Do Implicit and Explicit Racial Biases Influence Autism Identification and Stigma? An Implicit Association Test Study

We use identity-first language (e.g., autistic people) rather than person-first language (e.g., people with autism) because identity-first language is preferred by autistic people (Kapp et al.; 2013; Kenny et al., 2016) and may be less likely to contribute to stigma (Gernsbacher, 2017). As recommended by Dunn and Andrews (2015), the APA now recommends that researchers follow the preferences of minority groups by using the terms that they prefer to describe them (https://​www.​apa.​org/​pi/​disability/​resources/​choosing-words).

In order to be consistent with the terms used by the creators of the Child Affective Facial Expression database (CAFÉ; Lobue and Thrasher 2015), which we used to construct our racial bias IAT, we used the terms “European Americans” and “African Americans” in the Racial Bias IAT. Anonymous reviewers pointed out limitations with this choice of terms, specifically that participants had no reason to infer that a White child was necessarily of European descent or a Black child was necessarily African American. Given limitations in our choice of terms (which are discussed in more detail in the limitations section), the terms White and Black are used throughout the paper except when describing the construction of the racial bias IAT and the exact results obtained from it in the results section.

Reaction time data from Stage 1 of the Disability Valence IAT was used to explore the construct validity (i.e., internal consistency) of the target phrases. If all terms characterize the category equally well (i.e., have high construct validity) there should be similarity in the average latency to categorize different targets from the same category. Cronbach’s alpha was used to compare average latencies for target term for each participant.

For the improved D score algorithm, this process is completed again by subtracting participants’ average reaction time to sort target terms in Stage 7 by participants’ average reaction time to sort target terms in Stage 4.

There was also a difference in reaction times when participants sorted the ASD and CD phrases between the Disability Valence and Racial Bias IAT. Participants were significantly faster when sorting the ASD and CD phrases (M = 1.28 s; SD = 0.44) in Stage 1 of the Racial Bias IAT compared to Stage 2 of the Disability Valence IAT (M = 1.58 s; SD = 0.52), t(467) = 13.35, p < 0.001. Participants also made significantly fewer errors when sorting the ASD and CD phrases (M = 0.93 errors; SD = 1.31) in Stage 4 of the Racial Bias IAT compared to Stage 4 of the Disability Valence IAT (M = 1.13 errors; SD = 1.53), t(467) = 2.55, p = .011. These differences between the Disability Valence IAT and the Racial Bias IAT were likely the result of increased familiarity with the phrases since the Racial Bias IAT was always presented second.

We focus on spontaneous judgements in analyses as they were more sensitive than structured ratings of the likelihood of a given condition in prior research (Beeger et al. 2009). However, a similar pattern of findings was observed with spontaneous and structured ratings.

This relationship remained significant when the analysis was run with only participants who accurately identified both ASD and CD, Z = -16.79, p < 0.001.