Negatively phrased items of the Autism Spectrum Quotient function differently for groups with and without autism

The Autism Spectrum Quotient is a widely used instrument for the detection of autistic traits. However, the validity of comparisons of Autism Spectrum Quotient scores between groups may be threatened by differential item functioning. Differential item functioning entails a bias in items, where participants with equal values of the latent trait give different answers because of their group membership. In this article, items of the Autism Spectrum Quotient were studied for differential item functioning between different groups within a single sample (N = 408). Three analyses were conducted. First, using a Rasch mixture model, two latent groups were detected that show differential item functioning. Second, using a Rasch regression tree model, four groups were found that show differential item functioning: men without autism, women without autism, people 50 years and younger with autism, and people older than 50 years with autism. Third, using traditional methods, differential item functioning was detected between groups with and without autism. Therefore, group comparisons with the Autism Spectrum Quotient are at risk of being affected by bias. Eight items emerged that consistently show differences in response tendencies between groups across analyses, and these items were generally negatively phrased. Two often-used short forms of the Autism Spectrum Quotient, the AQ-28 and AQ-10, may be more suitable for group comparisons.