Empathic Accuracy and Cognitive and Affective Empathy in Young Adults With and Without Autism Spectrum Disorder

ASD is characterised by qualitative differences in social communication/interaction and a pattern of repetitive, restricted behaviours, interests and activities (American Psychiatric Association, 2013). People with a diagnosis of ASD¹ are commonly thought to have difficulties in empathy compared with typically-developing (TD) individuals; as such, ASD has been characterised as an “empathy disorder” (Gillberg, 1992). There are several theories which discuss possible mechanisms underlying empathy deficits in people with ASD. A leading theory of autism suggests that individuals with ASD lack Theory of Mind, meaning they find it difficult to infer the mental states of others (Baron-Cohen, 2000). People with ASD typically perform worse on ‘false-belief’ tasks which measure Theory of Mind by assessing one’s ability to distinguish between events in reality, versus how events may be (incorrectly) represented in another person’s mind (Baron-Cohen et al., 1985). In contrast, the social motivation hypothesis (Chevallier et al., 2012) proposes that individuals with ASD are less socially motivated than TD individuals, which contributes to social impairments such as empathic deficits. Finally, Baron-Cohen’s (2009) ‘Empathising-Systemising’ theory of ASD suggests that people with ASD have a preference towards tasks which involve systemising (e.g., solving maths problems, making lists, fixing bikes, etc.), because systems often change in lawful and predictable ways. However, the theory also indicates that people with ASD struggle with empathy because it is not ‘truth-oriented’; there are no laws which can be applied to understand emotions in others, as different people express emotions in different ways. Using self-report methods, Greenberg et al. (2018) measured the “brain types” (classifying them as either empathising or systemising) of over half a million people, including individuals with a diagnosis of ASD. They found that, when compared to TD participants, the brain types of people with a diagnosis of ASD were balanced in favour of ‘systemising’ at the expense of ‘empathising’ to either a strong or very strong degree.

The Empathy Imbalance Hypothesis (Smith, 2009) provides a more nuanced view on empathy by suggesting that people with a diagnosis of ASD experience “empathic overarousal”, leading them to experience increased distress in response to others’ emotions compared with TD individuals. Consequently, Smith (2009) argues that people with a diagnosis of ASD have heightened AE in comparison to TD individuals, despite showing impairments in CE. An abundance of research has provided evidence for impaired CE in ASD (Baron-Cohen et al., 1997; Dziobek et al., 2008; Baron-Cohen et al., 1997; Ozonoff et al., 1990; Rogers et al., 2007; Zalla et al., 2009), as measured via self-report questionnaires such as the IRI or experimental tasks such as the MET. However, the evidence regarding AE in ASD is mixed, with the majority of studies indicating that people with ASD have either higher or similar levels of AE compared with people without ASD (Dziobek et al., 2008; Murray et al., 2017; Rogers et al., 2007; Song et al., 2019).

Supporting the Empathy Imbalance Hypothesis, in a meta-analysis of 51 studies, Song et al. (2019) found that people with ASD scored higher on IRI Personal Distress, and the AE subscale of the Questionnaire of Cognitive and Affective Empathy (QCAE) (Di Girolamo et al., 2019) compared to TD individuals, suggesting higher trait AE. However, they found that people with ASD scored lower overall on the Empathic Concern subscale of the IRI compared to TD participants. Song et al.’s (2019) findings also indicated that people with ASD performed similarly to TD individuals in experimental tasks measuring AE—most commonly the MET. For example, Dziobek et al. (2008) used the MET in a sample of adults with ASD and found deficits in CE, but not AE, when compared to TD individuals.

More nuanced findings were reported by Mazza et al. (2014), who found that, in comparison to TD participants, individuals with ASD showed deficits in AE (measured by the MET) specifically for negative emotions (such as sadness, anger and disappointment); AE for positive emotions (such as happiness and positive surprise) was unimpaired in the ASD group. Contrary to this, Jankowski and Pfeifer (2021) found that ASD and TD adolescents reported similar levels of AE when watching videos of actors taking part in a singing competition, regardless of whether the emotion was positive or negative. Participants were required to rate the levels of pride and embarrassment felt by the actors and were also asked to rate the levels of pride and embarrassment they themselves felt when watching the videos. CE abilities of participants with ASD (how accurately they rated the emotions felt by the actors) were more strongly influenced by the level of contextual information available. When the context was incongruent with the emotion being displayed (e.g., the contestant performed well but felt embarrassed), participants with ASD were more likely than TD participants to reference contextual information and rate that participants additionally felt proud, although they rated that the participant felt embarrassed to a similar degree.

In contrast, specific impairments in processing negative emotions have been reported in studies measuring CE in ASD populations using facial emotion recognition tasks (e.g., Ashwin et al., 2006). These studies focusing on facial emotion recognition in ASD have suggested that people with a diagnosis of ASD have specific deficits in identifying fear (Howard et al., 2000; Pelphrey et al., 2002), disgust (Golan et al., 2006), sadness (Boraston et al., 2007) and anger from facial expressions (Bal et al., 2010).

There is some, albeit limited, evidence to suggest that individuals with ASD may show atypical neural activity when responding to negative emotions. Krach et al. (2015) measured brain activity whilst participants looked at drawings depicting a protagonist within social scenarios, some of which were neutral, and some of which were likely to induce embarrassment. Compared with TD individuals, participants with ASD showed deficits in AE (measured by the levels of vicarious embarrassment they felt on behalf of the protagonist within the scenarios). Relative to TD controls, the ASD group showed reduced anterior insula and anterior cingulate cortex activity, but greater hippocampus activity, when viewing these scenarios. This suggests that whilst TD individuals are likely to rely on social cues (e.g., facial expressions, body language), people with ASD are more likely to rely on memories of previous social encounters to inform their AE responses.

Further to this, Baron-Cohen et al. (1999) found that individuals with ASD showed significantly less activation in the amygdala overall than TD controls during a task in which participants were required to infer what a person was thinking/feeling by looking at photographs of their eyes. The researchers did not explore whether there was an interaction between group status and emotion on amygdala activity (i.e., whether the ASD individuals showed reduced activation to specific emotions or emotions in general). However, amygdala activation seems to play a particularly important role in the recognition of threat-based emotions such as fear (Adolphs et al., 2005; Baron-Cohen et al., 2000) and anger (Scott et al., 1997). Thus, abnormal amygdala activity in people with a diagnosis of ASD may account for the findings discussed above, which suggest that empathic deficits in ASD might be disproportionately seen for negative emotions; however, further research is required to clarify this. Despite the evidence discussed above which indicates that empathy amongst individuals with ASD may vary across different emotions, few studies have explored this issue whilst measuring multiple components of empathy in parallel. Consequently, the literature may have provided an over-simplified view of empathic abilities in people with a diagnosis of ASD.

Whilst there are a large number of studies examining CE and AE in people with a diagnosis of ASD, primarily using self-report measures, fewer studies have measured empathic accuracy (EA) in ASD. Three studies by the same research group (Demurie et al., 2011; Ponnet et al., 2008; Roeyers et al., 2001) measured EA by asking participants to infer the thoughts and feelings of targets in video-taped social interactions which were either ‘structured’ (playing a board game) or ‘unstructured’ (an unplanned ‘getting acquainted’ conversation). The targets listed the thoughts and feelings they had experienced during the interactions and independent judges assessed how closely the participants’ responses matched the target’s actual thoughts and feelings, yielding an EA score. The authors found that participants with ASD only showed significant impairments when rating the unstructured, but not the structured, social interactions. It should also be noted that these three studies relied on predominantly male samples, which is a common feature of research measuring empathy in ASD populations (Song et al., 2019). This is a key limitation of the existing research, as it means comparatively little is known about empathy in females with ASD.

A recent study by Santiesteban et al. (2021) used an adapted version of the Empathic Accuracy Task (EAT) to measure empathy in 21 adults with ASD and 45 adults without ASD. The EAT assesses AE, CE and EA by measuring participants’ responses to videos of narrators describing autobiographical events. In Santiesteban et al.’s (2021) study, videos depicted either ‘affective’ (sad) or ‘neutral’ events. The narrators provided continuous ratings of how they felt whilst recording the videos. In half of the trials, participants continuously rated the strength of their own emotions whilst watching the video (measuring ‘online’ AE) and, immediately after the video, rated the strength of the actor’s emotions (assessing ‘offline’ CE). In the other half, participants continuously rated the strength of the actor’s emotions during the video (measuring ‘online’ EA), and afterwards gave a single rating reflecting the strength of their own emotions (measuring ‘offline’ AE). Participants with ASD showed deficits in ‘offline’ CE and lower AE compared to TD participants, particularly when viewing ‘affective’ (sad) videos. However, the groups did not significantly differ on the ‘online’ measures of EA and AE. Unfortunately, this study only measured AE, CE and EA in response to sad or neutral events, and did not explore empathic abilities in response to other discrete emotions as anger, happiness or fear.

Santiesteban et al.’s (2021) study also controlled for alexithymia, a trait which is increasingly thought to have an important relationship with empathy (Grynberg et al., 2010). Alexithymia, which is defined as problems in identifying and describing one’s own emotions (Sifneos, 1973), commonly co-occurs with ASD (Hill et al., 2004), and is independently linked to deficits in empathy (Guttman & Laporte, 2002; Prkachin et al., 2009). Research investigating the neural correlates of AE found that differences between people with versus without ASD in anterior insula activity when witnessing a friend experiencing pain were rendered non-significant when alexithymia was accounted for (Bird et al., 2010). More recently, Shah et al. (2019) found that alexithymia was associated with empathy deficits (measured by the Questionnaire of Cognitive and Affective Empathy) in people with autistic traits (but no confirmed diagnosis), but that autistic traits were a more important predictor of empathy than alexithymia. Given the mixed findings in the literature, it is unclear whether empathy deficits in ASD are explained by alexithymia. Thus, we investigated this issue by measuring the impact of alexithymia on different forms of empathy.

In summary, empathy is a multi-faceted construct with cognitive and affective components which can be expressed and measured at trait and state levels using self-report and experimental methods. Amongst people with a diagnosis of ASD, research has provided consistent evidence of impairments in CE, at both state and trait levels. There is more mixed evidence regarding AE and EA, which appears to be partly accounted for by the variance in measurement methods and lack of granularity in empathy measurement as related to discrete emotions. Furthermore, the generalisability of the existing literature is limited due to the under-representation of females with ASD in studies within this field.

This study therefore aims to improve upon previous research on empathy in ASD by investigating multiple facets of empathy using self-report and experimental methods in a mixed gender sample, while also taking account of alexithymia. We included stimuli depicting happiness, a positive emotion, and a range of negative emotions (anger, sadness and fear) to determine whether deficits in empathy in ASD vary across emotions or are specific to negative emotions. Based on previous research, we predicted that individuals with ASD would show deficits in CE compared to TD participants (Dziobek et al., 2008; Rogers et al., 2007). Although the evidence regarding AE is more mixed, there is no consistent evidence to suggest that AE is impaired in people with ASD. With the exception of trait Empathic Concern (a facet of AE where people with ASD show deficits), the majority of studies using trait measures of empathy indicate that AE is higher in people with ASD compared with TD individuals (Song et al., 2019). Furthermore, experimental studies suggest that people with and without ASD do not differ in state AE (Song et al., 2019). Taking the experimental evidence into account, we predicted that the groups would not differ in AE. It was more difficult to make predictions regarding EA, as this component of empathy has not been widely investigated. Nevertheless, based on studies exploring CE in people with ASD (Jolliffe & Baron-Cohen, 1999; Zalla et al., 2009), and evidence indicating that similar brain regions are involved in CE and EA (Mackes et al., 2018), we hypothesised that participants with ASD would show deficits in EA relative to TD participants. We also predicted that empathic deficits in the ASD group would be partly explained by alexithymia.

The study used a between-groups, cross-sectional design.

G Power was used to conduct an a priori power analysis, based on Dziobek et al.’s (2008) study which had a similar two-groups design. This suggested that a total of 54 participants (27 per group) was required to detect large effects at an α = 0.05 level of statistical significance and 95% power.

A total of 61 people consented to participate and were divided into two groups; people with a diagnosis of ASD (n = 29) and a control group of TD individuals (n = 32). Participants were matched at a group level in terms of age and gender (ASD Group: 17 males, 12 females; TD Group: 13 Males, 18 females). Participants were all aged 16–25 years and native English speakers. The majority of participants were White British.

Participants were excluded if they had a suspected intellectual disability or current diagnosis of a psychotic disorder. Participants in the ASD group were only included if they had a clinical diagnosis of ASD (confirmed via clinical report or equivalent). Participants in the TD group completed a brief version of the Autism Quotient (AQ-10) (Baron-Cohen et al., 2001) to screen for traits of ASD; they were excluded if they scored above the screening cut-off (≥ 6). One participant was excluded on this basis.

Participants were recruited via the University of Bath’s Research Community Participation Pool, the Centre for Applied Autism Research database and local educational programmes, such as a summer school for autistic students considering going to university. Poster advertisements were also placed in several university departments. Participants were reimbursed five pounds for taking part.

The IRI (Davis, 1980) was used to measure trait empathy. This is a 28-item self-report measure which consists of four subscales; ‘Empathic Concern’ and ‘Personal Distress’ (thought to measure AE); ‘Perspective Taking’ and ‘Fantasy’, (thought to measure CE). The IRI has adequate internal consistency with alpha coefficients ranging from 0.68 to 0.79 (Baldner & McGinley, 2014; Davis, 1980). The IRI has previously been used to measure trait empathy in people with ASD across numerous studies (Song et al., 2019).

The Toronto Alexithymia Scale (TAS-20) (Taylor et al., 1985), a 20-item self-report questionnaire, was used to measure trait alexithymia. Research indicates that this scale has moderate to high reliability when used in ASD populations (Cronbach’s α = 0.86) (Samson et al., 2012).

The presence of ASD traits in the TD group were measured using a brief version of the Autism Quotient (AQ-10). The AQ-10 has good sensitivity and specificity (Allison et al., 2012). Booth et al. (2013) found that the AQ-10 had an area under the curve of 90.3%, indicating excellent predictive validity.

Following data collection, the researchers conducted reliability analyses which showed that both the IRI (α = 0.845) and TAS-20 (α = 0.851) fell into the ‘good’ range for reliability.

Ethical approval was gained from the University of Bath’s Psychology Research Ethics Committee. During the debrief, the experimenter asked the participant how they were feeling and administered a mood repair task if the participant felt distressed.

Participants completed the self-report measures prior to the EAT (the AQ-10 was completed by the TD group only). Participants were debriefed after the session, which lasted up to one hour. No participants reported feeling distressed following the EAT and thus the mood repair task was not used.

The IRI and EAT data were not normally distributed, therefore Mann–Whitney U tests were used to compare IRI scores between groups, applying a Holm-Bonferroni correction (Holm, 1979) to control for familywise error rates. The EAT data were analysed using the procedure outlined by Martin-Key et al. (2017). The emotional intensity ratings provided by each participant were separated into two second bins. Correlations between these ratings and the narrators’ own ratings were then determined. These correlations were transformed using Fisher’s Z, resulting in one EA score per clip. Average total EA scores and EA scores for each emotion were then calculated for each participant. Mann–Whitney U tests were used to compare groups in terms of EA, AE and CE scores (total scores and scores for individual emotions). The same method was used to explore whether the ASD group’s EA, AE and CE scores (total scores and scores for each emotion) differed depending on whether they watched the old or new clips.

Preliminary analysis of the EAT data revealed that participants (n = 7, ASD Group) who watched the ‘old’ film clips (the original male and female narrators, before the female video clips were re-filmed) had significantly lower EA scores compared with those who watched the ‘new’ clips (with the original male narrator and new female narrator) [n = 22 (ASD Group), n = 31 (TD Group)], suggesting that it may not be valid to combine the data from participants who watched both types of clips (old and new). The seven participants who watched the old clips were therefore excluded from further analyses comparing the groups’ EA scores, meaning the ASD Group’s sample size for the EA analyses is 22 rather than 29. Analyses of EA data including the seven participants who watched the original clips is available in the Supplementary Material. As there were no significant differences in the CE or AE scores of participants who watched the old film clips versus those who watched the new clips, all (n = 59) participants were included in the CE and AE analyses.

To investigate whether alexithymia was associated with empathy in the ASD group, total TAS-20 scores were correlated with IRI and EAT scores for the ASD group only (n = 29 for IRI, AE and CE scores, n = 22 for EA Scores).

In addition to the 7 participants who were excluded from analyses of the EAT on the basis of viewing the original film clips, one additional TD participant was excluded from the analyses of Total EA scores and EA for Fear. This is because they did not change the default emotional intensity rating for the frightened clips, meaning their EA score for this clip could not be calculated. Consequently, it was not possible to determine their Total EA score. Additionally, EAT data was missing for one TD participant due to technical issues. Therefore for the EAT analyses, the final sample was n = 29 (ASD Group) and n = 30 (TD Group) for the CE and AE scores, and n = 22 (ASD Group) and n = 30 (TD Group) for the EA scores.

There were no differences between the ASD and TD groups in age [t(58) = 1.11, p = 0.272], gender (\(\chi^{2}\) (1) = 1.669, p = 0.196) or highest education qualification achieved (\(\chi^{2}\) (1) = 4.518, p = 0.211) (see Table 1). Significantly more participants in the ASD group than the TD group reported current mental health difficulties, \(\chi^{2}\) (2) = 10.043, p = 0.007.

In terms of trait CE, alexithymia scores were not significantly correlated with IRI Perspective Taking, r(28) = − 0.075, p = 0.699, or Fantasy, r(28) = − 0.074, p = 0.704. There was a marginally significant positive correlation between alexithymia scores and CE for anger [r(28) = 0.366, p = 0.051] in the ASD group. However, alexithymia did not correlate significantly with any other experimental CE scores. For trait AE, there was a significant positive correlation between alexithymia and IRI Personal Distress subscale of the IRI in the ASD group, r(28) = 0.484, p = 0.008 (see Supplementary Material for a scatterplot). However, the correlation between alexithymia and IRI Empathic Concern was non-significant, r(28) = − 0.327, p = 0.083, as were the correlations between alexithymia and all experimental AE measures. EA Scores for Anger were not significantly correlated with alexithymia scores, r(21) = − 0.041, p = 0.858. Similarly, no significant correlations were found between alexithymia and EA scores for the other emotions. Please see Supplementary Materials for details of the non-significant findings (p values ranged from 0.057 to 0.980; r values ranged from − 0.026 to 0.366). This pattern of correlations suggests that alexithymia does not explain the deficits in trait empathy or EA for anger observed in the ASD group.

The contrast between the findings of the trait and experimental empathy measures may be explained by individuals with ASD having a negative (or more realistic) view of their empathic abilities, leading them to report lower levels of empathy on the IRI. Alternatively, these differences may be linked to the sensitivity of the CE and AE components of the EAT, which was restricted in comparison to the IRI. The nuanced findings of this study and apparent discrepancies between different measures of empathy reinforce the conclusion that empathy is a complex construct, which is highly dependent on the context in which it is measured. By measuring EA and using a range of discrete emotions, the EAT has improved upon previous experimental measures in its ability to capture this complexity; however, further work is required to develop more dynamic measures of empathy. It may also be informative to use the EAT in a sample of children or adolescents with and without ASD, to explore whether differences in empathy are more pronounced at a younger age (and the specificity of the deficits in EA for anger observed here).

In addition, more research focusing on CE, EA and AE in people without ASD across varying contexts may be helpful in challenging the common narratives surrounding ASD and empathy. For instance, the ‘double empathy’ hypothesis (Milton, 2012) is based on research which indicates that people without ASD also show difficulties in empathy when responding to emotions displayed by people with ASD (Sasson et al., 2017; Sheppard et al., 2016). It would be interesting to explore this in relation to EA, AE and CE, to examine whether similar empathic deficits are observed when TD individuals view autobiographical experiences recounted by those with ASD. This could help in challenging the simplistic view that people with ASD have impaired empathy, and in promoting a more nuanced view, that people with differing brain types and different experiences of the world may, in some scenarios, have difficulties empathising with each other (Milton, 2012).

The finding that the ASD group showed deficits in EA when responding to anger has implications for social interactions, e.g., people with ASD may find it harder to detect changes in anger displayed by others; consequently, they may respond inappropriately, inadvertently causing the person to become more irate. This may put them at higher risk of experiencing interpersonal conflict, physical harm and lead to difficulties maintaining relationships. In order to understand why people with ASD show this deficit, further research examining brain activity in people with and without ASD during EAT performance is required. Furthermore, as people with ASD may require more information to detect anger in others, it may be helpful to develop interventions to enhance EA in this population.

One of the main strengths of this study is that it employs a more naturalistic and dynamic measure of empathy than the tasks that have been used previously. Furthermore, assessing empathy using both an experimental task and a validated self-report questionnaire allowed for a richer exploration of empathy as a multidimensional construct. The measurement of variables which have been shown to influence empathy (such as alexithymia and mental health problems) is a further strength. Possible confounding variables were also addressed by matching groups for age and gender; this is also one of the first experimental studies on empathy to include a well-balanced sample of male and female participants with ASD.

A limitation of the study is that a lack of statistical power may have reduced the likelihood of detecting statistically significant group differences, particularly for EA. The power calculation was performed using data from a study which used a different experimental measure of empathy, which may have implications for the validity of the power calculation. If a larger sample had been used, it is possible that more widespread empathy deficits would have been found within the ASD Group. A larger sample would have allowed for more meaningful exploration of gender differences; with greater statistical power, we could have adopted a four-group design and investigated whether empathy differed according to diagnostic group and gender. Unfortunately, the moderate sample size in the present study means that we would have been underpowered to detect interactions between gender and group on empathic accuracy, CE or AE. This is nevertheless an important area for future research given prior research reporting gender differences in the relationship between empathy and other clinical disorders such as Conduct Disorder (Martin-Key et al., 2020).

Furthermore, the CE and AE scores, in which participants are categorised as either ‘right’ or ‘wrong’, may result in an overly-simplified measurement of these constructs. More nuanced measures of CE and AE could be developed in the future; for example, by asking participants to rate the intensity of the narrators’ emotion on a nine-point scale (CE). A further limitation is that the authors failed to systematically collect information about the participants’ ethnicity. This information would have been useful given that all of the narrators in the EAT were White British.

The current findings suggest that people with a diagnosis of ASD show selective deficits in EA for anger, but no deficits in other aspects of empathy (AE or CE) on the experimental task. In contrast, some deficits were found in trait CE and AE (as assessed using the IRI) in the ASD group, whereas there was a non-significant tendency toward the ASD participants reporting higher levels of personal distress (a measure of AE). These findings challenge the notion that ASD is an ‘empathy disorder’ as they indicate that people with ASD are capable of recognising dynamic emotions and the emotional states of others. Furthermore, the findings highlight the need for further research exploring empathic responses to anger in ASD and the importance of assessing empathy to discrete emotions in future studies in this field.