 0.02), and a bell shape (Quadratic
Coef. = −0.29 [−0.31~−0.27], SE = 0.02), flatter in
the autistic versus nonautistic males (Coef. = 0.05
[0.03 ~ 0.08], SE = 0.01; see Figure 3 right bottom
panel). In females, we saw a net increase by the end of
the scene (Slope Coef. = 0.26 [0.22 ~ 0.31], SE =
0.02), and a bell shape (Quadratic Coef. = −0.33
[−0.37~−0.30], SE = 0.02), that was marginally flat-
ter in autistic females versus nonautistic females
(Coef. = 0.04 [−0.003 ~ 0.08], SE = 0.02).
Dimensional variation
In neither sex did SRS-2 and RBS-R relate to
individual differences in Social Attention (see
Appendix S1 Table S28).
In females, both the ADOS SA-CSS (Coef. = −0.18,
95% CI = −0.28~−0.07) and the ADOS RRB-CSS
were related to the Intercept (Coef. = −0.22, 95%
CI = −0.32~−0.11). In males, the relationships
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
doi:10.1111/jcpp.13630 Dynamic Profiles of Social Attention in Females 7
between Intercept and ADOS SA-CSS (Coef. = 0.02,
95% CI = −0.05 ~ 0.08) and RRB-CSS (Coef. = 0.02,
95% CI = −0.04 ~ 0.08) were not significant. The
Quadratic Component was related to both ADOS
domain scores in both females (ADOS SA-CSS
Coef. = 0.17, 95% CI = 0.07 ~ 0.27; ADOS RRB-
CSS Coef. = 0.11, 95% CI = 0.01 ~ 0.22) and males
(ADOS SA-CSS Coef. = −0.11, 95%
CI = −0.17~−0.04; ADOS RRB-CSS Coef. = −0.08,
95% CI = −0.14 ~−0.01). In both cases, the contrasts
between males and females were significant (ADOS
SA-CSS, Intercept, Contr. = 0.19, SE = 0.06, p-
value = .01; Quadratic, Contr. = −0.28, SE = 0.06,
p-value < .01; ADOS RRB-CSS, Intercept, Contr. =
0.24, SE = 0.06, p-value < .01; Quadratic,
Contr. = −0.19, SE = 0.06, p-value = .01).
Regarding the difference score, for both intercept
and quadratic, ADOS SA-CSS had an inverse rela-
tion: when the influence of sex differences on social
attention was relatively bigger than the influence of
diagnostic differences, symptoms were low; i.e.,
females with fewer autistic symptoms tended to be
more similar to neurotypical females than they were
to autistic males for both linear (Coef. = −0.001,
95% CI = −0.002~−0.0008; see the full list of coef-
ficients Appendix S1 Table S29) and quadratic
aspects of social attention (Coef. = 0.0006, 95%
CI = 0.00004~−0.001). In other words, when face-
looking was more influenced by sex than a diagnosis
in autistic females, social-communication symptoms
were lower.
Discussion
In this large sample of autistic and nonautistic
individuals aged 6–30 years, we found that (a)
females showed more social attention than males,
in both autistic and nonautistic groups; (b) social
attention in autistic females significantly differed
from nonautistic females, but differently depending
on stimulus (see below); and (c) in females, more
observed autistic symptoms were associated with
poorer social attention, and a greater effect for sex
differences relative to diagnostic group differences
Figure 3 Bar plot of the intercepts – left side; representing the estimated average PLT by group – and comparisons of the raw (top right)
and model estimated dynamics (bottom right) for autistic and nonautistic males in the Dynamic Video. We illustrated the comparison
between autistic and nonautistic males because the estimates significantly differ from each other. In the right plot, the colour codes of
the curves represent age, while the black lines represent the average estimate for each age group
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
8 Teresa Del Bianco et al. J Child Psychol Psychiatr 2022; 0(0): 1–13
on social attention was associated with fewer social-
communication symptoms.
The patterns of our findings for averaged face
looking were similar to those found by Harrop et al.
(2019, 2020) in that autistic females showmore face-
looking than males. Another aspect that was repli-
cated was higher average face-looking in autistic
females compared to nonautistic females in the
leanest stimulus, the Face Pop-Out; Harrop and
colleagues’ suggestion that autistic females’ atten-
tion may fall on a continuum (from increased for a
lean scene, to decreased for a complex scene) is
supported here. Furthermore, autistic females
showed a later peak compared to nonautistic
females, influenced by age, flattening/growing in
autistic/nonautistic females respectively. As age did
not influence sex differences, and autistic males did
not strongly differ from nonautistic males for this
stimulus, the age effect seems specific to being
female and autistic. This might possibly explain the
consistency with previous studies, despite the wider
age range in our sample, and relate to sensitive time
windows of learning in females that do not overlap
across diagnostic groups.
In line with previous findings (Kaliukhovich et al.,
2020; Pierce et al., 2016; Tang, Chen, Falkmer,
Bo¨lte, & Girdler, 2019), in the dynamic video, the
social attention of autistic females diverged from that
of nonautistic females at the average level. However,
we found no difference at the dynamic level, so the
evidence for a different process is less strong. In
contrast, autistic males differed from nonautistic
males in this context and evaded or diverted their
attention from the face during the scene. This
pattern may be a sign of less contextual adjustment
of attention in autistic males – and since it is evident
in the dynamic stimulus only, it may relate to
differences in endogenous orienting that may only
be elicited under more naturalistic conditions (Che-
vallier et al., 2015) and prolonged viewing times (Del
Bianco et al., 2021). Future research could assess if
this pattern is associated with specific early-stage
processing differences of attentional control, coupled
with weaker activation of compensatory anterior
cortical systems (Johnson et al., 2021).
Of note, we did not find a consistent difference in
the Static Scenes, for which the explanation may
reside in the wider Confidence Intervals (~25% in the
Static Scenes, vs. <10% in the Face Pop-Out and
~15% in the Dynamic Scenes) that indicate increased
variability that may have shadowed sex and diag-
nostic group differences in the split samples. How-
ever, it is notable that the pattern of the diagnostic
group and sex differences were consistent across all
stimuli (see Table 2).
In general, this pattern of findings resonates with
accounts of different levels of sensitivity and effect
sizes between stimuli, interpreted as one’s social
attention adapting differently to different stimuli in
autistic people compared to nonautistic people.
Furthermore, our findings suggest that sex adds
an additional layer to the complexity, with social
attention to leanest stimuli being more sensitive to
differences between autistic vs. nonautistic females,
and dynamic stimuli eliciting different social atten-
tion behaviour in males. It is possible that eye-
tracking captures different processes in females
(e.g., adaptation and learning from simple social
information) and in males (e.g., preferential looking
and withdrawal with naturalistic input) that emerge
dynamically from genetic predisposition and in
response to social stimulation (Johnson et al.,
2021).
Dimensional variation
In females, average face-looking was inversely asso-
ciated with ADOS SA-CSS and RRB-CSS, meaning
that females with higher social attention had lower
severity scores. This pattern fits with the idea that
females with a diagnosis of ASD who have higher
levels of social attention may display fewer cardinal
autistic symptoms across domains. It is possible
that social attention may partially mitigate the
effects of yet unidentified etiological mechanisms
operating in the earliest postnatal developmental
stages (Chawarska et al., 2016), and early-stage
processing differences (Johnson et al., 2021) that
may otherwise broadly increase symptom load.
The relationship with the difference score shows
that this modulation holds stronger when an indi-
vidual autistic female is more different from males
than she is from neurotypical females (i.e., when the
effect of sex outweighs the effect of diagnostic group
on social attention). This observation further sup-
ports the idea that social attention may be a man-
ifestation of adaptation/learning in autistic females,
thus attenuating their autistic behavioural presen-
tation compared to that of the standard levels of
males.
Limitations
The implications of our cross-sectional findings are
limited to concurrent relationships with diagnostic
status and symptomatology rather than causal
inferences. Also, we did not have information about
the individuals’ gender identity, gender expression,
gendered socialisation experiences and other sex-
related biological factors, such as a pubertal stage.
Notably, we did not find associations between eye-
tracking metrics and parent-reported measures of
autistic characteristics, which may represent a more
comprehensive assay of everyday autistic presenta-
tion compared to ADOS CSS. Finally, the stimuli
used, although one of them was dynamic, may not
faithfully represent the experience of social attention
during everyday life; the fixed order of the stimuli
may not allow for disentangling the effect of stimulus
order from stimulus nature – even though we did not
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
doi:10.1111/jcpp.13630 Dynamic Profiles of Social Attention in Females 9
statistically compare the three tasks to each other –
and the gender imbalance (of the actor) in the SS
stimuli might have influenced visual preferences.
Conclusions
Overall, our results indicate that sex differences in
social attention exist. They differ between autistic
and nonautistic people, are context-dependent and
may be underpinned by multiple mechanisms
depending on stimulus type. Furthermore, females
with higher levels of social attention show fewer
observed autistic symptoms. These observations
suggest that social attention may be a candidate
modifier that ameliorates autistic social-
communication characteristics by granting occa-
sions for learning and enhancing the navigation of
the human social world: for example, it may provide
the autistic person with a tool for cultivating reward-
ing and desired social interactions, and/or avoiding
unpleasant and stressful ones. Since we found
associations between social attention and dimen-
sional symptoms across domains, future work could
re-evaluate the concept of core symptoms as a
continuous distribution of mild to elevated autism-
ness (Braithwaite, Gui, & Jones, 2020; Constantino,
2011). Ultimately, the modifying effect of social
attention that may influence the emergence of
social-communication disability should be examined
longitudinally and investigated in relationship with
phenomena such as compensation and camouflag-
ing (Lai et al., 2021).
Supporting information
Additional supporting information may be found online
in the Supporting Information section at the end of the
article:
Appendix S1. Results
Table S1. beta coefficients, standard errors (SE),
t-values and p-values of the multiple regression with
accuracy across the session
Table S2. Average percentage (%) of Missing Data, and
standard deviation, per time bin by stimulus, group and
sex. Cohen’s D provides the effect size of the between-
sex difference within each group
Table S3. Average percentage (%) of Missing Data, and
standard deviation, per time bin by stimulus, group and
sex. Cohen’s D provides the effect size of the between-
sex difference within each group
Table S4. Non-autistic group model selection output,
comparing the base model (i.e., including polynomials
of degree 3 as fixed effects and random effects), with
models with additional fixed effects. ‘+’ marks adding
the specified variable as a covariate; ‘*’ marks adding
the interaction between the polynomial components
and the specified variable. A p-value < 0.05 marks a
significant comparison, i.e., better explanatory power
compared to the base model
Table S5. Autistic group model selection output, com-
paring the base model (i.e., including polynomials of
degree 3 as fixed effects and random effects), with
models with additional fixed effects. ‘+’ marks adding
the specified variable as a covariate; ‘*’ marks adding
the interaction between the polynomial components
and the specified variable. A p-value < 0.05 marks a
significant comparison, i.e., better explanatory power
compared to the base model
Table S6. Non-autistic group model output, with ‘*’
marking Interactions. Significant β are marked with a P-
Value < 0.05, to be interpreted as different from the
reference level (female)
Table S7. Autistic group model output, with ‘*’ marking
Interactions. Significant β are marked with a P-Value
< 0.05, to be interpreted as different from the reference
level (female)
Table S8. Males model selection output, comparing the
base model (i.e., including polynomials of degree 3 as
fixed effects and random effects), with models with
additional fixed effects. ‘+’ marks adding the specified
variable as a covariate; ‘*’ marks adding the interaction
between the polynomial components and the specified
variable. A p-value < 0.05 marks a significant compar-
ison, i.e., better explanatory power compared to the
base model
Table S9. Females model selection output, comparing
the base model (i.e., including polynomials of degree 3
as fixed effects and random effects), with models with
additional fixed effects. ‘+’ marks adding the specified
variable as a covariate; ‘*’ marks adding the interaction
between the polynomial components and the specified
variable. A p-value < 0.05 marks a significant compar-
ison, i.e., better explanatory power compared to the
base model
Table S10. Males model output, with ‘*’ marking
Interactions. Significant β are marked with a P-Value
< 0.05, to be interpreted as different from the reference
level (non-autistic)
Table S11. Females model output, with ‘*’ marking
Interactions. Significant β are marked with a P-Value
< 0.05, to be interpreted as different from the reference
level (non-autistic)
Table S12. Non-autistic group model selection output,
comparing the base model (i.e., including polynomials
of degree 3 as fixed effects and random effects), with
models with additional fixed effects. ‘+’ marks adding
the specified variable as a covariate; ‘*’ marks adding
the interaction between the polynomial components
and the specified variable. A p-value < 0.05 marks a
significant comparison, i.e., better explanatory power
compared to the base model
Table S13. Autistic group model selection output,
comparing the base model (i.e., including polynomials
of degree 3 as fixed effects and random effects), with
models with additional fixed effects. ‘+’ marks adding
the specified variable as a covariate; ‘*’ marks adding
the interaction between the polynomial components
and the specified variable. A p-value < 0.05 marks a
significant comparison, i.e., better explanatory power
compared to the base model
Table S14 Non-autistic model output, with ‘*’ marking
Interactions. Significant β are marked with a P-Value
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
10 Teresa Del Bianco et al. J Child Psychol Psychiatr 2022; 0(0): 1–13
< 0.05, to be interpreted as different from the reference
level (female)
Table S15. Autistic group model output, with ‘*’ mark-
ing Interactions. Significant β are marked with a
P-Value < 0.05, to be interpreted as different from the
reference level (female)
Table S16. Males model selection output, comparing
the base model (i.e., including polynomials of degree 3
as fixed effects and random effects), with models with
additional fixed effects. ‘+’ marks adding the specified
variable as a covariate; ‘*’ marks adding the interaction
between the polynomial components and the specified
variable. A p-value < 0.05 marks a significant compar-
ison, i.e., better explanatory power compared to the
base model
Table S17. Females model selection output, comparing
the base model (i.e., including polynomials of degree 3
as fixed effects and random effects), with models with
additional fixed effects. ‘+’ marks adding the specified
variable as a covariate; ‘*’ marks adding the interaction
between the polynomial components and the specified
variable. A p-value < 0.05 marks a significant compar-
ison, i.e., better explanatory power compared to the
base model
Table S18. Males model output, with ‘*’ marking
Interactions. Significant β are marked with a P-Value
< 0.05, to be interpreted as different from the reference
level (non-autistic)
Table S19. Females model output, with ‘*’ marking
Interactions. Significant β are marked with a P-Value
< 0.05, to be interpreted as different from the reference
level (non-autistic)
Table S20. Non-autistic group model selection output,
comparing the base model (i.e., including polynomials
of degree 3 as fixed effects and random effects), with
models with additional fixed effects. ‘+’ marks adding
the specified variable as a covariate; ‘*’ marks adding
the interaction between the polynomial components
and the specified variable. A p-value < 0.05 marks a
significant comparison, i.e., better explanatory power
compared to the base model
Table S21. Autistic group model selection output,
comparing the base model (i.e., including polynomials
of degree 3 as fixed effects and random effects), with
models with additional fixed effects. ‘+’ marks adding
the specified variable as a covariate; ‘*’ marks adding
the interaction between the polynomial components
and the specified variable. A p-value < 0.05 marks a
significant comparison, i.e., better explanatory power
compared to the base model
Table S22. Non-autistic group model output, with ‘*’
marking Interactions. Significant β are marked with a
P-Value < 0.05, to be interpreted as different from the
reference level (female)
Table S23. Autistic group model output, with ‘*’ mark-
ing Interactions. Significant β are marked with a
P-Value < 0.05, to be interpreted as different from the
reference level (female)
Table S24. Males model selection output, comparing
the base model (i.e., including polynomials of degree 3
as fixed effects and random effects), with models with
additional fixed effects. ‘+’ marks adding the specified
variable as a covariate; ‘*’ marks adding the interaction
between the polynomial components and the specified
variable. A p-value < 0.05 marks a significant compar-
ison, i.e., better explanatory power compared to the
base model
Table S25. Females model selection, comparing the
base model (i.e., including polynomials of degree 3 as
fixed effects and random effects), with models with
additional fixed effects. ‘+’ marks adding the specified
variable as a covariate; ‘*’ marks adding the interaction
between the polynomial components and the specified
variable. A p-value < 0.05 marks a significant compar-
ison, i.e., better explanatory power compared to the
base model
Table S26. Males model output, with ‘*’ marking
Interactions. Significant β are marked with a P-Value
< 0.05, to be interpreted as different from the reference
level (female)
Table S27. Females model output, with ‘*’ marking
Interactions. Significant β are marked with a P-Value
< 0.05, to be interpreted as different from the reference
level (female)
Table S8. Contrasts specification, with coefficient esti-
mate of the slope (significant coefficients marked with
‘*’), and the difference to which the statistical test has
been applied
Table S29. Difference score/ADOS SA-CSS multiple
linear regression (significant coefficientsmarkedwith ‘*’)
Acknowledgements
The authors thank the participants and their families
for their participation.
This work was supported by the EU/EFPIA/SFARI/
Autistica/AUTISM SPEAKS Innovative Medicines Ini-
tiative 2 Joint Undertaking (AIMS-2-TRIALS grant n.
777394) and Innovative Medicines Initiative 1 Joint
Undertaking (AIMS-2-TRIALS grant n. 115300), and the
Medical Research Council for the projects “The Devel-
opment of Social Attention and Perception Abilities in
Typical and At-risk Infants” (grant n. MR/K021389/1)
and “Human neurocognitive development: Early-stage
processing, modifiers, and outcomes” (grant n. MR/
T003057/1). L.M., T.C., M.H.J. and E.J. received
funding from the UK Medical Research Council. M-
C.L. was supported by a Canadian Institutes of Health
Research Sex and Gender Science Chair (GSB 171373)
and an Academic Scholars Award from the Department
of Psychiatry, University of Toronto. The funders had no
role in the design of the study; in the collection,
analyses, or interpretation of data; in the writing of
the manuscript, or in the decision to publish the
results. Any views expressed are those of the authors
and not necessarily those of the funders.
J.T. has acted as a paid consultant and is a current
employee of F. Hoffmann-La Roche AG. T.C. has acted
as a paid consultant of F. Hoffmann-La Roche Ltd and
Servier and receives royalties for the publication of
textbooks for Guilford Press and Sage Publications. J.B.
has been in the past 3 years a consultant to/member of
advisory boards of/and/or speaker for Janssen Cilag
BV, Eli Lilly, Lundbeck, Shire, Roche, Novartis, Medice,
and Servier. S.B. receives royalties for the German and
Swedish KONTAKT manuals and adaptations of the
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
doi:10.1111/jcpp.13630 Dynamic Profiles of Social Attention in Females 11
ADI-R, ADOS, and SRS from Hogrefe Publishers. He has
in the last 3 years acted as an author, consultant, or
lecturer for Medice and Roche. T.B. has served in an
advisory or consultancy role for Actelion, Hexal
Pharma, Lilly, Medice, Novartis, Oxford Outcomes,
Otsuka, PCM Scientific, Shire, and Vifor Pharma. He
has received conference support or speaker fees from
Medice, Novartis, and Shire. He is/has been involved in
clinical trials conducted by Shire and Vifor Pharma. He
has received royalties from Hogrefe, Kohlhammer, CIP
Medien, and Oxford University Press.
The present work is unrelated to the above grants and
relationships. All other authors report no biomedical
financial interests or potential conflicts of interest.
Correspondence
Teresa Del Bianco, Centre for Brain and Cognitive
Development, Henry Wellcome Building, Birkbeck,
University of London, Malet Street, London WC1E
7HX, UK; Email: t.delbianco@bbk.ac.uk
Key points

 This work highlights the same patterns of sex differences in social attention in autistic and nonautistic people,
and distinct dynamic changes in social attention in autistic females, depending on age.

 The inverse association between social attention and symptoms load in autistic females only may inform
clinical interventions targeting subgroups with different liability.

 Autistic girls showed intensified social attention between 6-and 14 years. Educational models could flexibly
synchronise to autistic developmental spurts to facilitate learning and wellbeing.

 Association between social attention and symptoms load across domains suggests that future research should
try to overcome the conception of autism behavioural manifestations as an independent trait.
References
American Psychiatric Association. (2013). Diagnostic and sta-
tistical manual of mental disorders (5th edn). Washington
DC: Author.
Baron-Cohen, S. (2002). The extreme male brain theory of
autism. Trends in Cognitive Sciences, 6, 248–254.
Bedford, R., Jones, E.J.H., Johnson, M.H., Pickles, A., Char-
man, T., & Gliga, T. (2016). Sex differences in the association
between infant markers and later autistic traits. Molecular
Autism, 7, 1–11.
Bourreau, Y., Roux, S., Gomot, M., Bonnet-Brilhault, F., &
Barthe´le´my, C. (2009). Validation of the repetitive and
restricted behaviour scale in autism spectrum disorders.
European Child and Adolescent Psychiatry, 18, 675–682.
Braithwaite, E.K., Gui, A., & Jones, E.J.H. (2020). Social
attention: What is it, how can we measure it, and what can it
tell us about autism and ADHD? In S. Hunnius & M. Meyer
(Eds.), Progress in brain research (Vol. 254, pp. 271–303).
Amsterdam: Elsevier B.V.
Charman, T., Loth, E., Tillmann, J., Crawley, D., Wooldridge,
C., Goyard, D., . . . & Buitelaar, J.K. (2017). The EU-AIMS
Longitudinal European Autism Project (LEAP): Clinical
characterisation. Molecular Autism, 8, 27.
Chawarska, K., Macari, S., Powell, K., DiNicola, L., & Shic, F.
(2016). Enhanced social attention in female infant siblings
at risk for autism. Journal of the American Academy of Child
and Adolescent Psychiatry, 55, 188–195.
Chevallier, C., Parish-Morris, J., McVey, A., Rump, K.M.,
Sasson, N.J., Herrington, J.D., & Schultz, R.T. (2015).
Measuring social attention and motivation in autism spec-
trum disorder using eye-tracking: Stimulus type matters.
Autism Research, 8, 620–628.
Constantino, J.N. (2011). The quantitative nature of autis-
tic social impairment. Pediatric Research, 69(5 Part 2), 55–
62.
Constantino, J.N., Charman, T., & Jones, E. (2021). Clinical
and translational implications of an emerging developmental
substructure for autism. Annual Review of Clinical Psychol-
ogy, 17, 365–389.
Constantino, J.N., & Gruber, C.P. (2005). The Social Respon-
siveness Scale (SRS). Los Angeles: Western Psychological
Services.
Del Bianco, T., Mason, L., Charman, T., Tillman, J., Loth, E.,
Hayward, H., . . . & Zwiers, M.P. (2021). Temporal profiles of
social attention are different across development in Autistic
and Neurotypical people. Biological Psychiatry: Cognitive
Neuroscience and Neuroimaging, 6, 813–824.
Esler, A.N., Bal, V.H., Guthrie,W.,Wetherby, A., Weismer, S.E.,
& Lord, C. (2015). The autism diagnostic observation sched-
ule, toddler module: Standardized severity scores. Journal of
Autism and Developmental Disorders, 45, 2704–2720.
Floris, D.L., Filho, J.O.A., Lai, M.-C., Giavasis, S., Oldehinkel,
M., Mennes, M., . . . & Di Martino, A. (2021). Towards robust
and replicable sex differences in the intrinsic brain function
of autism. Molecular Autism, 12, 19.
Frazier, T.W., Strauss, M., Klingemier, E.W., Zetzer, E.E.,
Hardan, A.Y., Eng, C., & Youngstrom, E.A. (2017). A meta-
analysis of gaze differences to social and nonsocial informa-
tion between individuals with and without autism. Journal of
the American Academy of Child and Adolescent Psychiatry,
56, 546–555.
Gliga, T., Elsabbagh, M., Andravizou, A., & Johnson, M.
(2009). Faces attract infants’ attention in complex displays.
Infancy, 14, 550–562.
Gui, A., Mason, L., Gliga, T., Hendry, A., Begum Ali, J., Pasco,
G., . . . & Jones, E.J.H. (2020). Look duration at the face as a
developmental endophenotype: Elucidating pathways to
autism and ADHD. Development and Psychopathology, 32,
1303–1322.
Hammill, C., Lerch, J.P., Taylor, M.J., Ameis, S.H., Chakra-
varty, M.M., Szatmari, P., . . . & Lai, M.-C. (2021). Quantita-
tive and qualitative sex modulations in the brain anatomy of
autism. Biological Psychiatry. Cognitive Neuroscience and
Neuroimaging, 6, 898–909.
Harrop, C., Jones, D.R., Sasson, N.J., Zheng, S., Nowell, S.W.,
& Parish-Morris, J. (2020). Social and object attention is
influenced by biological sex and toy gender-congruence in
children with and without autism. Autism Research, 13,
763–776.
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
12 Teresa Del Bianco et al. J Child Psychol Psychiatr 2022; 0(0): 1–13
Harrop, C., Jones, D., Zheng, S., Nowell, S.W., Boyd, B.A., &
Sasson, N. (2018). Sex differences in social attention in
autism spectrum disorder. Autism Research, 11, 1264–
1275.
Harrop, C., Jones, D., Zheng, S., Nowell, S., Schultz, R., &
Parish-Morris, J. (2019). Visual attention to faces in children
with autism spectrum disorder: Are there sex differences?
Molecular Autism, 10, 28.
Hedger, N., & Chakrabarti, B. (2021). Autistic differences in
the temporal dynamics of social attention. Autism, 25, 1615–
1626.
Johnson, M.H., Charman, T., Pickles, A., & Jones, E.J.H.
(2021). Annual Research Review: Anterior Modifiers in the
Emergence of Neurodevelopmental Disorders (AMEND)—A
systems neuroscience approach to common developmental
disorders. Journal of Child Psychology and Psychiatry, 62,
610–630.
Kaliukhovich, D.A., Manyakov, N.V., Bangerter, A., Ness, S.,
Skalkin, A., Goodwin, M.S., . . . & Pandina, G. (2020). Social
attention to activities in children and adults with autism
spectrum disorder: Effects of context and age. Molecular
Autism, 11, 79.
Lai, M.C., Hull, L., Mandy, W., Chakrabarti, B., Nordahl, C.W.,
Lombardo, M.V., . . . & Livingston, L.A. (2021). Commentary:
‘Camouflaging’in autistic people–reflection on Fombonne
(2020). Journal of Child Psychology and Psychiatry, 62.
https://doi.org/10.1111/jcpp.13344. Online ahead of print.
Lai, M.-C., Lombardo, M.V., Suckling, J., Ruigrok, A.N.V.,
Chakrabarti, B., Ecker, C., . . . & Baron-Cohen, S. (2013).
Biological sex affects the neurobiology of autism. Brain, 136,
2799–2815.
Lai, M.-C., & Szatmari, P. (2020). Sex and gender impacts on
the behavioural presentation and recognition of autism.
Current Opinion in Psychiatry, 33, 117–123.
Loomes, R., Hull, L., & Mandy, W.P.L. (2017). What is the
male-to-female ratio in autism spectrum disorder? A
systematic review and meta-analysis. Journal of the
American Academy of Child and Adolescent Psychiatry, 56,
466–474.
Lord, C., Risi, S., Lambrecht, L., Cook, E.H., Leventhal, B.L.,
DiLavore, P.C., . . . & Rutter, M. (2000). Autism Diagnostic
Observation Schedule (ADOS). Journal of Autism and Devel-
opmental Disorders, 30, 205–223. https://doi.org/10.
1023/A:1005592401947
Loth, E., Charman, T., Mason, L., Tillmann, J., Jones, E.J.H.,
Wooldridge, C., . . . & Buitelaar, J.K. (2017). The EU-AIMS
Longitudinal European Autism Project (LEAP): Design and
methodologies to identify and validate stratification
biomarkers for autism spectrum disorders. Molecular Aut-
ism, 8, 24.
Maenner, M.J., Shaw, K.A., Baio, J., Washington, A., Patrick,
M., DiRienzo, M., . . . & Dietz, P.M. (2020). Prevalence of
autism spectrum disorder among children aged 8 Years-
Autism and developmental disabilities monitoring network,
11 Sites, United States, 2016. MMWR Surveillance Sum-
maries, 69, 1–12.
Mirman, D., Dixon, J.A., & Magnuson, J.S. (2008). Statistical
and computational models of the visual world paradigm:
Growth curves and individual differences. Journal of Memory
and Language, 59, 475–494.
Pierce, K.,Marinero, S., Hazin, R.,McKenna,B., Barnes, C.C., &
Malige, A. (2016). Eye tracking reveals abnormal visual pref-
erenceforgeometric imagesasanearlybiomarkerofanautism
spectrum disorder subtype associated with increased symp-
tom severity.Biological Psychiatry, 79, 657–666.
Posada de la Paz, M. (2018). Autism spectrum disorders in the
European Union, ASDEU. Technical report, European Com-
mission.
Tang, J.S.Y., Chen, N.T.M., Falkmer, M., Bo¨lte, S., & Girdler, S.
(2019). Atypical visual processing but comparable levels of
emotion recognition in adults with autism during the
processing of social scenes. Journal of Autism and Develop-
mental Disorders, 49, 4009–4018.
Appendix
The EU-AIMS LEAPGroup: Antonia San JoseCaceres,
DaisyCrawley, JessicaFaulkner, JessicaSabet, Claire
Ellis, Bethany Oakley, Rosemary Holt, Sara Ambro-
sino, Nico Bast, Sarah Baumeister, Annika Rausch,
Carsten Bours, Ineke Cornelissen, Daniel von Rhein,
Larry O’Dwyer, Jumana Ahmad, Emily Simonoff,
SarahDurston, Tobias Banaschewski, Andrea Persico
Accepted for publication: 25 March 2022
© 2022 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
doi:10.1111/jcpp.13630 Dynamic Profiles of Social Attention in Females 13