The Autism-Spectrum Quotient (AQ) : evidence from Asperger Syndrome / high-functioning autism, males and females, scientists and mathematicians

Text-only Preview

Published: Journal of Autism and Developmental Disorders, 31, 5-17 (2001)

The Autism-Spectrum Quotient (AQ):
evidence from Asperger Syndrome/high-functioning autism,
males and females, scientists and mathematicians.

Simon Baron-Cohen, Sally Wheelwright,
Richard Skinner, Joanne Martin, and Emma Clubley

Departments of Experimental Psychology and Psychiatry,
University of Cambridge, Downing St,
Cambridge, CB2 3EB, UK.


Acknowledgements: The authors were supported by a grant from the MRC, the McDonnell-Pew
Foundation, and the Three Guineas Trust during the period of this work. We thank the reviewers who
improved this work through their feedback. RS, JM, and EC carried out Study 2 in part fulfilment of the
BSc in Experimental Psychology, University of Cambridge. Thanks also to Jessica Hammer for initial
discussions, to Nick Stone for data entry, and to Hugh Osborne and Imre Leader for access to Group 4.

Currently there are no brief, self-administered instruments for measuring the degree to
which an adult with normal intelligence has the traits associated with the autistic
spectrum. In this paper, we report on a new instrument to assess this: the Autism-
Spectrum Quotient (AQ). Individuals score in the range 0-50. Four groups of subjects
were assessed: Group 1: n = 58 adults with Asperger Syndrome (AS) or high-functioning
autism (HFA); Group 2: n = 174 randomly selected controls. Group 3: n = 840 students in
Cambridge University; and Group 4: n = 16 winners of the UK Mathematics Olympiad.
The adults with AS/HFA had a mean AQ score of 35.8 (sd = 6.5), significantly higher
than Group 2 controls (x = 16.4, sd = 6.3). 80% of the adults with AS/HFA scored 32+,
vs 2% of controls. Among the controls, males scored slightly but significantly higher than
women. No females scored extremely highly (AQ score 34+) whereas 4% of males did
so. Twice as many males (40%) as females (21%) scored at intermediate levels (AQ
score 20+). Among the AS/HFA group, males and female scores did not differ
significantly. The students in Cambridge University did not differ from the randomly
selected control group, but scientists (including mathematicians) scored significantly
higher than both humanities and social sciences students, confirming an earlier study that
autistic conditions are associated with scientific skills. Within the sciences,
mathematicians scored highest. This was replicated in Group 4, the Mathematics
Olympiad winners scoring significantly higher than the male Cambridge humanities
students. 6% of the student sample scored 32+ on the AQ. On interview, 11 out of 11 of
these met 3 or more DSM-IV criteria for AS/HFA, and all were studying
sciences/mathematics, and 7 of the 11 met threshold on these criteria. Test-retest and
inter-rater reliability of the AQ was good. The AQ is thus a valuable instrument for
rapidly quantifying where any given individual is situated on the continuum from autism
to normality. Its potential for screening for autism spectrum conditions in adults of
normal intelligence remains to be fully explored.



Autism is defined in terms of abnormalities in social and communication development, in
the presence of marked repetitive behaviour and limited imagination (APA, 1994).
Asperger Syndrome (AS) is defined in terms of the individual meeting the same criteria
for autism but with no history of cognitive or language delay, and not meeting the criteria
for PDD (ICD-10, 1994). Language delay itself is defined as not using single words by
two years of age, and/or phrase speech by three years of age. There is growing evidence
that autism and Asperger Syndrome (AS) are of genetic origin. The evidence is strongest
for autism, and comes from twin and behavioural genetic family studies (Bailey et al.,
1995; Bolton & Rutter, 1990; Folstein & Rutter, 1977; Folstein & Rutter, 1988). Family
pedigrees of AS also implicate heritability (Gillberg, 1991). There is also an assumption,
still under debate, that autism and AS lie on a continuum of social-communication
disability, with AS as the ‘bridge’ between autism and normality (Baron-Cohen, 1995;
Frith, 1991; Wing, 1981; Wing, 1988). The continuum view shifts us away from
categorical diagnosis and towards a quantitative approach.

Currently there are no brief, self-administered instruments available for measuring where
any given individual adult, with normal intelligence, lies on this continuum. Existing
instruments, such as the ADI-R (Autism Diagnostic Interview) (Le Couteur et al., 1989;
Lord, Rutter & Le Couteur, 1994), the ADOS-G (Autism Diagnostic Observation
Schedule) are fairly time-consuming to administer, and the CARS (Childhood Autism
Rating Scale) which can be brief, is not self-administered (Schopler, Reichler & Renner,
1986). What is needed is a short, self-administered scale for identifying the degree to
which any individual adult of normal IQ may have ‘autistic traits’, or what has been
called ‘the broader phenotype’ (Bailey et al., 1995). This would be useful for both
scientific reasons (e.g., establishing who is “affected” and who is not, or the degree of


‘caseness’ of an individual, in scientific comparisons), and potentially for applied reasons
(e.g., screening for possibly “affected” individuals to assist in making referrals for a full
diagnostic assessment). For both of these reasons, we developed the Autism-Spectrum
Quotient (AQ). The instrument’s name was chosen because of the assumption, mentioned
above, that there is an autism spectrum (Wing, 1988)1.

Design of the AQ

The AQ was designed to be short, easy to use, and easy to score. It is shown in Appendix
1. It comprises 50 questions, made up of 10 questions assessing 5 different areas: social
(items 1,11,13,15,22,36,44,45,47,48); attention switching (items
2,4,10,16,25,32,34,37,43,46); attention to detail (items 5,6,9,12,19,23,28,29,30,49);
(items 7,17,18,26,27,31,33,35,38,39); imagination (items
3,8,14,20,21,24,40,41,42,50). Each of the items listed above scores 1 point if the
respondent records the abnormal or autistic-like behaviour either mildly or strongly (see
below for scoring each item; Abnormality = poor social skill, poor communication skill,
poor imagination, exceptional attention to detail, poor attention-switching/strong focus of
attention). Approximately half the items were worded to produce a ‘disagree’ response,
and half an ‘agree’ response, in a high scoring person with AS/HFA. This was to avoid a
response bias either way. Following this, items were randomized with respect to both the
expected response from a high-scorer, and with respect to their domain.

Instrument development

1 The term ‘quotient’ is not used in the arithmetic sense (the result of dividing one quantity by another) but


Items were selected from the domains in the “triad” of autistic symptoms (APA, 1994;
Rutter, 1978; Wing & Gould, 1979), and from demonstrated areas of cognitive
abnormality in autism. The AQ as shown in Appendix 1 is the outcome of piloting
multiple versions, over several years. The instrument was piloted on adults with AS or
high functioning autism (HFA), and age matched controls. An early version was also
interview-based, and required the coding of responses. Following piloting, items which
controls scored on as often, or more often, than did people with autism/AS were omitted.

Due to the concern over whether a condition like HFA or AS might impair one’s ability
to understand the items in the questionnaire, we checked comprehension with the patients
in our pilot study. We did this by calling some patients into our lab, selected at random,
where we had the opportunity to ask them about their responses. Comprehension of
wording might be a greater problem in a less able population, but this instrument is
designed for high functioning individuals who are perfectly able to read or discuss issues.
For caution, however, parents independently completed an AQ for their child with
AS/HFA. A related issue is whether a condition like AS or HFA might impair the
subject’s ability to judge their own social or communicative behaviour, due to subtle
mind-reading problems (Baron-Cohen, 1995; Baron-Cohen, Jolliffe, Mortimore &
Robertson, 1997). If this occurred, this would lead a person to score lower on the AQ,
rating their own behaviour as more appropriate than it might really be. Any inaccuracies
of this kind would therefore, if anything, lead to a conservative estimate of the person’s
true ‘AQ’ score. However, to guard against false negatives, we included questions in both
the social and communication domains that ask about the person’s preferences, rather

as derived from the Latin root quotiens (how much or how many).


than only asking them to judge their own behaviour. Piloting revealed that such able
subjects were certainly able to report on their own preferences and what they find easy or
difficult. Equally, items in the other domains ask about their attentional preferences or
focus of attention (e.g., to dates, numbers, small sounds, etc.,). There is no reason to
expect that a high functioning person with autism or AS would be at all impaired in being
able to report faithfully on such items. The final version of the AQ has a forced choice
format, can be self-administered, and is straight forward to score since it does not depend
on any interpretation in the scoring.


Four groups of subjects were tested: Group 1 comprised n = 58 adults with AS/HFA (45
males, 13 females). This sex ratio of 3.5:1 (m:f) is similar to that found in other samples
(Klin, Volkmar, Sparrow, Cicchetti & Rourke, 1995). All subjects in this group had been
diagnosed by psychiatrists using established criteria for autism or AS (APA, 1994). They
were recruited via several sources, including the National Autistic Society (UK),
specialist clinics carrying out diagnostic assessments, and adverts in newsletters/web-
pages for adults with AS/HFA. Their mean age was 31.6 yrs (sd = 11.8, range 16.5-58.3).
They had all attended mainstream schooling and were reported to have an IQ in the
normal range. See below for a check of this. Their mean number of years in education
was 14.2 (sd = 2.41). 32 had higher educational qualifications (university degrees). Their
occupations reflected their mixed socio-economic status (SES). Because we could not
confirm age of onset of language with any reliability (due to the considerable passage of
time), these individuals are grouped together, rather than attempting to separate them into
AS vs HFA. The final sample of 58 were those who responded from a larger sample of


63. Group 2 comprised 174 adults selected at random (n = 76 males and 98 females).
They were drawn from 500 adults sent the AQ by post, giving a return rate of 34.8%.
They were all living in the East Anglia area. Their mean age was 37.0 yrs (sd = 7.7, range
18.1-60.0). Their mean number of years in education was 13.9 (sd = 2.34). 89 had
university degrees, and their mix of occupations was similar to that of Group 1. In
Groups 1 and 2, 15 individuals were randomly selected from the individuals who had
returned an AQ and invited into the lab to check pro-rated IQ, using 4 subtests of the
WAIS-R (see below). Group 3 comprised n = 840 students in Cambridge University (n =
454 males, n = 386 females). Their mean age was 21.0 yrs (sd = 2.9, range = 17.6-51.1).
They were drawn from 4175 students sent an AQ, giving a return rate of 20.1%. The
return rates from the different disciplines did not differ significantly. Group 3 was
included to test if they showed a similar profile to the randomly selected controls (Group
2, above), despite the difference in both IQ and educational level of the two groups.
Group 3 also allowed us to test if scientists differed from students in the humanities,
given earlier reports (Baron-Cohen et al., 1998) suggesting that autism is more common
in families of physicists, engineers, and mathematicians. Finally, Group 4 comprised n =
16 winners of the UK Mathematics Olympiad (15 males, 1 female). They were included
as a re-test of this same association. Their mean age was 17.4 yrs (sd = 1.0, range = 15.3 -


Subjects were sent the AQ by post, and the subject was instructed to complete it as
quickly as possible (to avoid thinking about responses too long), and to complete it on


their own. Subjects in Group 2 had the option to complete this anonymously or not. To
confirm the diagnosis of adults in Group 1 being high-functioning, 15 of them were
randomly selected and invited into the lab for intellectual assessment using 4 subtests of
the WAIS-R (Wechsler, 1958). The 4 subtests of the WAIS-R were Vocabulary,
Similarities, Block Design, and Picture Completion. On this basis, all of these had a
prorated IQ of at least 85, that is, in the normal range (mean=106.5, sd-8.0), and did not
differ significantly from the subsample (n=15) selected from Group 1 (t test, p > 0.5),
(mean=105.8, sd=6.3).

Scoring the AQ

‘Definitely agree’ or ‘slightly agree’ responses scored 1 point, on the following items: 2,
4, 5, 6, 7, 9, 12, 13, 16, 18, 19, 20, 21, 22, 23, 26, 33, 35, 39, 41, 42, 43, 45, 46.
‘Definitely disagree’ or ‘slightly disagree’ responses scored 1 point, on the following
items: 1, 3, 8, 10, 11, 14, 15, 17, 24, 25, 27, 28, 29, 30, 31, 32, 34, 36,37, 38, 40, 44, 47,
48, 49, 50.


AS/HFA vs. controls, and sex differences
Mean total and sub-category AQ scores from each group are displayed in Table 1.
Comparing Groups 1 and 2 using an ANOVA of total AQ score by Group and Sex, we
found, as predicted, that there was a main effect of Group [F (1, 228) = 328.9, p =
0.0001], the AS/HFA group scoring higher than the controls, and a two-way interaction
of Group by Sex (F (1, 228) = 6.01, p = 0.015) the control males scoring significantly


higher than the control females (t = 2.56, df = x, p < 0.01). There was no difference
between mean AQ scores of males and females with AS/HFA. Group means on each
subscore are also shown in Table 1. See also Figures 1 and 2 for graphic displays of the
Group and Sex differences. The AS/HFA group differed from Group 2 on all subscores
(t tests, p < 0.0001). Comparing the students (Group 3) to the randomly selected controls
(Group 2), there was no main effect of Group (F(1, 1010) = 3.2, p = 0.07) and no Group
by Sex interaction (F(1, 1010) = 0.042, p = 0.84), but there was a significant effect of
Sex (F(1, 1010) = 19.4, p = 0.0001), males scoring higher than females. This means that
on the AQ the students do not differ from the general population sample, despite the
differences in IQ and educational level between the two groups. Combining Groups 2 and
3, males and females differed on all subscales except local details (t tests, all p < 0.0001).

insert Table 1 and Figures 1-2 here

Scientists vs non-scientists

Table 2 shows the AQ scores for subjects in Group 3, broken down according to their
Degree/area of study. We compared students studying Science (i.e., physical sciences2,
biological sciences3, mathematics, computer science, engineering, medicine4, and non-
specific science5), vs Humanities (i.e., classics, languages, law, architecture, philosophy,
English, theology, history, or music), vs the Social Sciences (i.e., geography, economics,

2 Physical sciences included physics, physical natural sciences, chemistry, geology, communications,
chemical engineering, mineral science, material science, and geophysics.
3 Biological sciences included experimental psychology, neurophysiology, biological natural sciences,
biology, bioanthroplogy, neuroscience, and molecular ecology.
4 Medicine included both medicine and veterinary science.
5 This last category included those subjects who simply listed their Degree as natural sciences, which could
have been any of the sciences.