Mimicry and social affiliation with virtual partner are decreased in autism

Abstract

Copying other people’s mannerisms (i.e., mimicry) occurs spontaneously during social interactions, and is thought to contribute to sharing emotions, affiliation with partners and interaction quality. While previous research shows decreased mimicry of emotional facial expressions in autism spectrum disorders (ASD), we know relatively little about how non-emotional, non- facial behavioural mimicry manifests and, more importantly, what it means for autistic individuals’ social interactions. In a controlled, semi-naturalistic interaction setting, this study examined how often autistic and neurotypical (NT) children mimicked a virtual partner’s non-facial mannerisms as they engaged in an interactive story-telling activity. Subsequently, children reported how affiliated they felt towards their interaction partner using an established implicit measure of closeness and a set of questions. Results revealed reduced mimicry (p =.001, φ =0.38) and less affiliation (p =.01, φ =0.33) in ASD relative to NT children. Mimicry was associated with affiliation for NT (r(23) =0.64, p =.0009), but not ASD, children (r(31) =0.07, p =.72). These results suggest an autism-associated reduction in mimicry and that mimicry during social interactions may not substantially contribute to affiliation in autism.

Research conducted with neurotypical (NT) populations shows that people often mimic each other – i.e., unconsciously copy facial expressions or postures of their interaction partners without any explicit instruction to do so (Chartrand & Lakin, 2011). This phenomenon, called “mimicry” or “automatic imitation”, is widely agreed to play a critical role in many aspects of social-cognitive development (Meltzoff, 2002). In particular, mimicry is thought to enable embodying other people’s emotions and mental states (Arnold & Winkielman, 2020) and convey the signal that “I am like you”, thus forging social affiliation and interaction quality in NT interactions (for reviews, see: Duffy & Chartrand, 2015; Over & Carpenter, 2013). A defining feature of autism spectrum disorders (ASD) is social-communicative difficulties, yet research is limited on the relationship between mimicry and social affiliation and interactions in ASD. In this study, we examined how autistic and non-autistic children mimic their interaction partner and how mimicry is associated with affiliation towards the partner.

Decades of research have consistently documented that autistic children show decreased alignment of their movements with others, both in the form of automatic, unconscious mimicry and in the form of explicit, instructed motor imitation (Rogers & Pennington, 1991; Rogers & Williams, 2006). Research into the neural underpinnings of (mostly explicit, instructed) imitation have revealed that imitation is enabled by an interaction between the mirror neuron system implicated in direct action-matching (Iacoboni, 2005; Rizzolatti & Craighero, 2004) and prefrontal regions implicated in top-down control of social-contextual cues (Hamilton, 2015). Thus, ASD-related differences in either brain system and/or their interaction may underpin the observed differences in behavioural mimicry and imitation (Southgate et al., 2008). Two lines of literature offer important insights into the behavioural profile of unconscious mimicry in development: (i) automatic mimicry of emotional facial expressions from computer stimuli, and (ii) imitation of seemingly meaningless actions following explicit instruction.

Electromyography studies using computerised stimuli found that infants and children mimic emotional facial expressions above chance levels (de Klerk et al., 2019; Skinner et al., 2020) and tend to mimic more after witnessing ostracism or if they have insecure attachment (de Klerk et al., 2020; Vacaru et al., 2019). Similar computer-stimuli paradigms found delayed mimicry of facial expressions (Oberman et al., 2009) and reduced frequency of yawns (Senju et al., 2007) and laughter (Helt et al., 2020) in autistic as compared to NT children. While these studies have advanced our understanding of the perceptual, cognitive and neuronal basis of mimicry and imitation, we know relatively little about how automatic mimicry of non-emotional, non-facial behaviours manifests in autism, and, importantly, how such mimicry contributes to social interactions. Unlike emotional facial mimicry, behavioural mimicry has lower temporal resolution, is more overtly visible and, crucially for ASD, is less emotionally salient (Eigsti, 2013).

The few studies with NT children on behavioural mimicry have found that 4-6 year-olds mimic a video model’s yawning, nodding, rubbing and scratching actions significantly above baseline levels, even when these actions appear in isolated video clips not embedded into an interactional setting (van Schaik & Hunnius, 2016). Moreover, children are more likely to mimic an in-group member (van Schaik & Hunnius, 2016) and someone who shares with others (Van Schaik & Hunnius, 2018). Such social modulation suggests that behavioural mimicry assessed using video paradigms taps into mechanisms relevant for social affiliation in NT children. Yet, given the importance of natural, spontaneous and active displays as stimuli (Buck et al., 2017), it is not clear how mimicry occurring in response to static and posed computerised stimuli would translate into real-world interaction settings, and what it would mean for social affiliation. This leaves a crucial gap in our understanding of the development of spontaneous mimicry of behaviours that are not emotional or facial in more naturalistic, social contexts.

Another line of research providing important insights for mimicry is motor imitation following explicit instruction. Despite the more deliberate nature of action-copying involved in motor imitation, sensorimotor integration is an important shared characteristic between unconscious behavioural mimicry and explicit, instructed motor imitation. Both incidental mimicry and deliberate imitation require: (i) processing another person’s motor action, and (ii) configuring one’s own body to produce the same action (Phillips-Silver et al., 2010). Research shows that autistic children copy actions less accurately even when they are explicitly instructed (McAuliffe et al., 2019; Tunçgenç et al., 2020), and especially so if these are only stylistic (e.g., speed, force) or seemingly meaningless actions (Hobson & Hobson, 2008; Rogers et al., 1996). Over-imitation studies, which involve actions irrelevant to attaining a goal (e.g., stroking a box before opening it), reveal that autistic children copy these seemingly meaningless actions less than NT children (Marsh, Pearson, Ropar, & Hamilton, 2013). Thus, actions unrelated to attaining an object-related goal, as is often the case in mimicry (Eigsti, 2013), seem to be copied less in ASD.

Multiple factors including sensorimotor integration, understanding social-contextual cues, social motivation and their underlying neural mechanisms may contribute to autism-associated differences in mimicry. Neuroimaging studies show that networks involved in sensorimotor integration are impaired in ASD, with decreased visuo-motor connectivity in ASD being linked to motor imitation deficits (Lidstone & Mostofsky, 2021; Nebel et al., 2016). In ASD, poorer performance in sensorimotor integration tasks predict socio-cognitive skill (Baillin et al., 2020; Haswell et al., 2009; Izawa et al., 2012)s, indicating the importance of sensorimotor skills for real-time social interactions. In addition to differences in neural systems supporting mimicry, there are autism-associated differences in processing context-specific aspects, such as which actions are mimicked, when they are mimicked, and how mimicry adopts social functions (Wang & de Hamilton, 2012). Thus, reduced mimicry may be relevant for social interaction difficulties in ASD via: (a) impeding social-emotional alignment between interactants and (b) yielding negative impressions about the autistic individual by partners who associate mimicry with affiliation (Arnold & Winkielman, 2020). This view would be in line with the double empathy hypothesis, which refers to the idea that social interaction difficulties in ASD stem from impaired communication among individuals with different ways of perceiving the world (Milton, 2012). Recent studies provide some support for this hypothesis (Crompton, Hallett, et al., 2020; Crompton, Ropar, et al., 2020; Crompton, Sharp, et al., 2020) by showing that interactions among ASD individuals are similarly rich in information transfer, rapport and interaction smoothness to interactions among NT individuals, with decreases observed only in mixed ASD-NT pairs. Thus, to better understand mimicry and social interactions in ASD, it is important to consider not just whether ASD individuals show reduced mimicry, but also what mimicry means for their social affiliation with others.

Despite the evidence for the importance of mimicry in social-emotional reciprocity and development, and for autism-associated difficulties with movement alignment, to our knowledge, there has been only one study that has assessed mimicry in a naturalistic social interaction setting with autistic children (Helt et al., 2010). This study showed decreased contagious yawning (a form of mimicry) in ASD during a live, 12-minute story-reading activity as the adult storyteller yawned surreptitiously. Several important gaps in knowledge remain, however, including examining actions that do not conjure an emotion (e.g., boredom in case of yawning), generalisability of the findings to non-facial actions, and relevance of mimicry to social affiliation for autistic people.

The current study examines mimicry in a semi-structured social interaction and tests its association with self-reported social affiliation with the interlocutor. Adapting Helt et al.’s (2010) social interaction paradigm, we presented the children a pre-recorded story narration. This video stimulus was designed to maximise the naturalistic, conversational aspects of an interaction, while maintaining experimental control not being biased towards NT interactions. The narrator produced non-facial, non-emotional actions (e.g., head and arm scratching) while telling the story. We predicted that: (i) ASD children would mimic their partner’s mannerisms less than NT children, and (ii) Mimicry frequency and affiliation would be positively associated in the NT group. We explored whether this association would be different within the ASD group and the association of mimicry with core autism symptoms.

Methods

Participants

A total of 74 children (42 ASD, 32 NT) participated. Out of this sample, 10 ASD and 4 NT children did not have valid mimicry data due to: data loss (4 ASD, 3 NT), children’s unwillingness to complete the study (4 ASD, 1 NT), and not meeting the inclusion criteria (2 ASD). The final sample used in mimicry analyses comprised 32 ASD children (1 girl) and 28 NT children (3 girls), with groups balanced on age, gender and IQ (see Table S1). Another 6 (1 ASD, 5 NT) children were dropped from the social affiliation analyses because time constraints on day of testing meant we could not administer the questionnaires. Thus, the sample used in social affiliation questionnaire analyses comprised 54 children (ASD group: n = 31, 1 girl, M_age = 10.32; M_IQ = 103.81; NT group: n = 23, 3 girls, M_age = 10.50, M_IQ = 107.70).

The sample size was based a priori on previous studies with similar designs that detected autism-associated deficits in action-copying (Helt et al., 2010; McAuliffe et al., 2019). We aimed for a larger sample size than these prior studies (i.e., N = 30 participants per group) given our planned testing of correlations between mimicry and affiliation.

Participant recruitment took place as part of a larger, longitudinal study through announcements sent out to local schools and an existing database of families interested in taking part in research studies. Despite efforts to over-recruit girls, the resulting sample reflects the 1-to-4 ratio of girls-to-boys in ASD (CDC, 2018). Full details of the inclusion/exclusion criteria and characterisation of the diagnostic groups can be found in Supplementary Methods and in Table S1. For children in the ASD group, diagnosis was confirmed by a child neurologist (i.e., senior author S.H.M) according to DSM-5 criteria using Autism Diagnostic Observation Schedule (ADOS-2) module-3 (Lord et al., 2000). Key inclusion criteria were being 8 – 13 years of age and having the necessary linguistic skills for the story telling task, the inclusion criteria included scoring ≥ 80 on the Full-Scale IQ or one of the indices (Verbal Comprehension, Visual-Spatial, or Fluid Reasoning) of Wechsler Intelligence Scale for Children (WISC-V) (Wechsler, 2014). NT children were eligible if they had no diagnosed developmental or psychiatric disorder. For all children, parent-report of Social Responsiveness Scale (SRS-2) (Constantino & Gruber, 2012) was obtained.

Compliance with Ethical Standards

This study was approved by Johns Hopkins University Institutional Review Board. In line with the Declaration of Helsinki, participants’ legal guardians provided written consent before the study, and children provided verbal assent before data collection. The study took place during two day-long visits to the Kennedy Krieger Institute; participating families were compensated $100 for their time.

Procedure

Children completed the following tasks in order: the story game (assessing mimicry), the theatre task (assessing social closeness), and the social affiliation questionnaire. An overview of the study procedure can be seen in Figure 1.

FIG. 1.

Study procedure with screenshots from the video stimuli showing the narrator perform the target actions during the test blocks. Speaking head icons indicate the children’s retell blocks. Dashed grey boxes show the periods during which mimicry was coded (i.e., excluding children’s retell blocks).

Story Game

Mimicry was assessed during a “story-telling game”, in which children watched a video of a woman telling them a story and then narrated the story back to her. The story was taken from an American children’s story website (http://www.short-story-time.com/the-night-i-had-a-fight-with-the-almosttooth-fairy.html) and chosen because it is relatable, fast-paced, and highly action-based with a series of events unfolding within the main storyline. We adapted the original story to remove euphemisms, metaphors or other figurative speech, and to ensure the events are told in plain language. The full text of the presented story can be found in Supplementary Methods.

The video of the narrator was presented on a large TV screen, with the narrator appearing life-size and in the same room set-up as the child. The task was presented as a story game to prevent the children from focussing on mimicry as the purpose of the task. Previous research shows that even brief (1-minute) video stimuli can evoke mimicry responses (Lakin & Chartrand, 2003; Stel et al., 2010; van Baaren et al., 2006). Similar target actions to those used in this study were previously successfully used with children using brief video stimuli (van Schaik & Hunnius, 2016; Van Schaik & Hunnius, 2018).

To increase engagement and decrease memory load, the 10-minute story was split into two baseline (2.5 minutes) and three test blocks (2.5 minutes each). Children had unlimited time to retell that portion of the story they had just heard. The baseline blocks, during which the narrator did not perform any target actions, assessed the children’s natural tendency to produce the target actions (i.e., head or arm scratching). In the test blocks, the narrator scratched her head and arm once per block. Mimicry scores were calculated by deducting the frequency of target actions performed spontaneously at baseline (if any) from the mimicry instances in the test blocks. If the children scratched their head or arm following the narrator’s corresponding action within the same test block, this was considered an instance of mimicry. If children scratched their head or arm before the actor did so within the same test block, it was not counted as an instance of mimicry. This meant that children had a maximum of 2 minutes to do the target actions in order for them to be counted as an instance of mimicry, though, in reality, most children (86%) who performed a target action did so within 10 seconds of the actor performing that action (M = 7.56, range = 0.63 – 116.41 seconds, M_ASD = 5.03, M_NT = 10.45, p = .29). Supplementary analyses excluding the children (3 ASD, 3 NT) with mimicry latency of more than 10 seconds did not change any of the findings (see Table S2).

In addition to head or arm scratching, the narrator adopted a different posture for each block and yawned and drank water once per block. These variables were not included in the current analysis due to very low instances after excluding cases where (a) postural mimicry could be visibly and reliably detected by two independent raters (n = 3 remaining cases of postural mimicry), and (b) children explicitly noticed the instances of yawning and drinking (remaining cases: n = 4 yawning, n = 7 drinking).

Several procedural measures were implemented to optimise maintaining strict experimental control, while facilitating the interactive nature of the video, and reducing social interaction pressures and biases on the participants. Video recording was used, rather than live interaction, to control for inadvertent social cues from the narrator (e.g., eye contact, smiles), which could be applied differentially to ASD vs NT children, which, in turn, could influence participants’ affiliation and mimicry. A similar video paradigm has been previously used for the assessment of mimicry in NT children (Van Schaik & Hunnius, 2018). To heighten make the interaction more naturalistic, the narrator and the activity were never referred to as a “video”. The narrator started by introducing herself saying “Hi! My name is Emma. Now, I’ll tell you one of my favourite stories”. As children retold the story, the narrator adopted an attentive listening face. Thus, the story game had a conversation-like turn-taking structure akin to real-world social interactions, while allowing us to maintain experimental control. Anecdotally, several children reported that they thought the narrator was in the adjacent room or that this was a live video feed.

At the end of the session, we asked the children “if they noticed anything unusual in the narrator’s behaviours or if anything attracted their attention in the way she did things”. If they said “No”, we further prompted, asking whether they noticed the narrator’s head or arm scratching. If children reported noticing either action, we asked whether they noticed the narrator do any other action. As is standard in behavioural mimicry research, incidents of that action were excluded from the child’s mimicry score (n = 1 NT head scratching, n = 4 ASD arm scratching excluded). Of note, none of these 5 children mimicked the other target action (i.e., the one they did not report explicitly noticing) and excluding these children from the analyses did not change any of the findings reported below.

Theatre Task

Immediately after the story game, children were asked a single-item question to gauge their feelings of social closeness towards the narrator from the story game. We showed children a picture of a “movie theatre” layout with 10 seats on a row (see Fig 1) and asked to indicate their preferred seat in that theatre by writing their own name on top of a smiley face sticker and placing it on their preferred seat. Next, we told the children to imagine that Emma, the narrator, was also at the theatre, and asked them to indicate their preferred seating location for Emma by writing her name on another sticker and placing it in the audience. The number of seats intervening between the child’s and the narrator’s locations (1–9) was taken as the measure of social closeness. This measure was adapted (i.e., by adding more seating options) from other similar measures used with children (Howard et al., 2021; Song et al., 2015; Tunçgenç & Cohen, 2016). To our knowledge, this was the first time this measure was used in an autistic population. We therefore conducted an initial pilot with N = 8 ASD children, which ensured that the children’s responses showed considerable variability in different conditions (e.g., when they were asked to imagine sitting with their best friend vs a stranger).

Social Affiliation Questionnaire

The last task in the study was a 16-item paper questionnaire, asking the children to rate their agreement with a series of statements on a Likert scale; see Table 1. The items were presented in a fixed-order and options ranged from 1 = Not very much to 7 = Very much. Children were told that there were no right or wrong answers, and the experimenter purposely turned away from the child to reduce experimenter demand effects. The questions were adapted from previous research assessing social bonding in children (Rabinowitch & Knafo-Noam, 2015; Tunçgenç & Cohen, 2016). We also added some questions assessing children’s likelihood of attributing positive qualities (e.g., honesty, generosity) to the narrator. Similar to the Theatre task, we are not aware of prior research using these questions with autistic children. Our pilot with N = 8 ASD children ensured that the questions and answer options were easy to understand for the children and that they showed considerable variability.

Table 1.

Social affiliation questionnaire.

Here are some questions about Emma (the narrator) and the tooth fairy story. How much do you agree with the statements? (Response options: 1 = Not Very Much, 7 = Very Much)

1 I think that Emma is a good storyteller

2 I would like to be friends with Emma

3 If I were to play another fun game now, I would like Emma to join me in this game

4 I like Emma

I think that Emma is…

5 caring

6 hardworking

7 kind

8 creative

9 friendly

10 clever

11 honest

12 athletic

13 generous

14 clean

15 I like the tooth fairy story

16 I understand the tooth fairy story

Data processing

Story Game

All children were video recorded throughout the narration activity for post-hoc coding of mimicry. Using the open-source ELAN software version 5.7 (ELAN, 2019), children’s behaviours during the blocks in which the narrator told the story were coded frame-by-frame by annotating the start- and end-points of all observed head and arm scratching actions as performed by the children. The start-point of a target action was determined as the first frame that the child initiated a move towards performing the target action, and the endpoint was determined as the last frame of that action before the child either stopped scratching or started moving their hand to another position. Out of the 60 videos, two raters naïve to condition each coded 24 videos, and reached excellent inter-rater reliability on mimicry frequency, r(23 ) = .86, p < .0001. The main rater’s codes (second author C.K.) were used in subsequent analyses.

Whilst coding the child’s actions, both raters were naïve to the timing of the narrator’s actions. Once the raters finished coding for the child’s actions, the video was unmuted and viewed again to annotate the timestamps of the blocks and the narrator’s target actions. The ELAN output was processed using a custom-written script on Matlab 2018b, which exported the timestamps of only those mimicry instances that occurred after the narrator had performed the same action within the same block. Quantification of what counts as mimicry has been widely varied in the literature. Our post-hoc video coding method can capture only overt behaviours, unlike electromyography studies that can detect minute behavioural changes unobservable to the human eye. This limits our ability to assess any potential carry-over effects from one test block to another. Given the possibility of carry-over effects across different test blocks, we included one instance of arm and one instance of face scratching action in each test block and evaluated these instances cumulatively in a single category of mimicry. Additionally, we coded the duration of time children spent looking at the video (versus elsewhere in the room) as a proxy for task attention and engagement.

Theatre Task

Since children always put the sticker representing themselves first and the narrator second, for each child, we first calculated the maximum possible distance between the child and the narrator given the child’s seat. For instance, if a child selected the leftmost seat for themselves, their distance value would range between 1 and 9. Next, we converted these raw distance values to percentages by mapping the minimum possible distance value of one seat to be equal to 100% and the maximum possible distance value for that child to be equal to 0%, such that higher percentage scores refer to a higher degree of closeness.

Social Affiliation Questionnaire

Items 2–4 in the questionnaire examined how affiliated the children felt towards the narrator. Due to their high internal consistency (Cronbach’s α = .89), these three items were averaged to comprise the children’s “affiliation” score. Of note, social affiliation questionnaire was significantly and positively correlated with the theatre task for both ASD (r(29) = .53, p = .002) and NT groups (r(21) = .64, p = .001), which corroborates the validity of these measures. The next 10 items (#5–14) asked the children what kinds of personal traits they thought the narrator might have. Originally, we construed half of these items to tap into more social traits (i.e., caring, kind, friendly, honest, generous), but the internal consistency analysis revealed that all 10 traits together had a higher consistency with a Cronbach’s alpha of .89 (as compared to Cronbach’s α = .85 for the five ‘social’ traits). Thus, we averaged all 10 traits to comprise the children’s “positive traits” score. The other three items (#1, #15, #16) served as control checks and were analysed to control for any unintended differences between the ASD and NT groups in terms of their understanding and enjoyment of the story. For all items of the social affiliation questionnaire, higher scores indicate a more positive attitude towards the narrator or the story.

Statistical Analysis

All analyses were completed in R Studio (R Core Team, 2019). Mimicry frequency was positively skewed due to high numbers of zero mimicry instances (i.e., reflecting floor effects); to address this, we implemented a zero-inflated negative binomial regression model, which can account for the possibility that zero values reflect a different underlying process compared to the count values. The social closeness and affiliation variables were also non-normally distributed; thus, we used the square root of the outcome variables in all subsequent analyses. These analyses included age and sex as covariates. In addition, for all analyses, we tested the effect of diagnosis using non-parametric Kruskal-Wallis tests.

Results

First, we conducted a series of confound checks using non-parametric Kruskal-Wallis tests. These tests showed there were no diagnostic groups differences in terms of children’s social attention to the narrator and understanding or enjoyment of the task (see Table S3 for full statistics).

The model testing our first hypothesis of whether the ASD and NT groups exhibited different amounts of mimicry revealed that children in the ASD group (M = 1.25, SD = 1.67; see Fig 2) mimicked the narrator significantly less than did their NT peers (M = 3.57, SD = 3.44; β = 0.83, SE = 0.24, p = .001). Overall, 79% of NT children and 56% of ASD children mimicked the narrator at least once. Note that a non-parametric Kruskal-Wallis test examining the effect of diagnosis (ASD vs NT) on the raw mimicry frequency values similarly showed significantly reduced mimicry in ASD as compared to NT (X²(1) = 8.85, p = .003, φ = 0.38).

Figure 2.

Violin plots showing differences between the autism (ASD) and the neurotypical (NT) groups in mimicry frequency (left panel) and social affiliation (right panel). Black diamonds show the mean, white dots show individual data points and horizontal lines show the quantiles

Next, we conducted two linear models with diagnosis (ASD vs NT) as the predictor variable and age and sex as covariates to assess their effects on (i) social closeness scores from the theatre task, and (ii) social affiliation questionnaire scores (i.e., Items 2–4 about children’s “affiliation” with Emma, and Items 5–14 about “positive traits” children attributed to Emma). We found significantly less social closeness in ASD children as indicated by their preference that the narrator sit further from themselves compared to the children in the NT group (M_ASD = 57.42% closeness, SD_ASD = 34.87, M_NT = 83.70% closeness, SD_NT = 12.72; β = 2.32, SE = 0.76, p = .005). Further, ASD children reported affiliating significantly less with the narrator than did NT children (M_ASD = 4.12, SD_ASD = 1.70, M_NT = 5.29, SD_NT = 1.16; β = 0.28, SE = 0.11, p = .01; Fig 2); however, children in the ASD and NT groups did not differ in terms of their attribution of positive traits to the narrator (M_ASD = 5.34, SD_ASD = 1.08, M_NT = 5.57, SD_NT = 0.99; β = 0.03, SE = 0.07, p = .66). Examining the effect of diagnosis on these three variables using non-parametric Kruskal-Wallis tests confirmed the same results (closeness: X²(1) = 9.30, p = .002; affiliation: X²(1) = 6.38, p = .01; positive traits: X²(1) = .57, p = .45).

To test the hypothesis that affiliation is positively related to mimicry, we examined bivariate associations among raw mimicry frequencies and social affiliation scores (Items 2 – 4 on the questionnaire). In line with our predictions, we found a positive association between mimicry frequency and social affiliation scores in the NT group, r(23) = .64, p = .0009; Fig 3. In contrast, mimicry frequency and affiliation were not meaningfully associated in the ASD group, r(31) = .07, p = .72; Fig 3.

Figure 3.

Associations of mimicry frequency with social affiliation (left panel) and Social Responsiveness Scale (SRS-2) responses (right panel) for autism (ASD) and neurotypical (NT) groups

We also examined how mimicry frequency was correlated with the severity of autism symptoms as assessed by the SRS-2 total T-scores (administered to the entire sample) and ADOS-2 (administered to the ASD group only). This analysis revealed that, across diagnostic groups, mimicry was negatively associated with SRS-2, r(53) = −.39, p = .004. However, when we unpack this within groups, the association did not reach significance within the ASD group (see Table 2). The association between mimicry and ADOS-2 total scores was also not significant (see Table 2), which may be partly due to the relatively small variation in ADOS-2 responses.

Table 2.

Bivariate Pearson correlations among mimicry, closeness, affiliation and core autism symptoms as measured by SRS-2 and ADOS-2, with higher scores indicating increased autism severity (overall, autism: ASD, neurotypical: NT).

	Social closeness (Theatre task)			Social affiliation questionnaire			SRS-2 total			ADOS-2 total
Mimicry	.33 *	.08	.50*	.43 **	.07	.64**	−.39 *	−.29	−.02	−.23
Social closeness				.61 **	.53**	.64**	−.26	.17	.37	.01
Social affiliation							−.35 *	.02.	.16	.07
SRS-2 Total										.07

Note: *p< .05, **p< .01. P values are Bonferroni-corrected for multiple comparisons.

Discussion

Prior research shows reduced mimicry of emotional, facial expressions in ASD, leaving open the questions of how behavioural mimicry manifests and relates to social interactions in autism. Extant studies often assessed mimicry in non-interactive settings using still images or posed video clips, with limited generalisability for real-life social interactions. Addressing these gaps, we examined how children with and without ASD spontaneously mimic others’ non-facial mannerisms in a naturalistic yet experimental interaction setting and how such mimicry is associated with social affiliation. Consistent with our hypothesis, we found that autistic children mimicked their partner significantly less than did NT children. Mimicry and self-reported affiliation with the partner were positively associated in NT children, while no association existed for the ASD group.

Our finding of reduced mimicry in ASD complements and extends prior research on reduced imitation of actions unrelated to an object-goal (Eigsti, 2013). Our paradigm makes at least two key contributions: Firstly, we used non-emotional, non-facial behaviours, which is important given that autistic individuals attend less to eyes and faces during social interactions (Akhtar & Jaswal, 2020). Such behaviours have the shared characteristics with other motor imitation and gesture paradigms of requiring perception of others’ body and movements, mapping them onto one’s own body and executing a motor response in return (Phillips-Silver et al., 2010). Thus, despite the differences in the contexts in which they occur or the functions they may serve (Eigsti, 2013), non-facial, non-emotional mimicry also relies on visuo-motor integration mechanisms, which are known to be impacted in ASD (Lidstone & Mostofsky, 2021). The second contribution of our paradigm is that the social interaction context was designed to be relatively naturalistic, while allowing us to maintain experimental control and not be biased towards NT interactions which are characterised by coordinated exchanges, e.g., mutual smiles or eye gaze (Akhtar & Jaswal, 2020). To attain this for both ASD and NT participants, the video stimuli were made as lifelike as possible, and the task was introduced as a semi-structured story game. At the same time, we endeavoured to reduce experimenter bias and performance demands (e.g., experimenter’s engagement level) through presenting the partner’s actions via a pre-recorded video. This study provides proof-of-concept for assessing autism-associated differences in mimicry in a semi-structured social interaction context, which children with and without ASD enjoyed and engaged with. Future research can look into adapting and improving this methodology to examine real-world social interactions in an experimentally controlled setting. Similarly, the social closeness task (i.e., the theatre task) and the affiliation questionnaire demonstrated high internal validity, wide data distribution and were strongly associated with each other indicating construct validity, making them promising assessment methods for future research.

Uniquely, and to our knowledge, for the first time, in this study, we examined how mimicry was associated with social affiliation in ASD. Despite similar levels of task enjoyment, understanding of the story and attribution of positive traits to the narrator across the ASD and NT groups, we found that autistic children felt less close to the narrator than did NT children. Results further revealed that increased mimicry frequency was linked to increased social affiliation for NT children, while no association between mimicry and affiliation was observed for autistic children. These findings are in line with traditional accounts of diminished social motivation in autism (Chevallier et al., 2012) and with findings suggesting divergent brain systems at work in ASD vs NT populations for processing positive emotions, social stimuli and mimicry (Mundy, 2018). Prior research on attachment with caregivers has demonstrated that while autistic children can and do form secure attachments, in the absence of an external threat, they are highly selective in their attachment behaviours and are less likely to initiate proximity-seeking with caregivers as compared to their NT peers, thereby reducing opportunities to form bonds and learn from others in everyday situations (Vivanti & Nuske, 2017). Following this reasoning, the relatively inconsequential, everyday social interaction context presented in our study may not have been sufficient to trigger social affiliation motives for ASD children, which may be already diminished and/or highly selective. To further establish the mechanisms underlying a disrupted link between mimicry and social affiliation in ASD, future research can examine how brain networks involved in mirroring actions, visuo-motor integration and top-down control of social cues are implicated in ASD during unconscious mimicry (Hamilton, 2015).

In addition to exploring mechanisms of a disrupted mimicry – social affiliation link, future research should examine the alternative ways through which autistic individuals form and express affiliation, if not through unconscious mimicry. Our null findings may reflect a tendency in ASD individuals to forge and show affiliation through other verbal and/or nonverbal behavioural strategies. It may also be that mimicry has an affiliative function in ASD only in certain circumstances, which our study did not tap into. For instance, it may be that differences in high-order time perception (e.g., temporal order of events, one’s location in passing time), in binding multisensory stimuli together in perception (Zhou et al., 2018) or the interaction partner not being a visibly autistic person (Crompton, Sharp, et al., 2020) may have resulted in decreased perception of the alignment between self and the narrator and thus, reduced rapport. Prior research has shown that autistic children’s friendships manifest reduced prosocial behaviours such as sharing or helping, less coordinated play and responsiveness (Bauminger et al., 2008) despite the ASD and NT children having similar degrees of social closeness with their friends (Bauminger & Kasari, 2000). Approaches that are based on lived experiences of autistic people and a focus away from deficiencies as compared to NT children may have a better chance at identifying what works for autistic children in maintaining close relationships. Several other factors can intervene with these questions of contextual effects on the mimicry – social affiliation link and of alternative behavioural strategies. Thus, further examinations are needed on younger samples with more diverse demographic characteristics (e.g., age, gender, IQ, comorbidities), as our sample, comprised mostly of 8- to 13-year-olds boys above certain IQ and verbal ability thresholds, are not necessarily representative of the highly heterogeneous autism population.

Conclusion

ASD is characterised by social-communicative difficulties, with early signs in childhood including reduced eye contact and social attention (Akhtar & Jaswal, 2020). Mimicking another person requires perceiving their actions and configuring one’s own body to perform the same actions. Research suggests that such sensorimotor integration can facilitate sharing emotions, goals and affiliating with interactants (Chartrand & Lakin, 2011) and that there are reliable differences in sensorimotor processing and responding in ASD (Lidstone & Mostofsky, 2021). Thus, in theory, these social, cognitive and motor differences can lead to reduced mimicry in ASD, which may then relate to social-communicative difficulties through (a) reduced opportunities for sharing emotions and affiliating with partners, and/or (b) leading autistic individuals to be perceived less favourably by their NT partners for whom mimicry is positively associated with affiliation. Our findings show that while autistic children mimic others less, their degree of mimicry is unrelated to their feelings of affiliation. Future studies should address how social motivation, sensorimotor integration and other mechanisms are associated with mimicry, affiliation and what, if any, alternative methods autistic individuals adopt for social affiliation.