ThesisPDF Available

AUTISM SPECTRUM CONDITIONS IN WOMEN: DIAGNOSIS, MENTAL HEALTH, AND THE ROLE OF CAMOUFLAGING

Authors:

Abstract

The female phenotype of autism may cause a delay in diagnosis for autistic women. Studies show autistic females may camouflage their autistic traits and may have more mental health difficulties as a result. It has also been hypothesised that autistic women might be misdiagnosed with other conditions. The current investigation aimed to explore social and behavioural factors that might delay or prevent diagnosis, and factors that may influence the mental health pathways to diagnosis for autistic women. In the first and second study a nationwide survey was conducted to identify potentially autistic individuals, defined as those who score highly for autistic traits on the Autism Quotient (AQ) screening tool but have no formal diagnosis of autism (Study 1 n = 834, Study 2 n = 88), and comparing them to diagnosed autistic individuals (Study 1 n = 179, Study 2 n = 121) on a number of questionnaires measuring emotional and social abilities and mental health. In Study 3, eighty participants (40 autistic and 40 non-autistic) completed a self-reported camouflaging measure, a battery of executive functioning tasks, and theory of mind test. They were also video-recorded having a natural conversation with a researcher, which a further 127 non-autistic participants rated using a first-impression scale. In Studies 1 and 2, potentially autistic women had a significant empathy and social functioning advantage over diagnosed women, and were more likely to be diagnosed with Borderline Personality Disorder. However, they were less likely to have other psychiatric diagnoses, and had similar difficulties in friendship, theory of mind, self-monitoring, anxiety, and depression. Strong correlations were not found between social performance and age of Autism Spectrum Condition (ASC) diagnoses, or with mental health traits. Diagnosed autistic women were more likely than men to have other psychiatric diagnoses, and these were more likely to be received prior to an ASC diagnosis. In Study 3, no differences on self-reported camouflaging were found between autistic men and women, although both groups scored more highly than non-autistic controls, and camouflaging was not associated with theory of mind or executive functioning. However, autistic people were rated less favourably on first-impressions than non-autistic people, and males were rated less favourably than females. Furthermore, male raters were harsher in their judgements of autistic males. These ratings correlated with age of diagnosis, but not with camouflaging scores. Findings suggest that a combination of factors may delay diagnosis in women. Clinicians may be biased towards diagnosing other psychiatric conditions before ASC is identified. This may be because women present less typically than males and are judged less harshly by peers.
ANGLIA RUSKIN UNIVERSITY
AUTISM SPECTRUM CONDITIONS IN WOMEN: DIAGNOSIS, MENTAL
HEALTH, AND THE ROLE OF CAMOUFLAGING
HANNAH LOUISE BELCHER
A thesis in partial fulfilment of the requirements of Anglia Ruskin University for the
degree of Doctor of Philosophy
Submitted: 09.03.2020
Table of Contents
Page
Acknowledgements i
Abstract iii
List of Figures iv
List of Tables v
Abbreviations and Symbols vii
Copyright Declaration xi
Chapter 1: Introduction 1
1.1 Definition of Autism Spectrum Conditions 1
1.2 Prevalence of Autism Spectrum Conditions 6
1.3 Diagnosis of Autism Spectrum Conditions 8
1.3.1 Diagnostic process 8
1.3.2 Gender differences in diagnosis 10
1.3.3 Psychiatric co-morbidities 14
1.4 Post-Diagnosis and Prognosis 17
Chapter 2: Gender Based Theories of Autism 23
2.1 Extreme Male Brain Theory 24
2.1.1 Empathising 25
2.1.2 Systemising 30
2.1.3 Additional limitations of the EMB theory 33
2.2 Female Phenotype Theory 34
2.2.1 Presentation of autistic characteristics in males and females 34
2.2.2 Gender socialisation and the presentation of autistic symptoms 40
2.2.3 Camouflaging autistic traits 42
2.2.4 Gender-distinctive cognitive strategies for camouflaging 49
2.2.5 Mental health repercussions of camouflaging 53
2.2.6 Misdiagnosis 56
2.3 Summary and Research Directions 58
2.4 Thesis Overview 62
Chapter 3: Study 1 67
3.1 Introduction 67
3.1.1 Aims and hypotheses 76
3.2 Methods 78
3.2.1 Participants 78
3.2.2 Measures 80
3.2.3 Design 81
3.2.4 Procedure 82
3.3 Results 83
3.3.1 Data checks and descriptive statistics 83
3.3.2 Proportion of potential ASC participants 84
3.3.3 Age of diagnosis 85
3.3.4 Group differences in EQ scores 85
3.3.5 Exploring the age of autism diagnosis 88
3.3.6 Group differences in mental health diagnoses 88
3.3.7 Differential psychiatric diagnoses in diagnosed ASC and
potential ASC
90
Page
3.4 Discussion 92
Chapter 4: Study 2 101
4.1 Introduction 101
4.1.1 Aims and hypotheses 105
4.2 Methods 106
4.2.1 Participants 106
4.2.2 Measures 108
4.2.3 Design 112
4.2.4 Procedure 112
4.3 Results 113
4.3.1 Data checks and descriptive statistics 113
4.3.2 Female group differences on questionnaire measures 116
4.3.3 Correlations between questionnaire measures for female groups 120
4.3.4 Predicting the age of autism diagnosis 124
4.3.5 Other mental health diagnoses in females 124
4.3.6 Exploratory comparisons between males and females in the
ASC groups
125
4.4 Discussion 127
Chapter 5: Study 3 135
5.1 Introduction 135
5.1.1 Camouflaging and associated traits in autism 136
5.1.2 The effects of camouflaging on impressions made on others 139
5.1.3 Aims and hypotheses 144
5.2 Part One 145
5.2.1 Method 145
5.2.1.1 Participants 145
5.2.1.2 Measures 148
5.2.1.3 Procedure 152
5.2.2 Results 153
5.2.2.1 Data checks and descriptive statistics 153
5.2.2.2 Effects of gender and autism on all measures 154
5.2.2.3 Correlation analysis 158
5.2.3 Summary 161
5.3 Part Two 162
5.3.1 Method 162
5.3.1.1 Participants 162
5.3.1.2 Materials 162
5.3.1.3 Procedure 166
5.3.2 Results 168
5.3.2.1 Participant-raters’ first-impressions of participant-stimuli 168
5.3.2.2 Correlation analyses 172
5.3.3 Summary 173
5.4 General Discussion 173
5.4.1 Part-one: Group and gender differences in self-reported
camouflaging
174
5.4.2 Part-two: Group and gender differences in first-impressions 179
5.4.3 Strengths and limitations 181
Page
5.4.4 Conclusions 183
Chapter 6: General Discussion 185
6.1 Key Findings from the Literature 185
6.2 Thesis Predictions and Current Findings 187
6.2.1 Comparison of diagnosed and potentially undiagnosed autistic
women on other mental health diagnoses
188
6.2.2 Comparison of diagnosed and potentially undiagnosed women
on social and behavioural measures
190
6.2.3 Self-reported camouflaging and peer-assessed judgements of
social behaviours in autistic males and females
191
6.3 A Reflection of the EMB and FPT Theories of Gender and Autism 193
6.4 Limitations and Strengths 196
6.5 Implications
6.6 Avenues for Future Research
201
205
6.7 Conclusions 208
References 211
Appendix 249
Acknowledgements
I very quickly learnt when I started my PhD that no researcher was an island, and the
journey I was embarking on would require the support and company of many people; all
of whom coloured my PhD experience in different and exciting ways. Completing a
PhD means much more to me than obtaining the academic objective of a doctorate. I
grew and developed in ways I never thought achievable, researching an area which
often mirrored my own experiences. For this I wish to thank and acknowledge the
following people.
Firstly, both my supervisors, Dr Ruth Ford and Dr Sharon Morein-Zamir - whilst
you were thrown into my PhD at different points, you both knew exactly what needed to
be done. I am incredibly grateful for your patience, enthusiasm, the generosity of your
time, and the invaluable guidance you gave throughout. I would also like to thank the
generous charities who awarded me tuition fee and maintenance grants, and therefore
enabled me to actually undertake this PhD: the Ruby and George Will Trust; Burwell
Churchlands Trust; Ferguson Trust; Yorkshire Ladies Council; and the J.C. Robinson
Trust.
Secondly, I wish to thank the following other academics who guided me:
Professor Will Mandy (UCL), for reigniting my passion in my research and for your
collaboration; Dr Fiona Ashworth, for your mentorship and for giving me a voice; Dr
Jane Aspell and Dr Mick Finlay, for your yearly feedback; Dr Charlie Nevison, for your
ongoing encouragement; Dr Ross Kemble, for your friendship and for not letting me
give up; and to the whole Psychology department at Anglia Ruskin University,
particularly those who gave me the opportunity to teach and who provided their
assistance along the way. I would also like to thank the following support staff at ARU,
who gave me some of the most important things during this journey: Martine Vanwyk,
1
for your compassion, consistency, and ability to listen without judgement - you have
been a rock against the waves; Suzanne Drieu, for always giving me a safe place to go,
for helping me to grow, and for instilling in me the confidence to continue; Lisa Scott-
Donkin, for being such a patient and positive sounding board whenever I needed it; and
to the Students’ Union, for all the incredible opportunities you have given me
-particularly former advisor Burcin King, for your advice, friendship, and support in
helping me stay the course.
I would also like to thank various friends and family who carried me through on
rainy days: Megan Bennett, my soulmate and soon to be wife - thank you for being my
number one fan, for filling my life with laughter, and for all the sacrifices you make for
me; my parents, for seeing my autism as a strength - you have always dreamt big for
me, supporting me through the dark days; Duncan Fulton, for showing me nothing but
unconditional support and kindness from the very start to the very end; my friends, who
gave me some much needed relief and kept me grounded; my PhD buddies who I
journeyed with, particularly Fiona Howe and Sam Martin - you opened my eyes to new
perspectives, supported me through difficult times, and kept me entertained; and Diana
Watts, for setting me on this autism path - you continue to provide me with valuable
insights both professionally and personally.
Lastly, I wish to thank and acknowledge all the autistic individuals I have met,
spoken to, or whose stories I have heard. You have all been my inspiration, you never
give up, and you have given me a space where I can truly belong.
2
Abstract
The female phenotype of autism may cause a delay in diagnosis for autistic
women. Studies show autistic females may camouflage their autistic traits and may have
more mental health difficulties as a result. It has also been hypothesised that autistic
women might be misdiagnosed with other conditions. The current investigation aimed
to explore social and behavioural factors that might delay or prevent diagnosis, and
factors that may influence the mental health pathways to diagnosis for autistic women.
In the first and second study a nationwide survey was conducted to identify
potentially autistic individuals, defined as those who score highly for autistic traits on
the Autism Quotient (AQ) screening tool but have no formal diagnosis of autism (Study
1 n = 834, Study 2 n = 88), and comparing them to diagnosed autistic individuals (Study
1 n = 179, Study 2 n = 121) on a number of questionnaires measuring emotional and
social abilities and mental health. In Study 3, eighty participants (40 autistic and 40
non-autistic) completed a self-reported camouflaging measure, a battery of executive
functioning tasks, and theory of mind test. They were also video-recorded having a
natural conversation with a researcher, which a further 127 non-autistic participants
rated using a first-impression scale.
In Studies 1 and 2, potentially autistic women had a significant empathy and
social functioning advantage over diagnosed women, and were more likely to be
diagnosed with Borderline Personality Disorder. However, they were less likely to have
other psychiatric diagnoses, and had similar difficulties in friendship, theory of mind,
self-monitoring, anxiety, and depression. Strong correlations were not found between
social performance and age of Autism Spectrum Condition (ASC) diagnoses, or with
mental health traits. Diagnosed autistic women were more likely than men to have other
psychiatric diagnoses, and these were more likely to be received prior to an ASC
diagnosis. In Study 3, no differences on self-reported camouflaging were found between
autistic men and women, although both groups scored more highly than non-autistic
controls, and camouflaging was not associated with theory of mind or executive
functioning. However, autistic people were rated less favourably on first-impressions
than non-autistic people, and males were rated less favourably than females.
Furthermore, male raters were harsher in their judgements of autistic males. These
ratings correlated with age of diagnosis, but not with camouflaging scores.
Findings suggest that a combination of factors may delay diagnosis in women.
Clinicians may be biased towards diagnosing other psychiatric conditions before ASC is
identified. This may be because women present less typically than males and are judged
less harshly by peers.
Key words: ASC; female phenotype of autism; late diagnosis; camouflaging; psychiatric
comorbidity; misdiagnosis.
3
List of Figures
Page
Figure 5.1 Example of video clip and survey layout on Qualtrics 167
Figure 5.2 Average first-impression scores of non-ASC females, non-
ASC males, ASC females, and ASC males for male and
female participant-raters with SD bars.
171
4
List of Tables
Page
Table 3.1 Descriptive statistics of each group stratified by gender and
means for AQ and EQ
84
Table 3.2 Frequency of individuals in each diagnostic group diagnosed
with one or more psychiatric disorders other than ASC
89
Table 3.3 Frequency of autistic individuals versus potentially autistic
individuals reporting specific psychiatric diagnoses
91
Table 4.1 Means and standard deviations on all measures for female
participants, stratified by ASC diagnostic group
115
Table 4.2 Means and standard deviations on all measures for male
participants, stratified by ASC diagnostic group
116
Table 4.3 Correlations between continuous measures for females in the
ASC group
121
Table 4.4 Correlations between continuous measures for females in the
potential ASC group
122
Table 4.5 Correlations between continuous measures for females in the
no ASC group
123
Table 5.1 Average predicted WAISS full-scale, verbal, and
performance IQ scores from NART errors and standard
deviations per group
147
Table 5.2 Means and standard deviations on all measures as a function
of group and gender
155
Table 5.3 Correlations between continuous measures for all
participants
159
Table 5.4 Correlations between continuous measures for autistic
participants
160
Table 5.5 Correlations between continuous measures for non-autistic
participants
161
Table 5.6 Means and standard deviations for the first-impression
scores as a function of group and gender
169
5
Abbreviations and Symbols
AAA Adult Asperger’s Assessment
ADDM Autism and Developmental Disabilities Monitoring
ADHD Attention Deficit Hyperactivity Disorder
ADI Autism Diagnostic Interview
ADOS Autism Diagnostic Observation Schedule
ANOVA Analysis of Variance
APA American Psychological Association
APMS Adult Psychiatric Morbidity Survey
AQ Autism Quotient
ASC Autism Spectrum Condition
ASD Autism Spectrum Disorder
ASD-DC Autism Spectrum Disorders – Diagnostic – Child Version
ASSQ Autism Spectrum Screening Questionnaire
BAP Broader Autism Phenotype
BCST Bergin Card Sorting Task
BPD Borderline Personality Disorder
CAST Childhood Autism Spectrum Test
CAT-Q Camouflaging Autistic Traits Questionnaire
dCohen’s D effect size
DASS Depression, Anxiety and Stress Scale
DSM Diagnostic Statistical Manual of Mental Disorders
EF Executive functioning
EMB Extreme Male Brain Theory
EMG Electromyography
EQ Empathy Quotient
FF-statistic, analysis of variance
FPF Female Protective Factor
FPT Female Phenotype Theory
FQ Friendship Quotient
FQS Friendship Qualities Scale
fT Foetal Testosterone
GAD Generalised Anxiety Disorder
GAD-7 Generalised Anxiety Disorder Scale
GHQ-12 General Health Questionnaire
GP General Practitioner
HADS-A Hospital Anxiety and Depression Scale
ICD International Classification of Diseases and Related Health Problems
IQ Intelligence Quotient
IRI Interpersonal Reactivity Index
KDEFT Karolinska Directed Emotional Faces Tasks
MMean
MET Multifaceted Empathy Test
nNumber of participants
NART National Adult Reading Test
NAS National Autistic Society
NHS National Health Service
NICE National Institute for Health and Care Excellence
OCD Obsessive Compulsive Disorder
PProbability
6
PD Personality disorder
PDD-NOS Pervasive Development Disorder Not Otherwise Specified
PHQ-9 Patient Health Questionnaire
rCoefficient of correlation
RMET Reading the Mind in the Eyes’ Test
RRBI Restricted, repetitive behaviours and interests
RQ Relatives Questionnaire
SD Standard deviation
SEN Special Educational Needs
SFS Social Functioning Scale
SLC Skin conductance level
SMS Self-Monitoring Scale
SQ Systemising Quotient
SQC Social Communication Questionnaire
SRS Social Responsiveness Scale
SST Short Story Task
STEM Science, technology, engineering and mathematics
ToL Tower of London task
ToM Theory of mind
U Mann-Whitney U
WAIS Wechsler Adult Intelligence Scale
WASI Wechsler Abbreviated Scale of Intelligence
WHO World Health Organization
Z Z-score
z z-value for Wilcoxon Signed Ranks
Symbols
X² Chi-square statistic
φ Cramer’s phi effect size
ŋ2Eta squared effect size
α Chronbach’s alpha
7
Copyright Declaration
This work may:
(i) Be made available for consultation within Anglia Ruskin Library, or
(ii) Be lent to other libraries for the purpose of consultation or may be
photocopied for such purposes
(iii) Be made available in Anglia Ruskin University’s repository and made
available on open access worldwide for non-commercial educational
purposes, for an indefinite period
8
CHAPTER 1
Introduction: Prevalence, Diagnosis, and Prognosis
1.1. Definition of Autism Spectrum Conditions
Autism Spectrum Condition (ASC), clinically referred to as Autism Spectrum Disorder
(ASD), is a neurodevelopmental condition describing a collection of social and
communication difficulties that typically result in impairments of everyday functioning.
Throughout this thesis ‘ASD’ will be referred to as ‘ASC’, and the term ‘autistic person’
will be used rather than ‘person with autism’, except where discussion relates to the
wording used in clinical documents. This is in line with recent evidence showing that
the autistic community prefers identity-first language rather than person-first language,
as ASC is not considered an illness that needs curing but as a different way of operating,
and as a collection not only of impairments but also of abilities (Gernsbacher, 2017;
Kenny et al., 2015).
Autism was first referred to as a distinct condition in 1943, by Leo Kanner; at
the time this was labelled ‘Kanner’s Syndrome’, which later became ‘Early Infantile
Autism’. Around the same time Hans Asperger described a similar disorder, which he
labelled ‘Asperger’s Syndrome’ (Asperger, 1944); however he identified individuals
with no language deficits and a higher IQ than those with ‘Early Infantile Autism’.
The diagnosis and definition of ASC has undergone considerable change since
this first identification (Baron-Cohen & Wheelwright, 2003). The Diagnostic and
Statistical Manual of Mental Disorders (DSM), which was developed and first
published in the USA in 1952 by the American Psychological Association (APA), is the
handbook used by many health professionals worldwide to diagnose mental health
disorders (Daniels & Mandell, 2014). The DSM is periodically reviewed and updated in
1
order to ensure that the diagnostic criteria used are consistent with current research and
clinical practice. Previously, the DSM IV (APA, 2000) used the term ASD as an
umbrella term to describe five sub-disorders, which included Autistic Disorder (divided
into high functioning and low functioning), Asperger’s Disorder, Rett’s Disorder,
Childhood Disintegrative Disorder, and Pervasive Development Disorder - Not
Otherwise Specified (PDD-NOS). Whilst these disorders shared common
symptomology in social and communication difficulties, they were differentiated by
other symptoms and developmental trajectories. For example, the difference between a
diagnosis of Asperger’s and Autistic Disorder was that those with Asperger’s would
have had no clinically significant delays in language, and the differences between a
PDD-NOS and Autistic Disorder diagnosis were that those with PDD-NOS might have
a late age onset or atypical or sub-threshold symptomology. In 2013, the DSM IV was
updated to DSM 5 by a large team of researchers and clinicians, in order to improve
how disorders are characterised and defined (APA, 2013). These changes had large
ramifications for the classification of ASC. The DSM 5 combines four of the separate
disorders (Autistic Disorder, Asperger’s Disorder, Childhood Disintegrative Disorder,
and PDD-NOS) recognised by DSM IV and instead refers to a single condition: Autism
Spectrum Disorder. This change was made in order to better capture the concept of ASD
being a spectrum condition, whereby autistic individuals share common core features
but to different levels of severity. The APA found that there was not enough empirical
evidence to justify the sub-disorders that were currently being used, namely Asperger’s
Syndrome and PDD-NOS, and in the USA individuals with these diagnoses were not
eligible for some autism related benefits or services (Lord & Jones, 2012). According to
the DSM 5, the condition can be characterised better by different levels of severity of
two key symptoms: deficits in social communication and social interaction, and
restricted repetitive behaviours, interests, and activities (RRBIs).
2
Social communication and social interaction difficulties can manifest in social
emotional reciprocity deficits, for example, a persistent reduced ability to initiate or
respond to various social interactions such as sharing of interests or emotions. They also
include nonverbal and communicative behaviour deficits, for example, a reduced ability
to integrate verbal and nonverbal communication, abnormalities in making eye contact,
a lack of facial expressions, and difficulties interpreting others’ gestures. Finally, there
are likely to be deficits in developing, maintaining, and understanding relationships, for
example, a complete disinterest in peers, sharing imaginative play, making friends, and
difficulties adjusting behaviour to different social contexts (APA, 2013).
RRBIs can manifest in stereotyped or repetitive physical movements, use of
objects, or speech; for example, lining up objects, repeating phrases, and flapping
hands. There is typically an insistence on sameness, with inflexibility to routine
changes, or ritualized patterns of verbal or nonverbal behaviour; for example, distress
caused by small changes, difficulties with transitioning, rigid thinking, maintaining
certain rituals, and sticking to a rigid routine such as eating the same food every day.
Other characteristics include restricted fixations on specific interests that are abnormal
in intensity and focus, for example, a strong attachment or preoccupation with specific
and sometime unusual objects. Finally, hyper- or hypo-reactivity to sensory input or an
unusual interest in sensory aspects of the environment can be reflected in a strong
aversion to certain sounds or textures, an obsessive need to feel or smell certain objects,
or to watch visual activity such as light movement, and apparent indifference to pain
and temperature (APA, 2013).
Unlike the DSM IV, which describes separate neurodevelopmental conditions,
the DSM-5 categorises these social impairments and RRBIs into three levels of severity,
namely: level 1 - “requiring support”, level 2 – “requiring substantial support”, and
3
level 3 – “requiring very substantial support”. Key specifiers, in addition to severity of
ASD include: a) with or without accompanying intellectual impairment, and b) with or
without accompanying language impairment.
Regardless of these changes to diagnostic criteria, many clinicians in the UK
continue to differentiate between the different categories of ASC, particularly between
Autistic Disorder and Asperger’s, and such diagnoses are still considered valid and are
embraced by the autism community (National Autistic Society [NAS], 2016). This is
partly due to professionals in the UK more commonly using the International
Classification of Diseases (ICD), which has only recently been updated to reflect
changes in the DSM regarding the diagnosis of ASC. The ICD was first developed and
published by the World Health Organization (WHO) in 1948. Whilst new versions are
released only periodically, the WHO make minor updates annually. The previous
version used was the ICD-10, first published in 1990, and most recently updated in
2018. The 2018 version of the ICD-10 uses the umbrella term ‘Pervasive Development
Disorders’ to describe “a group of disorders characterized by qualitative abnormalities
in reciprocal social interactions and in patterns of communication, and by a restricted,
stereotyped, repetitive repertoire of interests and activities. These qualitative
abnormalities are a pervasive feature of the individual’s functioning in all situations”
(WHO, 2018). Eight sub-disorders are described under this umbrella, including
Childhood Autism, Atypical Autism, Rett’s Syndrome, Other Childhood Disintegrative
Disorder, Overactive Disorder - associated with learning disability and stereotyped
movements, Asperger’s Syndrome, Other Pervasive Developmental Disorders, and
Pervasive Developmental Disorder Unspecified. However, the ICD-11, recently
released in 2019, like the DSM collapses sub-disorders of autism into the one disorder:
Autism Spectrum Disorder (WHO, 2019). The ICD-11 characterises ASD by
impairments in initiating and sustaining reciprocal social interactions and
4
communications, and by RRBIs, acknowledging that these impairments may be present
in early childhood but also may not be apparent until later in adolescence when social
demands increase; these impairments must also affect the individual across situations
and settings. A diagnosis is made either with or without intellectual development
disorder and also with mild or no impairment of functional language. Unlike the DSM
5, the ICD-11 does not require that a person must meet certain criteria to meet the
threshold for an autism diagnosis. Instead it lists different features which may be
present, allowing a clinician to decide whether or not autism is an appropriate diagnosis.
As well as this, the ICD-11 provides more detailed guidelines for differentiating
between autism with and without intellectual disability, whilst the DSM 5 only
acknowledges that there may be differences. These features may prevent individuals
from slipping through the net, for example those who previously would have been
diagnosed as having Asperger’s, whose characteristics and behaviours may not be seen
as ‘severe’ enough to warrant diagnosis under the new DSM-5 criteria.
The changes in both the DSM and the ICD show a move away from
conceptualising ASC as a disorder that is either present or not present, and towards
conceptualising it as an expression of several neurobiological pathways of development
with behavioural dimensions. It is thought that these behavioural dimensions will be
better indicators of each individual’s needs (Lord & Jones, 2012). However, early
evidence has suggested that the sensitivity of the new DSM criteria may be poorer than
previous versions, especially for those with Asperger’s and PDD-NOS, suggesting that
the new criteria may exclude a large proportion of autistic individuals who are less
cognitively and intellectually impaired (Kulage et al., 2014; McPartland et al., 2012).
Whilst contributing to this important debate on the classification of diagnosis is outside
the realms of this thesis, it is important to note that the present research targeted autistic
adults who do not have additional intellectual or language impairments, regardless of
5
whether they were diagnosed according to the DSM IV or DSM 5 criteria, or those of
the ICD-10 or ICD-11.
1.2. Prevalence of Autism Spectrum Conditions
Early research on the prevalence of ASC suggested that the condition was extremely
rare, with 0.02% to 0.05% of children diagnosed with infantile autism (Burd et al.,
1987; Steinhausen et al., 1986; Wing et al., 1976). However, by the 1990’s these figures
had risen, with the prevalence of infantile autism found to be around 0.1% (Gillberg et
al., 1991) and Asperger’s found to be at its highest around 0.36% (Ehlers & Gillberg,
1993). However, Fombonne (2003) argued that the ratio of Asperger’s diagnoses to
autism diagnoses is much lower (4:1), this figure may be due to the lack of
epidemiological studies on Asperger’s around this time, given that it was only officially
added to the DSM IV in 1994. Generally, prevalence rates have risen as both the DSM
and ICD developed to describe autistic conditions as a syndrome with multiple
aetiologies, rather than as a unitary disorder, suggesting that autism was not as rare as
had previously been believed (Gillberg & Wing, 1999). Looking at 32 studies on the
prevalence rates in autism published between 1966 and 2001, Fombonne (2003) found a
significant correlation between the prevalence rates and the year of publication. When
dividing these studies into two groups based on their year of publication, the 16 studies
published between 1966-1991 had a median prevalence rate of 4.4/10,000 (0.04%),
whilst the 16 studies published between 1992-2001 had a median prevalence rate of
12.7/10,000 (0.13%).
Baird et al. (2006) suggested that the prevalence of ASC may be even higher
than had previously been recognised. In a population cohort of 56,946 children, who
were all born between 1990 and 1991 in South Thames, researchers screened all
children with a clinical autism diagnosis and any judged to be at risk. The prevalence
6
for childhood autism, the diagnosis that previous studies had used to calculate
prevalence, was 38.9/10,000 (0.39%), and the prevalence for other autism conditions
was 77.2/10,000 (0.77%). Combined, the prevalence of all ASCs was 116.1/10,000
(1.16%), which is significantly higher than that previously reported. Similar findings
were found by Baron-Cohen et al. (2009) who screened all schools within the UK
county of Cambridgeshire. The ratio of known to unknown cases of autism was
established as 3:2, with the overall prevalence of both known and unknown ASC
estimated to be 1.57%. More recent epidemiological research, looking at larger
geographical areas, found similar prevalence figures. For example, Christensen et al.
(2016) conducted research using the Autism and Developmental Disabilities Monitoring
(ADDM) network, which has an active surveillance system that monitors and evaluates
eight-year-old children across 11 different states in the USA. They estimated that in
2012 around 1 in 68 children had an ASC. Amongst those children identified by the
network as having an ASC, 82% had a previous ASC diagnosis. Similarly, the 2007
Adult Psychiatric Morbidity Survey (APMS) estimated a prevalence of between 1.1%
and 1.2% (National Statistics, 2009). Additionally, Russell et al. (2013) found a
prevalence rate of 1.7%, using data from the Millennium Cohort Study (MCS), a UK-
representative birth cohort study examining children born between September 2000 and
January 2002. Whilst this prevalence rate is slightly higher than others it should be
noted that their data was based on parents’ reports of whether they had been told by a
doctor or healthcare professional that their child had an ASC, meaning that some of
these children might not have had an official diagnosis. On the whole these studies point
towards an increase in the prevalence of ASC over the time since autistic conditions
were included in the diagnostic manuals. The reason for this increase could be the result
of a number of factors, including the broadening of the diagnostic criteria of the
condition. For example, the inclusion of Asperger’s and PDD-NOS allowed ‘higher-
7
functioning’ autistic individuals to receive diagnoses. Additionally, greater prevalence is
likely due to growing awareness around the condition (Fombonne, 2005; Gillberg &
Wing, 1999; Rutter, 2005).
1.3. Diagnosis of Autism Spectrum Conditions
1.3.1. Diagnostic process. For diagnosis in under 19 year olds, The National Institute
for Health and Care Excellence (NICE) guidelines specify that an autism specific local
pathway should be set up, which includes a multi-disciplinary team (NICE, 2011). The
core membership of this team should be a paediatrician and/or child and adolescent
psychiatrist, a speech and language therapist, and a clinical and/or educational
psychologist. After screening for possible autistic traits, a GP or health visitor should
refer a child/adolescent to this pathway. The team will consider whether to carry out an
autism assessment based on the severity/duration of symptoms, whether these
symptoms are present across different environments, the impact they have on the young
person and family, the level of concern of the child and parents, any factors increasing
the probability of autism, and the likelihood of an alternative diagnosis. If an assessment
is followed through, then a report is sought from the child/adolescent’s school as well as
any other addition health or social care information. A formal diagnosis should include
detailed questions about a parent/carer’s concerns and those of the child/adolescent,
details of their experiences in different environments, a developmental history focussing
on the ICD or DSM criteria, an assessment through interaction and observation with the
child/adolescent of social and communication skills and behaviours focussing on the
ICD or DSM criteria, a full medical history, a physical examination, the consideration of
other diagnoses, systematic assessment for co-morbid conditions, profiling of the
child/adolescent’s strengths, skills, impairments, and needs, culminating in a written
report communicating assessment findings.
8
For diagnosis in adults, the NICE guidelines recommend GPs or other health
professionals use the Autism Quotient (AQ) (Baron-Cohen et al., 2001) to screen for
autism if adult patients have persistent difficulties in social interaction, and/or persistent
difficulties in social communication, and/or stereotypic behaviours, resistance to change
or restricted interests, as well as problems in employment/education, and/or difficulties
initiating or sustaining relationships, and/or contact with mental health or learning
disability services, and/or a history of a neurodevelopmental condition or mental health
problem (NICE, 2012). They should then be referred to an autism diagnostic service,
which should involve a team of different professionals, and should be formally assessed
by a professional who is trained and competent in autism diagnosis. Where possible this
assessment should involve a family member or someone who has known the person
being assessed from a young age, in order to determine a full development history. A
diagnosis should include assessing the core signs and symptoms of autism, which
should have been present since childhood and have continued into adulthood, an early
developmental history, any behavioural problems, the person’s ability to function in
different environments, past and current physical and mental disorders, any other
neurodevelopmental conditions, and sensory issues.
There are several recommended formal assessment tools for both children and
adults. These include the Adult Asperger’s Assessment (AAA) (Woodbury-Smith et al.,
2005), which uses the AQ, the Empathy Quotient (EQ), and the Relatives Questionnaire
(RQ) self-report measures as well as a clinical assessment of key domains, the Autism
Diagnostic Interview (ADI) (Le Courteur et al., 1989), which is a structured interview
focussing on the core three domains (communication, social, and RRBIs), and the
Autism Diagnostic Observation Schedule (ADOS) (Lord et al., 1989), which involves
several structured and semi-structured social interaction tasks between the assessor and
the person being assessed.
9
1.3.2. Gender differences in diagnosis. One striking feature in the diagnosis of
ASCs is the prevalence of male cases. Since Kanner’s first report of autism where he
identified 11 case studies, of which 8 were boys (Kanner, 1943), autism has consistently
been found to be more common in males than females. Both early and more recent
studies report a male to female prevalence ratio of 3-4.5:1 (Baio, 2012; Bryson & Smith
1998; Christenson et al., 2016; Fombonne, 2003; Russel et al., 2014; Yeargin-Allsopp et
al., 2003).
To some extent it appears that this gender ratio can depend on the autism
spectrum condition subtype and its severity, with a lower gender ratio in those with
intellectual impairments than those without intellectual impairments (Fombonne, 2003;
Saemundsen et al., 2003; Wing, 1981). Examining 32 surveys published between 1966
and 2001 on the epidemiology of Pervasive Developmental Disorders, Fombonne
(2003) found that the gender ratio was actually much lower in those studies looking at
individuals with intellectual impairments (1.9:1 males to females) than studies
investigating individuals without intellectual impairments (2.75:1 males to females).
More recent studies have also found similar findings (Brugha et al., 2016; Lin et al.,
2011). Brugha et al. (2016) suggests that previous research relied on the projections of
research on children, or only on adults who had the capacity to consent to take part in
prevalence surveys, whereas their research examined adults of all ages and abilities to
determine a more representative prevalence rate. Looking at the clinical diagnostic
assessments of 7,274 adults living in private households in the UK and 290 adults
registered with intellectual disabilities, they found that being male was only a strong
predictor of autism in those with no or mild intellectual disability. The general trends
suggest that autism is generally more likely to occur in males, and that when it is
unaccompanied by intellectual impairment it is even more likely to occur in males.
However, whilst the presence of intellectual disability may influence the gender ratio,
10
there are other important factors that may also affect this, which will be discussed
below.
Much of this research into prevalence rates and gender ratios of autism
investigates highly probably or already diagnosed cases, and does not account for
unidentified cases of autistic individuals. When unidentified cases are taken into
account by assessing the general population, not only does the prevalence for ASCs
without intellectual disabilities increase, but the gender disparity is also diminished.
Kim et al. (2011) found the prevalence of ASC to be 0.75% amongst high-probability of
autism children, who were considered more likely to be autistic because they were in
special needs schools and/or on the disability register, and 1.89% in the low-probability
of autism children, who were considered less likely to be autistic as they had no known
disabilities; finding that over two thirds of the ASC cases they identified were actually
undiagnosed. Several studies have found that including unidentified cases lowers the
male to female ratios that have been reported previously (Kim et al., 2011;
Zwaigenbaum et al., 2012). Ehlers and Gillberg (1993) initially found a gender ratio of
4:1 males to females in those diagnosed with ASC; however, when possible and
suspected ASC cases were included this ratio dropped to 2.3:1. More recently Loomes
et al. (2017) conducted a meta-analysis of 54 studies conducted since the DSM-IV/ICD-
10’s release, which included 13,784,284 participants, of whom 53,712 had a diagnosed
ASC (43,972 males and 9,740 females). They found a general male-to-female ratio of
4.20:1, however in the studies which screened the general population for ASCs
regardless of ASC diagnosis, the male-to-female ratio was lower (3.25:1). These
findings suggest that there may be many more females without intellectual disability
with autism than previous prevalence studies have estimated, and it may be the case that
females with the condition are more likely to be detected and diagnosed if they also
have intellectual disabilities and potentially missed altogether if they do not.
11
In support of the hypothesis that autistic females are not being detected at the
same rate as autistic males are findings that females are diagnosed with ASC later than
males. Calculating the average age of ASC diagnosis across all genders from 42 studies
published between January 1990 and March 2012 revealed a mean age of between 38 –
120 months (Daniels & Mandell, 2014). Several large scale studies have identified that
this variability is largely due to varying levels of symptom severity, with ‘lower
functioning’ and more intellectually impaired individuals being diagnosed earlier than
‘higher functioning’ and less intellectually impaired individuals (Brett et al., 2016;
Crane et al., 2015; Daniels & Mandell, 2014; Howlin & Asgharian, 1999; Mandell et
al., 2005; Williams et al., 2008). However, there is emerging evidence that being female
is also a significant factor in having a later ASC diagnosis.
Shattuck et al. (2009) used data from a 2002 multi-site ongoing autism
surveillance programme, which included the data of 2,568 children aged 8 years (491
females and 2,077 males) who were either diagnosed with an ASC or who met criteria
for the condition but who had not been classified, to determine the prevalence and age
of ASC diagnoses in children. They found that whilst autistic females had a greater
likelihood of having a cognitive impairment, they were also diagnosed later than males.
Within the group of autistic participants who had an average to above average IQ the
median age of diagnosis for females was 7.1 years, compared to 6.5 years for males, and
within the group with below average IQ the median age of diagnosis was 5.5 years for
females and 5.1 years for males. Giarelli et al. (2010) investigated the same surveillance
data, observing differences between males and females who had been classified versus
those who had not. They found that girls with an IQ of 70 or less were significantly less
likely to have a diagnosis than boys with an IQ of 70 or less (odds ratio = 0.70), and that
a similar odds ratio was observed in girls with an IQ of 70 or more in comparison to
boys with an IQ of 70 or more (odds ratio = 0.60). When divided into impairment
12
severity levels (mild, moderate, and severe impairment), these findings did not differ. In
the case of boys, by contrast, having a cognitive impairment seemed to increase the
likelihood of receiving a diagnosis. These results suggest that girls, regardless of
severity of impairment, appear to be less likely to receive a diagnosis than their male
counterparts. However, both these studies only looked at children with diagnoses and
those likely to have a diagnosis. It is possible that if females are identified later, then
many more might not receive a diagnosis until adolescence or even adulthood (Lai &
Baron-Cohen, 2015).
Begeer et al. (2013) sampled a non-clinical population of both autistic children
and autistic adults (n = 2,275) derived from the general population. Generally, autistic
females took significantly longer to be diagnosed after initial signs of the condition
were identified (M = 2.3 years) compared to autistic males (M = 1.9 years), although the
difference appears to be rather small. However, when the sample was divided into adults
and children and also by diagnostic group (Asperger’s, autistic disorder, and PDD-NOS)
a larger difference emerged. For children, girls had on average a 1.8 year delay in
diagnosis for Asperger’s compared to boys, whilst no differences were found for autistic
disorder or PDD-NOS. For adults, women had on average a 4.3 year delay in diagnosis
for autistic disorder compared to men, whilst no differences were observed for
Asperger’s or PDD-NOS. The authors warn readers not to over interpret the differences
in diagnostic categories as these may reflect historical changes in how autism is
diagnosed. For example, the majority of adults were diagnosed according to DSM-III
criteria, which did not include broader criteria diagnoses.
In support of these findings Baldwin and Costley (2016) analysed data from an
Australia-wide self-report survey, which was completed by 82 women with ‘high
functioning’ ASC. They found that the mean age of diagnosis was 25, and 58% did not
13
receive a diagnosis of ASC until after 18 years of age. In 2012, NAS commissioned a
large scale survey, which received over 8,000 responses, to better understand what life
is like for autistic people in the UK (Bancroft, 2012). They reported that only one fifth
of the girls who took part in their study were diagnosed before the age of 11, compared
to over half of boys. It is evident from these findings that more research on the age of
diagnosis in adult females is required to better understand this gender disparity in ASC
diagnosis.
1.3.3. Psychiatric co-morbidities. Other psychiatric conditions frequently co-
occur with an ASC diagnosis. These include both internalising problems, whereby
difficulties are turned inwards and overly-inhibited, manifesting in disorders such as
depression and anxiety, and externalising problems, whereby difficulties are expressed
outwardly and are disinhibited, manifesting in more overt challenging behaviour and
disorders such as Attention Deficit Hyperactive Disorder (ADHD) (Gillberg & Billstedt,
2000; Ghaziuddin et al., 1998; Hofvander et al., 2009; Mazzone et al., 2012; Mukaddes
et al., 2010; Tarazi et al., 2015). Russell et al. (2016) retrospectively reviewed co-
morbid psychiatric conditions in 859 adults (645 males and 214 females) who were
referred for an ASC diagnosis. Of those diagnosed with ASC (n = 474), significantly
more (17.9%) were diagnosed with Obsessive Compulsive Disorder (OCD) compared
to the non-ASC group (13.2%), and whilst not significant there was a trend towards
more diagnosed participants having an anxiety disorder (39.2%) compared to those not
diagnosed with an ASC (32.9%). Whilst again not significant, the non-ASC group
showed a higher prevalence of Bipolar Affective Disorder and alcohol dependency. No
differences were found between the two groups for other conditions such as ADHD and
depression. However, because the comparison group was initially referred for an ASC
assessment they are not entirely representative of the general population, as they will
have exhibited some ASC traits causing them to be put forward for psychiatric
14
assessment. When the diagnosed ASC group was compared to a general population data
pool from the UK National Psychiatric Morbidity Survey (McManus et al., 2009), the
ASC group more frequently reported phobias (16.8% vs 1.4%), generalised anxiety
disorder (GAD) (11.8% vs 4.4%), OCD (17.9% vs 1.1%), depression (15.8% vs 2.3%),
ADHD (9.7% vs 2.3%), and psychotic disorders (2.1% vs 0.4%) than the general
population.
Internalised symptoms, such as anxiety and depression, may be the result of
difficulties with ASC traits, particularly the social stigma and isolation associated with
the condition, the need to maintain routines and avoid change, and also sensory
sensitivities (Portway & Johnson, 2005; Stewart et al., 2006; Whitehouse et al., 2009;
Wood & Gadow, 2010). For example, in one study 43% of 171 autistic children met the
screening criteria cut-off for an anxiety disorder. These symptoms appeared to be related
to stereotyped behaviours, however they were also related to higher IQ and the presence
of functional language use. In another study, 43% of 46 autistic adult participants
reported depressive symptoms; however, these symptoms appeared to be worse in those
with less social impairment, higher cognitive ability, and with higher rates of other
psychiatric symptoms (Sterling et al., 2008). It may be the case that these participants
had more insight and were therefore more aware of their difficulties, or alternatively
they may be a consequence of less help and support due to ‘milder’ impairments.
Barnhill (2001) studied 33 autistic adolescents, finding a significant positive correlation
with depressive symptoms and an ability attribution for social failings, meaning
attributing social failure to one’s own abilities, rather than external factors. The higher
the intelligence of these autistic adolescents the more likely they were to attribute social
success to their own abilities, rather than to change or task difficulty. The ability to
socially compare oneself to others, as well as social perception, understanding, and
15
negative past experiences have also been found to contribute to internalising symptoms
(Hedley & Young, 2006; Meyer et al., 2006).
The common occurrence of psychiatric co-morbidities in autism is concerning
because of the risk it poses to autistic people’s lives. Camm-Crosbie et al. (2018)
conducted a qualitative analysis on two hundred autistic adults’ (122 females and 77
males) experiences of mental health support, finding common themes of difficulties
accessing treatment and support, a lack of understanding and knowledge of autistic
people with co-morbid mental health difficulties, and that a lack of appropriate
treatments and support contributed not only to low wellbeing but also to suicidal
thoughts. Self-harm and suicide are at an elevated risk in autistic people (Cassidy et al.,
2014; Chen, et al., 2017; Maddox et al., 2017; Hannon & Taylor, 2013; Segers &
Rawana, 2014; Takara & Kondo, 2014; Zahid & Upthegrove, 2017). Cassidy et al.
(2018) found that out of 164 autistic adults (99 females and 65 males), 72% scored at or
above the cut off for the Suicide Behaviours Questionnaire, which was significantly
more than people in the general population (33.7%). Furthermore, on a measure of non-
suicidal self-injurious behaviours the autistic participants were significantly more likely
to report lifetime symptoms (65%) than those in the general population (29.8%). Whilst
there were no differences between autistic males and autistic females in suicidal
behaviours, significantly more autistic females (74%) reported self-injurious behaviours
than autistic males (53.8%). Key risk factors found to be associated with suicide in
autistic people included autistic traits, self-injurious behaviour, depression, anxiety,
satisfaction with living arrangements and employment. When these key factors were
controlled for, deliberately hiding autistic traits and unmet needs also significantly
predicted suicidal behaviours. Furthermore, in a study by Pelton and Cassidy (2017),
which investigated the suicidal behaviours of 163 young autistic adults (106 females
and 55 males), feelings of burdensomeness and thwarted belonging significantly
16
interceded the relationship between autistic traits and suicidal behaviours. These studies
suggest that greater insight into one’s difficulties increases the risks associated with
mental health difficulties, putting autistic adults without intellectual impairments at a
greater risk.
1.4. Post-Diagnosis and Prognosis
For young people, under 18 years of age, NICE (2011) guidelines suggest that a report
of the findings and an evaluation of these are provided without delay to the person being
assessed and their parents/carers. A follow-up appointment should be made within six
weeks of the assessment with a member of the autism team to discuss the results.
Advice should also be given on where these young people and their families can access
support and advice. Every child/adolescent diagnosed with autism should be given a key
worker to manage and co-ordinate their support (NICE, 2013). The local autism team
should deliver/co-ordinate specialised care and interventions; advice, training, and
support for other professionals involved with the young person; advice and
interventions to aid general life functioning skills; assessing and managing challenging
behaviour and coexisting conditions; reassess needs throughout childhood and
transitioning to adult services; support the young person to access leisure activities, in
education, and with housing and employment services; and provide support for families
and carers. If local services cannot provide the interventions and support required then
the team should refer the young person instead to national services. Furthermore,
anyone working with an autistic child/adolescent should have training in autism
awareness and skills in managing autism. Autism teams should prepare to support
autistic children/adolescents and their families during times of increased need, such as
during major life changes (puberty, changing schools, birth of a new sibling etc.). A
collaborative approach should be offered if the young person and their families want to
be involved in shared decision-making about their support (NICE, 2013).
17
For adults obtaining an autism diagnosis, NICE guidelines (2012) state that
irrespective of whether further care/support is needed, a follow-up appointment should
be made to discuss the diagnosis. Within the assessment report a care plan should be
made, which incorporates risk management and the individual and their family’s
specific needs. Where there are coexisting mental health difficulties a 24-hour crisis
management plan should be developed in conjunction with mental health services. A
‘health passport’ should also be issued which includes information for all staff in contact
with the autistic person with their needs. The guidelines go on to suggest a number of
individual and group-based psychosocial interventions for the core ‘symptoms’ of
autism, life skills, managing challenging behaviour, and coexisting mental disorders.
These should be delivered by the local pathway, who are in turn advised by an autism
strategy group who should promote access to services for all autistic adults.
As ASC is a life-long condition with a spectrum of different traits, abilities, and
impairments, the prognosis of autism is varied and affected by individual differences.
Studies have shown that some autistic adolescents and adults improve significantly and
some show a stable course of maturation, however others show a deterioration in
functioning. Autism severity, cognitive functioning, language development, co-morbid
psychopathology and access to interventions are thought to affect outcomes in
adulthood but there is a lack of research investigating changes in traits from childhood
to older adulthood to determine exactly what effect these have (Levy & Perry, 2011).
Qualitative evidence indicates that getting a diagnosis is of real benefit, with many
autistic people feeling relieved to receive this. However, when diagnosis is gained in
adulthood this is often tainted with grief and anger that a diagnosis was not made sooner
so support could be accessed (Bancroft, 2012; Baldwin & Costly, 2015; Jones et al.,
2014; Stagg & Belcher, 2019).
18
Whilst ASC is not considered ‘curable’, evidence does show that early
diagnosis, and thus early interventions and support, can help autistic people greatly
(Elder et al., 2017). Howlin (1997) explored numerous findings on various types of
interventions and found that the most effective of these had the following in common:
they used behavioural oriented strategies; recognised that many undesirable behaviours
were the result of communication impairments; used the autistic child’s rituals and
obsessions to help reduce anxiety and as a reward; created structured teaching
environments that used visual cues rather than verbal cues; focussed on the development
of social-communication and play activities; recognised the importance of early
diagnosis and sharing of information and support for the parents; and were family-
orientated rather than solely being focussed on the autistic child. Howlin (1997)
suggests that such early interventions can have a considerably beneficial effect on the
quality of life in adulthood and are more cost effective than managing crises later in life.
Fernell et al.’s (2013) review of recent autism interventions in childhood suggests that
the most important outcome of an early autism diagnosis is the creation of an autism-
friendly environment around an autistic person, in order to help them overcome any
barriers they may face due to communication differences and problems with
understanding and interacting with others. As autistic people age they may have
different needs and require different support. Elder et al. (2017) highlight the
importance of family support, and that families learn to shift the focus of the support
needed as their autistic children develop into adults.
The purpose of these post-diagnosis interventions is to ensure support is in place
which addresses the complex nature of ASC. However, evidence suggests that autistic
people often do not receive the support they should after their diagnoses. Crane et al.
(2016) identified 559 services for parents of autistic children in the UK, recruiting from
these 1,047 parents who filled in a questionnaire on their experiences of their child’s
19
diagnosis and after care. On average there was a delay of 3.6 years between a parent
initially registering their concerns with a health professional and their child receiving a
diagnosis; children diagnosed with Asperger’s waited significantly longer (4.4 years)
than those with autistic disorder (3.7 years). Furthermore, despite NICE guidelines
stating that parents should receive support and advice, a report, and a follow-up
appointment, 15% of parents did not receive a report, 44% received no follow-up
appointment, 62% were not signposted to any advice or help, and 35% received no offer
of help or assistance.
Studies examining how satisfied autistic adults themselves were with the
diagnostic and post-diagnostic services revealed similarly poor outcomes. Bancroft
(2012) reported that 64% of the autistic adults who took their survey had to wait
between one and three years for a diagnosis after first raising concerns, leading to 55%
of their sample reporting that the process was too stressful for them. Furthermore, only
28% reported receiving useful information about further help and support post-
diagnosis. Jones et al. (2014) describe how many adults have to endure multiple
referrals to different health professionals before receiving their ASC diagnosis. In their
sample of 128 autistic adults, 42.2% were referred more than once; of these, 48.1%
received a diagnosis at the third referral, 20.4% at the fourth referral, 13% at the fifth
referral, while 18.5% attended six or more referrals before being diagnosed. A large
proportion of those diagnosed received no form of post-diagnostic support (41.9%).
Despite many scoring highly for anxiety and depression, 78.6% said they did not know
where to go to access support to help with these symptoms. Satisfaction with the
diagnostic process was most affected by this lack of post-diagnosis information. A
longer time taken to get a diagnosis, a greater number of different professionals seen,
and a higher frequency of referrals all increased overall dissatisfaction. Finally, there is
some evidence that autistic females may be particularly vulnerable to disappointing
20
post-diagnostic support. Bancroft (2012) reported that once diagnosed, 49% of autistic
females said their diagnosis made no difference to the support they received, compared
to 39% of males who also felt this.
These findings raise concerns about the wellbeing of autistic people in the UK,
and point towards a need for earlier identification and the provision of more timely and
appropriate support post-diagnosis, in order to ensure a better quality of life for autistic
adults. This is especially important for individuals receiving diagnosis only in
adulthood, and particularly for females who are more likely to be diagnosed later, and
who will therefore not have received early intervention support. Why females are likely
to be diagnosed later than males, and the impact this diagnostic delay has on them, will
be discussed in the following chapter.
21
22
CHAPTER 2
Gender-Based Theories of ASC
The consistently higher ratio of males to females in prevalence studies led many to
believe that autism was predominantly a ‘male condition’. It was thought that females
had a reduced susceptibility to autism (as described by the Female Protective Factor
[FPF] theory), and that in order to develop autism they needed a greater ‘genetic hit’
(Lord & Schopler, 1985; Robinson et al., 2013; Skuse, 2000). This was supported by
studies demonstrating that autistic females tended to have more autistic relatives than
autistic boys, suggesting that the girls had inherited more ‘severe’ autistic traits than
boys (Tsai et al., 1981; Werling & Geschwin, 2015), and that autistic girls have a greater
resistance to genetic causes of autism (Levy & Perry, 2011). Jacquemont et al. (2014)
analysed the DNA samples of just under 24,000 families affected with either autism or
other neurodevelopmental disorders, finding that females diagnosed with either of these
conditions had a higher number of damaging gene mutations than males. From this
theory another theory was born, ‘The Extreme Male Brain’ (EMB) theory (Baron-
Cohen, 1999), which has become one of the most prominent theories explaining gender
differences in autism. The EMB theory builds on the former FPF theory to suggest that
autistic traits are gender specific and are extreme versions of typically male
traits/behaviours, and that therefore females need a greater genetic hit than males in
order to develop autism. This theory again suggests that when females are affected they
may be affected to a greater extent, thus explaining why there is less of a gender
disparity in the frequency of autistic individuals with intellectual impairments and
comorbid disabilities. However, a newer theory (the Female Phenotype Theory [FPT])
(Kopp & Gillberg, 1992) suggests that there are actually more autistic females than
previously thought, and that the gender disparity in diagnosed cases is due to autistic
23
females manifesting autistic traits in a different way to autistic males. Currently,
diagnostic assessments and criteria are based on the pattern of traits observed in autistic
males, which may mean clinicians are biased towards looking for these and may miss a
different presentation of autistic traits in females.
This chapter will focus on reviewing these two dominant ideas, a) that autism
could be an extreme version of the male brain which females are biologically less likely
to be susceptible to, and b) the idea that autistic females are not being identified
correctly due to having a different presentation of autistic traits.
2.1. Extreme Male Brain Theory
One of the most influential accounts of the gender disparity in autism is the EMB
theory. According to the EMB theory, autism is an extreme version of the male brain
such that sexually dimorphic traits which are particularly strong or weak in non-autistic
males are accentuated in autistic people (Baron-Cohen, 2012). The cause of this is
thought to be foetal testosterone (fT). Hormonal influxes during certain critical periods
of a foetus’s life can significantly alter cognitive development, and testosterone in
particular can produce permanent behavioural changes if a foetus is exposed to it during
critical periods of gender development (Hines, 2006). For fT, this critical period is
thought to be when there is a surge occurring between weeks 8 to 24 of gestation
(Baron-Cohen, Knickmeyer, et al., 2005). FT therefore plays an organizational role in
the development of masculine and feminine traits, in that it has a permanent effect on
early development. Some studies have found fT to be elevated in both autistic males and
females (Bejerot et al., 2012; Ingudomnukul et al., 2007; Tordjman et al., 2006) and
another study found fT to be correlated with autistic traits in the general population
(Auyeung et al., 2010). However, as this review will go on to explain, the evidence
supporting the link between fT and autism is highly inconsistent. The theory proposes
24
that as males already have more testosterone, they are considered to be more vulnerable
to elevated levels leading to autism. Females are less susceptible to autism as a result of
lower testosterone and, as a consequence, when they are affected it is to a much greater
extent. This partly explains why the gender ratio at the lower end of the spectrum,
where individuals often have accompanying intellectual impairments, is much lower
(Lord & Schopler, 1985; Tsai et al., 1981).
The EMB theory states that the two sexually dimorphic traits that are integral to
autism are systemising and empathising. Autistic people are found to show greater
abilities to systemise, which is the ability to analyse and construct systems, and reduced
ability to empathise, which is the ability to understand and feels others’ emotional
states. These two dimensions are viewed as distinct, although there is generally a mild
negative relationship between them such that higher levels of systemising are associated
with lower levels of empathising and vice versa (Greenberg et al., 2018). In fact, some
studies have suggested that there may even be a neurobiological link whereby there is a
trade-off between the two abilities in non-autistic males and females (Goldenfield et al.,
2005), which has been found to be even more pronounced in autistic people (Baron-
Cohen et al., 2003; Wheelwright et al., 2006).
2.1.1. Empathising. Empathising is the ability to identify and understand
another’s emotional state (cognitive empathy) and to feel what others may be feeling
(affective empathy). Non-autistic females typically demonstrate higher empathy
abilities than non-autistic males (Manson & Winterbottom, 2011; McClure, 2000;
O’Brien et al., 2013; Reniers et al., 2010; Thompson & Voyer, 2014), and autistic
individuals demonstrate a deficit (Baron-Cohen et al., 2003; Hoffman, 1977; Krajmer et
al., 2010). Baron-Cohen and Wheelwright (2004) created the Empathy Quotient (EQ)
self-assessment questionnaire. A factor analysis has established that it measures both
25
affective and cognitive aspects of empathy, as well as social skills, in adults (Lawrence,
et al., 2004). In a recent large-scale study, which tested the EQ alongside other measures
in more than 670,000 people, non-autistic females scored on average higher than non-
autistic males, with a medium effect size (d = 0.39), and autistic people scored
significantly lower than the non-autistic participants, also with a medium effect size (d =
0.41). These findings have been replicated in several smaller studies (Auyeung et al.,
2009; Baron-Cohen et al., 2003; Baron-Cohen & Wheelwright, 2004; Lawrence et al.,
2004; Sucksmith et al., 2012; Wheelwright et al., 2006). In those studies which used a
representative sample of autistic females as well as autistic males, no gender differences
were found on the EQ, contrary to the previous prediction of the EMB theory (Auyeung
et al., 2009; Greenberg et al., 2018; Wheelwright et al., 2006). However, Sucksmith et
al. (2012) did find that autistic girls scored higher than autistic males on the Karolinska
Directed Emotional Faces Tasks (KDEFT), where participants had to guess what people
in photographs were feeling. In another study where teachers rated empathic traits in
children, the autistic girls were rated as being more empathic (Peterson, 2014).
When empathy is broken down into its two main components, affective and
cognitive empathy, it appears that rather than a global deficit in autistic people, there
may be a specific difficulty in cognitive empathy (i.e., interpreting and reading emotion)
while affective empathy may remain intact (Mazza et al., 2014; Mul et al., 2018).
Cognitive empathy, the ability to read and understand what others may be
thinking and feeling, has been linked to Theory of Mind (ToM), which itself refers to
the ability to recognise and attribute mental states to others (perspective taking).
Researchers have described how the process of cognitive empathy may rely on ToM, as
it requires one to take another’s perspective in gauging their current emotion (Stietz et
al., 2019). However, it should be noted that ToM comprises different factors also, and
26
whilst one part of ToM may involve the ability to infer what others may be feeling, a
distinct part of ToM is the ability to infer another person’s beliefs, thoughts, and
intentions. Indeed, studies have found that individuals may perform differently on these
distinct elements of ToM, and that different brain regions may be involved (Dvash &
Shamay-Tsoory, 2014). ToM is commonly found to be impaired in certain degrees in
autistic people, which may contribute to difficulties with empathising, particularly with
cognitive empathy (Brewer et al., 2017; Happé, 1994; Joliffe & Baron-Cohen, 1999;
Mathersul et al., 2013; Mazza et al., 2014). Baron-Cohen et al. (1997) studied 50 non-
autistic adults and sixteen adults with ‘high functioning’ ASC or Asperger’s (13 males
and 3 females), using tasks that require the inference of ToM from photographs of a
person’s eyes (Reading the Mind in the Eyes’ Test [RMET]). Their results showed that
non-autistic females performed significantly better than non-autistic males and that non-
autistic subjects performed significantly better than the autistic subjects, indicating that
autistic people performed significantly lower than non-autistic males, which is in line
with the EMB theory. Whilst these studies had quite low participant numbers, which
reduced their power, Baron-Cohen et al. (2015) tested 395 autistic adults (178 males,
and 217 females) and 320 non-autistic controls (152 males, and 168 females) in an
online study using the EQ, AQ, and the RMET. As predicted, the autistic participants
scored significantly worse than the controls on the RMET. In terms of gender, control
males performed significantly worse than control females on this task (d = 0.47), but
there was no difference between autistic males and autistic females. An interesting
finding was that the difference between control females and autistic females had a
greater effect size (d = 0.69) than between control males and autistic males (d = 0.35),
which the authors suggest may be because females need to have a higher number of
autistic traits to get diagnosed. When assessing the association between RMET scores
and self-reported empathy and autistic traits on the EQ and the AQ respectively, only
27
autistic females’ scores showed a significant correlation, which the authors suggest may
indicate a heightened self-awareness of cognitive empathy difficulties in autistic
females.
Affective empathy on the other hand, appears to remain relatively intact in
autistic individuals (Mul et al., 2018). For example, Dziobek et al. (2008) tested 17
autistic adults (13 males and 4 females) using the Multifaceted Empathy Test (MET),
and the Interpersonal Reactivity Index (IRI). The MET uses a series of photos of people
in emotional states; participants are asked to label the mental state of the person
(cognitive empathy) and also to rate their own emotional reaction to the picture
(affective empathy). They found that the autistic participants scored significantly lower
than non-autistic controls on the cognitive empathy part of the MET and IRI, but scored
similarly to non-autistic controls on the affective empathy part of the MET and the IRI.
As well as this, measures of the participants’ arousal when looking at the stimuli were
similar for the two groups. These results, however, may have been due to a response
bias in how autistic people rated their own emotional state in response to the images, as
between judging the mental state and responding with their own emotional reaction,
they were told the correct emotional state in the photograph.
A study which used a comprehensive set of physiological markers to determine
affective empathy is that by Trimmer et al. (2017), who evaluated the relationship
between self-reported empathic responses and physiological responses, as well as how
these related to self-reported trait empathy in ASC. They showed 10 video clips (half
emotional and half neutral) to 25 ‘high-functioning’ autistic participants (21 males and 4
females) and 25 non-autistic participants (20 males and 5 females). Whilst participants
were watching the clips, the researchers tested their automatic responses using skin
conductance level (SCL) and facial electromyography (EMG), which measures muscle
28
activity in the face for automatic emotional contagion response. Self-rated mood and
arousal, and IRI and EQ scores were also assessed. The findings revealed that the
autistic participants scored lower on both the cognitive and affective factors of the EQ
and IRI, and these participants also reported a reduced emotional response to the clips.
However, the autistic and non-autistic participants did not differ in their physiological
responses to the emotional stimuli, nor did their ratings of perceived arousal. These
findings suggest that the empathy deficit in autism may actually lie in autistic
individuals’ ability to interpret the emotional salience of the physiological response they
have experienced, rather than their ability to experience it.
These empathy differences do not appear to be very strongly related to fT. There
is some evidence that in non-autistic populations, scores on the EQ and RMET correlate
with levels of fT in the amniotic fluid of mothers (Chapman et al., 2006; Knickmeyer, et
al., 2005), but these results could reflect general gender differences rather than fT. Other
evidence demonstrates that injecting non-autistic women with testosterone results in a
reduction of empathic behaviours (Hermans et al., 2006; van Honk et al., 2011),
however, these findings represent temporary changes and not permanent and lifelong
developmental changes. In autistic populations evidence indicates that fT is not linked
to empathy deficits or other autistic traits (Bakker-Huvenaars et al., 2020; Honekopp,
2012; Krajmer et al., 2011; Kung et al., 2016; Voracek & Dressler, 2006; Whitehouse et
al., 2012). This calls into question whether the EMB can claim that empathy
impairments or autistic traits in autistic people are the result of an ‘extreme male brain’
caused by excess fT. Furthermore, it is not yet possible to test the hormonal levels of an
unborn foetus, and thus the direction of cause and effect regarding the relation of fT to
early development cannot be determined (Fine, 2010). A study by Bejerot et al. (2012)
even found an opposite pattern of findings; whilst the sample of 24 autistic females did
demonstrate elevated levels of testosterone and masculinised characteristics, such as
29
less feminine facial features, the sample of 26 autistic males displayed more feminised
characteristics, such as less masculine body types and voice quality.
This section has discussed findings which indicate that certain aspects of
empathising may be impaired in autistic people. However, the evidence does not
strongly support some aspects of the EMB theory of autism and there are some
conflicting findings. Furthermore, the empathy deficits observed in autistic people may
have different causes to the disadvantage that non-autistic males show on empathy
measures compared to non-autistic females (Bird et al., 2010).
2.1.2. Systemising. Systemising is the second sexually dimorphic trait in the
EMB theory. Systemising involves being able to analyse and construct systems that take
in inputs and produce outputs based on their operation and the rules that govern them.
This ability shows the opposite pattern to empathising: it is thought to be heightened in
non-autistic males relative to non-autistic females, and even more so in autistic
individuals (Krajmer et al., 2010; Manson & Winterbottom, 2011). Large scale surveys
using the Systemising Quotient (SQ) have indicated that males in the general population
score higher than females, and that autistic people score even higher, with no significant
difference between autistic males and autistic females (Baron-Cohen et al., 2003;
Greenberg et al., 2018; Wheelwright et al., 2006;). Further research has shown that non-
autistic males and autistic people perform better than non-autistic females on tasks such
as mental rotation and figure disembedding, which require a systemising approach to
identify a specific shape from a larger image (Baron-Cohen & Hammer, 1997; Collins
& Kimura, 1997; Jolliffe & Baron-Cohen, 1997; Voyer et al., 1995).
Autism has also been found to be associated with STEM fields of study and
work, which are typically male-dominated fields thought to involve high levels of
systemising (Baron-Cohen, 1999; Beede et al., 2011; Sassler et al., 2017; Weelwright et
30
al., 2006). For example, Baron-Cohen (1998) screened families of students studying
either maths, physics, and engineering (STEM students) or literature (non-STEM
students) for autistic relatives, finding that 6/641 STEM students had autistic relatives,
and only 1/652 literature students had an autistic relative. However, it is important to
note that the prevalence of autistic relatives in the STEM subjects was only 0.94%,
which is no higher than the general prevalence rates discussed previously. As literature
was the only non-STEM subject tested, it is difficult to conclude that generally students
in non-STEM subjects are less likely to have autistic relatives. Furthermore, it may not
be the case that certain subjects involve more systemising than others, particularly as
studying all subjects in academia involves some level of systemising (Fine, 2010). For
example, Ruzich et al. (2015) found in their large sample of 450,394 adults that careers
in STEM areas were associated with increased AQ scores in both non-autistic males and
females, and that males scored significantly higher on the AQ than females. However,
non-STEM careers included business, sales, transport, finance and banking amongst
others, which could be said to require high levels of systemising. Wei et al. (2013) also
found a gender difference between males and females in STEM and non-STEM fields.
However, this was in autistic participants, with 39% of male autistic students majoring
in a STEM field and only 3% of females majoring in a STEM field, compared to 29%
of non-autistic male college freshmen and 15% of non-autistic female college freshmen.
Furthermore, in the large-scale study on 670,000 autistic and non-autistic people by
Greenberg et al. (2018), autistic people were not more likely to enter STEM fields,
suggesting that an ‘extreme male brain’ may not be the cause of some autistic people’s
preference for STEM subjects. Others have questioned gendering fields and skills as
being ‘male-minded’, as the EMB theory promotes, on the basis that more males are in
them or better at them (Ridley, 2016). It may be the case that socialisation and a
society’s gender norms affect the number of females entering STEM careers (Charles &
31
Bradley, 2009; Milkman et al., 2012; Moss-Racusin et al., 2012; Xu, 2008), or it may be
a combination of both nature and nurture factors.
The evidence that systemising in autism is an ‘extremely male’ trait linked to
excess fT is also inconsistent. For example, Falter et al. (2008) found that the aspect of
the mental rotation task autistic people seemed to excel at was different to that of non-
autistic males, and they did not find a link between testosterone and performance on
these tasks. However, Brosnan et al. (2010) did find a correlation between ‘time awake’,
which is used as a proxy for circulating testosterone with peak levels occurring in the
morning and declining throughout the day, and both systemising and mental rotation in
a non-autistic population. Note, though, that the direction of cause and effect between
time awake, circulating testosterone, and systemising is unclear. There were no
statistically significant differences between non-autistic males and females on time
awake, and measuring time awake could introduce many other confounding variables,
such as concentration and fatigue levels, as well as exercise, protein intake, and time of
reproductive cycle, which are all known to affect levels of circulating testosterone
(Hulmi et al., 2008; Schoning et al., 2007).
Whilst the EMB theory does, once again, raise important findings highlighting a
difference in both empathising and systemising ability in the autistic population, the
evidence that systemising is an example of an ‘extreme male brain’ caused by excess fT
is uncertain. Furthermore, there may be other reasons why autistic people systemise, for
example, repetitive and restrictive behaviours may favour a systemising approach, and
systemising may also help autistic people manage confusing and complicated social
structures and systems. As suggested previously, systemising may also be used as a
trade-off for impairments in empathising (Goldenfield et al., 2005).
32
2.1.3. Additional limitations of the EMB theory. Based on the evidence
discussed in this section it is highly likely that other factors may also be at play in the
development of autism. Whilst there do appear to be differences in empathising and
systemising ability between those who are autistic and those who are not, these are not
core impairments featured in the DSM criteria for ASC (APA, 2013). Ridley (2019)
argues that collating empathising with systemising is not justified, likening describing
an autistic woman as having an ‘extreme male brain’ because she scores highly on
systemising and poorly on empathising is similar to describing an extremely tall female
as having ‘extreme male tallness’, because men are more likely to be tall. To take this
analogy further, an extremely tall woman may have an abnormality, which has increased
her height compared to the average female. It is an essentialistic fallacy to describe this
woman as having ‘male-tallness’, particularly as the reason for her height is different to
the reason why an average male is generally taller than an average female. In a similar
respect, the reason an autistic woman may have a similar cognitive profile to the
average non-autistic male may be for very different reasons, and it is limiting to
categorise this as an ‘extreme male brain’.
Furthermore, Ridley (2016) stresses the importance of taking into account that
no research on gender and brain anatomy has identified exactly what a ‘male brain’ or
‘female brain’ looks like. Instead, research by Daphna et al. (2015) suggests that the
human brain is a ‘mosaic’ of different unique features, which cannot be categorised as
either ‘male’ or ‘female’. Similarly, Ridley (2016) argues that autistic traits can be the
product of any brain, regardless of gender, and that we should broaden our investigation
into autism beyond gender. However, Greenberg et al. (2018) have stressed that the
EMB theory merely describes averages, and inferences should only be made about
males and females as groups rather than for individuals. Whilst this may be true, Krahn
and Fenton (2012) warn that an adverse effect of categorising autism as an ‘extreme
33
male brain’ is that it may have led to many autistic girls not being diagnosed, as
clinicians may have been biased in looking for ‘male’ signs of the condition. The
following section will address possible differences in how autistic males and autistic
females present on a behavioural level, offering an alternative theory that may explain
the gender disparity found in autism.
2.2. Female Phenotype Theory
The FPT suggests that rather than males being more likely to develop autism, autistic
females are instead going unidentified due to presenting differently with a number of
different and disguised observable characteristics (Kopp & Gillberg, 1992). Due to
current diagnostic criteria and measures being based primarily on male samples, it is
argued that many clinicians are unable to detect the phenotype seen in many autistic
females, explaining figures discussed earlier showing later diagnosis in females
(Baldwin & Costley, 2016; Shattuck et al., 2009). There could be a number of reasons
why autistic women present differently with the same condition, including both
biological and environmental causes. These will be discussed in more detail later in this
chapter.
2.2.1. Presentation of autistic characteristics in males and females. There is
conflicting evidence regarding differences in the autistic traits and symptoms displayed
by males and females. An early study by McLennan et al. (1992) testing 42 autistic
females and males (equally split) with a mean age of 14-15 years, using the Autism
Diagnostic Interview (ADI), found that parents of autistic daughters reported that their
child was less affected by social and communication behaviour difficulties than parents
of autistic sons. This was particularly prominent in the areas of social initiative play and
also comfort-seeking and offering. However, when these children became adolescents
this pattern was reversed, with autistic females demonstrating more severe social
34
difficulties, predominantly in peer relationships, compared to autistic males. The authors
suggest that this may be due to the greater social demands placed upon adolescent girls,
whereby peer activities rely on social communication and interest. However, it should
be noted that in this study the autistic girls had spent a significant amount of time in
special needs classrooms, which may have hindered their ability to learn socially from
non-autistic girls. As well as this, slightly different measures had to be used for different
time periods, as the younger and older versions of the ADI did not align at that time,
which may have led to some discrepancies.
More recent studies have supported the finding that girls may present with fewer
social communication difficulties. For example, Hsiao et al. (2013) evaluated social
deficits in autistic children and adolescents. A sample of 1,321 students aged 6-15 years
from schools in Taiwan were tested, with an equivalent number of males and females.
Generally, the study discovered that autistic children and adolescents were more likely
to exhibit social deficits than their non-autistic peers. However, autistic boys of all ages
were significantly more impaired than autistic girls on social awareness, with older girls
being more impaired on social emotion than younger girls. Likewise, Hiller et al. (2014)
found subtle differences in how autistic boys and girls behaved socially. They tested a
sample of 69 autistic girls and 69 autistic boys (M = 8-9 years) and measured how the
children met the broad social criteria on the DSM-5 using both clinician and teacher
reports. Findings showed that autistic girls were 14 times more likely than autistic boys
to engage in typical reciprocal conversation; a much larger percentage of girls (35%)
than boys (9%) showed virtually no impairments in their ability to integrate nonverbal
and verbal communicative behaviours; girls were 3.5 times more likely to engage in
imaginative play typical for their developmental level than boys; and finally girls were 6
times more likely than boys to show some adjustment of their behaviours across
situations, such as monitoring voice volume, avoiding inappropriate comments, and
35
hiding emotional meltdowns. In a study of 16 autistic girls and 17 autistic boys aged
between 5-10 years, Rynkiewicz et al. (2016) found that the autistic girls also tended to
use nonverbal gestures more vividly than autistic boys when assessed using the ADOS-
2. Finally, research by Parish-Morris et al. (2017) found that school-aged autistic girls
(n = 16) used more pragmatic language markers than autistic boys (n = 49), and at a
level similar to that found in non-autistic children, which may normalise the way
autistic girls sound when communicating and thereby disguise communication
difficulties. On the whole, autistic girls do appear to show an advantage over autistic
boys in social communication skills, which may be part of the female phenotype of
autism.
In contrast to these findings, there are many studies which show that for autistic
children without intellectual disability, autistic girls appear to experience the same
severity of autistic traits on assessments used to diagnose autism as autistic boys (May
et al. 2014; McLennan et al. 1993; Postorino et al., 2015). For example, Rivet and
Matson (2011) found no gender differences in autism symptomology on the Autism
Spectrum Disorders – Diagnostic – Child Version (ASD-DC) or the DSM-IV-TR/ICD-
10 Checklist for 37 autistic girls and 37 autistic boys (ages 3-17 years), as rated by
parents, caregivers, and teachers on several domains (nonverbal
communication/socialisation, verbal communication, social relationships, and insistence
on sameness/restricted interests). Similar findings were made by Reinhardt et al. (2015)
using 54 young autistic girls and 234 young autistic boys who were recruited from
paediatric patient lists, those with older autistic siblings, and those referred because of
suspected autism. They used a variety of measures to determine gender differences in
early social communication abilities, an infant cognitive functioning measure, and a
parent interview to assess different domains of adaptive behaviour (communication,
daily living skills, socialisation, and motor skills), finding no differences. These results
36
were supported by similar studies using smaller numbers of young participants
(Postorino et al., 2015). Furthermore, Harrop et al. (2015) found no differences between
40 autistic girls and 40 autistic boys aged 36-48 months in spontaneous play with a
stranger and non-verbal and verbal communication.
Whilst these studies predominantly used measures and scales that rely on
parental report, other studies have used the Autism Diagnostic Observation Schedule
(ADOS). For example, Hartley and Sikora (2009) tested 157 autistic boys and 42
autistic girls between the ages of 1.5-3.9 years using several parent-report measures of
adaptive behaviour traits and cognition alongside the ADOS, finding similar patterns of
traits and behaviours across girls and boys. Furthermore, in a study by Mussey et al.
(2017), for which 113 autistic females and 566 autistic males were tested on the ADOS,
the Childhood Rating Scale, and a developmental measure, no gender differences were
found in overall scores or in age of diagnosis (M = 10-11 years of age).
These conflicting findings may be due to the young ages of the samples used
and also the origin of the samples. Whilst some autistic girls may present typically,
others may have the female phenotype and may not present typically. Also this age
group is less likely to capture those with the female phenotype as they may have been
diagnosed later. Investigating the presentation of autistic characteristics in undiagnosed
autistic girls and in autistic adults reveals that autistic females may develop less overt
autistic characteristics, as described next.
Lai et al. (2011) tested 45 autistic males and 38 autistic females presenting at a
diagnostic clinic for adults in Cambridge on both the ADI-R and the ADOS. Males and
females were similar in terms of childhood autistic symptoms, as found previously,
although the researchers did select only those participants who had the same
behavioural criteria, e.g. reached the same ADI-R cut offs. However, whilst no
37
differences were found between males and females in empathising, systemising, or
mentalising (ToM), females demonstrated less severe socio-communication difficulties
on the ADOS and more lifetime sensory issues, and during immediate interpersonal
interactions the females also showed fewer autistic behaviours in the socio-
communication (r = 0.41) and RRBI domains (r = 0.50). A more recent study by Wilson
et al. (2016), reported similar findings. They tested 935 adult males and 309 adult
females referred for autism assessments by their GPs, finding a pattern of greater social
and communication difficulties and RRBIs in males who were subsequently diagnosed
with autism compared to females who were subsequently diagnosed with autism. These
findings suggest that, compared to autistic females, autistic males present with more
overt autistic behaviours, such as RRBIs, and greater social difficulties, which make
them stand out more for diagnosis. Indeed, evidence that RRBIs appear to a much
greater extent in autistic males than autistic females has been found consistently in a
large body of research (Duvekot, 2017; Frazier et al., 2014; Hartley & Sikora, 2009;
Hattier et al., 2011; Hiller et al., 2014; Hsiao et al., 2013; Lai et al., 2011; Mandy et al.,
2012; May et al. 2014; Park et al., 2012; Ratto et al., 2018; Sipes et al. 2011).
Looking more closely at research investigating the autistic behaviours and traits
of males and females it would seem that a key difference lies in externalising and
internalising traits. For example, findings cited earlier suggest that males have more
RRBIs than females, which includes more visible external traits. Other studies support
these findings, showing that generally autistic boys display more externalising
challenging and hyperactive behaviours (Giarelli et al., 2010; Levy et al., 2005), and
also that higher levels of reported emotional and behavioural problems predict an ASC
diagnosis more often in girls than in boys (2.44 times) (Duvekot et al., 2017).
Dworzynski et al. (2012) suggest that in order for girls to be diagnosed with autism they
require a greater number of external behavioural problems than boys. Their study drew
38
on a large data pool of approximately 11,000 families from TEDS, which is a UK based
study of twins born between 1994 -1996, and focussed on 189 autistic children who met
diagnostic criteria when they were between 10-12 years of age (29 females and 160
males), and a group of 174 children (55 females and 119 males) who scored above the
cut-off on the Childhood Autism Spectrum Test (CAST) but who did not meet the full
diagnostic criteria; this sample was referred to as the “high-CAST” group. The
diagnosis rate for boys with high CAST scores who went on to be diagnosed was found
to be 56%, however it was significantly lower for girls at 38%. For both genders, “high-
CAST” children had significantly fewer social autistic traits than diagnosed children,
demonstrating that better social skills may hinder diagnosis for both genders. However,
“high-CAST” girls were significantly more prosocial than “high-CAST” boys. They
also had significantly lower reports of hyperactivity and behavioural problems than
diagnosed girls, whereas there were no differences between “high-CAST” boys and
diagnosed boys in these domains. Furthermore, diagnosed girls were 8.4 times more
likely than “high-CAST” girls to show cognitive and behavioural difficulties. This
suggests that in order for girls to be diagnosed they require more overt challenging
behaviours and problems, and that their internalising of traits may contribute to them
missing diagnosis.
These studies stress the importance of investigating undiagnosed females with
high levels of autistic traits, who may be undiagnosed due to exhibiting less challenging
and external behaviours. The majority of studies investigating differences between
autistic males and females rely on already diagnosed individuals, which means that the
females will have displayed enough autistic traits to be sent for diagnosis. This may bias
the findings as greater differences may be found if females scoring highly on measures
of autism but who do not have a diagnosis are investigated as well. Also, it should be
noted that there were significantly fewer autistic girls tested in many of these studies
39
compared to autistic boys (e.g, Parish-Morris et al., 2007; Reinhardt et al., 2015), which
affects the overall power of these findings and may lead to incorrect rejection of the null
hypothesis (Type 1 error) (Rusticus & Lovato, 2014). Although, other studies have used
equal numbers of autistic boys and girls, and therefore support the conclusions made
from these more gender-biased studies (e.g. Lai et al., 2011). Due to fewer females
being diagnosed with autism, gaining equal numbers of autistic males and females
remains a methodological challenge for studies looking at gender differences in autism.
It is therefore important that future studies attempt to gain equal sample sizes, and to
ensure equal variance between these groups before comparisons are made.
2.2.2. Gender Socialisation and the presentation of autistic symptoms. It has
been suggested that one of the reasons that autistic girls exhibit better social
communication skills and more internalised difficulties than autistic boys is because of
gender socialisation pressures (Krahn & Fenton, 2012). In the development of social
skills for all children, socialisation plays a key role in gender differences in behaviours
(Bem, 1981). Ryle (2011) describes gender socialisation as a learning process of
understanding both gender norms and one’s own gender identity. Gender norms refer to
sets of rules about what society believes is masculine and what is feminine, whilst
gender identity refers to how individuals think of themselves as male or female (John et
al., 2017). Bandura (1963) developed the theory of social learning, part of which
involves the learning of ‘sex-typical’ behaviours. Children are often rewarded when
they conform to the correct sex-typical behaviour for their gender, which reinforces
these behaviours. The gender norms in Western cultures have historically stereotyped
males as being aggressive, dominant, leaders, independent, decisive, assertive, and self-
reliant, amongst other traits (Bem, 1974). In contrast, females have typically been
stereotyped as being gentle, sympathetic, shy, sensitive to others’ needs, compassionate,
soothers of hurt feelings, affectionate, and even childlike, amongst other traits (Bem,
40
1974). Miller et al. (1981) describes how the female sense of self is often derived from
how she is connected to others, whilst the male sense of self is often derived from his
independence from others. Although the feminist movement has meant society is
becoming more aware of the possible social construct of gender, it remains ingrained in
much of our society (Fine, 2010). Therefore, it is likely that just as the general
population experiences social learning of gender norms that affect behaviour, autistic
males and autistic females also experience this, shaping how their autistic traits manifest
themselves at a behavioural level. This may mean that autistic females are motivated to
fit in more socially, to behave better, and to be more introverted and empathic towards
others than autistic males might be.
Evidence of heightened expectations for autistic girls to behave in a socially
acceptable manner comes from studies that have found that parent ratings of their
child’s social functioning are often lower for autistic girls than autistic boys.
Specifically, even in the absence of gender differences detected by the researchers, or
with females demonstrating enhanced abilities compared to males, parents of autistic
daughters often rate their child as having more severe social problems than parents of
autistic sons. For example, Holtmann et al. (2007) did not find any significant
differences between 23 autistic girls and 23 autistic boys, with a mean age of 11 years,
on the ADI-R, the ADOS, or the Child Behaviours Checklist. However, parents reported
significantly more social problems in girls than in boys, suggesting some bias in the
level of social competence expected in daughters by their parents. Similarly, in
Rynkiewicz et al.’s (2016) study, despite autistic girls performing better than autistic
boys on social nonverbal communication aspects of the ADOS-2, in the Social
Communication Questionnaire (SCQ) the parents of autistic girls rated them as having
significantly poorer social skills than the parents of autistic boys. Ratto et al. (2018)
investigated this phenomenon further, comparing gender differences in the ADOS and
41
ADI-R with parental reports, in 114 school-aged autistic girls and 144 IQ and aged
matched autistic boys. Approximately 90% of the girls and 94% of the boys met the cut-
off criteria for autism on the ADOS, with similar scores across all domains. The girls
and boys also scored similarly on the ADI-R, although fewer numbers of both met the
cut-off criteria on this (73% of girls and 76% of boys). However, on the Social
Responsiveness Scale (SRS), which was completed by the parents, the girls were rated
as being significantly more impaired across all domains, including social awareness,
social information processing, capacity for reciprocal social communication, social
anxiety/avoidance, and autistic preoccupations and traits. The authors suggest that it
may be the case that parents expect girls to be more socially competent than boys, and
therefore any impairments may be emphasised more severely. A potentially interesting
secondary finding was that the girls who had higher cognitive abilities were more likely
not to meet the ADI-R criteria, particularly girls of higher intelligence, once again
suggesting that many girls with autism may fail to be diagnosed due to not meeting
diagnostic thresholds as they have a different manifestation of autistic traits.
2.2.3. Camouflaging autistic traits. A potential consequence of socialisation
pressures in autistic girls is that they may feel it is necessary to mask their autistic traits,
compensate for them, and act in a more desirable way by camouflaging. An emerging
area of research in support of the FPT suggests that one of the primary reasons that
females do not appear ‘autistic’ to others, and therefore why they may be undiagnosed
or diagnosed much later, is that females camouflage their autistic traits. This can be seen
in the masking of autistic characteristics and in the act of camouflaging to fit in with
others socially (Attwood & Grandin, 2006). Initially this theory was grounded in a large
body of qualitative data and anecdotal evidence from autistic females and their parents,
but more recently attempts have been made to measure camouflaging empirically and
the characteristics and skills associated with it. Livingston and Happé (2017) have
42
recently proposed a transdiagnostic framework to conceptualise compensation in ASC,
which will help ground further empirical research into the camouflaging effect in
autism, of which compensation is a large part. This acknowledges the research finding
that the core autistic difficulties are the same for all genders, but suggests that
compensation may affect the presentation of these in various situations. Three
hypothetical features of compensation are outlined in this framework, namely,
compensation may be shallow or deep, it may be modulated by the environment, and it
may come at some cost. These features will be discussed later on in the chapter.
Tierney et al. (2016) conducted interviews with ten autistic adolescent females
and analysed their responses using Phenomenological Analysis to investigate the girls’
experiences of managing social relationships. The majority of the girls mentioned some
form of imitation, for example, carefully observing peer interactions to build a social
repertoire and rules they could follow. They would often copy facial expressions,
postures, tone of voice, topic of conversation, and choice of interests in order to fit in.
Masking was reported by many of the girls, describing how they would often ensure
they maintained either happy or blank facial expressions when socialising in order to
hide how unhappy and anxious they often felt; this mask was maintained even in close
friendships out of a fear of losing their friends. These strategies appeared to be so
successful in hiding external signs of distress that those around them were surprised to
find out they were in fact struggling. Similarly, Bargiela et al. (2016) found a common
theme of ‘pretending to be normal’ from 14 autistic women (aged 22-30) who were
diagnosed in late adolescence or adulthood. Many of these women struggled with
socialising but had coped by ‘wearing a mask’, which they described as a conscious
effort to hide their autistic traits, as well as reporting social mimicry, which they
described as being more automatic. Furthermore, Baldwin and Costley (2016) found in
the open comments section of their survey on 82 autistic women that a large number
43
suggested they had purposefully learned aspects of socialising to enable them to act
appropriately. Furthermore, a study by Hull, Petrides, et al. (2017), focussing on adults,
examined the qualitative camouflaging experiences of 55 autistic women, 30 autistic
men, and 7 autistic individuals identifying as ‘other gender’, with a mean age of 43.
They discovered common themes of motivation to camouflage, which included a need
to ‘blend in with the ‘normals’’, which they felt was an expectation of them made by
others, as autistic behaviours were viewed as ‘unacceptable’. As well as this, many saw
camouflaging as a way to overcome social hurdles in forming the relationships they
desired with others. In order to mask autistic traits, many reported mimicking the
behaviour of others during social situations, some even copying social interactions from
television programmes and films. Additionally, many reported developing behaviours to
compensate for social communication difficulties, for example, using non-verbal
gestures such as maintaining appropriate levels of eye contact, avoiding dominating
conversations with details about themselves and interests, and practising conversations
beforehand so that they could maintain a social script. Imitation of social behaviours has
also been reported in semi-structured interviews by the mothers of autistic adolescents,
who believed that they found the process of obtaining diagnoses for their daughters
more challenging as a result (Cook et al., 2017; Cridland et al., 2014; Rabbitte et al.,
2017). These qualitative reports demonstrate camouflaging as an important aspect of the
female phenotype of autism. However, from these studies alone it is difficult to
determine if camouflaging is a female specific strategy and whether it does contribute to
a delayed diagnosis for women.
Lai et al. (2017) were the first researchers to attempt to quantify camouflaging.
They used a sample of 60 age and IQ matched adult autistic males and females to
determine the difference between their external behaviours in a social context (as
measured with the ADOS) and their internal and self-reported traits (as measured with
44
the AQ and RMET). Two scores were calculated from this, the first was the difference
between self-rated autistic-like traits and external behaviours (AQ – ADOS), and the
second the difference between mentalising and external behaviours (RMET – ADOS).
The study found that females had a significantly higher score than males, a group
difference that had a very large effect size (d = 0.98). The authors suggest that this
difference is most likely due to gender specific socialisation pressures in girls. However,
this study has several limitations; for example, previous research has shown that women
generally tend to rate themselves as being higher on the AQ, which could be because
they are more self-aware (Lenhardt et al. 2016; Lai et al., 2013; Lai et al., 2011).
Finally, as the study does not directly measure camouflaging; there may be other factors
responsible for this discrepancy between external and internal scores.
Dean et al. (2017) used an observation method to determine whether 96 autistic
and non-autistic elementary school children (48 girls and 48 boys) showed
camouflaging type behaviours in the playground. They found that generally both autistic
girls and non-autistic girls participated in significantly more ‘joint engagement’ than
boys and little time in ‘game’, with talking being the preferred activity for autistic girls.
However, autistic girls still spent significantly more time in ‘solitary’ than non-autistic
girls, and flitted between activities. This was considered by the authors to be evidence
of social compensation; for example, the girls may flit between ‘joint engagement’ and
‘solitary’, demonstrating that they are struggling socially but still attempting to fit in
with the ‘normal’ girls’ activity. During ‘game’, they were also witnessed as always
having a background role, which meant they were taking part but often from the side-
lines. In contrast, autistic boys tended to spend a significantly larger proportion of time
in ‘solitary’ and the non-autistic boys spent more time in ‘game’. The social
environment provides more opportunity for the girls to fit in, and girls tended to
maintain close proximity to where the social groups were forming. This made it difficult
45
from an outsider’s perspective to notice that autistic girls were struggling at all and thus
masking their social impairments, whereas autistic boys situated far away from their
peers and on their own were much easier to spot. These findings are supported by
Sedgewick et al. (2016) who assessed 13 autistic girls, 13 non-autistic girls, 10 autistic
boys, and 10 non-autistic boys aged between 12-16 for gender differences in friendship
motivation and experience. Key findings included autistic girls having similar scores to
non-autistic girls on the social motivation subscale of the Social Responsiveness Scale
(SRS-2), whilst autistic boys had significantly lower scores than non-autistic boys (d =
1.72) and autistic girls (d = 0.89), indicating lower social motivation. This same pattern
was observed on the subscale of closeness using the Friendship Qualities Scale (FQS),
with autistic boys reporting less intimacy with their best-friends than did autistic girls (d
= 1.15). Furthermore, in qualitative interviews with the participants, the girls described
their friendships as focussing on shared talk significantly more than shared activities,
which was not apparent for the autistic boys. It should be noted that this study had quite
a low number of participants, though it does show a similar picture to Dean et al.’s
(2017) findings.
Moving forward, some researchers are attempting to develop self-assessment
measures which will help to better conceptualise camouflaging behaviours and the FPT,
and will more directly measure camouflaging behaviours. For example, Kopp and
Gillberg (2011) have developed the Autism Spectrum Screening Questionnaire –
Revised Extended Version (ASSQ-REV), which uses an additional 18 items (ASSQ-
GIRL) reflecting characteristics seen in the female phenotype of autism. When tested on
71 autistic girls, 62 autistic boys, and 58 non-autistic girls (all aged between 6-16
years), the new revised version of the ASSQ reliably discriminated between autistics
and non-autistics, although it showed no differences between autistic males and
females. When considered in detail, however, some of these items were found to be
46
more commonly rated highly in autistic girls than autistic boys, for example, the item
“Copies you (can be in a very discreet way)”, which demonstrates that these autistic
girls may be deliberately copying the behaviours of others to fit in. One of the reasons
this study may not have found a significant gender difference overall could be because it
tested early-diagnosed girls, whereas many of these specific female phenotype
characteristics will only be apparent in later-diagnosed girls and women. The scale was
also rated by parents and does not focus solely on camouflaging behaviours, unlike a
more recent survey created by Hull, Mandy, et al. (2019) who developed the self-
reported adult Camouflaging Autistic Traits Questionnaire (CAT-Q). The CAT-Q is a 25
item scale, with items developed from previous qualitative findings by Hull, Petrides, et
al. (2017). The scale was found to measure three factors, which were a) ‘compensation’,
for example, the item “When I am interacting with someone, I deliberately copy their
body language or facial expressions”; b) ‘masking’, for example, the item “I adjust my
body language or facial expressions so that I appear relaxed”; and c) ‘assimilation’, for
example, the item “In social situations, I feel like I’m “performing” rather than being
myself”. The scale was found to have good reliability and validity when tested on 354
autistic and 478 non-autistic adults, and it significantly correlated with traits of anxiety
and depression. In a follow-up study Hull, Lai, et al. (2019) tested gender differences on
the CAT-Q between 182 autistic females, 108 autistic males, 16 non-binary autistic
people, and 472 non-autistic controls, with a total mean age of 34.56. Autistic
participants scored significantly higher on the CAT-Q than non-autistic participants (p
< .001), and autistic females scored significantly higher than autistic males (p < .001, d
= .65). However, autistic females only scored higher than autistic males on two of the
three subscales; ‘assimilation’ (p < .001, d = 0.51) and ‘masking’ (p = 0.001, d = 0.43).
The authors conclude that autistic females are under more pressure to adapt their
behaviours to assimilate with others and to use more masking strategies, although
47
compensation may be used by both genders to some extent. Whilst the study was
limited to adults, as demonstrated by the high mean age, and may have therefore
attracted more late diagnosed and higher-camouflaging autistic people, it does offer a
unique and novel insight into the act of camouflaging, which has not been captured
previously.
It should be noted, however, that there are inconsistencies in the data on whether
there are differences between autistic females and males in the presentation of
camouflaging. For example, in a study by Cassidy et al. (2018) there was no evidence
that 99 autistic females attempted to camouflage more than 65 autistic males on a four-
item scale that was developed for the purposes of their study, but there were some
gender differences in terms of the quality of camouflaging. The scale asked participants
if they had “ever tried to camouflage or mask [their] characteristics of ASC to cope with
social situations? For example, have [they] ever tried to copy or mimic other people’s
behaviour to try and fit in, or tried to mask or hide [their] symptoms of ASC from other
people?” If participants answered yes to this they were then asked to specify in which
areas of their life they camouflaged, how frequent this was on a scale of 1 (never) to 6
(always), and lastly the overall amount of the day they spent camouflaging on a scale of
1 (none of my waking time) to 6 (all of my waking time). An overall score was
calculated which consisted of the sum of areas where camouflaging took place
(maximum 8), the overall frequency (maximum 6), and overall amount (maximum 6).
89.2% of autistic females attempted to camouflage, which was similar to the 90.9% of
autistic males. However, the overall scores on the camouflaging scale were significantly
higher for autistic females (M = 14.7) than autistic males (M = 12.95), which had a
medium effect size (d = .47). This study suggests that whilst both genders may attempt
to camouflage, the effort put into camouflaging is higher in autistic females than autistic
males.
48
2.2.4. Gender-distinctive cognitive strategies for camouflaging. Several
studies have begun to determine the traits and skills necessary for social camouflaging,
which may explain why autistic females have a relative advantage; i.e. there may be
gender-distinct cognitive strategies which enhance camouflaging abilities in females
(Livingston et al., 2018). In particular, there has been interest in the importance of
differences between autistic males and females in executive functioning (EF). It has
been suggested that better EF skills may enhance camouflaging; that is, in order to
camouflage one needs to inhibit inappropriate social responses, play and script social
interaction beforehand, and have a certain level of flexibility in order to handle
unexpected social situations (Sedgewick et al., 2016). For example, Lenhardt et al.
(2016) investigated EF differences between 71 autistic females and 144 autistic males
recruited from an adult autism diagnostic centre. They administered the AQ, EQ, SQ,
RMET, WAIS, and a battery of EF tasks testing visuospatial and psychomotor speed
abilities, multiple conceptual tracking, cognitive flexibility, set-shifting, and verbal
fluency. The autistic females had significantly fewer processing speed and cognitive
flexibility impairments than autistic males, suggesting that this may enable autistic
females to observe and learn social behaviours quicker and adapt better to new social
situations. However, females rated themselves higher on autistic traits, which as
mentioned previously could reflect better self-awareness that in turn might motivate
more camouflaging behaviours. Similar findings were made by Lai et al. (2012), who
studied 33 non-autistic men, 35 non-autistic women, 45 autistic men, and 38 autistic
women. Whilst both autistic men and women showed similar deficits in ToM (as seen
using the RMET), facial emotion perception (as seen using the KDEF), as well as in a
battery of EF tasks measuring signal detection and response inhibition, autistic females
performed equally well to non-autistic females on attention to detail and dexterity-
involved EF, whilst autistic men were impaired on this compared to non-autistic men.
49
Finally, autistic males had slower reaction times on EF tests for phonological working
memory and word generativity than non-autistic males, but autistic females and non-
autistic females were comparable, suggesting that visuospatial attention deficits may
characterise autistic males but not autistic females. Finally, Bolte et al. (2011) compared
visual attention to detail and EF in 35 autistic males and 21 autistic females and their
non-autistic siblings (n = 58), with a mean age of 14-15. A battery of EF tasks were used
including set shifting, planning, cognitive flexibility, speed of attention and multiple
conception tracking capacities. The autistic females once again demonstrated better EF
skills on the cognitive flexibility task, which was associated with fewer RRBIs.
A recent study by Livingston et al. (2018) found that heightened levels of IQ,
EF, and anxiety were all linked to a greater ability to compensate for underlying deficits
in ToM. Testing a sample of 136 adolescents (112 males and 24 females) aged between
10-15 years who either had a diagnosis of ASC (n = 101) or had the Broader Autism
Phenotype (BAP) (n = 35), compared with 67 unaffected co-twins, the authors measured
autistic symptoms on the ADOS, IQ (using the Wechsler Abbreviated Scale of
Intelligence [WASI]), ToM (using the computerised Frith-Happé Animations test), and a
battery of EF tasks measuring inhibition, set-shifting, and planning, and anxiety (using
the Revised Child Anxiety and Depression Scale). Participants were divided into four
groups (Low Compensation, High Compensation, Deep Compensation, and unknown)
based on median ToM scores (‘Good ToM’ versus ‘Bad ToM’), and by median social
ADOS scores (‘Good ADOS’ versus ‘Poor ADOS’). This meant that those with poor
ToM scores but with good ADOS scores could be classified as having high
compensation abilities, those with both good ToM and good social ADOS could be
classed as having deep compensation abilities, those with poor ToM and poor social
ADOS could be classed as having low compensation, whilst those with good ToM but
poor social ADOS were considered unknown. The findings suggested that the High
50
Compensators had higher verbal IQ, better EF scores, and higher levels of anxiety
compared to the Low Compensators. However, the Deep Compensation and the
Unknown groups showed a similar pattern on these variables, leading the researchers to
conclude that the factors involved in compensation were specific to good performances
on the ADOS despite poor ToM. Furthermore, all groups were equally likely to have a
co-twin who also had ASC, meaning that the genetic ‘hit’ for ASC was not greater in
any of the groups. This suggests that the High Compensators did not have a ‘milder’
form of ASC, because they had the same autistic traits as Low Compensators. Whilst the
study did not find that females were more likely to be High Compensators, as the FPT
would predict, the study included quite a low number of females (n = 24). The authors
suggest that future studies would benefit from investigating these differences in non-
clinical populations using self-assessment methods.
Another skill which may aid in better compensation behaviours is
autobiographical memory; this could be considered important for remembering social
scripts and learning from previous social interaction. Goddard et al. (2014) assessed
autobiographical memory in 12 autistic males, 12 autistic females and 24 non-autistic
children aged between 8-16 years on the Social Communication Questionnaire (SCQ),
the WASI, the British Picture Vocabulary Scale, the Memory Measures
Autobiographical Memory Cueing Task, which required the children to retrieve specific
memories in response to 15 word cues, the Recent and Remote Memory Tasks, which
included 12 questions designed to provoke memories from the past week and events
from early childhood, and finally the Verbal Fluency task, which tests the number of
items generated within certain categories. Autistic males tended to generate fewer
specific memories than non-autistic males, whereas non-autistic and autistic females
performed similarly. Autistic females also demonstrated better recall of recent events,
which were remembered in greater detail than their remote memories; this was not seen
51
in autistic males, and both non-autistic and autistic females described memories with
more references to emotional states than all groups of males. The autistic girls also
performed better on the SCQ than autistic boys, which when combined with their
enhanced ability to recall autobiographical memories suggests that females may be
better at compensating for social and communication impairments as a result of better
innate cognitive skills.
Finally, there is some evidence for camouflaging and improved sociability in
autistic females compared to autistic boys as seen by friendship motivation. For
example, Head et al. (2014) compared 25 autistic females to 25 non-autistic females, 25
autistic males, and 26 non-autistic males, aged between 10-16 years, on the Friendship
Questionnaire (FQ), which measures how much individuals enjoy close, empathic,
supportive, and caring friendships, how interested they are in people, and how much
they enjoy interacting with others for its own sake. Generally, autistic participants
scored worse than non-autistic participants, although autistic girls performed better than
autistic boys, and equivalent to non-autistic boys. However, it should be noted that the
original study by Baron-Cohen and Wheelwright (2003) did not find any differences
between autistic males (n = 51) and females (n = 17) on the questionnaire, whilst they
did find differences between non-autistic males (n = 27) and females (n = 49). Head et
al. (2014) argue that this could be due to the wide ranges of age seen in the original
study (14-64 years), though this study also used a smaller sample of autistic women. In
Head et al.’s (2014) study, parents rated their children on the scale, whereas the original
measure was intended for adult self-assessment, which may also explain the discrepancy
in findings. Future studies should look to examine the FQ further in a larger sample of
the autistic adult population.
52
2.2.5. Mental health repercussions of camouflaging. A consequence of
camouflaging and/or the subsequent later ASC diagnosis could be an increased risk of
mental health difficulties. As described in Chapter 1, autistic individuals are already at
an increased risk of mental health concerns. Females in particular seem to be susceptible
to co-morbid mental health difficulties as a result of internalising their difficulties. For
example, Stewart (2012) reports anxiety in autistic girls, manifesting in chronic
insomnia, regular emotional outbursts, self-harm, and school refusal. Similarly, Baldwin
and Costley (2015) reported heightened levels of mental illness in autistic females; 73%
of their sample were in need of ongoing mental health support. Mandy et al. (2012)
found that parents reported their autistic daughters to have worse emotional difficulties
than autistic sons. Additionally, mental health difficulties have been found to be
prominent in autistic people diagnosed later in life, most of whom previous research has
indicated are women, with affective disorders being one of the main reasons for referral
of ASC in adults (Lehnhardt et al., 2016). In interviews with fourteen women diagnosed
in late adolescence or early adulthood, Bargiela et al. (2016) found that 92.9% of their
participants scored above the clinical cut-off on the anxiety subscale of the Hospital
Anxiety and Depression Scale (HADS-A), 21.4% scored within the clinical range for
depression on the HADS-D, and 35.7% scored within the ‘distress’ and ‘severe’ range
on the General Health Questionnaire (GHQ-12). In their interviews almost all reported
experiencing one or more mental health problems, particularly anxiety, depression, and
eating disorders.
Camouflaging has been found by several studies to be linked to heightened
mental health difficulties. A consequence of camouflaging is increased exhaustion
leading to anxiety and depression. Livingston et al. (2018) explained how the process of
masking autistic traits and camouflaging to appear ‘normal’ uses up valuable resources,
which would otherwise be used elsewhere, resulting in exhaustion and breakdown. For
53
example, Tierney et al.’s (2016) study found that the ten adolescent autistic women
reported emotional consequences of camouflaging, including severe depression and
anxiety, with five participants using self-harm to cope. This is supported by qualitative
findings by Hull, Petrides, et al. (2017) who found that the most common consequence
of camouflaging reported by participants was exhaustion, with many feeling mentally,
physically, and emotionally drained as a result. Stress and anxiety were experienced
both during and after situations involving camouflaging. As well as exhaustion, acting
in ways contrary to ones ‘true’ self while camouflaging may have a damaging effect on
self-esteem and feelings of authenticity (Kernis and Goldman, 2006). Goffman (1969)
describes how maintaining a ‘show’ and behaving in ways incongruent to one’s own
beliefs can cause feelings of alienation from oneself and others.
Quantitative studies have made similar findings regarding the detrimental effects
of camouflaging to mental health. For example, as briefly discussed in Chapter 1,
Cassidy et al. (2018) found that camouflaging, as measured using a four item
questionnaire with high internal consistency, significantly predicted suicidality in
autistic participants (65 males; 99 females). This finding was made after controlling for
age, sex, presence of at least one developmental condition, depression, anxiety,
employment, and satisfaction with living arrangements. Furthermore, camouflaging
explained a significant amount of variance in suicidality above depression and anxiety,
suggesting that the association between camouflaging and suicidality may be partially
independent of mental health problems. In a recent study by Cassidy et al. (2019), the
link between suicidality (measured using the Suicidal Behaviours Questionnaire) and
autistic traits (measured with the AQ), was significantly mediated by camouflaging
(measured using the CAT-Q) and thwarted belonging (measured using the Interpersonal
Needs Questionnaire). Whilst these findings were made in a sample of 160 non-autistic
young adults, they highlight the general risk that high levels of camouflaging pose.
54
In other studies using the CAT-Q, higher camouflaging has been linked with
more mental health difficulties. Hull, Mandy, et al. (2019) found that total scores on the
CAT-Q, as well as scores on the ‘assimilation’ factor, were significantly negatively
correlated with wellbeing in autistic participants, and that total scores on the CAT-Q and
all three subscales were positively correlated with depression and generalised anxiety. In
addition to these findings, Cage and Troxell-Whitman (2019) investigated the mental
health consequences of camouflaging in 262 autistic adults (135 females, 111 males,
and 12 non-binary) using the CAT-Q and the Depression, Anxiety and Stress Scale
(DASS). They also asked participants to rate 21 reasons for camouflaging on how much
they agreed it was a reason for them to camouflage, as well as to rate 22 contexts for
camouflaging on how often they camouflaged in that context. They found that those
who camouflaged highly in both formal and interpersonal contexts, and those who
switched between camouflaging in one context but not in the other, experienced more
anxiety and stress than those who reported low levels of camouflaging in both settings.
However, no significant differences between high and low camouflagers were found in
depression scores. Given the higher rates of suicidality reported in Cassidy et al.’s
(2018) study it is vital that this should be investigated further.
In contrast, Lai et al. (2017) found greater camouflaging to be associated with
more depressive symptoms in autistic men (n = 30) but not in autistic women (n = 30),
and they also reported no significant relationship between camouflaging and anxiety in
either gender, as tested using the 21-item Beck Anxiety/Depression Inventory. The
authors concluded that camouflaging may be an ingrained strategy that has perhaps been
practised by autistic women for longer over their lifetimes than it has for autistic men,
leading to less negative emotional consequences. However, this study had relatively low
numbers of autistic participants compared to those that did find that camouflaging has
significant negative consequences for mental health. Additionally, as discussed earlier,
55
this study did not directly measure camouflaging; instead, the camouflaging score was
derived from the discrepancy between internal autistic traits and external behavioural
traits.
On the whole it would seem that mental wellbeing is a concern in autistic
women who use camouflaging to hide autistic traits, as can been seen from the lived
experiences of those with the condition reported in qualitative studies, as well as the
self-reports of autistic women who have consistently rated themselves as being high
camouflagers in several studies. This may be due to the social demands and the
subsequent exhaustion experienced from using this strategy and hiding ones true-self, or
it may be due to the consequences of later diagnosis in these individuals, which would
deny them necessary support and therapeutic intervention growing up.
2.2.6. Misdiagnosis. One final important point to discuss when looking at
gender differences in the presentation of autism is misdiagnosis. Although no research
to date has directly investigated cases of misdiagnosis in autistic women, evidence
exists to suggest that it warrants further investigation (Brugha et al.,2016). In their
article addressing the ‘lost generation’ of autistic adults, Lai and Baron-Cohen (2015)
describe how many psychiatric conditions have overlapping symptoms dimensions to
ASC, for example OCD, or overlapping diagnostic criteria, for example personality
disorders. They describe how the difficulty in diagnosing adults with ASC is
determining which co-morbid mental health issues are differential diagnoses. For
example, those with overlapping diagnostic criteria but with key differences to ASC,
which are true comorbidities, and those with overlapping behavioural features, which
are differential diagnoses. Differential diagnoses appear to be the most likely candidates
for misdiagnosis. For example, symptoms of Schizoid Personality Disorder include
social-detachment and restricted affectivity, and symptoms of Schizotypal Personality
56
Disorder include eccentricity, which overlaps with key features and behaviours
observed in ASC. The obvious difference between the conditions is that ASC is present
in early development, and autistic people will present with RRBIs and sometimes
language delays in addition to these symptoms. It is therefore important that clinicians
investigate this before diagnosing with differential conditions. This might be of
particular concern to autistic females, who present with fewer RRBIs and who
camouflage their autistic traits, thus hiding their impairments. Lai and Baron-Cohen
(2015) specifically mention Borderline Personality Disorder (BPD) as a differential
diagnosis of particular concern for autistic women, as they may be misdiagnosed with it.
This could be due to similarities in secondary features of ASC, such as problems with
relationships, identity, affect regulation, and increased self-harm and suicidal
behaviours. Fitzgerald (2005) describes further the overlapping features in ASC and
BPD, including these and many others, such as impulsivity, gestures or threats, chronic
feelings of emptiness, inappropriate intense anger and/or difficulty controlling anger,
and stress-related paranoid ideation.
The overlap between BPD and ASC in women has been noted in other research.
For example, Bargiela et al. (2016) found in a group of late-diagnosed autistic women
that many had been misdiagnosed, and several mentioned that personality disorders
were preferred over ASC diagnoses by clinicians. Furthermore, Rabbitte et al. (2017)
found that parents of autistic girls frequently reported that it was difficult to get
clinicians to believe their daughters might have an ASC, many seeing signs of anxiety
and self-harm as the result of mental health conditions rather than a consequence of an
undiagnosed ASC. Ryden et al. (2008) investigated adult psychiatric patients in
Stockholm attending mentalisation-based therapy, who had been consecutively referred
and diagnosed with BPD, for autistic traits. Forty-one participants were assessed, with a
mean age of 29, and of these 15% fulfilled the criteria for ASC. Of particular concern
57
was the heightened rates of suicide attempts in those with BPD and ASC, compared to
those with just BPD, which supports research described earlier regarding the
consequences and risk of camouflaging and delayed diagnosis in autistic people.
Kreiser and White (2014) warn that there are adverse consequences associated
with misdiagnosing autistic women. For example, they may not receive the correct
treatment for their condition, or receive treatment that does not accommodate for
autistic differences. Furthermore, these women may lack the insight into their
difficulties which gaining a diagnosis gives, and as such this may lead to further mental
health difficulties. This could present as a vicious cycle; autistic girls camouflage their
impairments, they miss a diagnosis in childhood, and they develop mental health
difficulties as a result. When they present to clinicians their autistic traits may be
ignored and mental health difficulties focussed on, increasing the likelihood of a
misdiagnosis with a different condition, further delaying an ASC diagnosis.
2.3. Summary and Research Directions
In summary, autistic females are likely to receive their autism diagnosis later than
males, which may partly explain the gender disparity in the prevalence of autism.
Whilst the EMB theory does explain a number of traits (primarily systemising and
empathising) that seem to occur to a greater/lesser degree in autistic people, the
evidence provided does not consistently support the idea that these are extremely male
characteristics and that autistic people have an extremely male brain, with girls being
less likely to be affected. Much of the early research supporting this idea was focussed
on autistic males, and newer research has tended to investigate only those who already
have an autism diagnosis, usually given to them in childhood. Therefore, the theory
does not adequately account for the many autistic females diagnosed late whose autistic
traits may present differently to males. The FPT on the other hand does go some way to
58
explain why so many autistic females are diagnosed late, and also why the gender
disparity in ASC is not as wide as previously thought. Evidence has shown that although
many of the core impairments are the same in autistic boys and girls, females appear to
show more positive social behaviours and less externalising behaviours, such as RRBIs
and hyperactivity, and instead may internalise their difficulties. This may make
identification of autism more difficult. There is also evidence that autistic females may
have sex-distinctive cognitive skills and socialisation pressures which might facilitate
the use of camouflaging as a strategy to hide impairments and to ‘fit in’ socially.
However, camouflaging is likely to have mental health consequences, putting those who
use this strategy at greater risk of affective disorders and suicidal behaviours.
Furthermore, clinicians may interpret internalised emotional difficulties, behaviours
resulting from camouflaging, and co-morbid mental health difficulties as other
disorders, which have overlapping features; in particular, undiagnosed autistic girls may
be at risk of being misdiagnosed with personality disorders. It is therefore important that
further research investigate this population of late diagnosed and undiagnosed autistic
women, in order to improve identification and the support available to help tackle co-
morbid mental health difficulties resulting from camouflaging and missed diagnosis.
Whilst the research is expanding in the area of diagnosis of autism in women
and the use of camouflaging strategies, several key gaps in the literature remain, which
this thesis will address. These include:
1. Lack of information about differences between undiagnosed autistic women and
diagnosed autistic women on ASC screening measures and the number and nature of
co-morbid mental health conditions. This will shed light on the current measures for
screening autism and the potential consequences for mental health of living with
diagnosed- versus undiagnosed autism.
59
2. Lack of information about differences between undiagnosed autistic women and
diagnosed autistic women on standardised self-report measures of social and
emotional functioning, including camouflaging. This evidence will help to evaluate
the female phenotype theory, which suggests that autistic women often evade
diagnosis due to better social skills than autistic men. Currently we only know about
those autistic women who have been identified, and it remains to be seen whether as
predicted by the theory the phenotype is even more apparent in those who still
remain unidentified.
3. Lack of information about which measures best predict the age of ASC diagnosis in
autistic women, and how the age of ASC diagnosis compares to the ages of
diagnosis of co-morbid mental health conditions. This evidence will help to identify
risk factors for late or missed diagnosis of autism in women, such as greater
empathy, superior social functioning, or deliberate camouflaging. By documenting
the trajectory of mental health diagnoses over time for autistic women, it will also
be possible to highlight common misdiagnoses that occur prior to the autism
diagnosis.
4. Lack of experimental research that evaluates observable social behaviours in autistic
individuals as a function of self-reported camouflaging. This evidence would show
for the first time whether self-reported camouflaging is actually predictive of the
social skills of autistic individuals as judged by other people.
This thesis refers throughout to ‘potentially autistic’ individuals, which refers to
participants who do not currently have an autism diagnosis, but who score above the set
criteria for autism traits on autism screening measures. There currently exist only two
validated autism screening tests, the Autism Quotient (AQ) and Ritvo Autism Asperger
Diagnostic Scale-Revised (RAADS-r). The AQ was chosen for the purpose of the
studies conducted in this thesis, as it is recommended for screening under NICE (2012)
guidelines. Furthermore, the authors of the RAADS-R emphasise that whilst the
60
RAADS-R can go beyond the AQ in also being used as a diagnostic tool rather than just
for screening, it needs to be administered by a clinician in a clinical setting (Ritvo et al.,
2011). Given that the purpose of this thesis is to identify individuals in the general
population who may be potentially undiagnosed, and due to resourcing constraints, it
would not be possible to conduct the RAADS-R in a clinical setting. Also, the AQ has
been tested wide in large samples from the general population, demonstrating good
validity with this audience (e.g. Ruzich et al., 2015). Additionally, the AQ has a higher
specificity than the RAADS-R (70% vs 58%) (Sizoo et al., 2016). This means that the
AQ is more accurate when it comes to non-autistic individuals screening negatively.
Additionally, the positive predictive value of the AQ is slightly higher than the RAADS-
R (79% vs 77%), and its negative predictive value lower (45% vs 53%) (Sizoo et al.,
2016). This means that the AQ may be slightly better at predicting individuals who will
go on to receive an ASC diagnosis and those who will not, which will be advantageous
for screening a general population. However, there still remain flaws with this measure.
By using this screening tool it is likely that a proportion of potentially autistic
participants will not be identified correctly, but it will allow for the identification of the
majority of potentially autistic individuals sampled.
Another area of concern is the validity of the instrument for autistic females,
particularly those diagnosed late. Sizoo et al. (2016) did not explore differences in
predictive value between males and females, and 75.7% of their clinical sample were
male. Furthermore, the instrument was created and developed on a predominantly male
sample (45 males vs 13 females), and a gender difference between non-autistic males
and females was found, with men generally scoring higher (Baron-Cohen et al., 2001).
Some items could be argued to reflect a more male-typical presentation of autism. For
example, item 15 (‘I find myself drawn more strongly to people than things’) may be
less likely to be endorsed by autistic women who are motivated to socially camouflage
61
and assimilate with others (Sedgewick et al., 2016). Also, item 41 (‘I like to collect
information about categories of things, e.g. types of car, types of bird, types of train,
types of plant, etc’) may not reflect autistic female specific interests, which tend to be
perceived as more typical of non-autistic female interests, for example fictional
characters and psychology (Hull et al., 2020). Murray et al. (2016) tested 557 autistic
females and 680 autistic males, as well as 4,462 non-autistic females and 2,894 non-
autistic male controls, in order to determine whether the AQ-10 is an accurate screening
tool for both genders. Only two items demonstrated significant differential item
functioning between the genders, however one of the items favoured males and the
other females, balancing the bias out and eliminating any overall differential test
functioning between males and females. These findings support the use of the AQ for
both genders, and given this is the most accurate tool available for screening autism in
the general population, it will be used throughout this thesis to determine potentially
autistic participants.
2.4. Thesis Overview
This thesis aims to fill the gaps in the literature, identified above, in three studies.
Chapter 3 describes a nationwide questionnaire study (Study 1) that aimed to
identify women with high autistic traits, which may be indicative of potential autism,
across the UK, and to compare these women to already diagnosed autistic women. In
particular, this study examined differences between potentially autistic and diagnosed
autistic women in scores on the EQ and the relation between EQ and age of ASC
diagnosis. It also examined group differences in co-morbid mental health diagnoses to
see whether certain mental health conditions are more common in potentially autistic
women than diagnosed autistic women, which might indicate misdiagnosis or a
prevalent vulnerability. Results from this study showed that potentially autistic women
62
scored significantly higher on the EQ than those with a diagnosis, although they still
demonstrated a significant impairment compared to non-autistic women. This pattern
was not observed for males, with both diagnosed and potentially autistic men scoring
similarly lower than non-autistic men. The study also found different types of
psychiatric diagnosis to be more common in diagnosed woman compared with
potentially autistic women, and vice versa. For example, potentially autistic women
were more likely to be diagnosed with BPD, whilst significantly more diagnosed
autistic females were diagnosed with affective disorders, ADHD, and OCD.
Chapter 4 reports an extension of the initial survey study that looks in greater
detail at differences in presentation between potentially autistic women and diagnosed
autistic women (Study 2). In particular, it investigated whether there are differences in
self-reported social behaviours, social relationships, self-monitoring (a proxy measure
of camouflaging), ToM, and anxiety and depression symptoms. For the diagnosed
autistic women, whose age of ASC diagnosis was known, the study collected
information about ages of co-morbid diagnoses in order to shed light on the typical
history of mental health diagnoses. Finally, Study 2 examined whether the age of ASC
diagnosis was predicted by the measures of social functioning and camouflaging. This
study showed that diagnosed and potentially autistic women performed similarly on
measures of friendship, self-monitoring, ToM, and traits of anxiety and depression.
However, potentially autistic women did score higher on social functioning, although
this was significantly impaired compared to non-autistic women. Furthermore, this
study found that diagnosed autistic women received significantly more psychiatric
diagnoses than diagnosed autistic men prior to their autism diagnosis being made.
Chapter 5 reports an experimental study that investigated differences between
autistic females and autistic males in self-reported camouflaging, and whether executive
63
functioning and ToM affect the probability that individuals use camouflaging as a
strategy to hide autistic traits (Study 3). This investigation was only made possible by
the invention of the CAT-Q (Hull et al., 2019) that was published after Study 2 taking
place. Study 3 also explored whether external observers do indeed tend to form a more
favourable impression of autistic women than autistic men based on their social skills,
and whether this is related to higher levels of self-reported camouflaging among autistic
women. Specifically, participants were filmed in ‘everyday’ conversation and, after
viewing each video, non-autistic peers rated each videoed participant on their first
impressions and their willingness to socialise with that person. Findings from this study
demonstrated that autistic people significantly camouflaged more than non-autistic
people, however no gender differences were found. No differences between any groups
were found on EF or ToM. However, on first-impression ratings autistic people were
rated less favourably than non-autistic people, males were rated less favourably than
females, and male raters were harsher in their judgements, particularly of autistic men.
This meant that autistic women did make significantly more favourable first-
impressions than autistic males, and whilst first-impressions did not correlate with
camouflaging, they did correlate positively with age of autism diagnosis.
Together these three studies make important and novel contributions to the
existing literature by investigating a hidden population of potentially autistic women
who have not previously been explored in detail. This will help provide new evidence as
to whether autistic women do have a different phenotype of autism, which may make
them harder to identify, more likely to be misdiagnosed, and more vulnerable to mental
health difficulties. Furthermore, this research will provide specific evidence as to
whether camouflaging is a successful strategy for autistic women, in one of the first
quantitative studies of its kind.
64
65
66
CHAPTER 3
Study 1: Screening and Identifying Potentially Autistic Women across the UK
3.1. Introduction
The overarching purpose of this thesis is to explore the reasons why autistic females are
often diagnosed later than autistic males or fail to receive an autism diagnosis altogether
(Bancroft, 2012), with a focus on the Female Phenotype Theory (FPT) of autism. FPT
suggests that autism in women is often missed by clinicians due to autistic females
displaying behavioural traits which are different from those displayed by autistic males,
and are not the typical traits associated with autism (Kopp & Gillberg, 1992). The main
aims of Study 1 were, first, to shed light on the prevalence of undiagnosed female
autism in the general population using a large-scale online survey, and second, to
compare levels of empathy between diagnosed autistic, potentially autistic, and non-
autistic women and men. Participants with diagnosed autism were asked to report the
age at which they received their diagnosis. Additionally, participants were asked to list
whatever other formal psychiatric diagnoses they had ever received (e.g., GAD, Eating
Disorder, BPD). This was to see whether potentially autistic women were more likely to
report psychiatric problems, as might occur due to the stress of living with an
undiagnosed ASC, the stress of attempting to hide ASC traits, or from being
misdiagnosed with other conditions by clinicians who misinterpret their autistic traits.
Baron-Cohen et al. (2009) found evidence that current statistics regarding the
prevalence of autism may be grossly under-estimated. They suggested that this is due to
the majority of investigations only considering those with diagnoses and/or those
considered as more likely to have the condition, such as those whose relatives are
autistic or who have children with additional needs (Ehlers & Gillberg, 1993; Gillberg
67
et al., 1991). For example, Baron-Cohen et al. (2009) surveyed the Special Educational
Needs (SEN) register for known cases of ASC as well as screening the mainstream
primary school population in Cambridgeshire for unknown cases. The screening
involved a diagnostic survey, which was sent to all participating schools to be
completed by all parents of 5-9 year old children. The CAST, which is a 37-item
screening tests to be completed by parents, was used, and suspected cases were
followed up with full ASC assessments using the ADI and the ADOS. Results showed
that 0.94% of the SEN population and 0.99% of the mainstream population had an ASC.
Further analysis revealed that for every three known cases of ASC there were two
unknown cases. These findings suggest that there may be quite a significant number of
autistic individuals who remain undiagnosed. No differences were found in the number
of unknown cases of boys versus girls, despite finding a prevalence rate of 1.53% in
male known cases and only 0.42% in female known cases. However, this result may
have occurred because only children were tested, and it could be the case that females
are more likely to go into adulthood with undiagnosed ASC compared to males.
To date, few studies have been able to provide estimates for the gender difference in
undiagnosed cases of ASC in adulthood. This is probably due to the difficulty in
identifying these individuals, given that most may not present in a typical way. Self-
assessment screening measures, which can be used in the general population, may
therefore be of value in identifying missed cases. Baron-Cohen et al. (2001) developed
the AQ to screen for autism. In the process of validating their measure, they tested 174
randomly selected non-autistic controls drawn from 500 adults who were sent the AQ
by post to fill in, all living in the East Anglia area (mean age = 37). Using the cut off of
≥32 to determine possible cases of autism, which was derived from testing the measure
on autistic participants, the study was able to determine the number of non-autistic
adults in the population who were potentially autistic but who were not diagnosed. 40%
68
of males scored at or above the intermediate point of the scale (20+) compared to 21%
of females, and only 1% of the females scored above the clinical cut off points
compared to 3.9% of males. These findings would suggest that whilst there is a
possibility of a missed diagnosis for both genders, there are likely to be more males that
fit this category than females. This conflicts with the FPT, which suggests that females
are more likely to be missed for diagnosis. It should be noted that screening with the AQ
cannot give a definite answer as to whether a person is autistic or not, and does rely on
the person’s own awareness of their difficulties. However, Sizzo et al. (2015) found that
shortened versions of the AQ (AQ-28 and AQ-10) correctly identified cases of autism
70% – 72% of the time amongst a sample of 285 adults referred for ASC assessments.
This demonstrates that the AQ could be used cautiously to estimate incidences of
potential autism. Indeed, NICE guidelines recommend the Adult Asperger’s Assessment
(AAA) (Baron-Cohen, Wheelwright, et al., 2005) for the diagnosis of adults, which uses
the AQ as one of its key tools alongside the EQ and RQ.
Sizzoo et al. (2015) found a higher number of males referred for assessments in their
sample (75.7%), which could indicate a gender bias in referrals. It is unknown whether
those females referred for assessment were more or less likely to receive an ASC
diagnosis after scoring above the cut-off on the AQ. Dworzynski et al. (2012) have
suggested that girls who score above thresholds for autistic traits (according to the
CAST) are less likely to receive a diagnosis than their male counterparts.
While the studies by Baron-Cohen et al. (2001; 2009) suggest that autistic females are
not more likely to be undiagnosed, results nevertheless indicate that the prevalence of
autism may be higher, and the gender ratio of autistic males to females lower, than
originally thought. Furthermore, more recent research has provided support for the
FPT, demonstrating that autistic females are indeed diagnosed later (Begeer et al., 2013;
69
Lai & Baron-Cohen, 2015; Shattuck et al., 2009). For example, Baldwin and Costley
(2015) found the mean age of diagnosis to be 25 years amongst a sample of 82 autistic
women, and Bancroft (2012) found that only one fifth of girls who took their survey
were diagnosed before the age of 11 years, compared to over half of boys. However,
these latter studies did not aim to determine prevalence rates or use the random
participant selection methods used by Baron-Cohen, et al. (2001; 2009). It is possible
that there is a gender bias in these studies examining the FPT, such that late diagnosed
autistic women are more motivated to seek information and engage with such studies in
order to better understand themselves. Regardless, it is clear that there is great
variability in the age of diagnosis for autistic individuals, and that females may be
particularly susceptible to being missed in early childhood for reasons discussed next.
The FPT suggests that the reason for the frequently later or missed diagnosis of autistic
women is the differences in behavioural manifestation of autistic traits (Kopp &
Gillberg, 1992). McLennan et al. (1993) reported that autistic girls were less affected by
social and communication behaviour difficulties than autistic boys, a finding which has
been supported by more recent research on the subtle social behaviour differences
between autistic males and females (Hiller et al., 2014; Hsiao et al., 2013; Rynkiewicz
et al., 2016). In particular, Lai et al. (2011) found many similarities between 45 autistic
males and 38 autistic females in terms of childhood autistic symptoms, and difficulties
with empathising and mentalising. However, the autistic girls were less impaired in
socio-communication and demonstrated fewer RRBIs, findings which have also been
supported by several other studies (Hiller et al., 2014; Mandy et al., 2012; Ratto et al.,
2018; Wilson et al., 2016). Superior functioning in social areas may therefore act as a
mask for other autistic traits and hinder diagnosis. Additionally, research has suggested
that autistic females may deliberately camouflage their social behaviours in order to ‘fit
in’ and appear less atypical (Bargiela et al., 2016; Hull, Lai, et al., 2019).
70
There are two consequences hypothesised to be the result of this atypical ASC
presentation in autistic females, namely, increased mental health issues and increased
likelihood of misdiagnosis with other psychiatric conditions. Livingston et al. (2018)
have suggested that the process of masking and camouflaging autistic traits uses up
valuable resources, resulting in exhaustion and breakdown. This is supported by
findings by Cassidy et al. (2018) that self-reported camouflaging traits significantly
predicted suicidality in 65 autistic males and 99 autistic females. Hull, Mandy, et al.
(2019) also found that self-reported camouflaging traits were significantly, negatively
correlated with wellbeing and positively correlated with anxiety and depression. In
qualitative studies where autistic women were interviewed regarding their camouflaging
behaviours, it has been found that such women often report great emotional
consequences of attempting to hide their autism, including exhaustion, depression,
anxiety, and self-ham (Hull, Petrides, et al., 2017; Tierney et al., 2016;). Additionally,
having a late diagnosis presents its own issues in terms of gaining the correct support
and having an unknown condition regardless of the presence of camouflaging. For
example, Stagg and Belcher (2019) interviewed nine autistic adults between 52 and 54
years of age (5 females and 4 males) who had received a diagnosis later in life. These
participants commonly referred to feelings of alienation as a result of living with a
condition they had little or no knowledge about. These findings are supported by Jones
et al. (2001) who examined written first-person accounts of the emotional experiences
of autism, finding that depression could be caused from not understanding one’s
differences in comparison to others.
Furthermore, Taylor’s (1983) cognitive adaptation model could partially help us to
understand why a later diagnosis of autism is so detrimental to mental health, as a
diagnosis requires the individual to re-evaluate who they are and rebuild their self-
esteem. In conflict with these findings are those by Cassidy et al. (2018), who did not
71
find significant correlations between age of autism diagnosis and camouflaging,
depression, or anxiety. However, it is important to note that the participants in this study
were all adults and the mean age of ASC diagnosis was 34. Given the conflicting
findings, more studies are needed to explore the link between age of ASC diagnosis,
camouflaging of ASC, and mental health.
Another consequence of an atypical ASC presentation in autistic females is likely to be
misdiagnosis with other psychiatric conditions. Before exploring this possibility, it is
important first to understand the issue of co-morbidity and autism in general, as autistic
people are thought to be at a heightened risk of psychiatric illness. Russell et al. (2016)
retrospectively reviewed 474 autistic people who had received an ASC diagnosis and
compared co-morbid psychiatric diagnoses against those seen in the general population
from the UK National Psychiatric Morbidity Survey (McManus et al., 2009). The ASC
group were more frequently diagnosed with phobias (16.8% vs 1.4%), generalised
anxiety disorder (GAD) (11.8% vs 4.4%), OCD (17.9% vs 1.1%), depression (15.8% vs
2.3%), ADHD (9.7% vs 2.3%), and psychotic disorders (2.1% vs 0.4%) than the general
population.
Because psychiatric co-morbidity is high and camouflaging can cause mental health
issues, it has been hypothesized that late and missed diagnosis may be the result of
misdiagnosis. For example, Lai and Baron-Cohen (2015) suggested that difficulties may
arise due to overlapping symptom dimensions to ASC and determining which co-
morbid mental health issues are differential diagnoses. Differential diagnoses, whereby
a condition has overlapping but also distinct features, could lead to misdiagnosis when
the typical behavioural characteristics of autism are hidden. A number of conditions
which have overlapping features with autism have been discussed in the literature,
including schizophrenia, personality disorders, ADHD, OCD, and affective disorders. It
72
is possible that without obvious signs of the social impairments characteristic of autism,
clinicians may mistakenly diagnose other conditions, which are discussed in turn below.
The original diagnostic criteria for Schizophrenia included many of the same features
as autism, such as social withdrawal, flattening affect, eccentricity, having a narrow
circle of interests, and lacking sympathy (Bleuler, 1911; Kraepelin, 1919). Whilst the
criteria have changed, there are still overlapping attributes. For example, Leitman et al.
(2014) found deficits in ToM for both autistic and schizophrenic patients, and catatonic
behaviour has been found in 17% of adolescent and adult autism referrals (Wing &
Shah, 2000). Furthermore, Aggarwal and Angus (2015) found that 12% of their sample
of 31 adults referred for ASC assessments presented with psychotic symptoms, and that
childhood ASC and autistic traits increased the likelihood of having psychotic
symptoms. Both Fitzgerald and Corvin (2001) and Dossetor (2007) suggest, however,
that psychotic symptoms may be misinterpreted in autistic patients by clinicians. Due to
difficulties in concrete thinking and ToM, autistic patients may answer that they do hear
voices, when they are actually referring to background noises or their own internal
voices.
Lehnhardt et al. (2013) conducted a literature search of articles on PubMed that
discussed autism and differential diagnoses. They found that personality disorders (PDs)
were the most common differential diagnoses made in autistic people. Hofvander et al.
(2009) found that 19-32% of autistic patients met the criteria for compulsive PD, 21-
26% for schizoid PD, 13-25% for avoidant PD, and 3-13% for schizotypal PD. This
supports evidence presented in the previous paragraph regarding the overlapping
features of schizophrenia, as both schizoid and schizotypal PD are considered to be
associated with schizophrenia. Fitzgerald and Corvin (2001) discussed how schizoid
symptoms such as solitariness, empathy deficits, lack of attachment to others, paranoia,
73
and special interests are all also characteristic of autism. Wolff (Chapter 10, 1998) even
described autistic children and those with ‘cluster A PDs as belonging to the same
group behaviourally.
Another PD which has frequently appeared in the literature on misdiagnosis is
BPD. This may be a more common differential diagnosis for girls and women (Bargiela
et al., 2016; Lai & Baron-Cohen, 2015), particularly as in the general population the
ratio of females to males with BPD is thought to be 3:1 (APA, 2000). Parents of autistic
girls have reported that they had a difficult time getting clinicians to believe their
daughters may have ASC, as many focussed instead on signs of mental illness, such as
anxiety and self-harm (Rabbitte et al., 2017). Bargiela et al. (2016) found in a group of
late-diagnosed autistic women that many had been misdiagnosed before getting their
diagnosis, with several mentioning BPD diagnoses being preferred by clinicians over
ASC diagnoses. Ryden et al. (2008) found that 15% of their sample of women with
BPD also fulfilled criteria for ASC. Fitzgerald (2005) described further the overlapping
features in ASC and BPD, such as impulsivity, relationship difficulties, gestures or
threats, chronic feelings of emptiness, inappropriate intense anger and/or difficulty
controlling anger, and stress-related paranoid ideation.
Another differential diagnosis that may result in misdiagnosis is OCD. Between
2.6% and 37.2% of autistic children and adolescents are thought to have OCD (van
Steensel et al., 2011). Ivarsson and Melin (2008) investigated 109 children with OCD
using the Autistic Symptom/Syndrome Questionnaire and found that they had a
significant number of autistic traits, accounting for 40% of the variance in the model.
Fitzgerald and Corvin (2001) likened the OCD traits of repetitive obsessions and
compulsions to the repetitive routines seen in autism. However, Postorino et al. (2017)
74
pointed out that autistic individuals find comfort in their repetitive activities and are not
usually distressed by them.
Fitzgerald and Corvin (2001) also described ADHD as a differential diagnosis
that has many overlapping features with autism. In particular, impulsivity may make
individuals with ADHD appear to be lacking in empathy, and distractibility may be
found in autistic people who are highly sensitive to sensory information around them or
who are fixated on attending to their special interest above all else. Gillberg and Ehlers
(1998) wrote that children who meet criteria for ADHD might also meet those for
autism, and Russell et al. (2016) found the prevalence of ADHD to be higher in the
autistic population than it was in the general population (9.7% vs 2.3%).
Finally, anxiety and depression also present with some overlapping features with
autism. As discussed previously, these two disorders are more common in autistic
people than in the general population (Russell et al., 2016). Symptoms which may
overlap include social withdrawal and anxiety, flattening affect, and a loss of interests
and in relationships (Fitzgerald & Corvin, 2001). Lehnhardt et al. (2013) listed social
anxiety, in particular, as one of the most common differential diagnoses with autism.
This ties in with evidence regarding the camouflaging of autistic traits by girls and
women, who say that they want to be able to ‘fit in’ better socially (Tierney et al., 2016;
Hull, Petrides, et al., 2017).
The consequences of misdiagnosis are likely to include a further delay in gaining an
autism diagnosis, which as discussed previously may lead to further mental health
problems. Kreiser and White (2014) highlighted the lack of correct treatment and
support that individuals with a misdiagnosis will experience. However, no studies to
date have explored whether misdiagnosis is indeed common in autistic females,
presumably because of the difficult nature of identifying those who might have a
75
misdiagnosis and in determining whether the misdiagnosis is really a misdiagnosis or,
alternatively, a co-morbid diagnosis.
3.1.1. Aims and hypotheses. Currently research into autistic women has
focussed on individuals with a clinical diagnosis of ASC and little is known about
women who meet criteria for ASC but have not received a diagnosis. In a paper
addressing evidence gaps and emerging areas of priority in the research of sex
differences in autism, Halladay et al. (2015) stressed the need for studies to look at non-
clinical samples of undiagnosed autistic females.
In the first instance it would be useful to attempt to replicate those findings made
previously by Baron-Cohen et al. (2009), in order to examine whether in the last decade
there have been any changes in the number of potentially autistic women compared to
potentially autistic men amongst a non-clinical sample. Furthermore, very few studies to
date have explored the characteristics of this hidden population, which might explain
why they are undiagnosed. Evidence supporting the FPT has largely looked at
diagnosed autistic women; but it is important that we understand the profile of
potentially autistic women too. If the FPT is accurate then we would expect to see
differences in the behavioural manifestations of autism between undiagnosed women
and diagnosed men and women, as well as in differential mental health diagnoses that
could indicate misdiagnosis.
Study 1 therefore represents a novel attempt to identify a large group of
potentially autistic females through a nationally distributed online survey advertised to
women and men aged 16-40 years in the general population, and to begin to build a
psychological profile of such women, which may lead to this group’s earlier
identification. This age range was chosen to ensure that findings were not reflective of
76
historical biases but rather current issues in the identification and diagnosis of autism.
Given that Asperger’s Syndrome was only introduced by the APA in 1994, and further
autism subtypes in 2000, it is reasonable to expect that autistic adults aged 16-40 years
would have been able to be identified with an ASC at some point in their childhood or
adolescence.
Specifically, Study 1 addressed the following questions and hypotheses:
1. What proportion of women in the sample have high autistic traits, which could
be indicative of potential autism but who have not have received a diagnosis? It
was predicted that there would be a higher proportion of women than men with a
potential ASC.
2. Can this study replicate findings that autistic women tend to be diagnosed with
an ASC at an older age than autistic men? It was predicted that autistic females
would be diagnosed later than autistic males.
3. Do potentially autistic women have impairments similar to those of diagnosed
autistic women on measures used for screening and assessment of ASC? It was
predicted that potentially autistic women would demonstrate less impairment
than diagnosed autistic women on the EQ and that, among diagnosed autistic
women, age of diagnosis would correlate positively with EQ scores. In
particular, it was predicted that cognitive empathy (as measured using the
‘cognitive empathy’ subscale of the EQ’) would be less impaired in potentially
autistic women, whilst no differences between groups would be found in
affective empathy (as measured using the ‘emotional reactivity’ subscale of the
EQ).
4. Are potentially autistic women more prone than diagnosed autistic women to
receive other mental health diagnoses? It was predicted that potentially autistic
women would be more likely to report other psychiatric diagnoses, perhaps due
77
to the difficulties of coping with an undiagnosed ASC, the stress of
camouflaging ASC traits, or from being misdiagnosed by clinicians. In
particular, it was expected that they may have more differential psychiatric
diagnoses, which have overlapping features with ASC.
3.2. Methods
The Checklist for Reporting Results of Internet E-Surveys (CHERRIES) (Eysenbach,
2012) was used, which has been established to ensure the quality of reports in the
medical literature that use online surveys to collect data.
3.2.1. Participants. The target population was young adults (aged 16-40) from
the UK without a diagnosis of ASC, and a comparison group of young adults with a
diagnosed ASC. Due to the nature and novelty of the research (our target group was
undiagnosed individuals) the required sample size could not be calculated. Initially, UK
universities were targeted for participants, as young adults make up the majority of their
populations. Heads of Department (or administrators) from every department in every
UK university were contacted requesting them to send the link to the survey and a
description of the study to their students. The study was also advertised with the same
description on social media via Students’ Union pages, and through Facebook
advertisements targeted at students aged 16+. Participants with diagnosed ASCs were
recruited via university disability services, autism Facebook pages, and through the
organisation ‘Research Autism’. Non-student participants were also recruited through
various media outlets, including in local newspapers. To ensure that a representative
sample of the general population was obtained, the adverts used for participant
recruitment purposefully did not mention autism, but instead called for participants to
take part in a ‘student screening study’ (see Appendix 1). This made it possible to fairly
78
assess the rates of potentially undiagnosed autism, rather than attracting only
respondents with autism or who thought they may be autistic.
In the demographic section of the survey, participants were asked to confirm
their age, any psychiatric diagnoses, and their country of birth in order to confirm that
they met the criteria for the study.
There were 8,731 responses recorded in total for the first question, which asked
participants for their age. Of these, 5,165 individuals completed the whole survey giving
a completion rate of 59.16%. Due to the nature of the web-based research, it was
impossible to ascertain the total number of individuals that the advertisements for the
survey reached, and therefore the response rate is unknown.
Of the participants who completed the survey, 1,324 (25.6%) were male, and
3,841 (74.4%) were female. Of those who reported having an ASC, 27 were male and
153 were female. The average age of diagnosed autistic females was 27.37 (SD = 7.193)
and for diagnosed autistic males it was 25.19 (SD = 6.027). Of those in the potential
ASC group, who scored above the clinical criteria on the AQ (≥32) but who did not
have a diagnosis (690 females and 144 males), the average age of females was 29.17
(SD = 6.759) and the average age of males was 27.58 (SD = 7.210). Of those with no
ASC (2,998 females and 1,154 males), the average age of females was 24.46 (SD =
6.451) and the average age of males was 22.93 (SD = 5.428).
Across the whole sample, 70.3% were students (college, undergraduate, and
postgraduate), whilst 24.6% were in employment, and 5.1% were unemployed.
Participants were recruited from across the UK and lived in over 70 different counties,
with the majority living in London (10.7%), Cambridgeshire (4.6%), West Midlands
(3.7%), Essex (3.3%), Strathclyde (3.2%), and Devon (3.1%).
79
3.2.2. Measures
Mental Health: Participants were given a checklist containing the common
mental health conditions according to the DSM 5 (APA, 2013), including ADHD,
Alcohol/Substance Abuse, Anxiety disorders, Bipolar Disorder, Depression, Eating
Disorder, OCD, Personality Disorders, and Schizophrenia. They were asked to select
any that they had been formally diagnosed with by a clinician, and given the
opportunity to select ‘other’ if they had any condition not listed. Participants were also
asked to select whether they had been clinically diagnosed with ASC and, if so, at what
age.
Autism Quotient: The full 50 item Autism Quotient (AQ) (Baron-Cohen, et al.,
2001) was used to screen participants for a potential ASC. The AQ is reported to have
good internal consistency and good test-retest reliability (r =.7, p = .002) and a cut off
score of ≥32 has been found to be accurate in identifying possible cases of ASC (Baron-
Cohen et al., 2001). Sizoo et al. (2016) recently reported 80% accuracy in an
undiagnosed population referred for diagnosis, and previously it has been used
successfully in large epidemiological studies in non-clinical samples to determine
autistic traits in the general population (Lai, et al., 2011; Ruzich et al., 2015).
Empathy Quotient: The 40 item version of the Empathy Quotient (EQ) (Baron-
Cohen & Wheelwright, 2004) was used to see whether potentially autistic women
possess similar impairments as diagnosed autistic women on another measure used for
screening and assessment of ASC. The EQ is included alongside the AQ when assessing
for ASC (Baron-Cohen, Wheelwright, et al., 2005). The EQ is reported to have excellent
test-retest reliability (r = .97, p < .001). A cut off score of < 30 has been found useful in
identifying those with empathy difficulties; 81.1% of adults with an ASC score below
this cut off. In adults without ASC, females typically score higher than males, indicating
80
less susceptibility to empathy impairments (Baron-Cohen & Wheelwright, 2004). The
survey has excellent test-retest reliability in both clinical and non-clinical populations
(Lawrence et al., 2004). Lawrence et al. (2004) also established reliable subscales for
the EQ, using 79 male and 93 females to factor analyse the scale. Three factors were
identified: “cognitive empathy”, which contains 11 items and pertains to an appreciation
of emotional states; “emotional reactivity”, which contains 11 items also and pertains to
the tendency to experience emotional states in response to others’; and “social skills”,
which contains 6 items. Significant gender differences on both empathy subscales were
identified but not on social skills. Different factors of the IRI showed concurrent
validity with some of the subscales of the EQ, so that ‘emotional reactivity”
significantly correlated with ‘empathic concern’ and ‘perspective taking’ on the IRI ,
and ‘social skills’ correlated with ‘perspective taking’ also, but none correlated with
‘cognitive empathy’. For the purpose of this study only the two emotional factors were
explored separately.
3.2.3. Design Participants were grouped by gender and autism status to
generate six groups: males versus females diagnosed with an ASC (‘diagnosed
autistic/diagnosed ASC’), males versus females without an ASC diagnosis who scored
above the criteria on the AQ (≥32) (‘potentially autistic/potential ASC’), and males
versus females without an ASC diagnosis who scored below the criteria on the AQ (<
32) (‘non-autistic/no ASC’). A between-subjects analysis was conducted on scores from
the questionnaires.
3.2.4. Procedure. The survey was designed online using Qualtrics, and tested
prior to distribution by three members of the research team who went through the
survey as though they were participants. The survey was set to open access allowing
anyone to take the survey. However, it allowed for only one response per participant;
81
this was achieved through the monitoring of cookies. All items were set to forced
response, and progression through the survey was dependent on all items being
answered (non-response options were provided throughout).
Full ethical approval for the survey and its contents was granted under the terms
of Anglia Ruskin University’s Policy and Code of Practice for Conduct on Research
with Human Participants. Participants were presented with an information page before
beginning the survey, which purposefully did not mention autism but instead described
the study as an investigation into a gender bias in empathy and behavioural responses;
this was to avoid demand characteristics and also to ensure we did not receive a biased
sample of only individuals who suspected that they may have autism. Participants were
informed that the survey would take around 20 minutes to complete, that an iPad prize
was being offered for completion of the survey, and they were also given the contact
details of the lead researcher. The first section of the survey collected demographic
information, any mental health information, and information about ASC diagnoses. This
was followed by two further sections measuring autistic traits and empathy. Finally,
participants were fully debriefed. They were informed that the study was specifically
looking at ASC and that the questionnaires they had filled out were commonly used as
preliminary screening tools, but that scores on these would not be sufficient for a
clinical diagnosis. For ethical reasons it was decided that individual scores would not be
released to individuals. This was to ensure the data remained anonymous and to avoid
causing distress. However, contact details of the National Autism Society were
provided. Finally, participants were given the opportunity to leave their email addresses
to be entered into the prize draw.
3.3. Results
82
3.3.1. Data checks and descriptive statistics. Inspection revealed some departure from
normality in the data. This was expected as the participants were assigned to groups
according to their questionnaire scores, which necessarily skewed the distribution of
their scores across the groups. Additionally, as the study could not control for the
number of participants in each group, uneven numbers can be seen across the six
groups. Non-parametric tests were therefore employed to analyse the data.
One-way ANOVAs using a Kruskal-Wallis H explored differences between all
groups on age and on the EQ, and Mann-Whitney U tests explored pairwise
comparisons of these. Mann-Whitney U was also used to explore differences between
males and females on age of ASC diagnosis. Bonferroni corrections were applied with
comparisons of more than three groups. Spearman’s correlation tests were performed to
determine correlations between AQ, EQ and age of diagnosis. Finally, Chi-Square tests
were used to explore differences in the frequency of other mental health diagnoses
across groups, and which specific diagnoses were more prevalent; for this latter analysis
particular attention was paid to the differential diagnoses types mentioned in the
introduction (Schizophrenia, Schizoid Personality Disorder, BPD, OCD, ADHD, and
affective disorders). Where cell counts were less than five, Chi-Squares could not be
performed due to problems with accuracy.
Table 3.1 shows group means (and standard deviations) for the AQ and EQ. For
the ASC group, the mean age of diagnosis is also presented.
Table 3.1
Descriptive statistics of each group stratified by gender and means for AQ and EQ
83
Diagnostic Group N Age of
ASC
diagnosis
AQ EQ
Total Cognitive
Empathy
Emotional
Reactivity
Females
ASC 153 23.57
(9.34)
39.75
(4.38)
19.13
(8.56)
2.48
(3.00)
7.03
(4.16)
Potential ASC 690 - 36.52
(3.79)
23.19
(10.25)
4.48
(4.15)`
7.84
(4.61)
No ASC 2,998 - 18.89
(6.89)
44.37
(12.93)
12.40
(5.11)
12.40
(5.11)
Males
ASC 26 16.92
(10.14)
39.50
(4.22)
16.54
(6.94)
2.00
(1.98)
5.38
(3.70)
Potential ASC 144 - 36.05
(3.39)
19.55
(9.05)
4.08
(3.99)
5.64
(3.92)
No ASC 1,154 - 19.00
(6.11)
38.17
(12.02)
11.70
(4.98)
9.57
(4.45)
3.3.2. Proportion of potential ASC participants. Of the 3,841 females who
took the survey, 17.96% (690) scored above the clinical cut off on the AQ (≥ 32) and
were classed as being potentially autistic, whilst 3.98% (153) were already diagnosed
with ASC. Of the 1,324 males who took the survey, 10.88% (144) scored above the
clinical cut off on the AQ and were classed as being potentially autistic, whilst 1.96%
(26) were already diagnosed with ASC. Chi-Square analysis revealed that there was a
significant difference in the frequency of participants in each group, X²(2) = 52.382, p
<.001, φ = .101. Odds ratios revealed that females were 2.3 times more likely than
males to be in the diagnosed ASC group and 1.8 times more likely to be in the potential
ASC group.
84
3.3.3. Age of diagnosis. As can be seen from Table 3.1 females were diagnosed
later than males. Using a Mann-Whitney U this was found to be significant with a
medium effect size: U = 1195.00), p = .003, d = 0.68.
Age of diagnosis was categorised as being made either in childhood/adolescence
(1-17 years of age) or in adulthood (18+ years of age) for each participant. 73.9% of
females were diagnosed at the age of 18 or later (n = 113) compared to 44% of males (n
= 11). This difference was found to be significant: X²(1) = 9.064, p = .003, φ = .226.
Autistic women were 3.6 times more likely be diagnosed in adulthood than autistic men.
3.3.4. Group differences in EQ scores. For both males and females, the
diagnosed ASC and potential ASC participant groups scored on average below the cut-
off on the EQ (< 30), indicating empathy impairments.
For females, the differences between the three groups on EQ were significant:
X2(2) = 1296.589, p < .001. Diagnosed ASC participants scored lowest, followed by
potential ASC participants, and no ASC participants. A Bonferroni corrected p value of
0.02 was established for pairwise comparisons, which found a significant difference
with large effect sizes between the diagnosed ASC and no ASC groups (U = 25660.00,
p < .001, d = 2.30), the potential ASC and no ASC groups (U = 216368.00, p <.001, d =
1.82), and a significant difference but with a smaller effect size between the diagnosed
ASC and potential ASC groups (U = 40045.50, p <.001, d = 0.43. A significant
difference was found between all three female groups on the ‘cognitive empathy’ scale:
X2(2) = 1235.11, p < .001. Diagnosed ASC participants scored lowest, followed by
potential ASC participants, and no ASC participants. Applying Bonferroni corrections, a
significant difference with large effect sizes was found between the diagnosed ASC and
no ASC groups (U = 23987.00, p < .001, d = 2.37), the potential ASC and no ASC
groups (U = 245633.00, p <.001, d = 1.70), and a significant difference but with a
85
medium effect size between the diagnosed ASC and potential ASC groups (U =
36615.00, p = .001, d = 0.55). A significant difference was found between all three
female groups on the ‘cognitive empathy’ scale: X2(2) = 671.409, p < .001. Diagnosed
ASC participants scored lowest, followed by potential ASC participants, and no ASC
participants. Applying Bonferroni corrections, significant differences with large effect
sizes were found between the diagnosed ASC and no ASC groups (U = 78200.50, p < .
001, d = 2.15), and between the potential ASC and no ASC groups (U = 451474.500, p
<.001, d = 0.94), but no significant differences were found between the diagnosed ASC
and potential ASC groups (U = 47663.500, p = .060).
A similar pattern was observed for the males, with a significant difference found
between the three groups: X2(2) = 286.995, p < .001. The ASC participant group scored
lowest, followed by the potential ASC participant group, and the no ASC participant
group. Applying Bonferroni corrections, significant difference with large effect sizes
were found between the ASC and no ASC groups (U = 1661.50, p < .001, d = 2.20), and
the potential ASC and no ASC groups (U = 17719.50, p <.001, d = 1.75), but no
significant differences were found between the ASC and potential ASC groups (U
=1527.50, p = .136). ). A significant difference was found between all three male groups
on the ‘cognitive empathy’ scale: X2(2) = 283.025, p < .001. The diagnosed ASC group
scored lowest, followed byt the potential ASC group, and the no ASC group. After
Bonferroni corrections, significant differences with large effect sizes were found
between the diagnosed ASC and no ASC groups (U = 1048.500, p < .001, d = 2.56), and
the potential ASC and no ASC groups (U = 19167.500, p <.001, d = 1.69), and a
significant difference but with a medium effect size was found between the diagnosed
ASC and potential ASC groups (U = 1251.00, p = .007, d = 0.66). A significant
difference was found between all three male groups on the ‘emotional reactivity’ scale:
X2(2) = 107.929, p < .001. The diagnosed ASC group and potential ASC group scored
86
similarly, and lower than the no ASC group. After Bonferroni corrections, significant
differences with large effect sizes were found between the diagnosed ASC and no ASC
groups (U = 7183.00, p < .001, d = 1.02), and the potential ASC and no ASC groups (U
= 42672.00, p <.001, d = 0.96), but no significant difference were found between the
diagnosed ASC and potential ASC groups (U = 1824.00, p = .835).
A Bonferroni corrected p value of 0.02 was established for pairwise comparisons
between males and females per group, revealing a non-significant difference in the EQ
scores for ASC participants (U = 1651.00, p = .166). However, a significant difference
with a small effect size was found between the potentially autistic males and females (U
= 3964.30, p <.001, d = 0.38), with potentially autistic females scoring higher than
potentially autistic males. Similarly, there was a significant difference between non-
autistic males and females (U = 1244328.00, p <.001, d = 0.50), with non-autistic
females scoring higher than non-autistic males. For the cognitive empathy subscale
there was no significant differences between males and females for the diagnosed ASC
group (U = 1953.50, p = .882) or the potential ASC group (U = 47137.50, p = .331), but
there was a significant difference with a small effect size between non-autistic males
and females (U = 1589362, p <.001, d = 0.14). On the emotional reactivity subscale
there was no significant differences between males and females in the ASC group (U =
1551.00, p = .072), but there was in the potential ASC group (U = 36179.50, p <001, d
= 0.51) and the no ASC group (U = 1022391.00, p <.001, d = 59).
3.3.5. Exploring the age of autism diagnosis. Correlations were performed to
determine whether later diagnosis was associated with higher EQ scores amongst males
and females in the diagnosed ASC group. Because age of diagnosis was significantly,
positively correlated with current chronological age for both genders, p values < .001,
age was entered as a control variable. Results were not significant when both males and
87
females in the ASC group were analysed together: partial r(175) = .033, p = .660.
Likewise, when considered separately, results were not significant for either females,
partial r(150) = .053, p = .519, or males, partial r(22) = -.250, p = .240. Furthermore,
age of diagnosis did not show a significant correlation for the cognitive empathy
subscale (partial r(175) = -.016, p = .832) or the emotional reactivity subscale (partial
r(178) = .074, p = .326). Likewise, when considered separately, results were not
significant for either females, partial r(150) = ..036, p = .656 and partial r(150) = .
079, p = .33, or males, partial r(22) = -.360, p = .077 and partial r(22) = -.159, p = .449.
3.3.6. Group differences in mental health diagnoses. As can be seen from
Table 3.2 a higher frequency of females in the diagnosed ASC group had one or more
‘other’ psychiatric diagnoses than females in the potential ASC and no ASC group,
whilst a higher frequency in the potential ASC group had one or more other psychiatric
diagnoses than females in the no ASC group. The difference between groups was found
to be significant, X²(2) = 246.686, p <.001, φ = .253. Females in the diagnosed ASC
group were 1.6 times more likely than those in the potential ASC group and 4.9 times
more likely than those in the no ASC group to have one or more other psychiatric
diagnoses. Females in the potential ASC group were 3.1 times more likely than those in
the no ASC group to have one or more other psychiatric diagnoses.
Table 3.2
Frequency of individuals in each diagnostic group diagnosed with one or more
psychiatric disorders other than ASC
Diagnostic Group 1 + Other Psychiatric
Diagnosis
No Other Psychiatric
Diagnoses
Females
ASC 102 (66.7%) 51 (33.3%)
Potential ASC 387 (56.1%) 303 (43.9%)
No ASC 872 (29.1%) 2126 (70.9%)
88
Diagnostic Group 1 + Other Psychiatric
Diagnosis
No Other Psychiatric
Diagnoses
Males
ASC 14 (53.8%) 12 (46.2%)
Potential ASC 49 (34.0%) 95 (66.0%)
No ASC 189 (16.4%) 965 (83.6%)
A similar pattern can be observed for male participants, with a higher frequency
of males in the diagnosed ASC group having one or more other psychiatric diagnoses
than males in the potential ASC and no ASC group, whilst a higher frequency in the
potential ASC group had one or more other psychiatric diagnoses than males in the no
ASC group. The difference between groups was found to be significant, X²(2) = 46.737,
p <.001, φ = .188. Males in the diagnosed ASC group were 2.3 times more likely than
those in the potential ASC group and 6 times more likely than those in the no ASC
group to have one or more other psychiatric diagnoses. Males in the potential ASC
group were 2.6 times more likely than those in the no ASC group to have one or more
other psychiatric diagnoses.
Comparing males with females in each group, there were no significant
differences between males and females in the diagnosed ASC group, X²(1) = 1.602, p = .
206, φ = .095. However, a significant difference between males and females was found
in the potential ASC group, X²(1) = 23.237, p <.001, φ = .167. Females in the potential
ASC group were 2.5 times more likely than males in this group to have one or more
other psychiatric diagnoses. A significant difference was also found between males and
females with no ASC, X²(1) = 70.738, p <.001, φ = .131. Females in the no ASC group
were 2.1 times more likely than males in this group to have one or more other
psychiatric diagnoses.
3.3.7. Differential psychiatric diagnoses in diagnosed ASC and potential
ASC. As can be seen from Table 3.3 a higher frequency of females in the potential ASC
89
group had a diagnosis of BPD compared to those in the diagnosed ASC group and the
no ASC group. Females in the diagnosed ASC group also had a higher frequency of
BPD diagnoses than those with no ASC. The difference between groups was found to be
significant, X²(2) = 47.719, p <.001, φ = .111. Females in the potential ASC group were
1.3 times more likely than those in the diagnosed ASC group and 5.7 times more likely
than those in the no ASC group to have a BPD diagnosis. Females in the diagnosed ASC
group were 4.6 times more likely than those with no ASC to have a BPD diagnosis.
Table 3.3
Frequency of autistic individuals versus potentially autistic individuals reporting
specific psychiatric diagnoses
Diagnosis Females Males
ASC Potential
ASC
No ASC ASC Potential
ASC
No ASC
Schizophrenia 1
(0.7%)
3
(0.4%)
2
(0.1%)
0
(0.0%)
1
(0.7%)
2
(0.2%)
Schizoid PD 0 1 2 1 1 3
90
Diagnosis Females Males
(0.0%) (0.1%) (0.1%) (3.8%) (0.7%) (0.3%)
BPD 5
(3.3%)
28
(4.1%)
22
(0.7%)
1
(3.8%)
0
(0.0%)
5
(0.4%)
OCD 13
(8.5%)
42
(6.1%)
53
(1.8%)
4
(15.4%)
3
(2.1%)
19
(1.6%)
ADHD 12
(7.8%)
13
(1.9%)
25
(0.8%)
2
(7.7%)
2
(1.4%)
14
(1.2%)
Affective
Disorder
97
(63.4%)
358
(51.9%)
756
(25.2%)
12
(46.2%)
38
(26.4%)
149
(12.9%)
A higher frequency of females in the diagnosed ASC group had an OCD
diagnosis than females in the potential ASC and no ASC groups, and those in the
potential ASC groups had a higher frequency than those in the no ASC group. The
difference between groups was found to be significant, X²(2) = 57.135, p <.001, φ = .
122. Females in the diagnosed ASC group were 1.4 times more likely than those in the
potential ASC group and 5.2 times more likely than those in the no ASC group to have
an ASC diagnosis. Females in the potential ASC group were 3.6 times more likely than
those in the no ASC group to have an OCD diagnosis.
A higher frequency of females in the diagnosed ASC group had an ADHD
diagnosis than females in the potential ASC and no ASC groups, and those in the
potential ASC groups had a higher frequency than those in the no ASC group. The
difference between groups was found to be significant, X²(2) = 57.885, p <.001, φ = .
123. Females in the diagnosed ASC group was 4.4 times more likely than those in the
potential ASC group and 10.1 times more likely than those in the no ASC group to have
an ADHD diagnosis. Females in the potential ASC group were 2.3 times more likely
than those in the no ASC group to have an ADHD diagnosis.
91
Lastly, a higher frequency of females in the diagnosed ASC group had an
affective disorder diagnosis than females in the potential ASC and no ASC groups, and
those in the potential ASC groups had a higher frequency than those in the no ASC
group. The difference between groups was found to be significant, X²(2) = 259.745, p
<.001, φ = .260. Females in the diagnosed ASC group were 1.6 times more likely than
those in the potential ASC group and 5.1 times more likely than those in the no ASC
group to have an affective disorder. Females in the potential ASC group were 3.2 times
more likely than those with no ASC to have an affective disorder.
Chi- Squares could not be calculated for Schizophrenia and Schizoid PD as the
frequency count was too low. Likewise, results for other psychiatric diagnoses for males
were not analysed as the frequency count was too low.
3.4. Discussion
Previous literature has suggested that autistic women may miss being diagnosed or be
misdiagnosed with other conditions. The FPT suggests that this is because autistic
females show fewer autistic characteristics than autistic males (Kopp & Gillberg, 1992).
However, very few studies examining this theory have explored non-clinical samples of
autistic women who do not have a diagnosis. It is vital that this population is explored,
as research is unable to confirm the FPT when only knowledge of those who have
received a diagnosis is available. Therefore, the aim of this study was to try to identify a
sample of women with high autistic traits indicative of a potential ASC diagnosis and to
compare them with women who had received a formal ASC diagnosis. As well as this,
the study aimed to examine the possible mental health implications of being
undiagnosed, and whether women with a potential ASC are more likely to report
psychiatric problems, which might occur due to the stress of living with an unknown
92
condition, the exhaustion of attempting to hide traits, or as a result of clinicians
misinterpreting symptoms.
Firstly, it was predicted that a larger number of women than men would be
identified as being potentially autistic. This hypothesis was supported, as it was found
that almost 18% of women and 11% of men were potentially autistic according to the
AQ screening tool. This is a much larger proportion than expected. Baron-Cohen et al.
(2009) discovered that 1% of the general population of children in their sample were
potentially autistic, and although it may be argued that parental assessments are less
accurate, Baron-Cohen et al. (2001) used the AQ on adults and found 1% of females and
3.9% of males in the general population were potentially autistic without a diagnosis.
Taking into account the identification accuracy percentage put forward by Sizzo et al.
(2015) of around 70%, these figures still remain high. It may be the case that with
growing autism awareness individuals now have more insight into their own autistic
traits. However, it is probable that the sample collected in the present study was heavily
biased given that a higher prevalence of diagnosed autistic women took part in the
survey than previous prevalence surveys on the general rates of autism diagnosis had
estimated (3.98% vs 1.7%) (Russell, 2014). This suggests that whilst measures were
taken to avoid sharing the wider aims of the research initially, the true purpose of the
survey was likely to be discovered by participants. This may have happened as result of
the debrief being given out prior to all participants completing the study, and the
subsequent media attention the study received. Thus the study may have attracted more
late-diagnosed women and women who might have been aware of their high autistic
traits but who may have not yet received a diagnosis. This limitation is discussed in
more detail in the General Discussion (Chapter 6). Regardless of concerns around
estimating the prevalence rates in this cohort, the aim of the study was to identify a
group of potentially autistic women, which this study has achieved.
93
As predicted, the diagnosed autistic women in this sample were diagnosed
significantly later than autistic men, around the age of 23.57 compared to 16.92. These
results confirm those made previously; for example, Bancroft (2012) found that 58% of
their sample did not receive a diagnosis until after the age of 18, with a mean age of
diagnosis around 25. It should be noted that the autistic men in this sample were also
diagnosed significantly later than previous studies have estimated. The average age of
diagnosis for autism has been found to be between the ages of 3 to 10 years (Brett et al.,
2016; Crane et al., 2016; Daniels & Mandell; 2014; Williams et al., 2008). It is likely
that due to the small sample size of autistic men, this figure has been skewed by several
late-diagnosed participants. However, findings from the current study, that the majority
of autistic women were diagnosed in adulthood and the majority of autistic men were
diagnosed in childhood, with autistic women being 3.6 times more likely to have
received their diagnosis in adulthood compared to autistic men, are clearly in line with
previous research.
In terms of the EQ scores, it was hypothesized that females in the potential ASC
group would demonstrate less impairment on the EQ than females in the diagnosed ASC
group, and that for females in the diagnosed ASC group, the age of diagnosis would
correlate positively with the EQ score. Findings only partially supported these
predictions; whilst age of diagnosis did not correlate with EQ score, a slight empathy
advantage was found for women in the potential ASC group. This was not the case for
males in the potential ASC group, who scored similarly to diagnosed autistic males.
Regardless of this slight advantage, both males and females in the potential ASC group
demonstrated empathy impairments relative to participants without an ASC. However,
both males and females in the potential ASC group showed a significant advantage on
the cognitive empathy subscale over participants in the diagnosed ASC group, but
similar levels of emotional reactivity. This is in line with previous findings, which have
94
suggested that it is cognitive empathy rather than affective empathy that is affected in
diagnosed autistic individuals (Mul et al., 2018). Nevertheless, it should be noted that
the emotional reactivity subscale does not fully measure affective empathy, as it fails to
take into account levels of personal distress, and therefore cannot determine whether the
reaction is self-orientated or a reflection of affective empathy for others (Lawrence et
al., 2004). It was argued earlier that empathy might be able to assist in improved
socialisation and help autistic individuals to mask their traits and ‘fit in’ with others,
which autistic females have been found to be better at than autistic males (Hiller et al.,
2014; McLennan et al., 1992). The current finding that there was no difference in
empathy between females and males in the ASC group are in line with those by Lai et
al. (2011), who also failed to uncover differences between autistic males and autistic
females on impairments in empathising. Despite this, their study still found less socio-
communicative difficulties in autistic women, suggesting that other factors are at play in
the later diagnosis of autistic women. It would appear that those with a potential ASC
are impaired on screening questionnaires relative to those without an ASC but may
demonstrate slight advantages relative to those with a diagnosed ASC.
In terms of mental health diagnoses other than ASC, it was predicted that more
females in the potential ASC group would have one or more psychiatric diagnoses than
those in either the diagnosed ASC group or the no ASC group. This was not found to be
the case. Whilst the potential ASC group reported more psychiatric diagnoses than those
in the no ASC group, those in the diagnosed ASC group were the most likely to have
other psychiatric diagnoses. Nevertheless, whilst there was no difference in the
frequency of psychiatric diagnoses between males and females in the diagnosed ASC
group, females in the potential ASC group were 2.5 times more likely than males in the
same group to have one or more psychiatric diagnoses. The same pattern was observed
when comparing males and females in the no ASC group. These findings appear to
95
conflict with previous literature suggesting that undiagnosed autistic females may be at
a raised risk of mental health problems due to the stress of camouflaging and masking
autistic traits (Hull, Mandy, et al., 2019; Livingston et al., 2018; Stagg & Belcher,
2019), although it is important to note that the current study is the first to compare
potentially autistic females with diagnosed autistic females. Possibly, women with an
ASC diagnosis had tended to collect other formal psychiatric diagnoses because they are
known to mental health services and may even have received other diagnoses at the time
of their ASC diagnosis. This suggestion is supported by the finding that autistic females
were not more likely than autistic males to have other psychiatric diagnoses, despite
females generally being more likely to have one or more psychiatric diagnoses in the
general population. Alternatively, it is possible that those with a diagnosed ASC may be
more vulnerable to mental health problems as a result of the stigma associated with
diagnosis, or due to more severe impairments.
In contrast, the prediction that females in the potential ASC group would be
more likely to have diagnoses that could be classed as differential diagnoses due to
overlapping features with ASC, in particular BPD, was supported. Females in the
potential ASC group were found to be 1.3 times more likely than females in the
diagnosed ASC group and 5.7 times more likely than females in the no ASC group to
have a diagnosis of BPD. This supports previous literature which has suggested that
clinicians may diagnose BPD over ASC due to a similarity in symptoms (Bargiela et al.,
2016; Lai & Baron-Cohen, 2015; Ryden et al., 2008; Rabbitte et al., 2017). For
example, both autistic women and women with BPD may demonstrate difficulties in
relationships, regulating their emotions, impulsivity, and stress-related paranoid ideation
(Fitzgerald, 2005). With classic signs of autism masked, such as RRBIs and socio-
communication problems, clinicians may favour diagnosing BPD, which is more
commonly seen in females in the general population (APA, 2000). However, without a
96
full ASC assessment of these potentially autistic women, we cannot determine for sure
if they have been misdiagnosed with BPD or whether this is a co-morbid condition.
All other differential psychiatric diagnoses (OCD, ADHD, and affective
disorders), were found to be more prevalent in women in the diagnosed ASC group.
There appeared to be no differences between groups for Schizophrenia or Schizoid PD
diagnoses, although numbers were too small to calculate significant differences. Rates
of OCD in the female diagnosed ASC group were slightly lower than those found by
Russell et al. (2016) (8.5% vs 17.9%), although higher than those found in this study in
the general population (4.4%). ADHD rates were more similar (7.8% vs 9.7%), and
again higher than found in the general population (2.3%). Affective disorders were
grouped together in the current study, making it difficult to compare to Russell et al.’s
(2016) figures, although when grouped together the current study’s appeared to be
higher (63.4% vs 44.4%). It should be noted that Russell et al.’s (2016) study was based
on both autistic males and females, whereas the current study has only been able to
examine the female data. This may explain some of the slight discrepancies in figures.
Higher rates of ADHD and OCD among those women who were diagnosed
than those who are were potentially undiagnosed are in line with arguments put forward
by Dworzynski et al. (2012), who suggest that in order for girls to be diagnosed with
autism they require a greater number of external behavioural problems than boys. In
their study, females who scored high on the CAST but who had less hyperactivity and
behavioural problems, possibly due to internalising of traits, were less likely to receive a
diagnosis than females and males with these presenting issues. This may explain why
diagnosed women in the current study were more likely to have ADHD and OCD than
potentially autistic women, as they possess some external behavioural symptoms.
97
Taken together, results of Study 1 provide some support for the FPT. In
particular, the types of other psychiatric diagnoses seen in females in the potential ASC
group compared to those seen in the diagnosed ASC group suggest different behavioural
manifestations of symptoms. However, further examination of potentially autistic
women is required to fully understand their profile. For example, the current study has
not tested whether those females in the potentially autistic group present with less social
impairments than those who are diagnosed, as the FPT would suggest, especially given
their slight empathy advantage. Accordingly, Study 2 explores differences in social
functioning between diagnosed autistic females and potentially autistic females, as well
as the association between social functioning and self-monitoring (a proxy for
camouflaging). Secondly, Study 1 only looked at diagnosed psychiatric conditions. As
already discussed, it is possible that those with an autism diagnosis are better known to
services and therefore more likely to receive other psychiatric diagnoses from clinicians.
Therefore, Study 2 compared diagnosed autistic and potentially autistic women for
undiagnosed mental health problems by administering self-report measures of
depression and anxiety. Finally, for diagnosed autistic women, information was
collected not only about the age of ASC diagnosis but the ages of all other psychiatric
diagnoses. In this way, Study 2 aimed to build a typical timeline of mental health
diagnoses among women with autism.
98
99
CHAPTER 4
Study 2: A Comparison of Social, Emotional, and Behavioural Traits between
Potentially Autistic Females and Diagnosed Autistic Females
4.1. Introduction
In Chapter 3 a large number of potentially autistic women without diagnoses were
identified. These women had a slight but significant empathy advantage relative to
diagnosed autistic females, specifically in cognitive empathy, were more likely to be
diagnosed with BPD than diagnosed autistic females, and were more likely to have one
or more other psychiatric diagnoses than their male counterparts. However, diagnosed
autistic females were equally as likely as diagnosed autistic males, and more likely than
potentially autistic women, to have one or more psychiatric diagnoses. Additionally,
they were more likely to be diagnosed with ADHD, OCD, and affective disorders than
potentially autistic women. This study left several key questions unanswered, which this
chapter aims to address. The first question is why might these potentially autistic
women be undiagnosed? More specifically, as well as a slight empathy advantage, do
these women also have better social skills and do they use camouflaging strategies to
mask autistic traits? Secondly, whilst potentially autistic women may have fewer mental
health diagnoses than diagnosed autistic women, might they still have higher traits of
anxiety and depression that have not been diagnosed? Finally, do diagnosed autistic
women tend to receive their other psychiatric diagnoses before or after their ASC
diagnosis?
The FPT (Kopp & Gillberg, 1992) suggests that one of the reasons why autistic
females may have a missed or late diagnosis is because they often have a different
manifestation of autistic traits, which acts as a mask. For example, Dworzynski et al.
100
(2012) suggest that in order for girls to be diagnosed with autism they require a greater
number of external behavioural problems than boys. Girls who scored above the cut-off
on the CAST, which was filled in by parents, but who did not meet the full diagnostic
criteria, were less likely to be diagnosed as autistic than their male equivalents (38% vs
56%). Additionally, these girls had fewer social autistic traits than diagnosed girls
(partial ŋ2 = .09). This study stresses the importance of investigating undiagnosed yet
high autistic trait scoring females, who may be undiagnosed due to exhibiting less
challenging and observable behaviours. The majority of studies investigating
differences between autistic males and females rely on already diagnosed individuals,
which means the females will have displayed enough autistic traits to be sent for
diagnosis (Halladay et al., 2015).
Women who are potentially autistic but undiagnosed may be more motivated to
intentionally camouflage in social situations to disguise their autism. Research
investigating the social behaviours of autistic females, has found that they show some
advantages over autistic males, which may support the FPT. For example, Hiller et al.
(2014) compared 69 autistic girls with 69 autistic boys (mean age 8-9 years) on clinician
and teacher reports about social functioning. The autistic girls were 14 times more likely
than the autistic boys to engage in typical reciprocal conversation, 3.5 times more likely
to engage in imaginative play typical for their developmental age, and 6 times more
likely to show some adjustment of their behaviours across situations. This included the
ability to monitor voice volume and avoid inappropriate comments and public
meltdowns. This may mean that the behaviour of autistic girls appears less atypical than
that of autistic boys to others observing them.
The ability to monitor social behaviours can be referred to as ‘self-monitoring’,
which Snyder (1974) developed a scale to measure. The Self-Monitoring Scale (SMS)
101
looks at individuals’ ability to monitor their own inner state, the social situations they
are in, and to change and monitor their own behaviour accordingly to fit into different
social contexts. Whilst the measure has not previously been used with autistic people, it
seems reasonable to suppose that it might be a useful tool to examine whether autistic
females try harder than autistic males to camouflage their autistic traits. For example,
Ickes and Barnes (1977) found that non-autistic females scored higher on self-
monitoring than non-autistic males, which therefore may indicate a general female
advantage. Furthermore, Snyder (1974) found that peers of individuals with high SMS
scores thought that they were good at learning how to behave in socially acceptable
ways in new situations and were good impression makers, and that high self-monitoring
scorers were more likely than low self-monitoring scorers to seek out social comparison
information about their peers. Estow et al. (2007) reported that students mimicked
videotaped individuals more if they were high self-monitors, and Schaffer et al. (1982)
found that high self-monitoring individuals were more likely than low self-monitors to
mimic a confederate. Given that social mimicking is thought to be a key strategy in
camouflaging by autistic females, who have been found to closely observe the
behaviour of others to copy in different social contexts (Atwood & Grandin, 2006;
Baldwin & Costley, 2016; Hull, Petrides, et al., 2017; Tierney et al., 2016), the SMS
could give some indication as to whether potentially autistic women are using social
strategies that mask their autistic traits.
Some studies have found that autistic girls also have an advantage over autistic
boys on measures of friendship, which may be related to a better ability to adapt in
different social settings, and reduced atypical behaviours. For example, Sedgewick et al.
(2016) compared 13 autistic girls with 10 autistic boys, 13 non-autistic girls, and 10
non-autistic boys on friendship motivation and experience using the Social
Responsiveness Scale (SRS-2) and Friendship Qualities Scale (FQS). Autistic girls were
102
found to score similarly to non-autistic girls on the social motivation (SRS-2) and
closeness (FQS) subscales, which was significantly higher than autistic males (d = 0.89
and 1.15 respectively). In addition to these findings, Dean et al. (2017) found that
autistic girls participated in more ‘joint engagement’ with other groups of girls during
play at school, whilst the autistic boys spent more time by themselves in ‘solitary’ play.
However, these autistic girls often appeared to take a background role, flitting between
activities to appear to be engaged, when actually they were spending more time than
non-autistic girls by themselves. These findings suggest that autistic girls have some
awareness of the social environment around them, and that they are more motivated to
try and ‘fit in’ than autistic males. This could again hide autistic girls’ social
impairments. However, in a previous study where autistic adults were tested using the
Friendship Quotient (FQ) (Baron-Cohen & Wheelwright, 2003), whilst autistic
participants were found to score significantly worse than non-autistic participants, no
gender differences were found between autistic males (n = 51) and autistic females (n =
17). Autistic females scored on average 59.8 (SD = 25.1) compared to autistic males
who scored on average 53.2 (SD = 18.3). This null finding could reflect the small
number of autistic females tested in comparison to autistic males, resulting in low power
to detect a group difference, or it may be the case that when this study was conducted in
2003, many autistic females with heightened social skills and better friendships were
not yet diagnosed. It would be useful, therefore, to investigate whether potentially
autistic women perform better on the FQ than diagnosed autistic women.
As discussed in previous chapters, a probable consequence of autistic females
camouflaging and masking their autism is greater mental health problems (Cassidy et
al., 2018; Hull, Mandy, et al., 2019). Livingston et al. (2018) suggested that this was
because techniques which mask autism use up valuable cognitive resources. Whilst
Study 1 looked at incidences of different types of mental health diagnosis in potentially
103
autistic women compared to diagnosed autistic women, as yet research has not
investigated whether potentially autistic females suffer more depressive and anxiety
related symptoms than diagnosed autistic women. If their autism is undiagnosed
because of greater camouflaging ability then we might expect better social functioning
at the expense of mental health, due to the increased stress of maintaining this mask.
There is currently a gap in the literature on the topic of the FPT. Several studies
have explored social behaviour differences between autistic males and autistic females,
but only one has considered the large number of potentially autistic females
(Dworzynski et al., 2012), who could be expected to be even better at hiding their
autistic traits than their diagnosed peers. However, this study looked at children only.
This chapter therefore aims to once again explore a group of potentially autistic women,
looking in more detail at what subtle differences in social behaviours they show
compared to diagnosed autistic females.
4.1.1. Aims and hypotheses. Previous literature has suggested that autistic
women may be diagnosed later due to a lack of social impairments and increased social
camouflaging. Furthermore, Study 1 uncovered a large number of potentially autistic
women who had a significant empathy advantage over those with a diagnosis. Several
key questions remain unanswered about this population, which Study 2 aims to address.
These include the following:
1. Do potentially autistic women demonstrate an advantage in social abilities
relative to diagnosed autistic women? It was predicted that potentially autistic
women would demonstrate better self-monitoring, friendship quality, social
functioning, and ToM.
2. Is greater empathy associated with better social abilities? It was predicted that all
three groups of female participants would show positive correlations between
104
empathy, particularly cognitive empathy, self-monitoring, friendship quality,
social functioning and ToM.
3. Is the age of autism diagnosis for autistic women predicted by social abilities? It
was predicted that age of autism diagnosis would be correlated positively with
measures of self-monitoring, friendship quality, social functioning, and ToM.
4. Study 1 found that autistic women were more likely to have other mental health
diagnoses than potentially autistic women, but might potentially autistic women
still demonstrate more depressive and anxiety symptoms? It was predicted that
potentially autistic women would score higher on self-report measures of
depression and anxiety.
5. In women with a diagnosed ASC, what is the typical timeline on which they
receive their additional mental health diagnoses? It was predicted that for most
such women, their other mental health diagnoses would tend to be received at a
younger age than their ASC diagnosis.
Although the main objective of Study 2 was to compare results for potentially
autistic and diagnosed autistic women, male participants and non-autistic women were
also included in the sample. Where numbers permitted, these groups were included in
the analyses.
4.2. Method
4.2.1. Participants. The current study had the same 2 (gender) x 3 (group) design as
used in Study 1, with a target population of young adults (aged 16-40) from the UK.
Some of the sample was derived from the previous study; all participants who left their
email addresses and gave consent to be re-contacted were sent the second survey. As the
number of males in the previous sample was quite low and numbers could be expected
to drop for the follow-up study, the new survey was also re-advertised through social
media and through autism groups and autism research centres in the hope of increasing
the number of males participating. Using G Power 3.1.9.2 with an alpha level of 0.05, a
105
power level of 0.95, and an effect size of 0.3, which was based on Study 1’s findings, a
minimum of 226 participants was required for conducting an ANOVA with six groups.
Again, the adverts used for participant recruitment purposefully did not mention autism,
but instead called for participants to take part in a study looking at ‘gender differences
in social awareness and motivation’.
1,005 individuals who met the criteria began taking the survey, 390 of these
responses came from participants emailed from the previous survey (10.14% of
previous participants re-contacted). A total of 513 people completed the entire survey, of
whom 372 were previous participants re-contacted and 141 were new participants.
Of the participants who completed the survey, 103 were males, 402 were
females, and 8 identified as ‘other’ or preferred not to say. Of all participants, 41
claimed the gender they now identified with was different to the gender they were
assigned at birth. Of those who reported having a diagnosed ASC, 90 were female and
27 were male. The average age of diagnosed autistic females was 28.84 (SD = 6.193),
and 26.56 (SD = 6.216) for diagnosed autistic males. Of those in the potential ASC
group, who scored above the clinical criteria on the AQ (≥ 32) but who did not have a
diagnosis (77 females and 9 males), the average age of females was 30.56 (SD = 5.819)
and the average age of males was 26.67 (SD = 7.517). Of those with no ASC (235
females and 67 males), the average age of females was 26.24 (SD = 5.574) and the
average age of males was 25.42 (SD = 5.252).
56.9% were either in full-time or part-time employment, 31.7% were in higher
education, and 11.5% were unemployed and not students. Participants were spread
across the UK and lived in over 60 different counties, with the majority residing in
Cambridgeshire (10.3%), Greater London (8.7%), Essex (4.5%), Surrey (4.3%), West
Yorkshire (4.1%), and Greater Manchester (3.7%).
106
4.2.2. Measures
AQ: The full 50-item Autism Quotient (AQ) (Baron-Cohen et al., 2001) was
used to evaluate autistic traits. A more detailed description of the measure can be found
in Chapter 3, section 3.2.2.
EQ: The 40-item version of the Empathy Quotient (EQ) (Baron-Cohen &
Wheelwright, 2004) was used to evaluate empathising. A more detailed description of
the measure can be found in Chapter 3, section 3.2.2. The EQ scores were again split
into two subscales reflecting cognitive empathy and emotional reactivity.
Self-Monitoring Scale: The Self-Monitoring Scale (SMS) was used, which is a
25-item scale yielding ‘yes’ or ‘no’ responses from participants on each item (Snyder,
1974). This scale looks at the self-control of expressive behaviours, which requires the
ability to monitor one’s own inner state and the social situations one is in, and to change
and monitor one’s own behaviour accordingly. Ickes and Barnes (1977) established a set
of norms for the scores, with 15-22 indicating a high score, 9-14 indicating an
intermediate score, and 0-8 indicating a low score. The scale has good reliability (r = .
70) and test-retest reliability (0.83) (Snyder, 1974). However, Briggs et al. (1980) have
suggested that rather than being one dimension, the SMS is made up of three distinct
dimensions (acting, extraversion, and other-directedness), which may conflict with each
other. For example, other-directedness correlates positively with shyness and
neuroticism, whereas extraversion correlates negatively with shyness and positively
with self-esteem and sociability. Therefore, as recommended by these authors, the
current study will consider scores on the full scale as well as those that could be
hypothesised to relate to social camouflaging separately (namely, other-directedness and
acting).
107
The Friendship Questionnaire: The Friendship Questionnaire (FQ) is a 35-item
scale (27 of which are scored) measuring an important part of normal social functioning,
the quality of participants’ friendships and relationships (Baron-Cohen & Wheelwright,
2003). There are a number of different response styles used within the survey, ranging
from Likert scales to rankings, with a maximum possible score of 135 in total. Higher
scores on the FQ indicate that the respondent values close, empathic, supportive, and
caring friendships, and that they enjoy the company of people, and interacting with
others for its own sake rather than for another purpose. Baron-Cohen and Wheelwright
(2003) found that generally non-autistic women score higher on the scale than non-
autistic men, and that autistic people without intellectual disabilities score lower than
non-autistic people. They found that the internal consistency of the scale was excellent,
with Chronbach’s alpha ranging from 0.75 – 0.84. Convergent validity has been found
with other scales related to the FQ, for example Lyons and Aitken (2010) found that
Machiavellianism was negatively related to the FQ.
Social Functioning Scale: Birchwood et al.’s (1990) Social Functioning Scale
(SFS) is a 79-item, 7 factor self-report assessment initially developed to assess social
functioning relevant to the needs and impairments of individuals with schizophrenia.
The questionnaire has been designed to be taken by both the person to whom it applies
and by a relative or someone in daily contact with the person. However, due to
accessibility of the online survey the current study only used the first part of the
assessment. In the initial validation of the scale by the authors, no differences in the
scores between the relative and the self-report were observed (inter-rater reliability, r =
0.94), suggesting that the scale is valid to be used on just the participant alone.
The SFS has good reliability (r = .80) and good internal consistency, as
demonstrated by item-total correlations (r = 0.71). Factor analyses revealed that it was
108
appropriate to obtain a mean score for the whole SFS scale, as well as on individual
factors. Birchwood et al. (1990) found that around 50% of participants in their study
with schizophrenia scored between 86-105, whereas those participants without
schizophrenia scored between 116-135, with none scoring below 86.
The 7 factors were based on the impairments and disability assessed by the
Disability Assessment Schedule (Ustan et al., 2010). They included social
engagement/withdrawal; interpersonal behaviour; pro-social activities; recreation;
independence-competence; independence-performance; and employment/occupation.
Whilst these factors are based on the defining characteristics observed in schizophrenia,
many of these can be seen to overlap with those experienced by individuals with autism;
for example, difficulties in interpersonal relationships and impairment in life-role
functioning (social activities and independence skills). Other available scales, such as
the Weiss Functional Impairment Rating Scale Self-Report (WFIRS-S), did not appear
to be as specific to the types of social impairment found in autistic individuals.
Moreover, Canty et al. (2017) further validated the survey in their study on ‘healthy’
participants, to test a new measure of ToM.
Reading the Mind in the Eyes Test (brief version): The current study used the
brief version of the Reading the Mind in the Eyes Test (RMET) (Olderbak et al., 2015),
which was initially developed by Baron-Cohen et al. (2001), in order to measure ToM.
The original RMET was designed to identify different clinical populations (mainly
autistic people) from non-autistic controls in ToM capabilities. The original RMET
presents subjects with 36 images of other peoples’ eyes and gives them a choice of four
terms to choose from, which could describe the person’s mental state. Whilst the full
revised version of the test reported adequate reliability, the new brief version of the test,
which includes just 10 of the items of the original test, reported better internal
109
consistency (α = 0.73). It is therefore a more precise measure of ToM and shorter to
administer.
The Patient Health Questionnaire – 9: The 9 item version of the Patient Health
Questionnaire was used (PHQ-9), which specifically measures depression using the 9
DSM-IV criteria (Kroenke et al., 2001). Participants rate each item as to how often they
experience the symptom from ‘not at all’ to ‘every day’. Scores ranging from 5-9
represent mild depression, 10-14 represent moderate depression, 15-19 represent
moderately severe depression, and 20 + represent severe depression. The internal
reliability of the scale is excellent, with Chronbach’s alpha ranging from 0.84 – 0.89.
The PHQ-9 also has excellent test-retest reliability (r = 0.84) and good construct
validity, with scores on the scale strongly associated with functional status, disability
days, and symptom-related difficulty. Furthermore, good external validity for the scale
was found by replicating the initial findings to a second sample, suggesting that the
PHQ-9 may be generalizable to outpatients in a variety of clinic settings (Kroenke et al.,
2001).
Generalized Anxiety Disorder – 7: The Generalized Anxiety Disorder – 7 (GAD-
7) scale has 7 items derived from the DSM-IV symptom criteria for GAD and from
other existing anxiety scales (Spitzer et al., 2006). Similarly to the PHQ-9, participants
rate each item as to how often they experience the symptom from ‘not at all’ to ‘every
day’. Scores ranging from 5-9 represent mild anxiety, 10-14 represent moderate anxiety,
and 15+ represent severe anxiety. The GAD-7 has excellent reliability (α = .92) and test-
retest (r = 0.83). The scale also has strong construct validity, with scores associating
strongly with scores from a functioning scale, and convergent validity, with scores on
the scale correlating strongly with two other anxiety scales (Spitzer et al., 2006).
110
4.2.3. Design. Participants were grouped by gender and also by autism status:
those diagnosed with an ASC (‘autistic/diagnosed ASC group’), those without an ASC
diagnosis who scored above the criteria on the AQ (≥32) (‘potentially autistic/potential
ASC group’), and those without an ASC diagnosis who scored below the criteria on the
AQ (≤ 32) (‘non-autistic/no ASC group’). A between-subjects analysis was conducted
on scores from the various questionnaires.
4.2.4. Procedure. The survey was designed on Qualtrics, and tested prior to
distribution by three members of the research team who underwent the survey as though
they were participants. The survey was set to open access allowing anyone to take it,
however it only allowed for one response per participant; this was achieved through the
monitoring of cookies. Items were set to forced response, and progression through the
survey was dependent on all items being answered (non-response options were provided
throughout).
Participants who took part in Study 1 were asked to enter a password they were
emailed using the email addresses they had left in the previous study, which enabled
them to skip the AQ and EQ measures. Alternatively, if they had not taken part
previously then they were asked to select this option and were directed to a version of
the survey which included the AQ and EQ. Participants were presented with an
information page before beginning the survey; this informed them that the online survey
was looking at gender differences in autistic traits, mental health, and individuals’ social
awareness and motivation. They were also informed that they would have a chance to
win a £100 Amazon voucher upon completion of the survey.
The main survey presented participants with 5 blocks containing 6
questionnaires: the AQ was used to screen for autistic traits; the EQ was used to
measure empathy; the FQ was used to measure quality and motivation of friendships;
111
the Self-Monitoring Scale (SMS) was used to measure how well participants could
adapt to different social situations; the Social Functioning Scale (SFS) was used to
measure social functioning; the brief version of the RMET was used to measure ToM;
the PHQ-9 was used to measure depression; and the GAD-7 was used to measure
anxiety. They were then asked to indicate any autism or mental health diagnoses they
had received and at what age, and to fill in a number of demographic questions about
their age, gender, country/county of birth, and employment status. Once the survey was
completed, the participants were fully debriefed and informed that the study was
“looking specifically at whether social motivation and awareness was related to high
scores on an autism screening tool in individuals who are not diagnosed with autism;
more specifically whether there are gender differences”. They were also made aware
that the AQ was not a diagnostic test and that it just looked at traits, and that we were
unable to disclose individual scores for ethical reasons, however advice and support
contacts were provided. Finally, they were given the opportunity to leave their email
addresses to be entered into the prize draw.
4.3. Results
4.3.1. Data checks and descriptive statistics. Group means and standard deviations on
all measures are presented for participants in the diagnosed ASC group, potential ASC
group, and no ASC group, separately for females (Table 4.1) and males (Table 4.2). Due
to the low number of participants in the potential ASC male group it was not possible to
conduct the same analyses for males. Secondary analyses on males in the ASC group
were conducted largely for descriptive and replication purposes.
Distributions for each of the three groups were visually inspected for normality.
These revealed some departure from normality on most variables tested and therefore
112
non-parametric tests were used throughout the analysis. Kruskal-Wallis H tests were
used to explore group differences and Mann-Whitney U tests were used to explore pair-
wise comparisons; the Wilcoxon Signed Ranks test was used for pairwise comparisons
for within subjects, both with Bonferroni corrections applied for multiple comparisons.
For correlation analysis Spearman’s was used, and for categorical variable analysis Chi-
Squares were used. The main analysis includes a section on group differences on all
questionnaire measures for female participants, a section on correlation analysis of the
continuous variables derived from the survey results for female participants, analysis of
mental health conditions and age of onset for females participants, and lastly an
exploratory analysis of group differences between males and females in the ASC group.
Table 4.1
Means and standard deviations on all measures for female participants, stratified by
diagnostic group
113
Measure ASC Potential ASC No ASC
n = 90 n = 77 n = 235
Mean (SD) Mean (SD) Mean (SD)
AQ 39.67 (5.13) 37.13 (4.10) 19.30 (7.35)
EQ 18.79 (8.23) 22.27 (10.17) 43.91 (12.34)
FQ 54.88 (21.35) 53.06 (18.36) 81.21 (19.84)
RMET 6.94 (2.40) 7.64 (1.91) 8.31 (1.41)
GAD-7 11.93 (6.02) 10.57 (5.94) 7.20 (5.65)
PHP-9 14.40 (6.30) 12.34 (6.46) 8.75 (6.09)
SMS 10.18 (4.94) 10.45 (4.42) 12.25 (3.92)
SFS 116.64 (26.24) 125.53 (20.29) 141.01 (21.48)
Table 4.2
Means and standard deviations on all measures for male participants, stratified by
diagnostic group
Measure ASC Potential ASC No ASC
n = 27 n = 9 n = 67
Mean (SD) Mean (SD)Mean (SD)
AQ 37.93 (4.86) 35.22 (2.22) 19.76 (6.43)
EQ 17.33 (7.98) 19.67 (6.04) 36.19 (11.10)
FQ 54.48 (25.80) 40.33 (13.64) 67.15 (19.94)
RMET 6.78 (2.03) 7.56 (2.35) 8.27 (0.99)
114
Measure ASC Potential ASC No ASC
n = 27 n = 9 n = 67
Mean (SD) Mean (SD)Mean (SD)
GAD-7 10.22 (4.87) 6.00 (3.71) 4.93 (5.24)
PHP-9 12.41 (5.75) 10.22 (4.68) 7.24 (5.95)
SMS 9.67 (4.38) 11.89 (3.06) 14.28 (3.85)
SFS 113.63 (19.22) 121.33 (13.99) 136.22 (24.51)
4.3.2. Female group differences on questionnaire measures.
EQ: There was a significant difference in empathic traits between female
diagnostic groups: X2(2) = 220.039, p <.001. Using a Bonferroni corrected alpha score
of .02, females in the diagnosed ASC group had a significantly lower EQ score than
those in the potential ASC group (p = .022, d = .38), and both groups had significantly
lower scores than those in the no ASC group (p <.001, d = 2.40 and 1.91). Looking at
the subscales, there was a significant difference in cognitive empathy between
diagnostic groups: X2(2) = 88.16, p <.001. Females in the diagnosed ASC group scored
lowest on this subscale (M = 2.33, SD = 2.48), followed by females in the potential ASC
group (M = 5.45, SD = 4.75), and females in the no ASC group (M = 11.78, SD = 5.19).
The difference between females in the diagnosed ASC group and potential ASC group
was significant and had a large effect size (U = 502.50, p = .009, d = 0.82), as was the
difference between females in the diagnosed ASC group and no ASC group (U =
243.00, p <.001, d = 2.32), and between females in the potential ASC group and no ASC
group (U = 511.00, p <.001, d = 1.27). A significant difference was also found between
diagnostic groups on the emotional reactivity subscale: X2(2) = 44.92, p <.001. Females
in the diagnosed ASC group scored lowest on this subscale (M = 7.16, SD = 3.86),
followed by females in the potential ASC group (M = 8.34, SD = 4.72), and females in
the no ASC group (M = 12.68, SD = 4.58). There was no significant difference between
females in the potential ASC group and those in the diagnosed ASC group (U = 619.50,
115
p = .184), however there were significant differences with large effect sizes between
females in the diagnosed ASC group and no ASC group (U = 835.50, p <.001, d = 1.30),
and between the potential ASC group and no ASC group (U = 688.00, p <.001, d =
0.93).
FQ: There was a significant difference in friendship scores between female
diagnostic groups: X 2(2) = 115.419, p <.001. Using a Bonferroni corrected alpha score
of .02, females in the diagnosed ASC and potential ASC groups scored similarly, and
both groups had significantly lower scores that those in the no ASC group (p <.001, d =
1.23 and 1.47, respectively).
Self-Monitoring: There was a significant difference in self-monitoring between
female diagnostic groups: X 2(2) = 18.832, p <.001. Using a Bonferroni corrected alpha
score of .02, females in the diagnosed ASC and potential ASC groups scored similarly,
and both groups had significantly lower scores that those in the no ASC group (p = .001,
d = 0.46 and p = .005, d = 0.43 respectively). There were no group differences on the
‘other-directedness’ subscale (X 2(2) = .404, p = .817) but there was a significant
difference on the ‘acting’ subscale (X2(2) = 15.50, p <.001) and the ‘extraversion’
subscale (X2(2) = 71.577, p <.001). Females in the diagnosed ASC and potential ASC
groups scored similarly on the acting subscale (M = 1.27 and 1.25 respectively), and
both groups had significantly lower scores than those in the no ASC group (M = 1.77)
(p = .017, d = 0.35 and p = .009, d = 0.39 respectively). Females in the diagnosed ASC
and potential ASC groups also scored similarly on the extraversion subscale (M = 1.39
and 1.56), and both groups had significantly lower scores than those in the no ASC
group (M = 2.83) (p <.001, d = 0.93 and p <.001, d = 0.82 respectively).
Social Functioning: There was a significant difference in social functioning
between female diagnostic groups: X 2(2) = 74.404, p <.001. Females in the diagnosed
116
ASC group had a significantly lower mean SFS score than those in the potential ASC
group, although this was not significant when Bonferroni corrections were applied with
a new alpha criteria of .02 (p = .025, d = 0.38), however the effect size was medium,
and both groups had significantly lower scores than those in the no ASC group (p <.001,
d = 1.02 and 0.74 respectively).
Examining each subscale on the SFS a significant difference between groups
was found for the majority of the subscales. There was a significant difference on the
‘engagement/withdrawal’ subscale between female diagnostic groups: X 2(2) = 78.702, p
<.001. Using a Bonferroni corrected alpha score of .002 throughout all comparisons
described below, females in the diagnosed ASC group scored on average lower (M =
8.29) than those in the potential ASC group (M = 9.39) (p = .002, d = 0.49), who scored
significantly lower than those in the no ASC group (M = 10.90) (p <.001, d = 0.63).
There was a significant difference between groups on the interpersonal communication
subscale: X2(2) = 65.497, p <.001. Females in the diagnosed ASC group scored on
average the same as those in the potential ASC group (M = 7.46 and 7.65 respectively)
but lower than those in the no ASC group (M = 8.44) (p <.001, d = 0.84 and 0.74
respectively). A significant difference on the ‘independence-performance’ subscale was
also found between female diagnostic groups: X 2(2) = 39.821, p <.001. No differences
were found between females in diagnosed ASC and potential ASC groups (M = 26.96
and 29.29 respectively) but both scored significantly lower than those in the no ASC
group (M = 32.74) (p <.001, d = 0.79 & 1.03). A significant difference on the
‘independence competence’ subscale was found between female diagnostic groups:
X 2(2) = 89.276, p <.001. Females in the diagnosed ASC group scored significantly
lower (M = 32.37) than those in the potential ASC group (M = 35.78) (p <.001, d =
0.63), and those in the potential ASC group scored significantly lower than those in the
no ASC group (M = 37.59) (p <.001, d = 0.46). The ‘prosocial’ subscale revealed a
117
significant difference between diagnostic female groups: X 2(2) = 63.834, p <.001. No
difference was found between females in the diagnosed ASC and potential ASC groups
(M = 14.89 and 14.61 respectively), however, both had significantly lower scores than
those in the no ASC group (M = 22.01) (p <.001, d = 0.77 & 0.90 respectively).
Significant differences between diagnostic female groups were found on subscale scores
for employment: X 2(2) = 31.875, p <.001. Females in the potential ASC group and no
ASC group scored similarly (M = 8.13 & 8.70), but both groups scored significantly
higher than those in the diagnosed ASC group (M = 6.59) (p = .001, d = 0.39 and p
<.001, d = 0.57 respectively). Finally, there was no significant difference between
diagnostic female groups on the recreation subscale (X 2(2) = .618, p = .734).
RMET: There was a significant difference in ToM between female diagnostic
groups: X 2(2) = 24.543, p <.001. Using a Bonferroni corrected alpha score of .02,
females in the diagnosed ASC and potential ASC groups scored similarly, and both
groups had significantly lower scores than those in the no ASC group (p <.001, d = 0.71
and p = .007, d = 0.41 respectively).
GAD: There was a significant difference in anxiety between female diagnostic
groups: X 2(2) = 47.328, p <.001. Using a Bonferroni corrected alpha score of .02,
females in the diagnosed ASC and potential ASC groups scored similarly, and both
groups had significantly higher scores than those in the no ASC group (p <.001, d =
0.81 and 0.58 respectively).
Depression: There was a significant difference in depression between female
diagnostic groups: X 2(2) = 55.509, p <.001. Using a Bonferroni corrected alpha score of
.02, females in the diagnosed ASC and potential ASC groups scored similarly, and both
groups had significantly higher scores that those in the no ASC group (p <.001, d = 0.91
and 0.57).
118
4.3.3. Correlations between questionnaire measures for female groups. As
can be seen from the Spearman correlations in Table 4.3, for females in the diagnosed
ASC group, the measures of social functioning were positively associated. Specifically,
with Bonferroni corrections applied due to multiple tests, the AQ was significantly,
negatively correlated with the EQ and FQ, and the EQ was significantly, positively
correlated with the FQ and RMET. The RMET was also significantly, positively
correlated with the FQ. Scores on the SMS and SFS were significantly, positively
correlated with the FQ. Both the GAD and PHQ were significantly, positively correlated
with each other but neither measure of mental health was associated with any of the
measures of social functioning. Examining the two EQ subscales for correlations
separately with a Bonferroni correction of p = .004, neither cognitive empathy nor
emotional reactivity were found to correlate significantly with any other variables (AQ,
FQ, SMS, RMET, SFS, PHQ-9, or GAD-7); all p values > .005.
Table 4.3
Correlations between continuous measures for females in the ASC group
Variable AQ EQ FQ SMS RMET SFS GAD PHQ
AQ -
EQ -.490* -
FQ -.461* .536* -
SMS -.155 .238 .335* -
RMET -.212 .324* .425* .140 -
SFS -.210 .183 .374* .158 .255 -
GAD .185 -.043 -.060 .122 -.030 -.252 -
PHQ .240 -.083 -.063 -.050 -.015 -.297 .835* -
* Correlation is significant at the p =.002 level (two-tailed) (Bonferroni corrected)
119
Table 4.4
Correlations between continuous measures for females in the potential ASC group
Variable AQ EQ FQ SMS RMET SFS GAD PHQ
AQ -
EQ -.492* -
FQ -.260 .457* -
SMS -.161 -.036 .064 -
RMET -.221 .437* .070 .025 -
SFS -.349* .172 .134 .248 .144 -
GAD .003 .129 .047 .182 .131 -.014 -
PHQ .117 -.025 .100 .089 -.001 -.185 .743* -
* Correlation is significant at the p =.002 level (two-tailed) (Bonferroni corrected)
As can be seen from Table 4.4, for females in the potential ASC group, and with
Bonferroni corrections applied, both EQ and SFS scores were significantly, negatively
correlated with the AQ, whilst both FQ and RMET scores were significantly, positively
correlated with the EQ. The GAD and PHQ were significantly, positively correlated
with each other but not with any of the measures of social functioning. Examining the
two EQ subscales for correlations separately, cognitive empathy significantly correlated
120
positively with RMET scores (r = .541, n = 30, p = .002) and emotional reactivity
significantly correlated negatively with AQ (r = -.587, n = 29, p = .001) and positively
with FQ (r = .661, n = 29, p < .001). All other correlations with other variables (SMS,
SFS, PHQ-7, GAD-5) were non-significant once Bonferroni corrections (p = .004) were
applied (all p values > .01).
Table 4.5
Correlations between continuous measures for females in the no ASC group
Variable AQ EQ FQ SMS RME
T
SFS GAD PHQ
AQ -
EQ -.499* -
FQ -.417* .424* -
SMS -.036 .424* .100 -
RMET -.022 .111 .036 .116 -
SFS -.394* .222* .428* .006 .055 -
GAD .416* -.154 -.252 .092 -.096 -.332* -
PHQ .368* -.145 -.282* .120 -.169 -.453* .752* -
* Correlation is significant at the p =.002 level (two-tailed) (Bonferroni corrected)
As can be seen from Table 4.5, for females in the no ASC group, and with
Bonferroni corrections applied, the AQ was significantly, negatively correlated with the
EQ, FQ and SFS. The EQ was significantly, positively correlated with the FQ, SMS,
and SFS, while the FQ was significantly, positively correlated with the SFS. Both GAD
and PHQ had significant, negative correlations with SFS, and significant, positive
correlations with each other. The PHQ also had significant, negative correlations with
the FQ and RMET. Examining the two EQ subscales for correlations separately,
cognitive empathy significantly correlated negatively with AQ scores (r = -.436, n = 93,
121
p < .001), and positively with FQ scores (r = .407, n = 88, p < .001) and emotional
reactivity significantly correlated positively with FQ (r = .554, n = 85, p < .001). All
other correlations with other variables (SMS, SFS, PHQ-7, GAD-5) were non-
significant once Bonferroni corrections (p = .004) were applied (all p values > .006).
4.3.4. Predicting the age of autism diagnosis. Correlations were performed to
determine whether later diagnosis was associated with higher scores for the measures of
self-monitoring, social functioning, friendship quality and motivation, and ToM
amongst males and females in the ASC group. Because age of diagnosis was
significantly, positively correlated with current chronological age for both genders, p
values < .001, age was entered as a control variable.
When males and females in the ASC group were analysed together, results
showed a reliable, positive correlation between age of autism diagnosis and self-
monitoring score: partial r(117) = .215, p = .019. However, the correlation failed to
reach significance when the two genders were considered separately, p values > .05. For
neither the group as a whole, or for the two genders considered separately, was age of
diagnosis predicted by any of the measures of social functioning, friendship motivation
and quality, or ToM; all p values > .05.
4.3.5. Other mental health diagnoses in females. Of females in the diagnosed
ASC group, 83.3% (n = 75) were diagnosed with a mental health condition, compared
to 57.1% (n = 44) of females in the potential ASC group, and 34.5% (n = 81) in the no
ASC group. Differences between the groups were significant: X²(2) = 64.240, p <.001, φ
= .400. Odds ratio calculations showed that females in the diagnosed ASC group were
3.75 times more likely than those in the potential ASC group and 9.50 times more likely
than those in the no ASC group to have a mental health diagnosis. Females in the
122
potential ASC group were 2.54 times more likely than those in the no ASC group to
have a mental health diagnosis.
A significant difference was found between female groups in the number of
mental health diagnoses they had: X2 (2) = 66.589, p <.001. Females in the diagnosed
ASC group had on average more mental health diagnoses (M = 1.87, SD = 1.47) than
females in the potential ASC group (M = 1.31, SD = 1.57). Using a Bonferroni corrected
alpha score of .02, this difference was significant: U = 2586.50, p = .004, d = 0.37.
Females in the potential ASC group had on average more mental health diagnoses than
those in the no ASC group (M = 0.63, SD = 1.03). This difference was significant: U =
6683.00, p <.001, d = 0.51.
No significant difference was found between female groups on age of first
mental health diagnosis made: X2 (2) = 1.341, p = .512. Females in the diagnosed ASC
group who had other mental health problems were diagnosed with their first mental
health condition on average at the age of 18.63 (SD = 6.05), those in the potential ASC
group were first diagnosed on average at the age of 19.75 (SD = 5.89), and those in the
no ASC group were diagnosed on average at the age of 19.02 (SD = 5.64).
4.3.6. Exploratory comparisons between males and females in the ASC
groups.
Age of ASC diagnosis: Females in the diagnosed ASC group were on average
diagnosed with ASC later than males: M = 24.88 (SD = 7.89) vs M = 18.96 (SD =
10.95): U = 793.500, p = -.008, d = 0.62.
Mental health: Females in the diagnosed ASC group were more likely to have
been diagnosed with another mental health condition than males (83.3% vs 55.6%):
X²(2) = 10.433, p = .005, φ = .294. Autistic females were 4 times more likely than
autistic males to have a mental health diagnosis. They also had more mental health
123
diagnoses than autistic males: M = 1.89 (SD = 1.47) vs M = 0.78 (SD = 0.85). However,
average age of first mental health diagnosis made was comparable between the two
groups (males M = 16.40, SD = 6.38 and females M = 18.63, SD = 6.05): U = 414.50, p
= .108.
To situate the age of ASC diagnosis within the context of all other mental health
diagnoses, for all participants with diagnosed autism the following two variables were
calculated: (1) the number of mental health diagnoses prior to ASC diagnosis, and (2)
the number of mental health diagnoses following the ASC diagnosis. In the rare cases
where another mental health diagnosis was concurrent with the ASC diagnosis, only the
latter was counted. For females, the number of earlier mental health diagnoses (M =
1.74, SD = 1.41) was significantly greater that the number of later mental health
diagnoses (M = 0.40, SD = 0.92), z = -4.798, p < .001. For males, in contrast, the
number of earlier mental health diagnoses (M = 0.80, SD = 0.86) was not significantly
different to the number of later mental health diagnoses (M = 0.53, SD = 0.52), z =
-.714, p = .475.
Additionally, a count was made of the number of times that the ASC diagnosis
was the only, first, middle or last diagnosis, separately for males and females. For
females, the ASC diagnosis was the last diagnosis on 51 of 89 occasions (57%). In
contrast, for males the ASC diagnosis was the last on 7 of 27 occasions (26%). Chi-
Square analysis revealed a significant difference between males and females: X²(2) =
9.137, p = .028, φ = .281. Autistic females were 3.8 times more likely than autistic
males to have received their autism diagnosis last.
Other Questionnaire scores: There were no significant differences between the
performance of autistic males and autistic females on any other scales (EQ, FQ, RMET,
GAD, PHP, SMS, or SFS); all p values > .05.
124
4.4. Discussion
Findings from Study 1 left several unanswered questions that needed to be researched
further in order to better explore the female autism phenotype. The aim of this current
study was to address these gaps by providing participants with a second survey that
would measure social abilities, traits of depression and anxiety, and ages of other
psychiatric diagnoses. Study 1 and Study 2 combined could provide a novel
contribution to our current knowledge of the manifestations of autism in women.
As predicted, potentially autistic women in the current study did have a
significant empathy advantage over diagnosed autistic women, consistent with findings
made in Study 1. When looking at the subscales this was again found only on the
cognitive empathy subscale and not the emotional reactivity subscale. It should be noted
that 72.51% of the sample for Study 2 were derived from Study 1, which therefore
explains this consistency in EQ scores across studies. No differences were found on the
RMET however, which is surprising given there were differences on the cognitive
empathy subscale, which ToM is thought to relate most closely to (Stietz et al., 2019).
Although, this is supported by research from Livingston et al. (2018), who recently
found that heightened levels of IQ, EF, and anxiety were all linked to a greater ability to
compensate for underlying deficits in ToM. These potentially autistic women may be
better able to mask their autistic traits and apparent ToM deficits than their diagnosed
autistic peers due to advantages in certain other areas, for example in empathy.
Although, Oakley et al. (2016) caution against over interpreting ToM based on the
RMET, as they found that rather than measuring ToM ability it instead measures
emotion recognition. They argue that emotion recognition may be affected by a sub-
clinical condition known as alexithymia, which affects the ability to describe and
recognise one’s own feelings, and that is relatively common in the autistic population
125
(Cook et al., 2013; Oakley et al., 2016). As this was not tested in this study, it is unclear
what other factors may have contributed to this null finding.
In terms of social performance skills, the prediction that potentially autistic
women would score more highly on social functioning than diagnosed autistic women
was confirmed. Moreover, both groups were more impaired on the SFS than the non-
autistic control participants. These findings lend support to the FPT, suggesting that
autistic females often miss receiving an ASC diagnosis due to less impaired social
difficulties than those receiving a diagnosis. In particular, this was seen on the
engagement and independence-competence subscales of the SFS, and evidenced
through similar employment scores to non-autistic women. It is possible that this may
be one of the reasons why these females have been missed by professionals. For
example, Dworzynski et al. (2012) found that potentially undiagnosed girls who had a
high number of autistic traits had significantly fewer social autistic traits and
challenging behaviours, and more prosocial behaviours than diagnosed autistic girls
compared to boys.
Despite this, the current study did not find that better social abilities among the
potentially autistic women resulted in increased friendship motivation or quality. This
conflicts with previous studies that had observed that autistic girls appeared to be better
at friendships than autistic boys (Dean et al.,2017; Sedgewick et al., 2016). However,
the current study measured adults only, and it is reasonable to expect that friendships in
adulthood are more complex, involving more than the playground interactions that these
previous studies had investigated. For example, Baron-Cohen and Wheelwright (2003)
did not find a difference on the FQ between autistic males and autistic females,
suggesting that the quality of friendship might not be an indicator of the female
phenotype of autism, or alternatively that autistic women may rate themselves more
126
harshly on these measures. Given that potentially autistic females are still impaired
socially in many areas, friendship may remain a difficult aspect of socialising to manage
for many.
The current study also did not find that potentially autistic women performed
any better than diagnosed autistic women on self-monitoring, a proxy measure for
camouflaging. This had not been measured before in an autistic population, but self-
monitoring has been argued to be linked to the ability to adjust in social situations and
to socially mimic others (Estow et al., 2006; Schaffer et al., 1982; Snyder, 1974). It
may be the case that the SMS is not sensitive to the subtle social differences between
different autism presentations; or given that it is a self-report, autistic women may be
more aware of their difficulties and so again rate themselves more harshly. For example,
autistic women often rate themselves higher on measures of autistic traits than autistic
males, despite not being observed to have more severe traits (Lai et al., 2013; Lai et al.,
2011; Lenhardt et al., 2016). Alternatively, the fact that many of the diagnosed autistic
women were diagnosed in later adolescence and adulthood could account for their
similar performance to potentially autistic women on the SMS.
In contrast, the prediction that empathy would positively correlate with ToM,
friendship, self-monitoring, and social functioning, was partially supported. For females
in the ASC group, both FQ and RMET scores significantly, positively correlated with
EQ scores, and both SMS and SFS correlated positively with FQ. For females in the
potential ASC group, positive correlations between the EQ and FQ and EQ and RMET
were found. These findings are consistent with the suggestion that better empathy skills
give rise to better friendship quality. However, empathy scores did not correlate with
social ability measures (SFS or SMS) for either group, suggesting other factors may
contribute towards the social functioning advantage seen in potentially autistic women
127
compared to diagnosed autistic women. In particular, higher cognitive empathy was
correlated with higher ToM scores on the RMET in this group, whilst lower emotional
reactivity was correlated with higher AQ scores and lower scores on the FQ. Different
correlations were observed for non-autistic females. For example, in this group self-
monitoring and social-functioning did correlate positively with empathy, and social-
functioning correlated negatively with traits of anxiety and depression.
The prediction that measures of social abilities would correlate with age of ASC
diagnoses was also partially supported. Across both men and women in the diagnosed
autistic group, age of ASC diagnosis was significantly, positively correlated with self-
monitoring. These findings suggest that the ability to adapt one’s behaviour in social
situations may delay identification of ASC. This could be the result of camouflaging of
autistic traits, caused by an autistic person’s ability to ‘fit in’ appropriately to social
situations. Nevertheless, no correlation was found between age of ASC diagnosis and
social functioning, friendship, or ToM across genders, and the correlation between self-
monitoring and age of ASC diagnosis was weak, suggesting that skills in these areas
may not be the most important factor delaying ASC diagnosis.
It had also been hypothesized that females in the potential ASC group would
have higher levels of anxiety and depression than females in the diagnosed ASC group,
and that this would be correlated to better social abilities. Whilst potentially autistic
females did not score higher on these measures than females in the diagnosed ASC
group, they did score similarly. This is in contrast to findings that females in the
diagnosed ASC group are more likely to be diagnosed with a psychiatric disorder and
have significantly more mental health diagnoses than females in the potential ASC
group. These findings raise the possibility that while diagnosed autistic women receive
more psychiatric diagnosis than potentially autistic women, they are not more likely to
128
suffer from mental health difficulties. Both females in the diagnosed ASC group and
those in the potential ASC group performed similarly on two of the social scales (FQ
and SMS), which might indicate that to some extent both groups are using
camouflaging strategies and learning social behaviours to ‘fit in’, which is thought to
increase mental health problems (Cassidy et al., 2018; Hull, Mandy, et al., 2019;
Livingston et al., 2018). However, anxiety and depression scores were not found to
correlate significantly with any of the social measures used in the ASC group or the
potential ASC group, whilst they did positively correlate with AQ scores and negatively
with SFS and FQ scores in the non-autistic group. This suggests that the autistic traits
and difficulties associated with being autistic increase the likelihood of having mental
health problems.
To explore the pattern of psychiatric diagnoses for diagnosed autistic females
and males, the current study also analysed the ages of other psychiatric diagnoses.
Autistic females had significantly more psychiatric diagnoses made prior to their ASC
diagnosis compared to after. For males no difference between the number of psychiatric
diagnoses made prior to or after their autism diagnosis was made. These findings
support the suggestion that diagnosis may be delayed for autistic females due to
clinicians’ diagnosis of other co-morbid or misdiagnosed conditions instead of ASC (Lai
& Baron-Cohen, 2015). Findings also revealed that an ASC diagnosis is more likely to
come last for women than it is for men, although this may be due to the later age of
ASC diagnosis in this group; autistic males were generally diagnosed earlier and
therefore have had more time to receive other psychiatric diagnoses. Finally, no
significant difference in the age of first mental health diagnosis between the potentially
autistic and diagnosed autistic women, or between diagnosed autistic men and women
was made. This suggests that earlier identification of other psychiatric difficulties may
not prompt diagnosis of autism by professionals.
129
Lastly, the current study compared the results for diagnosed autistic males and
diagnosed autistic females on all measures. It was found that whilst autistic women
were diagnosed significantly later and had significantly more mental health diagnoses
than autistic men, the groups scored similarly on measures of social abilities,
depression, and anxiety. This evidence does not provide support for the theory that
autistic women have a different phenotype than autistic males due to masking of
symptoms with better social abilities. As discussed above, though, it is possible that
self-report measures paint a false picture as individuals who are more aware of their
difficulties tend to rate their social abilities poorly. Additionally, it should be noted that
the small sample of autistic males in this study means that the statistical tests lacked
power. These limitations to the study are discussed further in the General Discussion
(Chapter 6). Additionally, the men were diagnosed on average later than previous
studies had found and therefore may be more like the females in this sample in their
presentation.
In conclusion, this study has explored the impact of social abilities on autism
diagnosis, as well as age of other psychiatric diagnoses. The study found that potentially
autistic women have an advantage over diagnosed women not just in empathy, but also
social functioning. Age of ASC diagnosis was found to be later across both autistic men
and women who showed greater self-monitoring, although this trend was relatively
weak. For diagnosed autistic women but not for diagnosed autistic men, significantly
more other psychiatric diagnoses were made prior to their autism diagnosis compared to
after; a diagnosis of autism was more likely to be the final psychiatric diagnosis for
women. However, against expectations there was no evidence that potentially autistic
women used self-monitoring more than diagnosed autistic women. As discussed, it is
possible that greater self-monitoring is associated with better self-awareness, and that
autistic women who have more insight into their difficulties tend to rate themselves
130
harshly on self-report measures of social abilities. Accordingly, Study 3 will follow-up
these findings by using a newly developed measure of camouflaging and objective
measures of social performance (i.e., peer ratings rather than self-report) to see whether
a link between camouflaging and social abilities can be demonstrated.
131
CHAPTER 5
Study 3: Differences in Self-Reported Camouflaging and Peer Judgements of
Social Abilities between Autistic Males and Autistic Females
5.1. Introduction
Studies 1 and 2 identified a group of potentially autistic women, comparing them to
diagnosed autistic women to determine what factors may contribute to their lack of
diagnosis. A significant empathy and social functioning advantage over diagnosed
autistic women was found in potentially autistic women, and self-monitoring was
significantly, positively correlated with age of ASC diagnosis across both diagnosed
autistic males and females. However, differences in self-monitoring (a proxy measure
for camouflaging) were not observed between potentially autistic and diagnosed autistic
women, and scores on the SMS did not correlate with social functioning, empathy,
depression or anxiety in these groups either. One possible explanation for these
conflicting findings is that self-report measures are not reliable, particularly as women
with greater insight into their difficulties might be overly severe in their self-ratings.
The primary aim of Study 3, therefore, is to use a more objective measure of social
performance, namely, peer ratings, and to examine the link between these, age of autism
diagnosis, and a more direct measure of self-reported camouflaging.
Since Studies 1 and 2 were conducted, a new self-report instrument measuring
camouflaging has been devised called the Camouflaging Autistic Traits Questionnaire
(CAT-Q) (Hull, Mandy et al., 2019). Using this instrument, recent research has explored
the theory that autistic women may deliberately camouflage their autistic traits more
than autistic males, which will be discussed in more detail later in this chapter.
Importantly, though, there is a gap in the literature as no studies have examined whether
camouflaging strategies by autistic women actually are successful in masking their
132
disorder. If autistic women are viewed more favourably than autistic men during social
interactions by observers who are not informed explicitly about their autism, then this
could explain why clinicians frequently miss it. Therefore, the main aims of Study 3
were (1) to compare the self-reported camouflaging behaviours of autistic women,
autistic men, non-autistic women and non-autistic men using the CAT-Q, and (2) to
examine whether scores on the CAT-Q are predictive of non-autistic observers’
impressions of the social skills and likability of the autistic participants during ordinary
social interactions.
5.1.1. Camouflaging and associated traits in autism. Livingston and Happé
(2017) describe camouflaging as a strategy utilised by those with a neurodevelopmental
disorder as part of a wider strategy to compensate for one’s disorder, in order to improve
the behavioural presentation of oneself despite cognitive impairments. As discussed in
Chapter Two, autistic women and girls have consistently reported using camouflaging
strategies as a way to manage social relationships (e.g. Tierney et al., 2016). In
particular, autistic females have reported using deliberate mimicry (e.g. Bargiela et al.,
2016), compensatory behaviours such as purposefully using non-verbal gestures,
maintaining appropriate levels of eye contact, avoiding dominating conversations, and
practising conversations beforehand to maintain a social script (Hull, Petrides, et al.,
2017). These reports are supported by findings of several studies that have compared the
social behaviours of autistic males and females. For example, Dean et al. (2017)
observed 24 autistic girls and 24 autistic boys during play with other children. They
found that the autistic girls were more likely to engage in ‘joint play’, which they
hypothesised may be due to better social camouflaging. Furthermore, Sedgewick et al.
(2016) found that 13 autistic girls scored higher than 10 autistic boys on social
motivation and friendship closeness.
133
Development of the CAT-Q has made it possible for researchers to evaluate
different facets of camouflaging. The CAT-Q asks 25 questions related to
‘compensation’ (strategies to compensate for social and communication difficulties),
‘masking’ (strategies to appear less autistic to others), and ‘assimilation’ (strategies to fit
into uncomfortable social situations). On this scale, self-reported camouflaging
behaviour has been found to be higher in autistic people than non-autistic people, and
higher in autistic females than autistic males. It was found that autistic females scored
on average 124.35 (SD = 23.27), autistic males scored on average 109.64 (SD = 26.50),
non-autistic females scored on average 90.87 (SD = 27.67), and non-autistic males
scored on average 96.89 (SD = 24.22) (Hull, Lai, et al., 2019). Note, however, that
these group differences were mainly apparent on the ‘assimilation’ and ‘masking’
subscales, where autistic females scores significantly higher than autistic males, and not
in the ‘compensation’ subscale, where no differences were observed. When compared to
non-autistic participants, autistic females scored significantly higher on all subscales
than non-autistic females, and autistic males scored higher on all subscales except for
‘masking’ than non-autistic males.
Several factors have been considered to relate to camouflaging, one of these
being executive functioning (EF). Better EF is thought to assist with camouflaging
because to camouflage one must inhibit inappropriate social responses, be able to script
social situations beforehand, and have the flexibility to deal with unexpected social
situations (Sedgewick et al., 2016). Some studies have found a female advantage among
autistic participants for cognitive flexibility and processing speed (Bolte et al., 2011; Lai
et al., 2012; Lenhardt et al., 2016). Other studies have linked better EF with better ToM,
which could be argued to aid in camouflaging as it would be beneficial to understand
the mental states of others in order to ensure one’s own behaviour is appropriate to the
situation. For example, Ahmed et al. (2011) found several ToM tests were related to
134
different aspects of EF when tested with 135 non-autistic participants. Verbal fluency
and problem solving were predictive of performance on the Strange Stories task and the
Faux Pas Test; verbal fluency was suggested to involve flexibility in initiating responses
such that in social situations one could generalise the basic concepts of social
interaction and apply these; and deductive reasoning was suggested to depend on one’s
ability to solve a puzzle from clues, which in social situations is required to figure out
why someone is behaving how they are. In a recent study by Livingston et al. (2018),
higher IQ, superior EF, and greater anxiety were all linked to a better ability to
compensate for underlying deficits in ToM amongst a sample of 136 autistic
adolescents. However, the study did not find a gender difference in compensation, and
there has been little to suggest that in clinical populations autistic females outperform
autistic males on ToM ability (Buitelaar et al., 1999; Happé, 1995).
As discussed in Chapter 2, another factor found to be associated with
camouflaging is poor mental health, including increased depression, anxiety, and
suicidal behaviours, thought to be due to the increased exhaustion of consciously
masking one’s autism (Livingston et al., 2018). Very few studies to date have measured
self-reported camouflaging traits in relation to mental health measures. Cassidy et al.
(2018) found that camouflaging, as measured with their four-item questionnaire,
significantly predicted suicidality even when depression and anxiety were controlled
for. In support of these findings, Hull, Mandy, et al. (2019) found depression and
generalised anxiety were positively correlated with the CAT-Q. However, somewhat
different findings were obtained by Cage and Troxell-Whitman (2019), who tested 135
autistic females and 111 autistic males on the CAT-Q, as well as developing their own
scales measuring 21 possible reasons for camouflaging and 22 possible contexts for
camouflaging, with mental health measured using the Depression, Anxiety and Stress
Scale (DASS-21). Out of the possible contexts for camouflaging, two broad categories
135
were determined: formal and interpersonal. Participants were categorised as
camouflaging consistently high for both contexts (high camouflagers), as being
significantly high in one context but low in the other (switchers), or as camouflaging
consistently low in both contexts (low camouflagers). Depression scores were not
significantly different between the three participant groups. However, consistently low
camouflagers had significantly lower rates of anxiety than high camouflagers, and also
significantly lower rates of stress than both high camouflagers and switchers. These
findings suggest that the mental health consequences of camouflaging may depend on
the context in which it is used.
As reviewed in this section, camouflaging by autistic adults has been linked
positively with EF and ToM, and negatively with mental health. A further objective of
Study 3 was therefore to attempt to replicate and extend these findings. Given that
Studies 1 and 2 found a slight empathy advantage in potentially autistic women, Study 3
examined whether empathy is also related to camouflaging ability. Therefore, in
addition to the CAT-Q, participants in Study 3 completed tests of EF, ToM, autistic traits
(AQ) and empathy (EQ). Given that autistic women tend to be diagnosed with autism
later than autistic males and are more likely to be misdiagnosed with other mental health
conditions, with greater camouflaging being suggested as a cause (Lai & Baron-Cohen,
2015), Study 3 collected information about participants’ various mental health
problems. It also examined the association between camouflaging and age of ASC
diagnosis.
5.1.2. The effects of camouflaging on impressions made on others. Research
on camouflaging in autism is still in its infancy, and there have been very few studies on
the topic. Most studies have investigated the first-person experience of camouflaging
through self-report questionnaires, in order to conceptualise the behaviours and
136
motivations associated with it as a strategy for autistic people. Others have made
observations hypothesised to be related to social camouflaging (e.g. performance on the
ADOS, friendship quality, and engagement in shared play). Lai et al. (2017) measured
the discrepancy between self-reported autistic traits and external behaviours observed
by a clinician, hypothesising that autistic females may report similar levels of autistic
traits as males but that they may score lower on clinician observations, causing a greater
discrepancy in scores between self-reported and observed autistic traits. They found that
autistic females did have a much greater discrepancy score than autistic males, with
autistic females being rated as performing better on social communication of the ADOS
Module 4 by clinicians but higher than males for self-reported autistic traits. As
discussed in Chapter Two, the study’s claim that this discrepancy score represents
camouflaging is somewhat problematic, given that camouflaging has not been measured
and a number of other factors could cause this discrepancy. However, a potentially
important aspect of this study is the use of observations by clinicians that determined
that autistic males scored higher for social communication difficulties (M = 8.5) than
autistic females (M = 4.3), which was significant and had a large effect size (d = 1.04).
This was despite autistic females scoring significantly higher on the AQ (M = 37.5) than
autistic males (M = 32.7), and similarly to autistic males on the ADI-R, which measured
reciprocal social behaviours, communication, and RRBIs. These findings suggest that in
social situations autistic females are viewed more favourably by clinicians, and this
might reduce the probability of those females receiving an ASC diagnosis. However, it
is unclear whether the results mean that autistic females camouflage their autistic traits
in social settings, and therefore appear less ‘autistic’, or whether there is a clinician bias,
specifically, such that clinicians are more used to associating social communication
difficulties with males and therefore may miss the autistic presentation demonstrated by
females.
137
Additionally, several studies have investigated differences between autistic and
non-autistic people in how they are perceived by others, which may be a useful method
in determining the success of camouflaging strategies. For example, Grossman (2015)
took short 1-3 second video clips of 9 autistic and 10 non-autistic children (17 male)
telling a made-up story. Eighty-seven non-autistic participants with a mean age of 23
(64 females and 23 males) were shown the clips, unaware of which children were
autistic, and asked if the child they saw appeared to be socially awkward. The autistic
children were rated as more socially awkward than the non-autistic children on both 3
second and 1 second clips, regardless of whether audio-visual clips, audio only clips, or
still images were used. However, it is unknown whether autistic females are rated as
less socially awkward than autistic males, which the FPT may suggest would be the
case if they are successfully camouflaging difficulties.
Sasson et al. (2017) conducted a number of experiments to evaluate the first
impressions of autistic adults and children by non-autistic peers using thin-slices of real-
life social behaviours. In their first study, 20 autistic participants and 20 non-autistic
participants (17 males in each), with a mean age of 25 years, were used as stimuli
(‘participant-stimuli’). They were recorded engaging in a mock audition for a
reality/game show, which was cut into 10-second clips and edited into five different
modalities (audio-only, visual-only, static image, and transcript of speech content). Non-
autistic participants were used as raters (participant-raters) and were shown the video
clips of each of the 40 participant-stimuli in one of the modalities. There were 214
participant-raters in total (164 females), with a mean age of 21. A rating scale was used
which listed six attributes found to be reliably perceived when forming first-
impressions, these were attractiveness, awkwardness, intelligence, likeability,
trustworthiness, and dominance/submissiveness. In addition to these items, four others
were measured that reflected behavioural intent towards the participant-stimuli
138
(willingness to live near, likelihood of hanging out in their free time, level of comfort
sitting next to, and likelihood of starting a conversation with). Autistic participant-
stimuli were rated less favourably overall that non-autistic participant-stimuli, and this
was the case across all modalities except in the transcript condition. Also, autistic
participant-stimuli were rated worse on the audio-visual modality than the others.
Looking at each item type, it was apparent that autistic participant-stimuli were rated
less favourably on all traits except trustworthiness, intelligence, and the raters’
willingness to live near them. For the autistic participant-stimuli, social awkwardness
was found to correlate negatively with raters’ intent to talk to and socialise with the
person. No differences were found between male and female participant-stimuli, though
it is worth noting there were only 3 autistic females included in this part of the study.
In a follow-up study conducted by the same authors, 12 autistic (10 male) and 16
non-autistic (9 male) participant-stimuli were presented to 37 participant-raters (19
male). Participant-stimuli were filmed engaging in natural conversation with an
experimenter who asked open-ended questions such as “have you seen any good movies
recently?” Unlike the first study, this study was filmed using video-recording glasses to
give a first-person viewpoint to the participant-raters. The recordings were edited into
10 still frames per participant-stimulus and shown to the participant-raters, who rated
them on three questions (“How socially awkward is this person?”, “How approachable
is this person?”, and “Would I see myself being friends with this person?”). Once again,
autistic participant-stimuli were rated less favourably than non-autistic participant-
stimuli, even though the raters were not aware that the participants had autism.
Better knowledge of autism has been found to be associated with more
favourable first-impression ratings, suggesting that harsh judgements may be reduced
when people are able to understand the persons’ appearance and behaviour in context of
139
their condition. For example, in a later study by Sasson and Morrison (2019), first-
impression scores improved when participant-raters were aware that the participant
stimuli had autism compared to when they did not know. The researchers used the same
participant-stimuli from their first study, which included 20 autistic and 20 non-autistic
participant-stimuli. When the participant-raters were provided the correct diagnosis of
the participant-stimuli, ratings were more favourable than when they were mislabelled
as either non-autistic or as having a schizophrenia diagnosis. The non-autistic
participants were also rated more favourably when they were mislabelled as autistic
compared to being labelled correctly or mislabelled as schizophrenic. These findings are
consistent with those of an earlier study by Matthews et al. (2015), who found college
students’ perceptions of peers with autism were more favourable when they knew they
were autistic.
Taken together, these studies suggest that autism affects the overt behavioural
appearance of an individual, and that others rate the traits displayed by autistic
individuals as less favourable. Moving forward, it would be beneficial to measure how
ordinary non-autistic peers (i.e. non-clinicians without training in autism) view autistic
males and females who they are unaware are autistic, and whether they view autistic
females more favourably than their autistic male counterparts. If being viewed more
favourably by these peers is associated with higher camouflaging scores, then this may
provide important evidence of the use and success of camouflaging as a strategy to ‘fit
in’ and evade diagnosis. On the other hand, if more favourable ratings are not associated
with self-reported camouflaging then this may suggest either that there is a societal bias
in the judgement of atypical behaviours, or that our current measures of camouflaging
are unable to detect the successful use of those strategies.
140
5.1.3. Aims and hypotheses. The first aim of Study 3 was to explore gender
differences in the use of self-reported camouflaging in autistic versus non-autistic
adults, and links between camouflaging and the AQ, EQ, EF, ToM, mental health
diagnosis, and age of ASC diagnosis. This aim was addressed by modelling the
procedures used by Hull, Lai, et al. (2019), which examined gender differences in
camouflaging, and the correlation between mental health and camouflaging. Study 3
extended Hull, Lai, et al.’s (2019) study by also investigating whether camouflaging
was correlated with better ToM, EF, and empathy, which has yet to be investigated using
the CAT-Q. It was predicted that autistic people would have lower EQ scores but higher
AQ and camouflaging (CAT-Q) scores than non-autistic people, and that autistic
females would score higher than autistic males on self-reported camouflaging. It was
also predicted that higher camouflaging scores would be associated with better EF
skills, better performance on tests of ToM, empathy, a later age of ASC diagnosis, and
also more mental health diagnoses. This was because previous studies have shown
camouflaging to be associated with enhanced cognitive abilities (which can delay
diagnosis) but poorer mental health.
The second aim of Study 3 was to extend the Sasson et al.’s (2017) first-
impression peer rating study by examining whether the social behaviours of autistic
adults are perceived less favourably than the social skills of people without autism by
non-autistic age-matched observers, whether results are affected by participant gender
or rater gender, and whether the first-impression scores correlate with camouflaging
scores and age of ASC diagnosis. Importantly, Study 3 used more naturalistic film clips
than Sasson et al. (2017) and included equal numbers of autistic males and autistic
females as participant-stimuli to enable a gender comparison. In terms of the first-
impression ratings, it was predicted that autistic males would be rated less favourably
than autistic females, and that both groups would be rated less favourably than non-
141
autistic males and females. Additionally, it was predicted that first-impression scores
would correlate positively with age of ASC diagnosis and camouflaging. This prediction
was made on the basis of the FPT, which suggests that camouflaging in autistic women
leads to later and missed diagnosis.
The remainder of this chapter is divided into two parts. Part 1 reports the
method, results and discussion relevant to the first aim, that is, to explore the relations
between autism, gender, empathy, EF, ToM, mental health diagnoses, age of ASC
diagnosis and self-reported camouflaging. Part 2 reports the method, results and
discussion relevant to the second aim, that is, to explore the first impressions made on
non-autistic peers by males and females with autism, and the relation between first
impression scores and self-reported camouflaging.
5.2. Part One
5.2.1. Method
5.2.1.1. Participants. The study was advertised in local universities and on social media
asking participants to take part in a study looking at differences in social behaviours
between autistic and non-autistic individuals. The majority of autistic participants were
recruited from advertisements placed in private autism groups on Facebook and in
community centres holding autism meetings/clinics. Participants were required to be
UK citizens and speak English as a first language; this was to ensure that any cultural
effects would not bias the second part of the study which would use the same group of
participants. Eighty participants were recruited for part one of this study. Forty of these
had an ASC diagnosis (20 males and 20 females) and 40 were non-autistic controls (20
males and 20 females). One female and one male autistic participant identified as
transgender and were grouped according to their currently defined gender. Participants
142
were required to be between the ages of 18-40 years (young adult) to limit the effects of
aging on autistic traits and EF, and also to ensure that in the second part of the study the
participant-stimuli and participant-raters would be equivalent in age. Age was
comparable between the four groups of participant-stimuli (autistic females = 25.45
years, autistic males = 25.85 years, non-autistic females = 27.75 years, non-autistic
males = 27.80 years; F(3, 76) = .753, p = .524).
The National Adult Reading Test (NART) (Nelson & Willison, 1991) was
administered to check that IQ was comparable between the groups. It comprises a list of
50 words which become progressively harder to pronounce as the list goes on.
Participants are instructed to read each of the words on the list aloud, and a point is
assigned if the word is pronounced correctly. NART error scores are used to predict
WAIS full scale IQ, verbal IQ, and predicted IQ (Bright et al., 2016). As can be seen
from Table 5.1 NART error scores were comparable between the four groups (autistic
females = 17.53, autistic males = 19.68, non-autistic females = 20.00, non-autistic males
= 19.42): F(3, 72) = .759, p = .386) .
143
Table 5.1
Average predicted WAISS full-scale, verbal, and performance IQ scores from NART
errors and standard deviations per group
Gender Predicted full-
scale IQ (SD)
Predicted verbal
IQ (SD)
Predicted
performance IQ (SD)
Autistic
Females 113.21 (4.34) 112.89 (4.99) 112.11 (3.53)
Males 111.37 (7.82) 111.00 (8.62) 110.89 (6.17)
Non-Autistic
Females 111.37 (4.44) 110.63 (4.88) 110.53 (3.44)
Males 111.63 (5.18) 111.16 (5.81) 111.00 (4.08)
ASC diagnoses were confirmed by requesting to see evidence, including
education and health statements and diagnostic reports. Whilst all autistic participants
reported having an ASC and gave details of how they were diagnosed, 11 failed to
submit their evidence. In most cases these reports remained with their guardians as they
were diagnosed as children, and the current research was unable to confirm diagnoses
by using methods such as the ADOS due to a lack of resources. However, there were no
differences in self-reported autistic traits on the AQ screening measure between those
who had submitted a report (M = 35.09, SD = 7.65) and those who had not (M = 35.00,
SD = 7.85), t(38) = .033, p = .974. Four of the latter group scored below the AQ criteria
(>32), the lowest scoring 23, but the remaining three scored above the less conservative
AQ criteria (>28) suggested by Baron-Cohen et al. (2001) for those in clinical settings
with an autism diagnosis. Therefore, it is reasonable to assume that these participants
were autistic and that they had similar levels of autistic traits to those who were able to
confirm their diagnoses, preventing any confounding effects from different levels of
autistic traits. Note also that the method of sampling autistic people without officially
confirming their diagnosis with tests undertaken by the researchers has been used
recently in other studies (Cage & Troxell-Whitman, 2019; Cassidy et al., 2018). The
144
advantages of this method are that it is not exclusive to a clinical population and it saves
the time and stress on participants associated with having to go through another
diagnostic assessment. None of the non-autistic participants reported an ASC diagnosis,
and only four reported having a first-degree family member with autism. Of these, one
non-autistic female and one non-autistic male had an autistic son, and one non-autistic
female and one non-autistic male had an autistic sister. Participants received £7 for their
time (1 hour) and all reasonable travel expenses were refunded.
5.2.1.2. Measures.
AQ: The full 50 item Autism Quotient (AQ) (Baron-Cohen et al., 2001) was
used to measure autistic traits. A detailed description of the measure can be found in
Chapter 3, section 3.2.2 .
EQ: The 40 item version of the Empathy Quotient (EQ) (Baron-Cohen &
Wheelwright, 2004) was used to measure empathy. A detailed description of the
measure can also be found in Chapter 3, section 3.2.2. The EQ scores were again split
into two subscales reflecting cognitive empathy and emotional reactivity.
CAT-Q: The Camouflaging Autistics Traits Questionnaire (CAT-Q) is a 25-item
self-report questionnaire developed from the theoretical model set out by Hull, Petrides,
et al. 2017), who provided a qualitative analysis of camouflaging by autistic
participants. The items in the questionnaire were intended to reflect two aspects of
camouflaging: first, compensation of social and communication difficulties, and second,
masking one’s presentation to appear non-autistic (Hull, Mandy, et al., 2019).
Participants answer each question on a seven point Likert scale from ‘Strongly
Disagree’ to ‘Strongly Agree’, with higher scores indicating higher camouflaging. The
scale was validated by the authors on 354 autistic participants and 478 non-autistic
participants (300 males and 434 females) with a mean age of 36. Factor analysis
145
revealed that the scale actually measured three factors: compensation and masking (as
described above), and assimilation, which involved strategies reflecting a need to fit in
with others socially. High internal consistency was found for the scale as a whole (α =   
0.94), as well as each of the three subscales (Compensation = 0.91, Masking = 0.85,
and Assimilation = 0.92). Test-retest reliability, as calculated from 30 autistic
participants who completed the questionnaire again three months later, was high (r = .
77). Furthermore, convergent validity was achieved because outcomes for the CAT-Q
were significantly, positively correlated with autistic traits and social anxiety in both
autistic and non-autistic samples, positively to wellbeing in both autistic and non-
autistic participants, and positively to depression and generalised anxiety in autistic
participants (non-autistic participants were not tested with depression and anxiety
measures) (Hull, Mandy, et al., 2019).
Executive Functioning: A battery of executive functioning (EF) tasks was
administered using PEBL software (Mueller & Piper, 2014). The tasks assessed set
shifting (Berg’s ‘Wisconsin’ Card Sorting Test), inhibition, cognitive flexibility, and
processing speed (Numerical Stroop Task), and problem solving and planning (Tower of
London).
The original Card Sorting Test (BCST) was created by Berg (1948) to test
peoples’ ability to respond selectively to one aspect of a situation and to shift attention
from one to another. The BCST presents participants with four cards each with an item
characterized by colour (red, green, yellow, or blue), shape (triangle, star, cross, or
circle), and number appearing on them (1-4). A series of cards are then presented to the
participant, with different shapes, colours, and number of shapes on them, and the
participant is required to sort them into one of the four piles according to an unwritten
rule; they may match on colour, shape, or number of shapes. Participants are told
146
whether they have guessed the rule correctly or incorrectly and must continue sorting
according to that rule until a new rule is required, prompting the participant to shift their
responses and attempt to determine through trial and error the new rule. There are 117
trials in total and the main score is taken from the number of errors made.
The Stroop task captures an effect that has been described as a mismatch in
stimuli resulting in a delay in reaction time on a task requiring cognitive inhibition
(Stroop, 1935). The current study used the Numerical Coding Stroop Task developed by
Windes (1968), which requires participants to select on their keyboard the number of
characters present on the screen for each trial. Each trial contains either neutral stimuli
(1-3 of the same letters are presented on the screen, e.g. ‘Z’, ‘ZZ’, and ‘ZZZ’),
congruent stimuli (1-3 of the same numbers are presented on the screen, and the number
will correspond to the number of characters, e.g. ‘1’, ‘22’, or ‘333’), and incongruent
stimuli (1-3 of the same numbers are presented on the screen, and the numbers will not
correspond to the number of characters, e.g. ‘11’, ‘222’, or ‘3’). Incongruent trials
generally take longer to respond due to a delay in response caused by cognitive
inhibition. Participants were given time to practise the task before being given 192
randomised trials, and both reaction time and accuracy were recorded for each trial.
The Tower of London (ToL) task is an adaptation of the problem solving puzzle
‘Tower of Hanoi’, which measures a person’s ability to solve a problem through forward
planning (Shallice, 1982). The task requires participants to mentally plan a sequence of
moves of three piles of different coloured disks in order to match a set of disks within a
certain number of moves. There are 12 trials in total and a score is accumulated for each
trial (3 points per successful trial, with a maximum of 46 points in total).
ToM: The Short Story Task (SST) was used to measure mentalising ability (also
referred to as ToM) (Dodell-Feder et al., 2013). This task has been specifically designed
147
to avoid ceiling effects and to assess the full range of ToM abilities, using multiple
levels of complexity of both first-order ToM (understanding another person’s thoughts)
and second-order ToM (understanding one other person is thinking about another
person’s thoughts). The task also tests ToM in a realistic social context, which requires
participants to understand the social landscape in order to make mental state inferences.
As Study 3 is concerned with social behaviours, it was decided that this measure of
ToM would best serve the study’s aims. The SST is also relatively quick and easy to
administer, requiring participants to read a short extract from the story ‘The End of
Something’ by Ernest Hemingway, and then answer 14 questions which relate to their
comprehension of the story, explicit mental state reasoning, and spontaneous mental
state reasoning. Spontaneous mental state reasoning was measured with one question
(participants were asked to summarise the story with no prompts); if participants
described the mental states of others in the story they were given one point, all other
responses scored 0. Comprehension was measured using five questions (e.g. “Nick and
Marjorie have a pail of perch for what purpose?”), with a possible two points assigned
for each (0 = inaccurate response, 1= partial understanding of non-mental story details,
and 2 = full understanding of non-mental story detail). Explicit mental state reasoning
was measured using eight questions (e.g. “Why does Nick say to Marjorie ‘you know
everything’?”), with a possible two points assigned for each (0 = no mental state
inference or inaccurate mental state inference, 1 = consideration of only one
perspective, or partially understood, 2 = consideration of several character’s mental
states (second-order mental state references), and accurate mental state reasoning).
Possible overall scores could be between 0 and 16.
Inter-rater reliability has been found to be relatively high for both mental state
reasoning (.98) and comprehension (.90) (Dodell-Feder et al., 2013). In the initial
testing of the measures scores ranged from 2 to 14, and there was no indication of a
148
ceiling effect. Concurrent validity was achieved by examining the relationship between
participants’ scores on other ToM measures, including the Interpersonal Reactivity
Index (IRI) and the RMET. Mental state reasoning on the SST demonstrated a
statistically significant relationship on the IRI ‘fantasy’ subscale, but not on the
‘perspective-taking’, ‘empathic concern’, or ‘personal distress’ subscales. A significant
relationship was found between SST mental state reasoning and the RMET.
5.2.1.3. Procedure. Prior to being tested, participants were fully informed about
what would happen in the study and were sent an online survey, accessed via Qualtrics,
which included a consent form for the study, the AQ, EQ, and CAT-Q. It also asked a
number of demographic questions, including confirmation of their age, gender,
nationality, first language, ASC diagnosis, age of ASC diagnosis, who their ASC
diagnosis was made by, any relatives with an ASC diagnosis, and if they were diagnosed
with any mental health problems or learning difficulties, and to specify what these were.
Once the survey was completed, participants were asked to attend a one-hour
testing session at the university. Informed consent was collected again and participants
were reminded of the testing that would take place. Initially, participants were filmed
having an everyday conversation with a research assistant (see Part Two, section 5.3.1.2
for more details). Following this, participants were given the computer battery of EF
tasks to complete, which were ordered randomly each time to avoid fatigue effects.
They were then asked to read out the list of words on the NART test, which was
recorded for later analysis. Lastly, they were asked to read the short story for the ToM
task, and were then recorded answering questions on the story they had just read.
5.2.2. Results
5.2.2.1. Data checks and descriptive statistics. A descriptive table was initially created
to examine group averages on each of the continuous variables (AQ, EQ, CAT-Q, EF,
149
and ToM), that is, for autistic females, autistic males, non-autistic females, and non-
autistic males. A two-way ANOVA was conducted for each of the measures to determine
if there was an interaction between gender and autism group. Pairwise comparisons
were made between groups using a Bonferroni correction for multiple comparisons.
Correlations were also calculated between all the variables and the CAT-Q, again with
Bonferroni corrections applied. Correlations were only carried out on samples with over
30 participants; any associations involving 30 or fewer participants were considered
exploratory due to limited power.
Prior to conducting the analyses, tests of normality were performed on
continuous variables to ensure these were not heavily skewed or abnormally distributed.
Examining histograms and employing the Kolmogorov-Smirnov (K-S) test indicated
slight departures from normality on EQ, ToL, and Stroop task though the K-S test
results were not significant. BCST scores had a strong negative skew and significant K-
S statistic demonstrating abnormality in the distribution. The BCST scores were
therefore transformed using log transformations; this improved the skew of the scores
slightly although it did remain significantly abnormally distributed according to the K-S
test. However, ANOVAs with equal numbers remain relatively robust to departures of
normality.
5.2.2.2. Effects of gender and autism on all measures. Table 5.2 presents
descriptive statistics (means and standard deviations/frequency data) for all measures as
a function of gender and group. Scores for the AQ, EQ, and CAT-Q are averages of the
raw scores. The ToM measure has three scores: the percentage of each group who
demonstrated a spontaneous mental state inference, the average percentage of correct
comprehension answers given, and the average percentage of correct explicit mental
state answers given. The EF measure has four scores: the difference in reaction times on
150
the numerical Stroop task between the incongruent and congruent trials (higher scores
represent worse inhibition), the percentage of correct moves on the BCST, the total
score on the TOL, and the total EF score derived by summing the average Z scores for
the three tasks (after reverse-scoring inhibition), with higher values representing better
EF overall.
Table. 5.2
Means and standard deviations on all measures as a function of group and gender
Measure ASC Non-Autistic
Females Males Females Males
AQ 36.55 (7.55) 34.05 (7.52) 18.25 (8.99) 18.90 (7.22)
EQ 25.10 (10.80) 23.89 (10.56) 46.20 (14.03) 38.80 (11.81)
Cognitive 3.25 (3.49) 4.32 (5.89) 12.20 (5.55) 11.60 (4.51)
Reactivity 9.50 (4.71) 8.21 (3.29) 14.05 (4.63) 10.05 (4.17)
CAT-Q 123.20 (28.76) 114.47 (27.06) 89.95 (25.69) 88.90 (29.36)
151
Measure ASC Non-Autistic
Compensating 42.60 (12.68) 39.53 (11.40) 26.10 (10.94) 25.80 (12.46)
Masking 38.50 (11.17) 34.58 (11.93) 35.60 (10.42) 35.50 (7.26)
Assimilation 42.05 (12.25) 40.37 (8.45) 28.20 (8.76) 27.60 (12.29)
ToM
Spontaneous
mental state
inferences (%
who made)
10.53% 21.05% 10.53% 26.32%
Comprehension
(% correct)
68.42 (17.72) 65.79 (19.53) 66.32 (16.06) 72.63 (17.90)
Explicit mental
state (% correct)
49.67 (14.80) 41.78 (17.44) 51.97 (18.05) 49.67 (17.98)
EF (Z score) -0.12 (0.62) -0.05 (0.62) 0.19 (0.47) 0.01(0.66)
Stroop RT (ms) 68.70 (31.50) 68.90 (45.61) 73.18 (47.66) 66.59 (29.95)
BCST % correct 81.25 (7.35) 78.74 (12.85) 76.57 (11.28) 76.70 (13.01)
ToL 22.80 (8.67) 23.70 (8.25) 26.70 (6.07) 25.20 (7.93)
CAT-Q: As can be seen from Table 5.2, autistic females scored on average
highest on the CAT-Q, followed by autistic males, and non-autistic females and non-
autistic males who had similar average scores. A two-way ANOVA revealed a non-
significant interaction between gender and group on the overall CAT-Q score, F(1,76)
= .580, p = .556. However, there was a significant main effect for group reflecting
greater self-reported camouflaging in the autistic participants, F(1, 76) = 23.017, p
<.001, ηp2 = .23. When considering the individual scales of the CAT-Q, in no case was
there a significant interaction between gender and group, all p values > .02 (Bonferroni
corrected). However, there was a significant main effect for group, reflecting greater
camouflaging by the autistic participants for both compensation, F(1,76) = 32.524, p
<.001, ηp2 = .30, and assimilation, F(1,76) = 31.219, p <.001, ηp2 = .29, but not
masking, p = .02.
AQ: As can be seen from Table 5.2, autistic females scored on average highest
on the AQ, followed closely by autistic males, whilst non-autistic females and non-
autistic males had similar average scores that were much lower. A two-way ANOVA
152
revealed a non-significant interaction between gender and group on the AQ, F(1,76)
=1.096, p = .298. There was a significant main effect for group reflecting higher AQ
scores in the participants with an ASC diagnosis, F(1, 76) = 86.675, p <.001, ηp2 = 0.53.
EQ: As can be seen from Table 5.2, autistic males scored on average lowest on
the EQ, followed by autistic females, non-autistic males, and non-autistic females. A
two-way ANOVA revealed a non-significant interaction between gender and group on
the EQ, F(1,76) = 1.714, p = .194. There was a significant main effect for group
reflecting lower EQ scores in the participants with an ASC diagnosis, F(1,76) = 43.345,
p <.001, ηp2 = 0.38. A similar pattern was observed when the EQ subscales were looked
at separately. A non-significant interaction between gender and group was observed for
cognitive empathy, F(1,76) = .562, p = .456, but with a significant main effect for group
only, reflecting lower cognitive empathy scores in the participants with an ASC
diagnosis, F(1,76) = 53.367, p <.001, ηp2 = 0.42. A non-significant interaction between
gender and group was also observed for emotional reactivity, F(1,76) = 2.641, p = .108,
but with a significant main effect for group only, reflecting lower emotional reactivity
scores in the participants with an ASC diagnosis, F(1,76) = 9.424, p = .003, ηp2 = 0.11.
ToM: As can be seen from Table 5.2, all groups scored similarly in terms of
spontaneous mental state inferences, comprehension, and on explicit mental state
inferences in the SST. A Chi-Square analysis revealed that the number of participants
making a spontaneous mental state inference did not differ significantly by group, X2(3)
= 2.505, p = .474. A two-way ANOVA revealed a non-significant interaction between
gender and autism on comprehension on the SST, F(1,72) =1.194, p = .278, and on
explicit ToM on the SST, F(1,72) = .507, p = .479. There were no significant main
effects or interactions when considering percentage accuracy of comprehension and
explicit ToM.
153
Executive Functioning: As can be seen from Table 5.2, all groups scored
similarly on the EF battery. A two-way ANOVA found a non-significant interaction
between gender and autism on the percentage of correct moves on the BCST, F(1,76) = .
091, p = .764, scores on the ToL, F(1,76) = .474, p = .493, and on the reaction times
differences between congruent and incongruent trials on the numerical Stroop task,
F(1,76) = .148, p =.702. A non-significant interaction was also reported for overall EF
scores, F(1,76) = .596, p = .442. No main effects for gender or autism were observed in
any of the tests or in the overall EF score.
Mental health: More autistic women had a mental health condition than autistic
men and non-autistic participants, and more autistic males had mental health conditions
than non-autistic participants (autistic females = 12, autistic males = 8, non-autistic
females = 5, and non-autistic males = 1). A Chi-Square analysis revealed that the group
difference was significant: X2(3) = 17.582, φ = .469, p = .001. Odds ratios revealed
autistic females were 2.3 times more likely than autistic males, 5.6 times more likely
than non-autistic females, and 28.5 times more likely than non-autistic males to have a
mental health condition.
Autistic participants were divided into two groups, low and high camouflagers,
using their median camouflaging score on the CAT-Q (median = 118.50). It was found
that the number of participants with a mental health condition did not differ between
high- and low camouflagers (11 versus 9 respectively).
Age of ASC diagnosis: Autistic females received their diagnoses later than
autistic males (females: M = 22.25, SD = 10.00, males: M = 13.90, SD = 8.81), which an
independent measures t test found to be significant, t(38) = 2.802, p = .008, d = 0.89.
154
5.2.2.3. Correlation analyses. Pearson correlations were calculated between all
continuous measures, first for all participants and then for autistic and non-autistic
participants separately. Groups were collapsed across gender as no consistent
differences between males or females were found on the tests described above.
Bonferroni corrections were applied to control for multiple tests.
As can be seen from Table 5.3, across all participants CAT-Q scores were
significantly, positively correlated with AQ scores, and significantly, negatively
correlated with EQ scores. Looking at correlations between other variables, AQ was
significantly, negatively correlated with EQ. Separate analysis conducted using the two
subscales on the EQ and three from the CAT-Q, with a Bonferroni correction, revealed a
significant negative correlation between cognitive empathy and AQ scores (partial r(80)
= -.811, p < .001) and overall CAT-Q scores (partial r(80) = -.393, p < .001). In
particular cognitive empathy was negatively associated with compensation on the CAT-
Q (partial r(80) = -.435, p < .001) and assimilation (partial r(80) = -.518, p < .001), but
not with masking (p = .836). Whilst emotional reactivity significantly correlated
negatively with only AQ (partial r(80) = -.478, p < .001) and the CAT-Q assimilation
subscale (partial r(80) = -.382, p < .001).
Table 5.3
Correlations between continuous measures for all participants
Measures CAT-Q AQ EQ EF ToM
CAT-Q -
AQ .545** -
EQ -.469** -.800** -
EF -.032 -.116 .183 -
ToM -.042 -.156 .258 .208 -
*Correlation is significant at the p < .003 level (two-tailed) (Bonferroni corrected)
155
As can be seen from Table 5.4, when the correlations were examined just in the
autistic groups, none of the variables correlated significantly with CAT-Q scores and the
only significant relationship was between AQ scores and EQ scores (negative). Separate
analysis conducted using the two subscales on the EQ and three from the CAT-Q, with a
Bonferroni correction, revealed a significant negative correlation between cognitive
empathy and AQ scores only (partial r(40) = -.619, p < .001), which was the same for
emotional reactivity (partial r(40) = -.611, p < .001). Given the strong, positive
correlations between current age and age of ASC diagnosis for both genders, the
correlation between CAT-Q and age of ASC diagnosis was re-examined after controlling
for current age. However, with Bonferroni corrections applied, there were still no
significant correlations between CAT-Q scores and other variables for this group.
Table 5.4
Correlations between continuous measures for autistic participants
Measures CAT-Q AQ EQ EF ToM ASC diagnosis
age
CAT-Q -
AQ .249 -
EQ -.070 -.810** -
EF .092 .196 -.109 -
ToM .080 -.104 .188 .244 -
ASC
diagnosis age
.187 .405 -.202 .305 .388 -
* Correlation is significant at the p < .003 level (two-tailed) (Bonferroni corrected)
As can be seen from Table 5.5, when the correlations were examined just in non-
autistic populations, the only significant relationship was between AQ scores and EQ
156
scores (negative). However, when separate analysis was conducted using the two
subscales on the EQ and three from the CAT-Q, with a Bonferroni correction, only
cognitive empathy was significantly correlated negatively to AQ scores (partial r(40) =
-.680, p < .001), and no other correlations between other variables or emotional
reactivity were found.
Table 5.5
Correlations between continuous measures for non-autistic participants
Measures CAT-Q AQ EQ EF ToM
CAT-Q -
AQ .400 -
EQ -.411 -.556** -
EF 0.15 -.185 .293 -
ToM -.012 -.038 .223 .137 -
*Correlation is significant at the p < .003 level (two-tailed) (Bonferroni corrected)
5.2.3. Summary
Autistic participants scored higher than non-autistic participants on measures of autistic
traits and camouflaging and lower on empathy. No group or gender differences were
found on ToM or EF, and no interaction between gender and autism, or main effect of
gender, was noted on any of the variables. In terms of mental health conditions, autistic
females were found to be significantly more likely to have them; however, this was not
found to be related to whether participants were high or low camouflagers.
When correlations were investigated, camouflaging was predicted by the AQ
and EQ only when the whole sample was considered. When the sample was divided into
autistic and non-autistic groups, this pattern was no longer significant. Together these
results suggest that camouflaging is a behaviour shown particularly by autistic
157
individuals, but that it does not vary according to gender or cognitive abilities as
previously thought. As predicted, higher camouflaging scores were linked with a later
age of ASC diagnosis – but only in males.
5.3. Part Two
5.3.1. Method
5.3.1.1. Participants. Participant-raters were recruited from the university, using both
online and physical posters asking participants to partake in a study looking at social
judgements of others based on first-impressions (note, no mention of autism was given
here). Course credits were offered as well as a place in a prize draw with a chance to
win a £50 Amazon voucher. In total, 53 males and 74 females were recruited; one male
was transgender and was therefore categorised as the gender they currently identified as
(male). Participants were aged between 18 and 40 years (males: M = 27.17, SD = 6.05,
females: M = 24.08, SD = 5.51). They were further required to not have an ASC, or any
uncorrected visual or hearing impairments, and they must speak English as a first
language. These criteria ensured that the participant-raters were similar to the
participants being observed (hereafter referred to as participant-stimuli) in terms of age
and cultural background, and therefore could be considered ‘peers’.
5.3.1.2. Materials. Video clips to be rated were created from the video-recorded
social interactions created during part one of the study; consent was gained from the
participant-stimuli to use their video clips in this way. Each of the 80 participants
described in part one were video-recorded having a conversation with a research
assistant. Following the procedures used by Sasson et al. (2017), the participant-stimuli
were recorded engaging in as natural a conversation as possible. Two female research
assistants aged in their early 20s met briefly with participants prior to recording, but
158
were not informed about group membership by the researcher beforehand as previous
research had found that this affects first-impression ratings (Grossman, 2015; Sasson &
Morrison, 2019). A similar number of participants across each of the four groups were
interviewed by each of the research assistants (RA 1 tested 10 non-autistic females, 10
non-autistic males, 8 autistic females, and 8 autistic males. RA 2 tested 10 non-autistic
females, 10 non-autistic males, 12 autistic females, and 11 autistic males). A Chi-Square
analysis found no significant differences between these frequencies (X2(3) = .659, p = .
883), and an independent-groups T-Test found no differences in the overall first-
impression ratings given to participants interviewed by either of the RAs (t(38) = -.800,
p = .429) .
Each interview was conducted by a single research assistant who sat directly
opposite the participant (approximately 1 meter away) and began by asking them a
number of open-ended questions about mundane topics (e.g., ‘what have you been up to
this summer?’ and ‘what do you like to do in your spare time?’). Subsequently, to
ensure consistency of content across participant-stimuli, the research assistants were
instructed to ask, at a natural and convenient point in the conversation, if the participant
could describe a film or book they had recently watched or read, or that was their
favourite. This meant that the participant-stimuli were all discussing similar topics and
were not disclosing any personal details about their lives or hobbies, which might bias
subsequent ratings.
Each research assistant wore a GoPro camera (Hero 4; recording in 1080p wide
at 60fps) on their head to record the conversation from a first-person point of view,
similar to the camera glasses used by Sasson et al. (2017). This enabled those
participant-raters later viewing the videos to observe the participant-stimuli as they
would if they were having a conversation with them themselves, from a natural angle
159
where the full face could be observed. The research assistants had been given training in
an interview technique that encouraged them to respond non-verbally as much as
possible (i.e., nodding and smiling), speaking only when needed to keep the
conversation going. Whilst the position of the GoPro camera may have felt intrusive to
the participants, the research assistants ensured that they had begun building a rapport
with the participants prior to attaching the headset, explaining to them why they would
be wearing it, making light of the unusual situation, and explaining that the conversation
was just a general informal chat and to try and ignore the camera. We did not reveal to
these participants exactly what participant-raters would be judging their conversations
on, so as not to influence the behaviours of the participant-stimuli. We also stressed that
we were not testing the content of the conversation, and that we just needed natural clips
of them having a ‘normal’ everyday conversation. Due to ethical considerations it was
important that participants knew they were being filmed, and had fully consented to
others viewing their conversations. In an attempt to mitigate from this distraction, video
clips of the recordings were taken after the participant-stimuli had been talking for over
one minute to give them time to feel more at ease with the unusual situation.
For each of the participant-stimuli, an excerpt of their recording lasting 10-15
seconds was extracted. These clips were always taken whilst the participant discussed a
book or film, which always occurred midway through or towards the end of the
conversation. The choice of 10-15 seconds was based on Sasson et al.’s (2017) study,
which used 10-second clips. Furthermore, Willis and Todorov (2006) found that
confidence in the judgements of others using the key trait assessments (social
awkwardness, attractiveness, trustworthiness, likeability, smartness, and dominance)
increased when the time of video clips increased from 1 second to 5 seconds, and from
5 seconds to 10 seconds. The precise point at which the clip was taken was selected
using a random number generator. However, these clips were also checked to ensure
160
that, where possible, they started and ended at a natural point in the utterance, for
example not in the middle of a sentence or word. No significant differences in the length
of videos was found between the four groups: F(1,36) = 1.352, p = .273.
Two independent raters reviewed clips to ensure the sound and picture quality
was consistent. There were 18 clips that were deemed of insufficient quality (non-
autistic females = 3, non-autistic males = 1, autistic females = 4, autistic males = 2). For
these clips, either the participant failed to engage in a sufficiently long enough
discussion of the topic (i.e., less than 10 seconds unbroken speech), or the research
assistant could be overheard responding to what the participant was saying (which could
potentially influence the participant-raters to view them as more sociable/friendlier). A
further 6 clips were discarded either because the participant-stimuli had visible
disabilities (two autistic female participant-stimuli and two autistic male participant-
stimuli), or because they had strong and sometimes incomprehensible regional accents
(two non-autistic females). Finally, one autistic male did not agree for filming to take
place. This left usable clips for 15 non-autistic females, 19 non-autistic males, 14
autistic females, and 14 autistic males. From this pool, ten clips were randomly selected
from each of the four participant-stimuli groups. The average age of the participant-
stimuli did not significantly differ between groups, F(3,36) = .231, p = .874 (M: non-
autistic females = 27.20, non-autistic males = 26.90, autistic females = 25.90, autistic
males = 25.20).
Video clips were uploaded onto the online survey platform Qualtrics and
presented to each participant-rater in a random order to avoid order effects. Each video
clip was accompanied by a short questionnaire on first-impressions derived from Sasson
et al.’s (2017) initial study. The questionnaire had 10 items, where participant-raters
rated how much they agreed with the behavioural intent and trait items for each of the
161
participant-stimuli on a four point scale from strongly agree (4) to strongly disagree (1)
with four items reverse-scored; larger first-impression scores therefore indicated more
favourable behavioural intent and trait assessments. There were six items related to
traits (social awkwardness, attractiveness, trustworthiness, aggressiveness, likeability,
and intelligence), found previously to reliably measure first-impressions (Grossman,
2014; Willis & Todorov, 2006). There were four items related to behavioural intentions
(willingness to live near the participant-stimulus, likelihood of hanging out with the
participant-stimulus in their free time, comfortableness sitting next to the participant-
stimulus, and likelihood of starting a conversation with the participant-stimulus), found
previously to reliably measure first-impressions (Campbell et al., 2004; Matthews et al.,
2015; Nevill & White, 2011). Sasson and Morrison (2019) found that averaging the 10
items into a single first-impression score indicated strong internal consistency
(Chronbach’s α = 0.82).
5.3.1.3. Procedure. Participant-raters completed the study online after being
provided with the link on request and instructions about how to open and view the
videos. Participants were informed that the study would involve watching and listening
to 40 videos and then rating these using a questionnaire. However, they were not
informed that some of the videos were of autistic people or that first impressions of
autistic and non-autistic people were being compared. They were told only that the
study was looking at the social judgements made when viewing short video clips of
strangers. Questions at the beginning of testing checked that the participants met the
inclusion criteria on age, were non-autistic, and that they didn’t have any uncorrected
visual or hearing impairments. A short test video was initially played where the
experimenter was seen verbally providing participants with a password to enter before
proceeding. This ensured that all participants were able to see and hear the videos they
were about to watch and rate. The actual test session was divided into two halves. Five
162
videos from each of the four participant-stimuli groups were played randomly in the
first half, followed by a five minute break, and then the final 20 videos. The First-
Impressions scale was presented after each video (see Figure 5.1 for an example).
Finally, participants were debriefed on the general aims of the study, which stated that it
aimed to, “investigate the first impressions of different groups based on short video
clips of social interactions, and whether this related to self-reported social
camouflaging, ToM, and empathy abilities.”
Figure 5.1
163
Example of video clip and survey layout on Qualtrics
164
5.3.2. Results
5.3.2.1. Participant-raters’ first-impressions of participant-stimuli. A 2 x 2 x 2
mixed ANOVA was conducted on the overall first-impression scores. Independent
variables included between-subject participant-rater gender (male versus female), and
within subjects participant-stimuli gender (male versus female), and participant-stimuli
group (autistic versus non-autistic). Distributions of first-impression scores were normal
for each condition group, and Levene’s test was non-significant.
165
Table 5.6
Means and standard deviations for the first-impression scores as a function of group
and gender
Autistic
Females
Autistic
Males
Non-autistic
Females
Non-autistic
Males
First-impressions 28.02 (2.70) 26.74 (2.92) 29.43 (2.85) 28.65 (3.06)
Behavioural-intent 11.28 (1.37) 10.83 (1.53) 11.98 (1.43) 11.57 (1.70)
Live near* 3.01 (0.42) 3.11 (0.53) 3.28 (0.48) 3.21 (0.50)
Hang out 2.48 (0.45) 2.29 (0.45) 2.66 (0.44) 2.54 (0.46)
Sitting next to* 3.14 (0.54) 2.94 (0.48) 3.21 (0.54) 3.11 (0.55)
Start
conversation
2.65 (0.45) 2.49 (0.48) 2.82 (0.44) 2.71 (0.46)
Traits 16.74 (1.53) 15.92 (1.59) 17.45 (1.64) 17.05 (1.70)
Socially
awkward*
2.34 (0.43) 2.20 (0.44) 2.85 (0.39) 3.05 (0.40)
Attractive 2.58 (0.44) 2.07 (0.44) 2.59 (0.42) 2.53 (0.43)
Trustworthy 2.91 (0.31) 2.86 (0.33) 2.96 (0.32) 2.82 (0.35)
Aggressive* 3.25 (0.42) 3.25 (0.42) 3.11 (0.46) 2.91 (0.48)
Likeable 2.88 (0.32) 2.78 (0.35) 3.06 (0.32) 2.96 (0.33)
Smart 2.78 (0.48) 2.76 (0.55) 2.88 (0.48) 2.77 (0.49)
* Reverse scored item as negatively worded (higher score = more favourable)
Main effects were found for participant-stimuli group, F(1,123) = 147.498, p < .
001, ηp2 = 0.55, participant-stimuli gender, F(1,123) = 55.110, p = .001, ηp2 = 0.31, and
for participant-rater gender, F(1,123) = 8.369, p = .005, ηp2 = 0.08. As can be seen from
Table 5.6 and Figure 5.2, autistic participant-stimuli were rated significantly poorer than
non-autistic participant-stimuli, males were rated significantly poorer than females, and
male participant-raters rated all participants significantly more negatively than female
participant-raters.
A significant 2-way interaction was found between participant-stimuli group and
participant-stimuli gender, F(1,123) = 11.086, p = .001, ηp2 = 0.08. However, non-
166
significant interactions were found between participant-stimuli group and participant-
rater gender, F(1,123) = .345, p = .558, ηp2 = 0.03, and between participant-stimuli
gender and participant-rater gender, F(1,123) = .326, p = .5691, ηp2 = 0.03.
Figure 5.2.
Average first-impression scores of non-ASC females, non-ASC males, ASC females, and
ASC males for male and female participant-raters with SD bars.
The 3-way interaction of participant-stimuli group x participant-stimuli gender x
participant-rater gender was significant, F(1,123) = 5.444, p = .021, ηp2 = 0.42. This was
followed up by two (2 x 2) simple repeated measure ANOVAs, to investigate the
interaction between participant-stimuli gender and participant-stimuli group separately
for male and female raters. For male raters, an interaction between participant-stimuli
gender and group was found (F(1,51) = 11.716, p = .001, ηp2 = 0.187). Moreover, main
effects were observed for both autism group (F(1,51) = 53.855, p <.001, ηp2 = 0.514)
and gender (F(1,51) = 16.354, p <.001, ηp2 = 0.243). Paired t tests, using a Bonferroni
correction due to multiple comparisons (p = .008), revealed significant differences in the
ratings between certain groups. Autistic females were rated significantly more
167
favourably (M = 27.45, SD = 2.33) than autistic males (M = 25.81, SD = 2.79) but
significantly less favourably than non-autistic females (M = 28.51, SD = 2.40), p <.001.
Autistic males were rated significantly less favourably than both non-autistic females
(M = 28.51, SD = 2.40) and non-autistic males (M = 27.90, SD = 2.96), p <.001. No
significant differences in first-impression scores were found between autistic females
and non-autistic males or between non-autistic males and non-autistic females.
For female raters, there was no significant interaction between participant-
stimuli gender and group, F(1,72) = .679, p = .413, ηp2 = .009. However, there was a
main effect of autism group (F(1,72) = 101.880, p <.001, ηp2 = .586), and gender
(F(1,72) = 53.920, p <.001, ηp2 = .428). Paired t tests, using a Bonferroni correction due
to multiple comparisons (p = .008), revealed significant differences in the ratings
between certain groups. Autistic females were rated significantly more favourably (M =
28.47, SD = 2.88) than autistic males (M = 27.42, SD = 2.85), but significantly worse
than non-autistic females (M = 30.09, SD = 2.98) and non-autistic males (M = 29.21,
SD = 3.03), p <.001. Whilst autistic males were rated significantly worse than both non-
autistic males and females, p <.001, and non-autistic males were rated significantly
worse than non-autistic females, p <.001.
Taken together the results indicate that non-autistic females scored most
favourably on overall first-impression scores, followed by non-autistic males, autistic
females, and then autistic males. Both male and female participant-raters rated non-
autistic males less favourably than non-autistic females, and male participant-raters
rated both males and females less favourably than female participant-raters. This pattern
is the same for male and female participant-raters when rating autistic stimuli. However,
it is also apparent that male participant-raters rated autistic males more harshly than
other groups.
168
5.3.2.2. Correlation analyses. The average first-impression score was
calculated for each of the participant-stimuli and entered into a correlation analysis with
each participant-stimulus’ camouflaging scores from the CAT-Q and the autistic-
stimulus’ age of autism diagnosis. No significant correlation between first-impression
ratings and camouflaging was found when including all participant groups, r(40) = .047,
p = .775, or for just autistic participant-stimuli, r(20) = .361, p = .117, and non-autistic
participant-stimuli, r(20) = .111, p = .641, when considered separately. However, a
significant, positive correlation was found between first-impression ratings and age of
diagnosis for the autistic-stimuli, r(20) = .505, p = .023. When autistic males and
females were considered separately no significant correlations was found (r(10) = .105,
p = .772 and r(10) = .535, p = .111 respectively.
5.3.3. Summary
As predicted, there was a significant interaction between group and gender on overall
first-impression scores. Significant differences were found between all four groups with
autistic males being rated least favourably, followed by autistic females, non-autistic
males, and finally non-autistic females. Importantly, gender of the rater was found to
moderate this pattern. Generally, both male and female raters rated males less
favourably than females and autistic participant-stimuli less favourably than non-autistic
participant-stimuli. However, male raters were significantly harsher in their ratings of
autistic males than females were. Therefore, the interaction between group and gender
of those being rated was being driven by male raters. Nevertheless, female raters
showed significant main effects for both gender and group, reflecting the fact that they
too rated autistic participants less favourably than non-autistic participants, and male
participants overall less favourably than female participants. Correlation analysis
169
revealed that while first-impression scores for the participant-stimuli showed a positive
correlation with camouflaging scores, it did not reach significance. However, first-
impression scores were significantly, positively correlated with age of ASC diagnosis in
the autistic participants.
5.4. General Discussion
Studies 1 and 2 highlighted a cohort of women with a potential ASC. Whilst these
women showed a slight advantage in empathy and social functioning over diagnosed
autistic women, they still demonstrated similar impairments on measures of friendship,
ToM, self-monitoring, and anxiety and depression. Study 3 therefore aimed to measure
self-reported camouflaging in autistic women (using a newly developed measure
designed specifically for this purpose) and to investigate how their social behaviours are
viewed by non-autistic peers, to determine whether their camouflaging is successful
and/or if they present less atypically than autistic males. The study was divided into two
parts. Part 1 explored gender differences in the use of self-reported camouflaging in
autistic versus non-autistic adults using the CAT-Q, and examined whether the use of
camouflaging strategies was related to the AQ, EQ, EF, ToM, mental health diagnoses,
or age of ASC diagnosis. Part 2 examined whether video clips of autistic males and
females having social conversations were rated more or less favourably than non-
autistic males and females on a first-impressions survey, and whether these ratings
correlated with self-reported camouflaging and age of ASC diagnosis.
5.4.1. Part-one: Group and gender differences in self-reported
camouflaging. In the first part of the study, self-reported camouflaging scores on the
CAT-Q were compared between autistic females, autistic males, and a control group of
non-autistic male and female participants. Additionally, correlations were examined
170
between camouflaging and empathy, EF, ToM, age of ASC diagnosis, and analysis was
also conducted to investigate if higher camouflaging led to more mental health issues. It
was predicted that autistic females would score higher on self-reported camouflaging
than autistic males, and that this would be associated with better EF, ToM, empathy, and
a later age of diagnosis, with a greater likelihood of mental health problems. These
hypotheses were not wholly supported by the results. Whilst autistic females were
diagnosed significantly later than autistic males and were more likely to have mental
health problems, there was no significant interaction between group and gender on the
CAT-Q or the EQ. Instead, regardless of gender, the autistic group scored higher on
camouflaging and lower on empathy than the non-autistic group, and this was true for
both the cognitive empathy and emotional reactivity subscales of the EQ. Furthermore,
there were no significant group differences on ToM or EF. CAT-Q scores were found to
correlate positively with the AQ and negatively with the EQ (as well as both cognitive
empathy and emotional reactivity separately) only when the whole sample was used, but
not when looking at autistic and non-autistic groups separately. The significant positive
correlation between the AQ and self-reported camouflaging across the sample most
likely reflects the fact that most autistic participants reported strong use of camouflaging
techniques. The correlation did not reach significance in the individual groups due to the
smaller sample size. It is worth noting that there was a moderate (albeit not significant)
correlation between AQ scores and self-reported camouflaging even in the non-autistic
group, which might mean that even neurotypical people with higher levels of autistic
traits tend to socially interact in more effortful ways.
The findings from this study are inconsistent with those of Hull, Lai, et al.
(2019), who found scores on the CAT-Q to be significantly higher in autistic females
than autistic males. However, both Hull, Lai, et al. (2019) and the current study’s
findings had a similar effect size in the difference between autistic males and autistic
171
females (d = 0.65 and 0.58 respectively). The mean CAT-Q score for autistic females in
Hull, Lai, et al.’s (2019) study was 124.35 (SD = 23.27) and the mean CAT-Q scores for
autistic males was 109.64 (SD = 26.50), compared to the current study which found a
mean score of 123.20 (SD = 28.76) for autistic females and 114.47 (SD = 27.06) for
autistic males. This previous study sampled a greater number of participants in total (n
=778) than the current study and therefore had more statistical power, leading to their
significant findings. On the other hand, Cage and Troxell-Whitman (2019) also did not
find a difference between autistic males (n = 111) and autistic females (n = 135) on the
CAT-Q: autistic females scored on average 118.90 (SD = 18.83) and autistic males
scored on average 114.25 (SD = 21.36). Findings regarding gender differences in
camouflaging scores on the CAT-Q therefore continue to be inconsistent. It may be the
case that both genders attempt to camouflage their autistic traits, but there may be subtle
differences in how this is achieved, which are not captured using the CAT-Q. For
example, Cassidy et al. (2018) also did not detect any differences between the
percentages of autistic men and autistic women who attempted to use camouflaging.
However, they used their own camouflaging questionnaire and not the CAT-Q. This
scale asked participants if they had “ever tried to camouflage or mask [their]
characteristics of ASC to cope with social situations? For example, have [they] ever
tried to copy or mimic other people’s behaviour to try and fit in, or tried to mask of hide
[their] symptoms of ASC from other people?” If participants answered yes to this then
they were then asked to specify in which areas of their life they camouflaged, how
frequent this was on a scale of 1 (never) to 6 (always), and lastly the overall amount of
the day they spent camouflaging on a scale of 1 (none of my waking time) to 6 (all of
my waking time). 89.2% of autistic females attempted to camouflage, which was similar
to the 90.9% of autistic males. In contrast, the overall scores on the camouflaging scale
were significantly higher for autistic females (M = 14.7) than autistic males (M = 12.95)
172
which had a medium effect size (d = .47). Moreover, they detected subtle gender
differences in the quality of camouflaging, for example, autistic women camouflaged
across more situations than autistic men. Unlike their scale, the CAT-Q may be unable
to determine the quality of camouflaging behaviours and the success of these.
With increased media coverage of the topic of late diagnosis in autism and
camouflaging, it is possible that more autistic people than before, both females and
males, are employing camouflaging strategies or becoming aware that they already use
them. The notion of the female autism phenotype started in the early 1990s (Kopp &
Gillberg, 1992); however, the idea of camouflaging in autistic females only became
popular over a decade ago, when autism professionals began to observe more autistic
females than had been seen previously camouflaging their autistic traits (Attwood &
Grandin, 2006). From this there grew an increasing body of autobiographical books and
online blogs from women who were diagnosed with autism in adulthood, describing
their attempts to ‘appear normal’ and to camouflage to fit in with others (Miller, 2003;
Simone, 2010; Willey, 1999). Qualitative studies explored the experiences of these
autistic women, where again camouflaging was flagged as a common theme (Hull,
Petrides, et al., 2017; Tierney et al., 2016). The concept of camouflaging in autism,
particularly in females, may have become a contagious concept. With increased
publicity around the topic it is likely that many undiagnosed autistic women became
more aware of their difficulties and social strategies and understood these better under
the concept of ‘camouflaging’. It is also possible that many young autistic females
growing up have learnt about the behavioural strategy from reading about other girls’
experiences, and therefore are more likely to use camouflaging strategies themselves. In
addition to this there has been an increase in social media use since the female
phenotype of autism was first conceived, and socialisation has altered as a result; many
people regularly ‘camouflage’ online, disguised as different people or present to others
173
how they wish to be seen (Aiken, 2017). Therefore, the concept of the camouflaging of
autistic traits may have changed since its conception. It is possible that it is not an
exclusively female trait, and many autistic males may have also utilised the strategy, or
themselves been diagnosed late because of it (as found in Part-Two of the current
study). None of the qualitative studies conducted considered the ‘male’ experience of
autistic camouflaging, and thus far this has only been framed from a female perspective.
There are only a handful of quantitative studies that have explored camouflaging in both
autistic men and women.
Further predictions for the current study that heightened camouflaging would be
correlated with better EF, ToM, and empathy were also not supported. Previous research
had suggested that autistic females may have sex-distinct cognitive abilities that
enhance their ability to socialise and camouflage autistic behaviours (Bolte et al., 2011;
Lai et al., 2012; Lenhardt et al., 2016). However, for the current sample this was not the
case. Furthermore, whilst Livingston et al. (2018) found superior EF, along with higher
IQ and greater anxiety, to be linked to a better ability to compensate for underlying
deficits in ToM, no gender differences in compensation were found. The current study is
the first to explore self-reported camouflaging behaviours and their link with EF, ToM,
and empathy. Future studies should attempt to replicate these findings on larger samples
of autistic people.
Finally, no support was found for the prediction that heightened camouflaging
would be associated with a raised likelihood of mental health problems. This finding is
inconsistent with previous literature which has found worse mental health in those who
camouflage (Cage & Troxell-Whitman, 2019; Cassidy et al. 2018). For example, Hull,
Mandy, et al. (2018) found mental health was positively correlated with the CAT-Q
using the Social Anxiety Scale, Warwick-Edinburgh Mental Wellbeing Scale, Patient
174
Health Questionnaire, and Generalized Anxiety Scale. However, the current study did
find that autistic females were more likely to have a mental health issue than autistic
males, which might indicate that rather than camouflaging being the direct cause of
mental health issues in autistic females, it may instead be a consequence of other
associated factors, such as later diagnosis and lack of support (Stagg & Belcher, 2019).
However, the current study did not measure mental health traits in the same way as
previous research, and instead relied on reporting of clinical diagnoses of mental health
problems. Therefore, it is not possible to say whether camouflaging scores correlate
with poorer mental health, only that those who are higher camouflagers are not more
likely than low camouflagers to have other mental health diagnoses. Livingston et al.
(2018) have suggested that camouflaging affects mental health because of the additional
mental resources required, which again conflicts with the current findings. It may be the
case that there is a ceiling effect in mental health issues caused as a result of
camouflaging, and as the current study found camouflaging to be a uniquely autistic
strategy, with most autistic participants using the strategy, it may not matter how high
participants scored; rather it is just the fact that they feel they have to use the strategy at
all.
5.4.2. Part-two: Group and gender differences in first-impressions. In the
second part of the study, a sample of video clips of natural conversations involving the
same participants used previously (participant-stimuli) were shown to non-autistic peers
(participant-raters). These participant-raters rated each video clip on the First-
Impressions scale, and these results were compared within the four different groups
(autistic females, autistic males, non-autistic females, and non-autistic males). It was
predicted that autistic female participant-stimuli would be rated more favourably than
autistic male participant-stimuli, due to their social presentation appearing more typical.
Secondly, it was predicted that participant-stimuli’s average first-impression scores
175
would positively correlate with their camouflaging scores on the CAT-Q and age of ASC
diagnosis.
The first prediction was supported by the findings. A group bias was observed,
with autistic participant-stimuli being rated more negatively than non-autistic
participant stimuli, and a gender bias was also observed, with males being rated more
negatively than females. The gender of the participant-rater was also found to have an
impact on the ratings. Male raters tended to rate participant-stimuli more harshly;
however, they were harsher on autistic males than they were on any other participant-
stimuli. Therefore, autistic males were rated significantly less favourably than autistic
females, as they appeared to have a triple hit of being autistic, male, and rated more
harshly by male raters. These findings are consistent with those of Sasson et al. (2017),
who found that autistic people were rated less favourably than non-autistic people by
their peers. In addition, by including equal numbers of autistic males and females as
stimuli and by analysing the effects of rater gender, the current study yielded the novel
finding that autistic males are rated less favourably than autistic females by their peers,