Neurotypical Peers are Less Willing to Interact with Those with Autism based on Thin Slice Judgments

Individuals with autism spectrum disorder (ASD), including those who otherwise require less support, face severe difficulties in everyday social interactions. Research in this area has primarily focused on identifying the cognitive and neurological differences that contribute to these social impairments, but social interaction by definition involves more than one person and social difficulties may arise not just from people with ASD themselves, but also from the perceptions, judgments, and social decisions made by those around them. Here, across three studies, we find that first impressions of individuals with ASD made from thin slices of real-world social behavior by typically-developing observers are not only far less favorable across a range of trait judgments compared to controls, but also are associated with reduced intentions to pursue social interaction. These patterns are remarkably robust, occur within seconds, do not change with increased exposure, and persist across both child and adult age groups. However, these biases disappear when impressions are based on conversational content lacking audio-visual cues, suggesting that style, not substance, drives negative impressions of ASD. Collectively, these findings advocate for a broader perspective of social difficulties in ASD that considers both the individual’s impairments and the biases of potential social partners.


Study 1
Method. Participants. Forty adults (20 ASD; 20 TD) served as stimulus participants. The stimulus groups were matched on gender (17 males in each group), and were comparable on both age (ASD mean = 24.5 years, TD mean = 25.0; p = 0.786) and full-scale IQ on the WASI (ASD mean = 106.4, TD mean = 110.5; p = 0.293). ASD diagnoses were confirmed by a certified clinician using the Autism Diagnostic Observation Schedule 33 . TD participants reported no ASD diagnosis in themselves or their first-degree family members. 214 undergraduates (164 female; mean age = 21.4) served as rating participants and completed informed consent. Power analysis using GPower 3.1 with an alpha level of 0.05, a power level of 0.8, and an effect of 0.3, which is comparable to related studies with a similar design 8 , determined a sample of 160 participants would be sufficient, but we oversampled relative to this number because previous studies included fewer stimulus participants and fewer presentation modalities. The Institutional Review Board at the University of Texas at Dallas approved this study and all methods were performed in accordance with the relevant guidelines and regulations.
Stimuli and Materials. Stimuli were created from video recordings of stimulus participants engaging in the "High Risk Social Challenge" task 34 , a 60 s mock audition for a reality/game show. Each video contains bodies and faces of stimulus participants and the first 10 s of social behavior occurring after any content-free introductions (e.g., "Hi TV. This is my audition tape") was edited into five presentation modalities: (1) audio-only (2) visual-only (3) audio-visual (4) static image and (5) transcript of speech content. Separation of the audio and visual components allows for examining the degree to which these social information sources independently contribute to first impression formation. The static image, taken from the first moment in the video in which the participant is sitting upright with eyes open and not speaking or gesturing, was used to determine whether first impressions of ASD and TD adults differ even in the absence of movement, speech, or conversational content. Finally, the transcript modality aimed to determine whether conversational content, which can differ in ASD and affect social interaction quality 35 , influences impression formation of ASD and TD adults independent of vocal and visual information.
Procedure. Rating participants provided their judgments individually on a desktop monitor using Qualtrics survey software. To avoid any carry-over effects, rating participants experienced only one presentation modality (e.g., transcript, static frame) for each of the 40 stimulus participants. Each stimulus was rated one at a time on ten items using a four-point scale (0-3). Six items related to traits that are reliably perceived when forming first impressions: attractiveness, awkwardness, intelligence, likeability, trustworthiness, and dominance/submissiveness 14,35 . The remaining four items related to behavioral intentions towards the stimulus participant as potential social partners: willingness to live near the stimulus participant, likelihood of hanging out with the stimulus participant in their free time, level of comfort sitting next to the stimulus participant, and likelihood of starting a conversation with the stimulus participant 28,30 . Items appeared in a random order between rating participants to account for potential order effects, but were presented in the same order across stimuli for each individual rater. Here, as in Studies 2 and 3, rating participants were not informed of the clinical status of stimulus participants, nor was any reference to autism made anywhere in the task.

Study 2
Method. Participants. A total of 12 adults with ASD (2 female; 10 male) and 16 TD participants (7 female; 9 male) served as stimulus participants. ASD diagnoses were confirmed by the ADOS-2 (Module 4) administered and scored by research reliable personnel. Groups were matched on age (ASD mean = 22.8 ± 3.5; TD mean = 22.8 ± 4.4; t(26) = 0.05, p = 0.96) and full-scale IQ (ASD mean = 117.8 ± 11.2; TD mean = 112.0 ± 14.9; t(26) = 1.12, p = 0.27). Thirty-seven undergraduate students (18 female, 19 male; mean age = 19.4 ± 1.3) served as raters and provided informed consent. They also received class credit for their participation. The Institutional Review Board of Indiana University approved this study and all methods were performed in accordance with the relevant guidelines and regulations.
Stimuli and Materials. Stimulus participants sat directly across from an experimenter while responding to a brief set of open-ended casual questions (e.g., "Have you seen any good movies recently?"). The experimenter was instructed to respond appropriately and attempt to engage the participant in a natural dialogue. All participants were familiar with the experimenter from having participated in previous studies in the laboratory. The experimenter recorded video using wearable point-of-view video recording glasses (Pivothead HD; recorded in 1080p), thus providing a direct first-person point of view of the social interaction. All but five participants also completed the UCLA Loneliness Scale.
We used MATLAB (v2014b; MathWorks, Natick, MA) to randomly select 10 still frames (henceforth, images) from each video to be used as stimuli. Images were only excluded if the subject was not near the middle of the frame or if the image was blurry (both due to head movement of the experimenter), or if the stimulus participant's eyes were closed. These images were then cropped to eliminate some of the background and show the person from approximately their torso and above.
Procedure. Raters were instructed that they would be viewing a variety of photographs of individuals taken during a typical social interaction, and their task was to make various judgments based on these images. Each rater completed 3 experimental blocks, with each block corresponding to a different rating question: (1) "How socially awkward is this person?" (2) "How approachable is this person?" (3) "Would I see myself being friends with person?" Stimuli within each block were randomized, and each stimulus participant was seen a total of 10 times per block. Male raters only rated male stimulus participants (190 stimuli per block), and female raters only rated female stimulus participants (90 stimuli per block).
Ratings were made using a non-graduated slider, with ends labeled and representing the extremes of the judgment (e.g., "not socially awkward" vs. "very socially awkward"). These extremes were quantified as '0' and '1' , with intermediate values falling in between this range corresponding to the location of the sliding bar. Participant responses were stored with a mouse click, and the next stimulus was shown after a 1 second inter-trial interval. There were no time constraints for responding. The experiment was presented using MATLAB and Psychtoolbox (PTB3) 37 .

Results
To examine initial first-impression judgments, we first limited our analysis to the rating on the first image from each stimulus participant. [Note that scores for the awkwardness were reverse scored so that higher scores reflect positive judgments for all questions.] Across all three We next asked whether ratings of an individual by others were stable across repeated exposure, or whether they changed over time. Each rater saw each stimulus participant 10 times over the course of each block. For each rating category and each rater, we fit a line across their 10 ratings of each stimulus participant. Then, we averaged the slopes of the lines across raters for each stimulus participant, and carried out a 3 (rating category) x 2 (group) repeated measures ANOVA as above, but with slope as the dependent variable. There was no effect of rating cate- Furthermore, for all ASD and TD stimulus participants, slopes were not different from zero (6 one-sample t-tests), demonstrating that ratings were reliable over repeated exposure [all ps > 0.11] (see Fig. 4). Finally, individuals with ASD reported greater levels of loneliness than TD control participants [ASD mean = 54.5 (11.5); Control mean = 40.2 (8.6); t(21) = 3.37, p = 0.0029; d = 1.41].

Study 3
Method. Participants. 14 boys, seven with ASD and seven TD controls, served as stimulus participants. The groups did not differ significantly on age (ASD mean = 12:1 [134.14 months], TD mean = 12:4 [147.43 months]; p = > 0.250), IQ on the Leiter 38 (ASD mean = 105.14, TD mean = 114.57; p = 0.200), or receptive vocabulary on the PPVT 39 (ASD mean = 119.14, TD mean = 131.71; p = 0.189). ASD diagnosis was confirmed by trained clinicians using the ADOS 32 , and all TD participants scored below the ASD threshold on the CARS 40 . 125 adults and 33 adolescents served as rating participants. Data from adult raters were collected via the online crowdsourcing platform Amazon Mechanical Turk (MTurk). Because facial expressions and prosody are highly dependent on cultural-linguistic context, we excluded participants who did not grow up and live in the US or spoke American English as their primary language, or who did not complete the task. The final sample of adult raters (n = 98) had an average age of 31:4 (range = 19-64), and a gender distribution of 53 male, 44 female, and one not specified. 33 TD adolescents (23 male, 10 female) with a mean age of 13:1 (range = 10-16:11) provided ratings in the laboratory. Data from this sample was analyzed separately to examine first impressions from raters more similar in age to the stimulus participants. The Institutional Review Board of Emerson College approved this study and all methods were performed in accordance with the relevant guidelines and regulations.
Stimuli and Materials. Stimulus participants were recorded during a task in which they were asked to retell brief stories containing happiness; surprise, fear, and anger (see ref. 9). Brief video clips (2-4 seconds) were extracted representing all four emotions and containing one complete sentence or phrase each. Video clips were selected based on video/audio quality, grammatical accuracy of the sentence productions, and were matched by expressed emotion across the two diagnostic groups. A maximum of two videos per stimulus participant were used with some children appearing in only one video due to availability of stimuli with acceptable video and audio quality, resulting in a total of 24 videos with 12 videos from each of the two subgroups. We created two pseudorandomized sequences, which were counterbalanced across participants.
Procedure. Adult rater participants completed the tasks remotely via M-Turk and electronically provided informed consent. All task directions were displayed before each video presentation. For each trial, participants viewed one of the 24 video clips and answered five questions using a non-graduated slider bar with "not likely" and "very likely" as the two anchor points. We chose a non-graduated slider to allow maximum freedom of judgment for participants, rather than forcing more categorical choices on a numeric scale. Participants were asked to rate the children's overall awkwardness, as well as their likelihood to start a conversation with others, have a lot of friends, get along well with others, and spend a lot of time by themselves. Average slider response was calculated for each prompt/question within each diagnostic group, using a range of 0-100 with 50 being the neutral or mid-point on the slider bar. Adolescent rater participants completed a similar task via a computer within the laboratory, but only rated three items provided by the adult raters: awkwardness, getting along well with others, and spending a lot of time by themselves.

Results
Scores for two items ("awkwardness" and "spending time by themselves") were reversed scored in all analyses so that higher ratings reflect more positive judgments for all items. Data from adult raters were analyzed first. We conducted a 2 (stimulus producer group: ASD vs TD) by 5 (question) repeated measures ANOVA, which revealed a main effect for stimulus participant group (F(1, 97) = 104.99, p < 0.001, η 2 = 0.52) and a main effect for question (F(4, 388) = 46.32, p < 0.001, η 2 = 0.32, as well as a significant interaction between question and stimulus participant group F(4, 388) = 4.98, p = 0.001, η 2 = 0.05). Ratings can be found in Fig. 5.
We conducted follow-up pairwise t-tests to compare item responses between groups, using Bonferroni adjusted alpha levels of.01 per test to adjust for multiple comparisons. Results show that children with ASD were rated as significantly less likely to start a conversation with others (t(97) = − 7.59, p < 0.001, 95% CI [− 13.47,  Furthermore, note that ratings of stimulus participants are highly consistent across these three judgments tasks (i.e., individuals who scored high on awkwardness scored low on approachability and friendship, and vice versa). For example, the one ASD participant with very typical ratings for awkwardness is the same participant that is rated high on the approachability and friendship ratings. To examine whether patterns held when raters were similar in age to the stimulus participants, a 2 (stimulus participant group: ASD vs TD) by 3 (question) repeated measures ANOVA was conducted using a sample of adolescent raters. We found a main effect for stimulus participant group (F(1, 32) = 15.97, p < 0.001, η 2 = 0.333), but no main effect of question (F(2, 64) = 0.12, p = 0.887, η 2 = 0.004), and no question by stimulus participant group interaction (F(2, 64) = 1.04, p < 0.36, η 2 = 0.032). Follow-up pairwise t-tests using Bonferroni adjusted alpha levels of.0167 per test compared responses to each question between stimulus participant groups and show that children with ASD were rated by adolescents as significantly less likely to get along well with others (t (

Discussion
Across three independent studies using distinct samples and a variety of methodological approaches, observers' first impressions of individuals with ASD engaging in real-world social behavior were found to be robustly less favorable than those of matched TD controls. These negative first impressions were consistent for both adults and children with ASD, for static as well as dynamic stimuli, for both brief (2-4 s) and longer (10 s) glimpses of social behavior, and did not change with repeated exposure. Further, because these impressions were associated with reduced intentions to socially engage by observers, they may reflect a previously under-recognized contributor to the reduced quantity and quality of social interaction experienced by individuals with ASD.
Our findings show that negative first impressions of adults with ASD occurred only when audio and/or visual information was present, and not when the transcript of their speech content was evaluated (Study 1). This discrepancy suggests that social presentation style rather than the substantive content of social speech drove negative impression formation of individuals with ASD. Supporting this conclusion, a static image was sufficient for generating negative first impressions of those with ASD and including additional information, such as body movement or voice, did not worsen them further. In contrast, first impressions of TD controls improved with the addition of a visual information, suggesting that unlike the ASD group, visual cues helped rather than hurt the impressions they made on observers. We also determined that negative impressions extended beyond perceived social competence to judgments of likeability (reduced), attractiveness (reduced), and submissiveness (increased) (Study 1). However, negative impressions did not occur for all evaluated traits, with the two groups not differing on ratings of perceived intelligence or trustworthiness. The lack of group differences on these traits suggest that the social presentation differences in the ASD group may lead to more negative evaluations of traits associated with social appeal and approach behaviors (e.g., awkwardness, attractiveness, likability), than those associated with competence (intelligence) and character (trustworthiness).
We also demonstrate that negative impressions remain stable across multiple thin-slice judgments. A series of randomly selected static images of college-aged individuals with ASD collected from a first-person perspective during social interaction were consistently rated as less approachable and more awkward than matched controls, with observers indicating a lower likelihood of being friends with members of the ASD group (Study 2). These ratings were stable across ten exposures of different randomly-selected images of each individual, suggesting that the negative impressions of the ASD group were not driven by poorer or non-representative image capture or by sequence effects, but rather by reliable differences in social presentation in ASD leading to consistent negative evaluation. This finding of negative evaluations of individuals with ASD is also emphasized by the consistency of results across all three studies, which include static and dynamic stimuli, as well as representations of different age groups.
Adult and school-aged observers also report negative perceptions of adolescents with ASD (Study 3), indicating that negative first impressions persist even when observers are similar in age to those being evaluated. The extension to school-aged children is important given that individuals with ASD generally receive the greatest amount of social skills intervention during this developmental period 41 , which decreases precipitously after high school 42 . Thus, despite being at an age when intense attention is often paid to social skills training, the ASD children in Study 3 are nonetheless rated poorly by both adults and same-age observers. Negative first impressions by adults may affect how children with ASD are perceived and treated by educators, and similar impressions by same-age observers may limit the formation of social networks and friendships. The establishment of a peer social network during early adolescence facilitates identity formation, access to resources, stress and coping support, and general well-being 43 , and the findings here may reflect another obstacle impeding the establishment of such social connections for youth with ASD above and beyond their already existing difficulties with social skills and social understanding.
Findings across the three independent studies were remarkably consistent despite using distinct samples and methods. Taken together, they offer strong evidence that the social presentations of individuals with ASD, particularly their non-verbal cues, including prosody, facial expressions, and body posture, are perceived less favorably and are associated with reluctance on the part of observers to pursue social engagement. This is particularly important given that individuals with ASD self-reported much greater feelings of loneliness than controls (Study 2). This is consistent with prior research indicating that individuals with ASD do not differ from their TD peers in their desire for relationships 4 , and suggests an unmet need for social experiences in ASD. Negative first impressions may serve as a barrier to fulfilling this desire for social interaction, as approach and withdrawal behavior towards novel social partners is based on subjective perceptions regardless of their accuracy 44 . In turn, this may limit opportunities in ASD for developing social connections and friendships, as well as the intergroup contact necessary for mitigating negative biases when present 45 . We explored the effect of repeated exposure to images of the same individual and found no changes in perceptual ratings (Study 2). However, repeated exposure to images is not equivalent to increased familiarity, and thus the current study cannot assess whether increased personal contact with individuals with ASD reduces negative impressions over time. Future studies are encouraged to Scientific RepoRts | 7:40700 | DOI: 10.1038/srep40700 explore this possibility, as such evidence would be consistent with a large literature in social psychology indicating that increasing intergroup contact reduces bias and prejudice 45 . In order to determine the effect of increased familiarity on these types of ratings, future studies would need to pursue a longitudinal approach, collecting interpersonal ratings at several time points over the course of weeks or months.
It is also important to explore specific components of visual and/or auditory presentation that may lead to negative impressions of young adults with ASD (e.g., body posture, prosody, grooming, and fashion). Several of the stimuli used across the three studies were videos and there is some evidence in the literature that the movement patterns of individuals with ASD are atypical and may represent a salient cue of awkwardness or difference to TD observers 46,47 . Kinematic analyses of facial expressions in this population point to subtle dynamic differences related to the complexity of dynamic transitions 45 as well as symmetry of movement patterns between face regions 48 , which could represent at least some of the cues potential conversation partners use to form their first impressions of individuals with ASD. However, negative judgments of people with ASD in our studies were not limited to video stimuli, but remained equally robust for static images, indicating that the rougher movement patterns of individuals with ASD 46,48 are not solely to blame for this phenomenon. It remains an open question whether videos would lead to more negative first impressions than static images within more socially demanding contexts such as the dynamic interaction in Study 2 that required spontaneous responses to questioning, rather than the one-sided social presentations elicited in studies 1 and 3.
Recent studies also suggest the presence of autism-specific dysmorphology, specifically related to distances between facial features 48 that could lead to perception of atypicality even when looking only at static images. However, significant dysmorphic features most commonly characterize individuals with more significant autism symptoms and severity 48 . This does not describe the individuals with preserved language and cognitive skills who appeared in the stimuli for the studies presented here and therefore cannot explain the rapid and robust negative evaluations reported here. Additionally, individuals with ASD are rated less favorably than their TD peers even in audio-only conditions (Study 1), indicating a significant contribution of prosodic features to the negative first impressions we report 14 . Based on evidence in the literature and the data presented here, we propose that negative first impressions of ASD are not founded on any one feature of expression, but rather represent an effect of subtle physical, dynamic, and auditory cues of presentation that can also include additional features, such as clothing choices, grooming habits, gaze patterns, or body posture.
The consistency of findings across these studies using different methodologies and stimulus types indicates that the foundation of these negative perceptions is complex and difficult to isolate. If our goal is to improve social interactions for individuals with ASD, it may therefore be equally important to educate others to be more aware and accepting of social presentation differences, rather than trying to change the many interwoven factors of self-presentation that mark the expressions of individuals with ASD as atypical. Given the social cognitive difficulties in perspective taking associated with autism 6 , some individuals with ASD may lack insight about how their social presentation is viewed by potential social partners. Others, however, may be more cognizant of these perceptions but are comfortable in their self-expression. For them, intervention strategies targeting awareness and acceptance among TD peers in their social environments may be a more sensitive and accommodating approach than encouraging impression management strategies.
The studies reported here should be viewed in the context of several limitations. Although the real-world social presentations used in each study provided a more naturalistic portrayal of actual social behavior in ASD compared to previous work using actors or vignettes, they may not fully reflect how impression formation occurs during live social interaction. We only included explicit judgments made of those with and without ASD. In addition to what participants indicate explicitly, their biases may also play out in more implicit ways. Future studies may seek to address whether implicit biases towards those with ASD parallel the more explicit findings reported here. Further, all participants with ASD in this study were intellectually-capable, and findings may not generalize to cognitively impaired populations. Additionally, while ratings did not differ for male and female stimulus participants in studies 1 and 2, sample sizes of females were small and future studies should examine whether patterns differ by gender. Finally, these studies present only group-wise comparisons and do not address individual differences among those with ASD, nor whether individual characteristics of the raters (e.g., gender, personality, etc.) affect the results reported here.
These limitations notwithstanding, the current project provides convergent evidence across three independent studies that first impressions of individuals with ASD are significantly less favorable than those of matched TD controls, and are associated with greater reluctance on the part of observers to pursue social engagement. The data demonstrate that negative impressions are formed across age groups and based on a range of features, including visual, auditory, dynamic, and static cues, indicating that these impressions are not context-or stimulus-dependent and are likely to persist across a range of real-world social situations. Taken together, these findings suggest that social interaction difficulties in ASD are not solely an individual impairment but also a relational one, and consideration of both of these factors is necessary for a full understanding of social impairment in ASD. The reluctance of TD individuals to engage in social interactions with their ASD peers further limits the opportunities for individuals with ASD to practice their already fragile social skills. This can have a significant negative impact on the ability of socially aware and socially interested individuals with ASD to improve their social communication abilities and work toward more successful social integration. Therefore, intervention and education approaches that target both those with ASD as well as their TD peers may offer a more comprehensive approach for improving social and functional outcomes in autism.