Re-Evaluating the Structure of the Autism Quotient: A Novel 3-Factor Solution

Background:  The Autism Quotient (AQ) is a widely employed 50-item self-report questionnaire that has been used to distinguish between Autism Spectrum (AS) individuals from those not diagnosed, along with assessing autistic behaviors in the general population (Baron-Cohen et al., 2001). The AQ was originally conceived as having five subscales: social skill, attention switching, attention to detail, communication and imagination. A number of subsequent studies have found better fit with 2-, 3-, 4- or 5-factor models compared with the original subscales; however, these models tend to have poor fit values. Thus, the factor structure of the AQ remains unresolved. Moreover, little research has examined whether individual factors load onto a single factor, which, theoretically, should be the case if autism reflects a single construct.

Objectives: This study evaluated whether a new factor structure for the AQ could be developed that would have acceptable measures of fit. We also assessed whether the factors loaded onto a single, higher-order factor.

Methods:  Participants were 1142 university students recruited from introductory Fall and Spring semester Psychology classes. Participants completed the AQ online as part of a battery of questionnaires. Our model was developed through a series of iterative steps, guided by the core features of autism spectrum disorders in the DSM-5 including deficits in social communication and social interaction along with the presence of restricted, repetitive patterns of behaviors, interests, and activities. A bottom-up construction rather than an elimination process was instituted such that we selected items that seemed most characteristic of each category until fit was no longer acceptable.

Results:  Confirmatory factor analyses were conducted to establish the factor structure guided by the DSM-5 categories. A 3-factor model emerged: verbal social communication (4 items, α = .85), nonverbal communicative behaviors (4 items, α = .66), and repetitive behavioral patterns (3 items, α = .56).  Moreover, these three factors loaded onto a higher order factor representing autistic behaviors (α =.83; see Figure 1).  The fit of the model was acceptable (χ² = 327.103, df = 92, RMSEA = .047, CFI = .928, TLI = .896).

Conclusions:  The 3-factor, 11-item model we developed provides an acceptable level of fit and an improvement on other published models. Moreover, this model demonstrates that these three factors load onto a single higher-order factor, consistent with the idea that these factors represent distinctive, core features of autism. This measure will be the first to have both strong statistical properties as well as provide an efficient screen for autistic behaviors through self-report. Although other brief AQ measures exist (Booth et al., 2013; Hoekstra et al., 2011), our measure has the strongest psychometric properties with a model derived from a theoretical framework.