How Reliable Is the Autism-Spectrum Quotient at Identifying Low and High Autistic Traits in College Students?

Background: The Autism-Spectrum Quotient, a measure of autistic traits, is commonly used to identify neurotypical adults (often college students) with low and high levels of autistic traits for comparison. However, researchers disagree on whether the Autism-Spectrum Quotient should be scored by converting items to a binary scale or maintaining the 4-point Likert scale. Furthermore, researchers use different methods to categorize adults as having low or high autistic traits which range from inclusive (e.g., median split) to restrictive (e.g., upper/lower deciles). Additionally, restrictive identification methods often require a time delay between administration of the Autism-Spectrum Quotient and the task(s) of interest. Research is needed to assess the reliability of the Likert scoring method and identification of low/high autistic traits in neurotypical adults.

Objectives: The present study systematically investigated the reliability of the Autism-Spectrum Quotient for identifying autistic traits in college students. In particular, this study aimed to elucidate whether reliability varies by scoring method (binary or Likert) or categorization method (median split, upper/lower tertiles, quartiles, or deciles).

Methods: Four hundred three college students completed a computerized version of the Autism-Spectrum Quotient while their eye movements were recorded. A subset of students (n = 178) completed a second version at least one week later (M = 15.17 days, SD= 5.53).

Results: Internal consistency of the total scores was acceptable (α ≥ .70) for both scoring methods. Internal consistency for subscales varied ranging from acceptable (social skills for Likert scoring) to poor (α ≤ .50; imagination for both methods). Internal consistency improved with Likert scoring for the total and all subscales scores except for attention switching (all others χ²(1) ≥ 6.74, p ≤ .009). Test-retest reliability was strong (r ≥ .80) for total scores for both methods, and all subscales for Likert scoring with the exception of imagination. Furthermore, test-retest reliability improved when using Likert scoring for total, social skills, attention switching, and attention to detail scores (zs ≥ -2.02, ps ≤ .04). Similar patterns of internal consistency and test-retest reliability were seen for men and women; however, men’s total scores decreased over time using binary scoring (t(54) = 2.44, p = .02), whereas women’s total scores increased over time using Likert scoring (t(120) = -3.45, p= .001). Categorizing individuals as low and high autistic traits was consistent across scoring method when using more inclusive methods (e.g., median split: 90.57%); however, agreement was reduced with more restrictive methods (e.g., deciles: 75.26%). Across time, the Likert method was equivalent or more stable than the binary method at categorizing individuals (e.g., median split: 85.80% versus 76.14%).

Conclusions: Overall, the Autism-Spectrum Quotient reliably assesses students’ autistic traits. The Likert method is superior in internal consistency and test-retest reliability. Furthermore, the Likert method may have better consistency at categorizing adults as low or high in autistic traits. Therefore, it is recommended that researchers move to using Likert scoring. However, researchers should be cautious about using restrictive identification methods because categorization at extremes may change over time.