Examining Potential Measurement Biases in the Autism Diagnostic Observation Schedule for Race, Ethnicity and Gender
The Autism Diagnostic Observation Schedule (ADOS) is widely used across cultural contexts to assess symptoms of autism spectrum disorders (ASD; Blacher et al, 2014). Little research has examined whether demographic characteristics such as race, ethnicity, and gender contribute to differential variability on ADOS score profiles, in spite of known cultural differences in social norms and variability in ASD symptom presentation across demographic groups (Bernier, Mao, & Yen, 2010). For example, cross-cultural variability exists regarding amount and type of social eye contact (e.g., Fugita, 1973; McCarthy, Lee, Itakura, & Mur, 2006) and facial expression (Vrana & Rollock, 2002), which may directly result in symptom variability in an assessment context depending on the diagnostic tests used. In potential support of this point, research shows variability exists with regard to ASD symptom presentation (Tek & Landa, 2012) and identified symptom profiles (Kharod, Giarelli, Blum, Hanlon, & Levy, 2012) across different demographic groups.
The objective of this study is to examine the ADOS to determine if item-level biases exist among distinct sociodemographic groups including race (Caucasian, African American, or Asian), ethnicity (Hispanic or non-Hispanic), and gender.
Participants included in this study (n = 2459) were part of the Simons Simplex Collection (SSC), had complete data and fell within one of the racial categories identified as sufficiently powered for analysis: White (n = 2245), Black/African American (n = 103), or Asian (n = 111). The majority participants classified their ethnicity as not Hispanic or Latino (n = 2165) rather than Hispanic or Latino (n = 294) and were male (n = 2129). We examined a subset of ten items from the ADOS that were included in at least three of the four ADOS modules and were worded the same in both the item description and the coding categories across all modules.
A measurement noninvariance analysis was used, and approached using a Multiple Causes analytic framework. The focus was limited to differences in item location. All analyses controlled for age and IQ. Holding level of overall ADOS score constant, we found significant item level bias for three items. We observed Black children we more likely have higher ratings on the ADOS items assessing levels of usual eye contact and stereotyped or idiosyncratic word use. Asian children were more likely to have elevated ratings on the unusual sensory interests ADOS item. In terms of ethnicity, Hispanic children were also more likely to have higher ratings on the ADOS item assessing levels of usual eye contact. No item level biases were observed for gender.
Conclusions: This study revealed that of the ten ADOS items examined, item level biases existed for race among three specific items. In a diagnostic assessment context, this variability within ADOS items may result in overestimation of impairment for specific racial groups. These findings speak to the need for more research assessing the need for specific norms for different racial and ethnic groups to aid in more accurate diagnosis.