Autism spectrum conditions (ASC) are diagnosed on the basis of behaviour, specifically difficulties in social and communication development, alongside repetitive behaviour and unusually narrow strong interests . The evidence for the genetic basis of ASC initially came from twin studies of classic autism [2, 3] and more recently twin studies of autistic traits [4–6]. Progress from these epidemiological findings to identifying specific DNA sequence variations that cause ASC has been slow: replication of results has been hampered by methodological issues such as limited power, varying designs and genotyping, along with imprecise phenotypic definitions . Another reason for limited progress is that although ASC has a high inheritance rate, it is genetically heterogeneous. Rare de novo mutations and chromosomal abnormalities could account for as many as 20% of ASC cases, but common allelic variation is also important, suggesting that a categorical approach to case ascertainment may not always be the best approach .
Indeed, a case-control or categorical approach to diagnosis ignores the view that autism is not just a spectrum within the clinical population, but that autistic traits are continuously distributed right through the general population [5, 9]. Taking seriously the dimensional view necessitates the use of techniques such as quantitative trait loci (QTL) analysis. Quantitative traits are characteristics that can be associated with a particular condition but are also continuously distributed in the non-affected population. If QTL is to be a successful tool in the search for autism susceptibility genes, then an instrument that can quantify autism traits in both affected and non-affected individuals is required. As well as quantifying autism traits within the whole population, such an instrument could also be used to define the broader autism phenotype (BAP).
The BAP is generally considered to be a subclinical set of characteristics or traits that index familiality and/or genetic liability to autism. This conception holds that the BAP is milder but qualitatively similar to the diagnosed autism phenotype. BAP characteristics were first observed by Kanner , and were also noted in Folstein and Rutter's early twin study , which found a higher concordance rate for a more broadly defined cognitive impairment.
Given the increasing importance of measuring the BAP, not only in genetic studies, but across autism research (for example, in neuroimaging and cognitive studies), it is important to select the right tool. There are already at least four instruments available for the assessment of the BAP: the Family History Interview (FHI), developed by Rutter and Folstein; the Social Responsiveness Scale (SRS) ; the Broader Phenotype Autism Symptom Scale (BPASS) ; and the Broad Autism Phenotype Questionnaire (BAPQ) .
The FHI measures autism-related traits in family members via interview with one member. The instrument covers social and communication skills and range of interests. An algorithm determines the presence or absence of the BAP in the three domains. The FHI was used by Bolton and colleagues  to demonstrate higher rates of the BAP in the parents and siblings of probands with autism compared with probands with Down's syndrome. This finding was replicated by Piven and colleagues .
The SRS is a 65-item questionnaire, which is completed by an adult informant. Its focus is on the ability of the subject of the questionnaire to engage in emotionally appropriate reciprocal social interaction and communication. Using the SRS, a higher number of autistic traits have been observed in the siblings of children with autism .
The BPASS was developed specifically for use in QTL studies . This instrument comprises seven semistructured interview items and six observation items, which together cover social motivation, social expressiveness, conversational skills and repetitive/restricted behaviours. The BPASS is for use with both adults and children.
The most recently published instrument, the BAPQ, comprises 36 items, 12 in each of three subscales: aloof personality, pragmatic language deficits, and rigid personality. Ideally, scores from self- and informant reports are combined to give a best-estimate rating. The BAPQ has been validated using direct clinical assessment of BAP .
We suggest that when choosing an instrument to measure autistic traits in both the affected and non-affected populations, the following qualities are important. (i) A good BAP measure should be quantitative, with a wide range of possible scores, avoiding both ceiling and floor effects. (ii) The content validity should be such that the instrument can distinguish people with from those without an ASC. (iii) Versions of the same instrument should be available for use with individuals of different ages, both adults and children. (iv) Ideally, the instrument should produce a normal or near-normal distribution in the general population. (v) The instrument should have good test-retest reliability. (vi) It should have good cross-cultural applicability. (vii) It should correlate with either biological and/or psychological measures of the broader phenotype (concurrent validity). Finally, (viii) it should be quick and easy to use: participants in research projects should be subjected to the least amount of testing possible, without compromising the validity of the research project.
Bearing in mind these qualities, each of the four instruments described above have some weaknesses: the FHI and BPASS are not truly quantitative and are both time-consuming to administer, the SRS is not normally distributed, and the BAPQ is currently limited to adults.
In this paper, we propose that an alternative to the four instruments already available is the Autism Spectrum Quotient (AQ) . The AQ was developed to assess where an individual lies on the autism spectrum (that is, how many autistic traits that person exhibits). Originally designed for adults, there are now child ( 4-11 years old)  and adolescent (12-16 years old)  versions available (both completed by parents to improve accuracy). The questionnaires are freely available to download http://www.autismresearchcentre.com.
The AQ for adults has the format of a self-report, forced-choice questionnaire and is at the reading level of a typical 10-year-oldl. It can be used by adults with an IQ in the average range, who can read and understand at least to this level. There are 50 items, covering behaviours across five domains: communication, social skills, attention switching, imagination, and attention to detail. Each item has the format of a statement with which the respondent rates how strongly they agree or disagree, using a four-point scale. In fact, each item is only scored as '1' if the person reports the autistic trait, and '0' if they do not, so total scores on the AQ can range from 0 to 50. Some researchers have used a four-point Likert scale for the AQ [19–21], but this does not significantly alter the pattern of results. The AQ is quick and easy to use and produces a near-normal distribution in the general population .
The adult AQ has been used extensively, and has been shown to have consistent results both across time  and culture [20, 23]. The AQ demonstrates high heritability . The AQ score is a good predictor of clinical diagnosis , correlates with brain function , single nucleotide polymorphisms (SNPs) in candidate genes , social attention [27, 28] and even prenatal testosterone levels .
A previous study, by Dorothy Bishop and colleagues, suggested that AQ scores differentiate parents of children with an ASC from control parents on two subdomains: communication and social skills . Although this was a pioneering study, its limitation was that it only tested 65 mothers and 46 fathers of children with a pervasive developmental disorder (PDD) diagnosis, and 48 mothers and 37 fathers of control children. Although this study indicates the potential of the AQ to serve as a measure of the BAP, it is not clear whether the lack of significance on three of the five subdomains of the AQ is because these play no role in the BAP or because the sample size was too small. Therefore, in the current study, we used the AQ with a much larger sample of parents.
Our first aim was thus to test if the results found by Bishop and colleagues could be replicated in a new, and considerably larger, sample. The second aim was to test the proportion of parents of children with an ASC who demonstrate the BAP, the medium autism phenotype (MAP) and the narrow autism phenotype (NAP). The MAP and NAP are new concepts. An individual with the NAP has a large number of autistic traits, and most (but not all) people with the NAP will also have a diagnosis on the autism spectrum. Not all individuals with the NAP will be diagnosed because there are many reasons why an individual may seek or receive a diagnosis: the NAP is one band within the phenotypic spectrum and does not take into consideration the individual's context, which may mediate whether they actually need or receive a diagnosis. Using the AQ, we define the NAP as those scoring ≥ 3 standard deviations (SDs) above the mean. An individual with the MAP has a medium number of autistic traits (defined as individuals scoring between 2 to 3 SDs above the mean on the AQ), and an individual with the BAP has even fewer traits, but still significantly more than average (defined as individuals scoring 1 to 2 SDs above the mean on the AQ). People with the MAP or BAP are unlikely to require clinical intervention, but their phenotypic status may be informative at a genetic level.