Journal of Autism and Developmental Disorders

, Volume 43, Issue 12, pp 2807–2815

Further Evidence on the Factorial Structure of the Autism Spectrum Quotient (AQ) for Adults With and Without a Clinical Diagnosis of Autism


  • Winnie Yu Pow Lau
    • School of PsychologyUniversity of Queensland
  • Adrian B. Kelly
    • Centre for Youth Substance Abuse ResearchMedical School of the University of Queensland
    • School of PsychologyUniversity of Queensland
Original Paper

DOI: 10.1007/s10803-013-1827-6

Cite this article as:
Lau, W.Y.P., Kelly, A.B. & Peterson, C.C. J Autism Dev Disord (2013) 43: 2807. doi:10.1007/s10803-013-1827-6


The Autism Spectrum Quotient (AQ) has been widely used for measuring autistic traits however its factor structure has been primarily determined from nonclinic populations. This study aimed to establish an internally coherent and reliable factor structure for the AQ using a sample of 455 Australian adults of whom 141 had autism spectrum disorder (ASD) diagnoses. Principal component analysis revealed a 39-item questionnaire with five-factors: Sociability, Social Cognition, Interest in Patterns, Narrow Focus and Resistance to Change. The revised AQ-39 had sound goodness-of-fit indices, good-to-excellent internal consistency and test–retest reliability, and scores for ASD and non-ASD participants were significantly different. The AQ-39 may be useful in screening and for guiding the focus of therapy.


Autism Spectrum QuotientFactor analysisAutism spectrum disordersMeasureAdult


Autism Spectrum Disorders (ASD) encompass a range of clinically-diagnosed disorders including autism, Asperger Syndrome (AS) and PDD-NOS (pervasive developmental disorders not otherwise specified). According to standard diagnostic systems like DSM-IV (APA 2000), cardinal markers of ASD are serious impairments in social understanding and communication. Rigid, repetitive behaviours and impaired imagination are also common, as are language delays in those with a specific diagnosis of autism (APA 2000). Autistic characteristics are often considered to be on a continuum of abilities (e.g. Baron-Cohen 2008b; Robinson et al. 2011) ranging from (a) impairment meeting the criteria for a clinically significant ASD diagnosis through (b) normative, non-clinical personality types involving low social motivation, narrow interests and poor communication to (c) normative personalities with exceptionally few ASD features and exceptionally high empathy and social aptitude (e.g. Ritvo et al. 2011). The notion that ASD characteristics are continuously represented through nonclinical and clinical populations highlights the need for a psychometric tool that can capture continuous profiles and potentially be sensitive to clinically significant ASD-related symptomatology.

To date, the most widely-used general-population-oriented ASD self-report measure is the Autism-Spectrum Quotient (AQ: Baron-Cohen et al. 2001). This 50-item questionnaire has shown great potential, but further work on the factor structure of this questionnaire is needed. In the original development of the AQ, factor analysis was not used. Instead, items were devised and grouped into five subscales based on the ‘triad’ of autistic symptoms (e.g. Wing and Gould 1979) and empirical research on other areas of cognitive non-normality in people with autism. The five domains on the original AQ were social skills, attention switching, attention to detail, communication, and imagination. The relatively low internal consistency of items within each of the domains (less than .70 for four out of five scales) in the original study points to the need for further psychometric refinement.

In recent years, several factor-analytic studies have been conducted to verify AQ factor structure and reliabilities, and key findings for available studies are summarized in Table 1. Most of the listed studies involve nonclinical samples (primarily university students), and results show divergence in identified factor from a priori scales proposed by Baron-Cohen et al. (2001). In particular, most studies have identified between two and four factors, and only one of the studies identified five discrete factors (Kloosterman et al. 2011). The latter study reported mostly suboptimal alphas, suggesting relatively poor internal consistency of scales. The reviewed studies relatively consistently identify discrete factors involving social skills and attention to detail, but on other potential dimensions, there is relatively little consistency across studies. Other challenges for the AQ as originally formulated are (a) many of the original 50 AQ items failed to load on any of the factors identified in past factorial solutions (see Table 1), (b) the “goodness of fit” of structural models is in some cases limited or not reported (see Table 1 footnote), and (c) little evidence is available on the extent to which scales arising from factor analysis are useful in distinguishing people on the autism spectrum from nonclinical samples. An important priority for AQ factor analytic studies is the inclusion of substantial numbers of people on the autism spectrum. Only one of the reviewed studies contained people with any autism spectrum disorder (Hoekstra et al. 2008) and this study had a comparatively low representation of people in this group (n = 12: less than .01 % of the sample). Conceivably, factorial structures based on non-ASD individuals may fail to generalise adequately to the unique profile of clinically significant ASD.
Table 1

Summary of previous factor-analytic AQ studies



Model (total items; total variance explained)a

Factors (with Cronbach Alpha)

Goodness of Fit Indicesb,c

Austin (2005)

201 non-ASD university students (age M = 20.9 years)

U.K. sample


26 items;

3 factors;

28 %

Social skills (.85)

Attention to details/patterns (.70)

Communication/mindreading (.66)

Three-factor AQ total (.82)

Not reported

Hurst, Mitchell, Kimbrel, Kwapil and Nelson-Gray (2007)

1,005 non-ASD university students (age M = 19.36 years)

American sample


26 items;

3 factors;

29 %

Social skills (.75)

Attention to details/patterns (.55)

Communication/mindreading (.40)

Three-factor AQ total (.66)

Not reported

Hoekstra, Bartels, Cath and Boomsma (2008)

961 non-ASD university students

302 parents of non-ASD twin families

36 teen and adult psychiatric patients (age M = 24.65 years)

Dutch sample


50 items;

2-hierarchical factors;

Social interaction (.84/.77)

Attention to details (.63/.68)

χ2/df = 8.75

SRMR = .095

GFI = .681

ECVI = 2.19

Stewart & Austin (2009)

536 non-ASD university students (age M = 24.3 years)

Scottish sample


43 items;

4 factors;

29 %

Socialness (.83)

Patterns (.69)

Understanding others/communication (.71)

Imagination (.55)

SRMR = .072

RMSEA = .055

Kloosterman, Keefer, Kelley, Summerfeldt and Parker (2011)

522 non-ASD university students (age M = 21 years)

American sample


28 items;

5 factors;

45 %

Social Skills (.85)

Communication/mindreading (.65).

Restrictive/repetitive behaviour (.40).

Imagination (.57)

Attention to details (.59)

χ2/df = 5.42

RMSEA = .052

SRMR = .074

CFI = .827

ECVI = 4.59

aType of factor analysis conducted (summed variance explained by all factors in model); PCA principal component analysis, CFA confirmatory factor analysis, EFA exploratory factor analysis

bGoodness-of-fit indices include Chi square (χ2/df); SRMR standardised root-mean-square residual, RMSEA root-mean-square-error of approximation, GFI goodness-of-fit index, ECVI expected cross validation index, CFI comparative fit index

cA value of χ2/df between 1 and 3 indicates the model fits the data well (Gable and Wolf 1993); a value of SRMR lass than .08 is generally considered good fit (Hu and Bentler 1999); a RMSEA value ranging from .05 to .08 indicates fair fit while ≥.10 indicates poor fit (Browne and Cudeck 1993); a value of CFI or GFI above .95 indicates adequate fit (Browne and Cudeck 1993)

The key aim of this study was to arrive at an internally coherent and reliable factor structure for the AQ based on a sample representing a wider continuum of the autism spectrum than has previously been studied. The ultimate objective of the study was to develop a brief and coherent tool that could be easily administered and potentially assist in screening and therapeutic planning for people with ASD. In this study, we evaluated the factorial structure of the AQ using a large sample (n = 455) of adults of whom 141 (31 %) had confirmed ASD diagnoses. In addition, we aimed to assess how well the resulting factorial model would discriminate adults with a DSM-IV ASD diagnosis from non-ASD controls.



The participants were 455 Australian men and women aged 19–71 years (mean = 40.6 years, standard deviations 7.8 and 8.8 years for ASD and non-ASD participants respectively). They were recruited over a period of 18 months through a variety of sources including (a) personal contacts, (b) staff and student mailing and email lists at a large university, (c) students volunteering from a large first-year psychology class, (d) parents attending play groups in the local community, and (e) adults listed (as parents and/or clients) on the databases of two specialist clinics in Queensland with a predominantly ASD clientele. Among the 455 respondents, 141 had received a confirmed clinical diagnosis of ASD (i.e. HFA, AS or PDD-NOS). We required that this be conferred, using DSM-IV (APA 2000) criteria, by a clinical psychologist, psychologist, general practitioner or psychiatrist operating independently of this research.

To exclude any risk biasing results by inclusion of adults with low intellectual functioning, we required all participants to be high-school graduates (i.e. they had completed 12 years of compulsory schooling). An exclusion criterion was that participants with ASD had received co-diagnoses of schizophrenia, personality disorders or other neuropsychological conditions (e.g. traumatic brain injury, Alzheimer’s disease). Further demographic details are presented in Table 2. In summary, most participants were Caucasian (78 and 86 % for ASD and non-ASD participants respectively), most were married or in a de facto relationship (77 % for both groups), and about half of participants had a university degree (55 and 51 % for ASD and non-ASD participants respectively). There were no significant differences between ASD and non-ASD participants on these demographic variables.
Table 2

Sample demographics (N = 455)


Clinically-confirmed ASD (n = 141)

Non-ASD group (n = 314)

Significance of group difference (p value)a

Mean age (years)

M = 40.56 (SD = 7.80)

M = 40.74 (SD = 8.84)

ns (.10)




43 (30.5 %)

71 (22.6 %)

ns (.86)


98 (69.5 %)

243 (77.4 %)





110 (78.0 %)

270 (86.0 %)

ns (.13)


31 (22.0 %)

44 (14.0 %)


Marital status



12 (8.5 %)

17 (13.9 %)

ns (.96)

 Married/de facto

108 (76.6 %)

95 (76.9 %)



21 (14.9 %)

11 (9.3 %)


Education attainmentb


 High school

64 (45.4 %)

154 (49.0 %)

ns (.25)

 University degree

77 (54.6 %)

160 (51.0 %)


aANOVA for age; χ2 for remainder of variables

bHighest educational level completed

Measures and Scoring

We used Baron-Cohen et al. (2001) original AQ, a self-report questionnaire consisting of 50 items describing behaviors, habits and preferences relevant to a clinical presentation of ASD. Half the items are formulated so that agreement indicates an autistic feature, the others are the reverse. On a priori grounds, Baron-Cohen et al. (2001) proposed five subscales (Communication, Social Skills, Attention Switching, Imagination, and Attention to Details), each with 10 items. Examples include: “Other people frequently tell me what I have said is impolite, even though I think it is polite” (Communication), “I am good at social chit-chat” (Social Skills), “I don’t particularly enjoy reading fiction” (Imagination), and “I often notice small sounds when others do not” (Attention to Details). Respondents rate their levels of agreement with each statement on a four-point Likert scale (“definitely agree” to “definitely disagree”). Scoring is such that higher scores depict closer proximity to prototypical ASD. Baron-Cohen et al. (2001) dichotomously scored responses “agree” (slightly or definitely) versus “disagree” (slightly or definitely) to yield a possible score range of 0–50. They proposed, based on their validation sample, that an AQ total score of 32 or higher indicates “clinically significant levels of autistic traits” (p. 12). The average score for their U. K. non-ASD community control group was 16.4. This was significantly lower than the diagnosed ASD group’s mean of 35.8. Following the precedent of past factor- analytic studies (e.g. Austin 2005; Hoekstra et al. 2008), we implemented a continuous scoring scheme (4 = definitely agree, 3 = slightly agree, 2 = slightly disagree to 1 = definitely disagree) to maximise sensitivity and likelihood of obtaining the continuous distribution as recommended for factor analysis (e.g. Gorsuch 1983; Swygert et al. 2001).

Overview of Statistical Analyses

The main factor-analytic analyses were conducted in two steps. Using LISREL 8.54, we used Principal Component Analysis (PCA) to establish the factor structure for a randomly selected sample of 300 participants, which permitted an adequate sample based on participant/item ratios. Oblique rotation was used to allow for the probability of significant correlations between factors, and the Scree plot was used to confirm an appropriate number of factors. Goodness-of-fit indices were based on Pearson’s Chi square (χ2) values and Root-mean-square-error of approximations (RMSEA). The χ2/df index is appropriate for models with relatively small sample size with normally distributed variables (MacCallum et al. 1996). As the χ2 is often sensitive to trivial deviations in model fit for samples over 100, the ‘normed’ χ2 [namely χ2/df (degree of freedom)] was also examined. A value of χ2/df between 1 and 3 indicates the model fits the data well (Gable and Wolf 1993). The RMSEA index measures the error of approximation in the population and determines whether the model, with unknown but optimally chosen parameter values, fits the population correlation matrix or covariance matrix (Byrne 1998). We adhered to Browne and Cudeck’s (1993) suggestions that a RMSEA value ranging from .05 to .08 is an adequate to fair fit, while values greater than .10 indicate poor fit. These fit indices were selected based on their frequent use and demonstrated robust performance under various data and model misspecification conditions (Bollen and Long 1993). The RMSEA was also chosen because it is the most widely used fit index in extant factor analytic studies of the AQ.

Cronbach’s alpha test of internal consistency was also computed for each derived factor. As in previous research, we interpreted α value of .70 or greater as indicating acceptable internal consistency (Kline 2005). Finally, Pearson correlations among the five subscales were computed to explore latent factors’ associations with one another. To confirm test–retest reliability, a random subsample of 67 adults (19 from the ASD group) completed the AQ for a second time 4 weeks after the first administration. Intra-class correlations (ICC) and paired t test (t) were calculated for the test–retest reliability checks. The ICC was chosen as it evaluates both correlations of the subscales with total scores and shifts in overall means across the test–retest interval (Koch 1982).


For the principal component factor analysis, the Kaiser–Meyer–Olkin test for sampling adequacy produced a value of .948, indicating high adequacy (.90 or greater is deemed “extremely good”). Bartlett’s Test of Sphericity indicated that the rotated solution was significant at p < .0001. In the rotated solution, nine factors emerged as having eigenvalues greater than 1.00. However, since the weakest four factors together explained only a further 9.01 % of the total variance, and respective eigenvalues were just over 1, we elected to discard these in favour of an initial five-factor solution. We used a combination of two tests to confirm the number of factors (Turner 1998). The first was the Scree test (Cattell 1966), which showed discontinuity between five factors and subsequent factors (Tabachnick and Fidell 2007). The second was Parallel Analysis (Horn 1965), which revealed four factors (factor 1—eigenvalue 19.17, random data eigenvalue 2.05; factor 2—eigenvalue 2.55, random data eigenvalue 1.93; factor 3—eigenvalue 2.40, random data eigenvalue 1.84; factor 4—eigenvalue 1.90, random data eigenvalue 1.77). Given that the scree test and parallel analysis may result in overfactoring and underfactoring respectively (Hayton et al. 2004), we retained a five-factor solution with the proviso that the fifth factor may or may not be replicable in subsequent research. Items that had a factor loading of less than .32 and/or strong cross-loadings were discarded. Only factors with three or more strongly loading items were accepted (Costello and Osborne 2005).

The five-factor model (see Table 3) explained 48.1 % of the total variance, and consisted of 41 items (See Table 3). To maximize the clinical utility of the scale, two of the 41 items were discarded because they had both very low loadings on their respective scale (marginally over the a priori cut-off of .32), had cross-loadings on other factors that were comparable in magnitude), and had clear conceptual incongruity with the other items on the respective scale. These items were (a) “I am often the last to understand the point of a joke” and (b) “I tend to have very strong interests which I get upset about if I can’t pursue” (see Table 3). The first item had its highest loading for Resistance to change (.35), but also loaded on Sociability (.33) and a low order factor that was discarded because of insufficient eigenvalues. This item was conceptually unrelated to the Resistance to Change factor, and the internal consistency of this factor was not meaningfully different when this item was excluded (α with this item = .810, α without this item .82). The second item loaded on Interest in Patterns but was only marginally over the cut-off of .32, and it loaded almost as well on other Social Cognition, the dropping of this item did not affect the internal consistency of the subscale in a meaningful way, and the item appeared conceptually unrelated to the Interest in Patterns subscale. The final number of items for the scale was 39.
Table 3

Factor loadings for principle component analysis (n = 300) and confirmatory factor analysis (n = 155) of the new model (AQ-39)




Item number and wording from original AQ

Fac 1

Fac 2

Fac 3

Fac 4

Fac 5

Fac 1

Sociability (Cronbach α = .906)



I find social situations easy








I enjoy social occasions








I enjoy meeting new people








I am good a social chit-chat








I enjoy social chit-chat








I would rather go to a library than a party








I frequently find that I don’t know how to keep a conversation going








New situations make me anxious.








I find it hard to make new friends








I am a good diplomat








I prefer to do things with others rather than on my own








I find myself drawn more strongly to people than to things








I find it very easy to play games with children that involve pretending







Fac 2

Social Cognition (Cronbach α = .867)



I find it difficult to imagine what it would be like to be someone else








I find it easy to ‘read between the lines’ when someone is talking to me








When I’m reading a story, I can easily imagine what the characters might look like








I find it easy to work out what someone is thinking or feeling just by looking at their face








In a social group, I can easily keep track of several different people’s conversations








I usually concentrate more on the whole picture, rather than the small details








I know how to tell if someone listening to me is getting bored








I find it easy to do more than one thing at a time








If there is an interruption, I can switch back to what I was doing very quickly








I find it difficult to work out people’s intentions








When I am reading a story, I find it difficult to work out the characters’ intentions







Fac 3

Narrow Focus (Cronbach α = .773)



I tend to notice details that others do not








I tend to have very strong interests which I get upset about if I can’t pursue








I often notice small sounds when others do not








I frequently get so strongly absorbed in one thing that I lose sight of other things








I notice patterns in things all the time








Other people frequently tell me that what I’ve said is impolite, even though I think it is polite








People often tell me that I keep going on and on about the same thing







Fac 4

Interest in Patterns (Cronbach α = .705)



I am fascinated by dates








I am fascinated by numbers








I like to collect information about categories of things (e.g. types of cars, birds, trains, plants, etc.)








I usually notice car number plates or similar strings of information








I tend to have very strong interests which I get upset about if I can’t pursue






Fac 5

Resistance to Change (Cronbach α = .733)



I like to plan any activities I participate in carefully








I enjoy doing things spontaneously








I prefer to do things the same way over and over again








It does not upset me if my daily routine is disturbed








I am often the last to understand the point of a joke






aReverse-scored items

bItem discarded on the basis of conceptual incongruity

Factor 1 consisted of 13 items depicting one’s interest in socializing with others as well as self-perceived social competency. This factor had an eigenvalue of 19.17, and accounted for 19.54 % of total variance. This factor was interpreted as Sociability. Factor 2 had an eigenvalue of 2.55, consisted of 11 items which explained 8.84 % of total variance. Focal items in this domain were: “I find it difficult to work out people’s intentions.” “I find it easy to ‘read between the lines’ when someone is talking to me.” This factor was interpreted as Social Cognition. Factor 3 consisted of 7 items, including “I frequently get so strongly absorbed in one thing that I lose sight of other things” and “I often notice small sounds when others do not.” These items were interpreted as having a Narrow Focus. This factor had an eigenvalue of 2.40 and accounted for 7.77 % of total variance. Factor 4 consisted of 4 items that described Interest in Patterns, including “I am fascinated by dates”, “I like to collect information about categories of things”, and “I usually notice car number plates or similar strings of information”. The eigenvalue was 1.89, and explained 7.02 % of total variance. Factor 5 consisted of 4 items that together accounted for 4.90 % of total variance and had an eigenvalue of 1.50. Items within Factor 5 included “I prefer to do things the same way over and over again” and “It does not upset me if my daily routine is disturbed” (reverse scored), and this factor was interpreted as Resistance to Change. The internal consistencies for each scale based on the five factors all exceeded α = .70 (see Table 3), indicating sound to excellent internal consistency (Kline 2005). Furthermore, examination of skewness and kurtosis statistics indicated that there were no significant departures from normality for any of the subscales or the total score for the 39 items. We ran a confirmatory factor analysis on the 5-factor solution separately for the ASD and non-ASD groups to test relative model fit. The fit of the model for the ASD group was good (χ2 = 305.61, χ2/df = .44, RMSEA = .001, CFI = 1.00, AIC = 538.65). The fit of the model for the non-ASD group was poorer but still acceptable (χ2 = 1,861.28, χ2/df = 2.92, RMSEA = .07, CFI = .97, AIC = 1,940.15). The coefficient of congruence across ASD and non-ASD groups was high (.99), indicating high factorial similarity (Van de Vijver and Leung 1997).

To cross-validate the five-factor solution we used CFA drawing upon the remaining 155 participants whose data had not been included in the original PCA. The smaller sample for the CFA was justified on the basis that the number of items had been substantially reduced from the original 50 items. Basic model-fitting techniques were applied. First, our iterative process of variable selection was performed in steps. We used Wald’s t test to drop non-significant structural parameters. Then we used the Modification Index (MI) to allow for the addition of meaningful structural parameters. As described in the procedure section above, the χ2/df, RMSEA and CFI were then examined. The measurement model based on 39 items (five factors) had a χ2/df was less than 3 (2.78), the RMSEA was less than .08 (it equalled .070), the CFI exceeded .95 (.977). These statistics indicated that the new factorial model for the AQ-39 represented the data very well, and fit indices were generally strong relative to reviewed studies where fit indices were reported (see Table 1).

Latent factors were expected to display some associations, given their communalities in clinical and research-based depictions of ASD symptomatology. Correlations between subscales were, as expected, significant, with all correlations differing significantly from zero (p < .001). The magnitude of inter-scale correlations was generally in the high range (.6–.7), suggesting high relatedness (see Table 4). Our repeat testing of a subgroup of the sample showed high test–retest reliability, with ICC indices greater than .86 for all factors. We also explored the discriminant validity of the AQ-39 subscale and total scale scores by comparing the ASD-diagnosed subset of our sample (n = 141) with the non-ASD control group (n = 314). Table 5 reports the means and the results of analyses of variance (ANOVA) and Cohen’s effect size analyses. As expected, the ASD group scored significantly higher on each subscale score and the total score (effect sizes in the medium range), suggesting that the AQ-39 subscales may be useful in distinguishing the two groups.
Table 4

Pearson correlations for the 39-item AQ factor solution



Social cognition

Narrow focus

Interest in patterns




Social cognition




Narrow focus





Interest in patterns





Resistance to change





** p < .001

Table 5

Means (and SDs) on new AQ-39 factors for participants with and without formal diagnosis of ASD




Group comparisons

Male (n = 43)

Female (n = 98)

Combined (n = 141)

Male (n = 71)

Female (n = 243)

Combined (n = 314)

Comparisons F value

Cohen’s da

Sociability (13 items)

37.65 (8.71)

34.37 (9.72)

35.44 (9.50)

29.36 (10.45)

29.39 (10.68)

29.38 (10.61)



Social cognition (11 items)

27.30 (7.89)

25.18 (8.44)

25.86 (8.50)

23.44 (7.21)

21.32 (7.50)

21.83 (7.48)



Narrow focus (7 items)

20.81 (5.26)

18.25 (5.65)

19.07 (5.64)

16.74 (4.92)

17.09 (5.40)

17.01 (5.29)



Interest in patterns (4 items)

10.53 (3.04)

9.02 (3.50)

9.51 (3.42)

7.81 (3.27)

7.70 (3.09)

7.73 (3.13)



Resistance to change (4 items)

11.55 (2.50)

10.43 (2.92)

10.78 (2.83)

9.67 (3.00)

8.91 (3.34)

9.09 (3.27)



Total AQ-39

108.00 (22.46)

97.38 (27.03)

100.86 (26.02)

87.13 (25.46)

84.41 (26.17)

85.06 (25.99)



p < .001

aAn effect size of .5 is considered large, .3 is moderate, and .1 is small (Cohen 1988)


The primary aim of this study was to examine factor-analytic structure of the AQ using a large sample that included a substantial proportion of people with an ASD diagnosis. This is one of the first factor analytic studies to achieve this. We took the conservative approach of initially developing a factor model using PCA with a subgroup of our sample and then cross-validating it with the remainder using CFA. The outcome was a 39-item, five-factor structure that had favourable psychometric properties, including high internal consistency, semantic consistency, good distributional properties, and good test–retest reliability. There was evidence yielded that the scale produced significantly different scores across clinic and nonclinical samples in the expected directions. The internal consistencies of the subscales derived in the present study were substantially higher than those reported in previous studies, and the proportion of variance accounted for appeared higher than most other studies that were reviewed in this area of research. Goodness-of-fit indices were comparable and in some cases notably improved relative to solutions derived in other studies.

The five factors derived in the present study mapped well onto some of the factors derived in earlier studies. In the present study, the five derived factors were sociability, social cognition, narrow focus, interest in patterns, and resistance to change. These partly mapped onto domains proposed in the original Baron-Cohen et al. (2001)—social skills, attention switching, attention to detail, and communication. The factors from the present study also consistently mapped onto some of the domains identified in available factor analytic studies subsequent to Baron-Cohen et al. (2001)’s study. Notably, the factors sociability and social cognition related well to social skill and communication/mindreading factors derived in the five prior studies (Table 1). The identification of the factor social cognition is consistent with earlier research showing that children and adults with ASD have difficulties with ‘theory of mind’, or the ability to infer the beliefs, desires and intentions of others (e.g. Baron-Cohen 2000, 2008a; Brown and Klein 2011; Peterson et al. 2012). The factors attention to detail and interest in patterns mapped onto related factors derived by Austin (2005), Hurst et al. (2007), Hoekstra et al. (2008), Stewart and Austin (2009), and Kloosterman et al. (2011). The similarity of these factors across the reviewed studies suggests a high level of consistency in ASD symptomatology across industrialized countries, notably Britain (Baron-Cohen et al. 2001), the Netherlands (Hoekstra et al. 2008) and the United States (Kloosterman et al. 2011).

Future research is needed on the extent to which the AQ-39 may be useful as a tool for assisting in screening, diagnosis, and early intervention for people on the autism spectrum. The AQ-39 factors resembled widely-used diagnostic guidelines for ASD, including the Diagnostic Criteria for Asperger’s Syndrome (DCAS; Gillberg 1991; Gillberg and Gillberg 1989). In the DCAS taxonomy, the five diagnostic criteria are social interaction impairment, all-absorbing narrow interest, imposition of routines, atypical speech and language profile, and non-verbal communication problems. Four of the five DCAS taxonomy map onto the AQ-39 factors, and the variation across the two measures related to atypical speech and language profile, a feature for which there is continuing debate on whether this a distinguishing feature of autism spectrum presentations (Gillberg and Ehlers 1998). The AQ-39 is likely to be a valid and reliable tool for quantifying the key dimensions of ASD, and it is quick and easy to use, making it useful in a busy clinical setting. This may be helpful in guiding the focus of therapeutic intervention for people with ASD. For example, a profile of factors that denotes particular challenges or relative assets in sociability, social cognition, narrow focus, interest in patterns, and resistance to change may guide the strategic focus of therapy. Further research on the potential utility of the AQ-39 as an indicator of clinical change during therapy is needed. The results of this study suggest that the AQ-39 factors are stable over time, so they should be able to pick up clinically reliable change in response to therapy.

Relative to earlier factor analytic studies on the AQ, a key strength of the study is its large sample of people with ASD. Nevertheless, the study has limitations. The number of factors varied across tests (Scree versus Parallel Analysis), and further research is needed to verify the fifth factor resistance to change. We relied on formal diagnostic information supplied to participants by health and medical professionals, rather than our own diagnostic interviews, and we did not screen for clinically significant problems in the nonclinical group. Generalizability to other sample populations may be restricted because of a reliance on convenience sampling in the nonclinical group. Participants were all unpaid volunteers and, given the length and content of the questionnaires, a certain level of interest and dedication was needed to complete them. This may have biased the sample towards those with less significant problems. The clinical group was recruited from two private clinics specializing in ASD, and this may have attracted a different population from that which might present at a nonspecialist clinic or community health services. The sample was relatively affluent and well-educated, potentially limiting generalization to people from lower socioeconomic backgrounds.


The AQ-39 had strong psychometric properties, including discrete factors that mapped onto earlier research, high internal consistency, and test–retest reliability. Next steps for research on the refined AQ might include an assessment of its capacity to capture clinical change, and the utility of targeting presenting problems drawing on AQ-39 profiles.


The authors wish to thank Dr. Michelle Garnett and Professor Tony Attwood for helpful discussion and assistance recruiting the sample. We are grateful to all participants for their fine cooperation.

Copyright information

© Springer Science+Business Media New York 2013