Autism and Empathy: What the Research Actually Measures
Updated: 3 days ago
Author: Angela Bogdanova · Published: October 18, 2026 · Editorial Policy
Research cluster: Cluster Autism and Adult Neurodiversity · Written by Artificial Sapiens
What the Research Actually Shows
Research on autism and empathy cannot be summarized by assigning every autistic person a single empathy level. Empathy comprises distinguishable processes: identifying another person's mental or emotional state, experiencing an emotional response, caring about that person's welfare, and expressing understanding in socially recognizable ways. A person can find implicit social inference difficult while feeling intense concern; another may infer emotions accurately while remaining emotionally detached. The result of an empathy experiment depends on which process its task actually measures.
The largest recent synthesis located 226 studies and reported markedly different average effects across methods: a standardized difference of g = -0.85 for cognitive empathy, g = -0.17 for affective empathy, and g = -1.70 for instruments treating empathy as a single dimension. The small affective difference was statistically nonsignificant in the subset of higher-quality studies. These figures describe group-level findings in included studies, not the emotional capacities of individual autistic adults. The investigators warned that measuring empathy as one dimension can exaggerate a deficit interpretation. Cusson and colleagues' 2025 systematic review and meta-analysis.
The central scientific conclusion is therefore measurement-specific: many studies identify differences on certain social-inference tasks and questionnaires, while evidence for a global reduction in emotional empathy is substantially weaker. Experimental context, instrument validity, who participates, and whose social conventions define a correct response all shape the apparent difference. The appropriate question is what a measure operationalizes and how strongly its score predicts understanding, concern, or action outside the laboratory.
Why Empathy Needs a Multidimensional Definition
Cognitive empathy usually means constructing an informed representation of another person's perspective, thoughts, intentions, or emotional state. It is often discussed alongside perspective taking, mentalizing, and theory of mind. These constructs overlap but are not identical: inferring a belief in a story need not involve concern for the character, and identifying a facial expression is narrower than understanding the person's situation.
Affective empathy refers to emotional responsiveness to another person's state. Researchers sometimes divide it into emotional resonance, empathic concern, and personal distress. Resonance concerns shared or corresponding feeling; concern concerns other-oriented compassion; personal distress is a self-oriented aversive response when witnessing another person's difficulty. These outcomes can move in different directions, including within the same participant.
Behavioral empathy refers to actions that convey recognition, support, or care. It depends on social expectations, available communication, relationship history, emotional regulation, sensory demands, cultural conventions, and opportunities to respond. Someone can understand another person's suffering yet choose silence because an immediate response would be intrusive, uncertain, or physically overwhelming.
The multidimensional approach predates contemporary autism debates. Davis's Interpersonal Reactivity Index separated perspective taking, empathic concern, personal distress, and fantasy rather than assuming a unitary empathy score. Later questionnaires also explicitly differentiated cognitive and affective factors. Davis's original measurement study and Reniers and colleagues' QCAE development paper illustrate why subscales require interpretation in their own right.
Cognitive Empathy and Social Inference
A cognitive-empathy study might ask a participant to identify emotion from a photograph, select a mental-state word for the eye region, interpret a conversation, predict a fictional character's intention, or infer what a narrator felt in a recorded account. These tasks differ in visual processing, reading comprehension, vocabulary, attention, familiarity with the scenario, cultural expectations, and the need to guess the experimenter's preferred answer.
Lower mean scores on some cognitive tasks are a real empirical finding in the literature. Their implications are nevertheless task-bounded. A lower probability of correctly assigning a standardized emotion label does not establish that the participant is indifferent, incapable of attachment, or unable to understand a familiar partner. Familiarity with the speaker, explicit information, and the time available for reflection may change performance.
Cognitive empathy also has two directions. Autistic participants may find some nonautistic signals difficult to interpret, while nonautistic participants may misunderstand the intentions or emotional expressions of autistic people. The ability to decode the majority group's conventions is frequently the outcome being scored, so the meaning of the score depends on the design of the communicative encounter.
Distinguishing automatic inference from deliberate understanding is useful. Rapid responses to ambiguous eye cues and slower interpretations after a full conversation are different cognitive events. A person may use conscious reasoning effectively while having greater difficulty when forced to infer feelings from an impoverished stimulus under time pressure. Neither task by itself measures a fixed moral or emotional capacity.
Affective Empathy, Concern, and Personal Distress
Affective empathy research asks whether another person's emotional state triggers emotional resonance, compassion, physiological arousal, or distress. These processes should be measured separately. Experiencing another person's distress intensely does not necessarily produce the comforting behavior expected by an observer, especially if the witness is trying to regulate their own intense response.
Cusson and colleagues found a small average affective-empathy difference overall, with the effect becoming nonsignificant when restricted to higher-quality studies. Crucially, the Interpersonal Reactivity Index components pointed in opposite directions: empathic concern averaged lower in autistic samples (g = -0.59), whereas personal distress averaged higher (g = 0.67). These were specific pooled subscale findings and cannot be converted into universal claims about compassion or distress among autistic people. The 2025 meta-analysis reports both effect sizes.
A questionnaire about whether one readily comforts a stranger can combine concern with conventions governing comfort. An item about being upset by another person's pain may capture emotional arousal more strongly than constructive helping. A participant who reports high distress may be emotionally reactive yet need additional time or environmental support to act. Treating both items as equivalent indicators of reduced or increased empathy obscures the mechanism.
The distinction also matters clinically. High interpersonal distress may contribute to exhaustion, withdrawal, or uncertainty in emotionally intense situations. Lower observed responsiveness may arise from anxiety, overload, learned suppression, or a mismatch in expected social signals. These possibilities require individual assessment and cannot be inferred from autism status alone.
What the 2025 Meta-Analysis Can and Cannot Establish
A meta-analysis pools published studies but does not erase the differences in how they define, sample, and measure their target. Standardized mean differences compare group averages relative to variability within studies. The negative sign in the 2025 synthesis indicates lower average scores on the measure as coded for the autistic group; it is not a percentage of people who experience a given emotion.
The review's 226 studies offered unusual breadth, but its central observation was heterogeneity between instruments, including scales nominally assigned to the same empathy domain. The authors' meta-regressions did not identify significant moderation of empathy differences by publication year, study quality, alexithymia, verbal IQ, or age in those analyses; sex moderated the unidimensional category. These statistical results should not be misread as proof that those characteristics never matter in an individual experiment or real-world interaction. Study abstract and bibliographic record.
There is a difference between a nonsignificant pooled estimate and evidence that every autistic and nonautistic participant responds identically. Confidence intervals, measurement error, subgroup coverage, and between-study heterogeneity matter. Likewise, an average cognitive difference does not determine whether it remains after richer contextual cues or reciprocal interaction are supplied.
The review's qualitative examination of ecological and neuroimaging work mostly found minimal autistic–nonautistic differences. Such tasks are promising because some demand richer responses than isolated questionnaire items, yet they also vary in stimuli, measurement precision, sample size, and interpretation. The aggregate literature supports a nuanced map of measured tasks rather than a single deficit narrative.
The Empathy Quotient: What a Questionnaire Score Means
The Empathy Quotient, or EQ, was introduced as a self-report measure of empathy and examined in adults described under the diagnostic terminology common in 2004. The original study is historically important for understanding how autism research came to treat certain social behaviors and attitudes as a single empathy construct. It should be read in light of subsequent psychometric critiques and contemporary diagnostic language. Baron-Cohen and Wheelwright (2004).
An EQ response is a person's endorsement of a written statement, not direct access to their emotional experience or a universal test of concern. Reading a sentence, deciding what a socially expected action means, recalling examples, and interpreting ambiguous situations all contribute to the answer. Scores can be influenced by self-knowledge, cultural norms, gendered expectations, and the tendency to interpret items literally.
Because an item can incorporate both social performance and emotional motivation, it is especially important to examine whether an instrument is measuring the same latent trait in autistic and nonautistic respondents. A lower total score cannot automatically identify which component—recognition, concern, arousal, or conventional expression—produced the difference. A person should never be assigned a clinical label or judged as compassionate or unfeeling from an EQ number.
The broader COSMIN appraisal of empathy questionnaires found inadequate or incomplete psychometric evidence across its inventory of 19 self-report measures. This finding applies to research inference and to careless individual interpretation; it is not a claim that every questionnaire response lacks information. Harrison and colleagues' systematic review.
Interpersonal Reactivity Index and QCAE: Why Subscales Matter
The Interpersonal Reactivity Index provides distinct windows onto perspective taking, fantasy, empathic concern, and personal distress. Its architecture makes it possible for the same individual to report strong distress but lower spontaneous perspective taking or show a different pattern on concern. Those profiles are analytically meaningful; combining them without justification destroys the distinction the instrument was designed to preserve.
The Questionnaire of Cognitive and Affective Empathy was developed to distinguish cognitive and affective facets and to explore subcomponents within them. The existence of those factors in an original validation sample does not guarantee identical measurement properties in every language, age group, disability group, or autistic population. Cross-group interpretation requires empirical validation rather than assuming that familiar scale labels make scores commensurable.
When research reports that an autistic group scores differently on 'empathy,' the first questions should be which instrument, which subscale, what response format, and whether the analysis accounted for multiple comparisons and participant characteristics. Reproducing a headline figure without those details can transform a narrow finding into an unsupported generalization. Original IRI source, QCAE development, and cross-group psychometric appraisal.
Psychometric Validity and Measurement Invariance
A psychological measure must support the interpretation researchers intend to attach to its scores. Content validity asks whether the questions cover the intended concept; structural validity asks whether the proposed factors fit the response patterns; reliability asks how much measurement noise may be present; and construct validity asks whether the measure relates to other observations as theory predicts.
Measurement invariance asks whether an instrument functions comparably across groups. Differential item functioning can occur when autistic and nonautistic respondents with similar underlying empathy give different answers because an item invokes different experiences or interpretations. Without adequate tests, an observed score gap can reflect some mixture of trait variation and instrument behavior.
Harrison, Brownlow, Ireland, and Piovesana evaluated 19 empathy self-report measures in autistic and nonautistic adults using COSMIN criteria. Five had high-quality evidence of insufficient measurement properties, and 14 had consequential evidence gaps. The reviewers specifically emphasized absent invariance and item-functioning evidence. This is a direct limitation on claims that one group has less of the trait solely because it has a lower score. Empathy Measurement in Autistic and Nonautistic Adults.
Good practice includes preregistered construct definitions, adequately powered comparisons, psychometric tests in the groups being compared, and reporting the results for the component being assessed. Converging evidence from different measurement modes is preferable to treating one self-report inventory as a clinical verdict.
Theory of Mind Is Related to Empathy but Remains a Different Construct
Theory of mind concerns interpreting other agents' beliefs, intentions, knowledge, and perspectives. A person might understand that a friend has a false belief without experiencing the friend's frustration. Conversely, a person may feel pain or tenderness in response to another person's suffering without being able to articulate that person's mental-state beliefs in an experimental story.
Autism research has often combined theory-of-mind tasks with cognitive empathy measures. The two can inform each other, but a task testing belief inference does not independently measure concern, moral conduct, emotional attachment, or capacity for mutual support. The task's accuracy criterion can also depend on a narrow interpretation of ambiguous social events.
A scientific account of empathy should state when evidence comes from explicit mental-state reasoning, rapid social cue decoding, emotional resonance, or interpersonal interaction. Such labeling clarifies why apparently contradictory studies can both be accurate descriptions of the particular task they performed.
Emotion Recognition and the Problem of Context-Free Stimuli
Emotion recognition is often tested using photographs of facial expressions, images of eyes, short audio clips, or carefully staged scenarios. Laboratory standardization increases control over the stimulus but may strip away contextual information that people normally use to interpret emotion: known relationships, speech, shared history, gestures, physical environment, and the person's own account.
A wrong response to a photograph can reveal a difficulty with that task, ambiguity in the stimulus, unfamiliar emotion vocabulary, or a different interpretation of the portrayed expression. In many real conversations, the most valid way to know how another person feels is to ask rather than infer from facial appearance. The laboratory's official answer does not always capture the only reasonable social meaning.
Future-facing interpretation of studies should therefore emphasize ecological validity. A useful result tells us which type of information was provided, which answer was scored, how long the respondent had, and how well the task generalizes to interactions with real partners. This is particularly important when the research outcome is used to make claims about friendships, parenting, clinical care, or social competence.
Self-Report, Informant Ratings, Behavioral Tasks, and Brain Measures
Self-report questionnaires reveal what respondents say they commonly think, feel, or do. Informant ratings reveal what another person observes or interprets. Behavioral tasks reveal performance under particular experimental constraints. Brain imaging reveals associations between neural activity and task conditions. These are distinct evidence streams, and none directly reads a person's moral character or complete private emotional life.
An observer may interpret restrained facial expression as low concern, while the person reports intense emotion and deliberate attempts to avoid intrusive comfort. Another respondent may rate their own perspective-taking highly yet struggle to predict a stranger's responses. A neural signal linked to a pain-observation task can be informative without functioning as a universal biomarker of empathy.
The strongest studies triangulate these approaches, specify which discrepancy is expected, and seek explanations rather than averaging incompatible outcomes together. The absence of one stereotyped outward sign is insufficient to infer absence of the corresponding inner state.
Alexithymia and the Interpretation of Emotional Responses
Alexithymia refers to difficulties identifying and describing one's own feelings and is associated with variable awareness of emotional states. It is conceptually distinct from autism, though the two can co-occur. A person who struggles to recognize or verbalize internal emotion may find self-report empathy items difficult to answer accurately, even when they respond strongly to another person's situation.
In a 2010 functional MRI experiment, Bird and colleagues examined empathy-related responses to another person's pain. The anterior-insula response covaried with alexithymia levels rather than diagnostic autism-group membership in that sample. This showed that co-occurring individual characteristics can explain findings otherwise attributed directly to autism. It did not establish that all empathy differences always originate in alexithymia. Bird and colleagues' study.
Bird and Cook subsequently advanced an influential interpretation of emotional-processing heterogeneity in autism that emphasized the role of alexithymia. Their account remains part of an active scientific discussion rather than a substitute for measuring the relevant dimensions directly. Mixed Emotions (2013).
The 2025 meta-analysis did not find significant moderation of pooled empathy effects by alexithymia in its meta-regressions. This result is compatible with the narrower 2010 experiment if the studies differ in measures, sampling, outcomes, statistical power, and aggregation. The appropriate synthesis recognizes the original experimental evidence and the later meta-analytic result without declaring either to answer every causal question. Cusson et al. (2025).
Sensory Processing, Overload, and Emotional Responsiveness
Autistic adults vary considerably in sensory sensitivity and regulation. Loud environments, confusing social cues, sustained eye contact, intense emotional exchanges, and unpredictable physical proximity can increase the effort required to process interaction. Under overload, a person may disengage, speak less, become slow to respond, or need to leave the setting, even when they care deeply about the person present.
An emotional encounter is therefore jointly shaped by available processing resources and the interpersonal environment. A snapshot of visible responsiveness is particularly ambiguous when sensory load is high. The scientific interpretation should separate the ability to infer another person's emotion, subjective emotional response, regulation under overload, and the outward form of support.
The evidence on sensory profiles and environmental accommodations is explored in the verified encyclopedia article Autistic Sensory Processing: Hypersensitivity, Hyposensitivity, and Accommodations. The connection to empathy is contextual: sensory load may affect observed behavior, and it should not be treated as a diagnosis of a distinct empathy disorder.
Masking, Camouflaging, and Learned Social Scripts
Some autistic adults consciously monitor and adapt facial expression, gesture, tone, and conversational timing to match anticipated expectations. These strategies are often described as masking or camouflaging. Successfully using a conventional empathic phrase does not prove that the emotion was experienced, and failing to display a socially conventional phrase does not prove that concern was absent.
A masked response may require active reasoning that is effortful but accurate. Conversely, someone may withdraw from a socially demanding encounter even when they understand the other person's feelings. Researchers who equate polished social performance with empathy and awkward performance with its absence risk confounding the construct with adaptive social strategies.
Camouflaging and its reported costs are examined separately in Autism Masking and Camouflaging: Strategies, Costs, and Context. Its existence reinforces the importance of distinguishing observable behavior, underlying inference, personal experience, and environmental demands.
The Double Empathy Problem: Interaction Between Different Perspectives
The double empathy problem, proposed by Damian Milton, frames some misunderstandings between autistic and nonautistic people as mutual difficulties arising when participants have different experiential and communicative expectations. This is a theoretical account of relational mismatch. It asks researchers to examine both members of an interaction rather than assigning explanatory responsibility automatically to one diagnostic group. Milton's original 2012 article.
Several experimental findings have motivated this approach. Crompton and colleagues reported effective information transfer in chains of autistic participants under their study conditions. Morrison and colleagues examined interactions with autistic and nonautistic partners and observed outcomes that depend on pairing and context. These findings challenge an assumption that all autistic-to-autistic communication must be poorer. Crompton et al. (2020) and Morrison et al. (2020).
A crucial limitation is that successful information transfer, rapport, and first impressions are not interchangeable with cognitive or affective empathy. Two people can communicate an instruction accurately without sharing emotion, and two people can care for each other despite imperfect information transfer. A design claiming to measure double empathy must identify what psychological outcome its experiment actually elicited.
In a 2026 methodological review, Fellowes assessed 20 empirical papers associated with the double empathy problem and found that many did not directly measure cognitive or affective empathy, used interaction settings poorly suited to eliciting those constructs, or allowed only brief encounters. The review did not demonstrate an inherent empathy deficit; it established a serious measurement gap in using some interaction findings to adjudicate empathy theories. Fellowes (2026).
Thus the double empathy framework is scientifically relevant to understanding relational asymmetry, but current evidence requires construct-specific interpretation. The present article owns empathy measurement; the broader theory of reciprocal cross-neurotype interaction remains a distinct research question.
What Empathic Accuracy Studies Add
Empathic accuracy designs compare a participant's judgment of another person's emotion or thought with the target's reported or otherwise defined state. They can use autobiographical narratives, continuous emotion ratings, or specific moments in an interaction. This approach examines inference under richer conditions than a static face photograph, although it still depends on how the target's 'correct' feeling is established.
Cheang and colleagues tested nonautistic adults varying in autistic traits as they interpreted autobiographical emotional accounts from autistic and nonautistic narrators. The researchers reported lower empathic accuracy for autistic narrators in their paradigm, alongside stronger bodily emotion intensity for some of those narratives. This illustrates the value of separately measuring understanding and emotional resonance, while remaining one study with its own sample and design constraints. Cheang and colleagues' study.
The findings suggest that difficulties in interpreting another person's expression may occur across neurotypes and that lower accuracy can coexist with stronger affective arousal. They do not measure every autistic person's empathy nor demonstrate that all cross-neurotype interactions have the same profile.
Empathic accuracy tasks also need more varied narrators, languages, neurotypes, and real-life relationships. A video of an unfamiliar narrator cannot reproduce the accumulated knowledge that supports a lasting friendship, partnership, or care relationship.
Interpersonal Rapport Is a Related but Separate Outcome
Rapport reflects perceived connection, comfort, reciprocity, and coordination in an interaction. Studies often measure it through partner ratings, observers, or post-conversation impressions. It may correlate with empathy but can also arise from shared interests, compatible conversational rhythms, predictability, or common communicative conventions.
Morrison and collaborators examined real-world interaction outcomes for autistic adults with autistic versus typically developing partners, and a related study analyzed how cognitive, social-skill, and motivation measures predict social outcomes. These studies show that relationship success is not reducible to an individual's isolated test score. Partner-comparison study and social-interaction prediction study.
Better rapport within a given pairing should be described as evidence about that pairing and outcome, not an all-purpose demonstration of superior or inferior empathy. Likewise, weak rapport does not establish that either party lacks compassion. Research gains precision when it treats rapport, recognition, concern, and helping behavior as potentially related but separately measurable variables.
Autistic Adults Are a Heterogeneous Research Population
Autistic adults differ in language, intellectual abilities, adaptive functioning, sensory experience, cultural background, clinical history, age, emotional insight, support needs, and opportunities for social participation. A population label does not provide a complete description of any one person's empathy profile. Study samples often contain more verbally fluent, research-engaged adults than the full clinical population.
When a task requires abstract verbal labels, rapid keyboard responses, or extensive written self-report, it may systematically exclude or disadvantage some participants. A study of university-recruited adults without intellectual disability cannot automatically represent non-speaking adults or individuals who need substantial communication support.
The findings should also be interpreted for the characteristics actually sampled. Reporting an autism-group average without subgroup coverage, uncertainty, and recruitment method hides important external-validity limits. Clinical support categories are addressed in Autism Support Needs and Diagnostic Levels: Meaning and Limitations.
Age, Sex, Gender, Culture, and Social Learning
Developmental experience can influence how people interpret and communicate emotion. Adults may develop explicit strategies through work, family relationships, education, or repeated social feedback. However, greater exposure to social conventions does not automatically make a questionnaire invariant across age groups or explain every difference in experimental performance.
Sex and gender may shape expected empathic behavior, self-description, willingness to disclose distress, and the consequences of social nonconformity. In the 2025 meta-analysis, sex moderated one unidimensional empathy category, but the review did not establish a universal sex-based mechanism for empathy. Researchers should avoid explaining individual behavior through stereotype. Cusson et al. (2025).
Culture affects when people express sympathy, how direct questions about feeling are perceived, whether eye contact is appropriate, and whether helping involves words, practical action, or respectful restraint. English-language items and majority-culture scenarios require thoughtful adaptation before group comparisons across languages or cultural contexts.
Sampling across demographic groups and verifying measurement invariance improve scientific interpretation. Importantly, a limited meta-regression result about age or verbal IQ is not proof that these variables cannot matter in any single participant or study setting.
Clinical Diagnosis and Empathy Are Separate Decisions
Autism is classified by a developmental pattern involving social communication and interaction as well as restricted or repetitive patterns of behavior, interests, or activities, with clinically relevant developmental and functional considerations. Empathy is neither one simple substance nor an independent yes-or-no test of autism. A profile on a self-report empathy questionnaire cannot confirm or exclude autism.
The adult assessment process integrates developmental history, current functioning, observation, differential diagnosis, co-occurring conditions, sensory and communication needs, and clinical judgment. NICE CG142 recommends comprehensive assessment rather than relying on a single test. It mentions the Empathy Quotient among instruments incorporated in one possible aid for complex assessment, which does not convert that subscale into a stand-alone diagnostic criterion. NICE adult autism guidance.
A formal diagnosis, a screening result, a questionnaire score, an impression about social behavior, and a person's account of their emotional experience answer different questions. For a diagnostic framework, see Autism Spectrum Disorder: DSM-5-TR and ICD-11 Diagnostic Criteria and Autism in Adults: Characteristics, Diagnosis, and Support.
Clinicians should avoid interpreting low visible affect or unfamiliar social expressions as evidence of callousness. Differential considerations include anxiety, depression, communication disorders, alexithymia, trauma-related responses, and environmental overload. Each requires its own assessment rather than automatic attribution to autism.
What Can a Clinician Assess in an Individual Adult?
A useful clinical question identifies the context of a concern: Is the difficulty inferring a partner's emotion from subtle cues, verbalizing feelings, tolerating emotionally intense situations, showing support in an expected form, or finding mutually workable communication? These are different assessment targets, each of which calls for different evidence.
A clinician can review developmental and relationship history, ask the autistic adult about their intentions and experiences, obtain consented collateral accounts where helpful, and observe communication under accessible conditions. Structured questionnaires may generate hypotheses, but they should be interpreted in relation to functional goals and validated score meanings.
If someone struggles in a particular relationship, assessment should consider both partners' interaction patterns. If distress is marked, mental health and safety needs must be assessed independently. The goal is actionable understanding and tailored accommodation, not ranking a person's humanity by a psychological score.
Clinical recommendations should stay close to what guidelines support: a respectful partnership, clear communication, environmental adaptations, appropriately assessed coexisting conditions, and individualized support. NICE CG142 explicitly emphasizes empathic professional relationships and accommodations.
Empathy, Ethics, and Moral Concern
An experimental cognitive-empathy score does not determine conscience, ethical conduct, capacity for affection, or the ability to be a supportive partner, friend, parent, or colleague. These outcomes require additional constructs and evidence. A person may use explicit ethical principles, attentive listening, remembered preferences, or practical help to respond with care even when emotional cues are ambiguous.
The reverse inference is also unsafe: apparent fluency in conventional social interaction does not guarantee compassionate conduct. Emotion recognition, empathic feeling, socially rewarded performance, and real-world behavior can diverge in every population.
Psychological descriptions become more accurate when they name observable phenomena and plausible mechanisms: a misread facial cue, a mismatch in conversational expectations, an overwhelming emotional setting, or an unmet communication need. These descriptions support cooperation while avoiding categorical judgments about someone's inner life.
Relationships, Friendships, and Mutual Understanding
In a close relationship, common sources of friction include indirect hints, uncertainty over whether practical help or verbal reassurance is wanted, differences in time needed to process emotional information, and assumptions about what silence means. Those difficulties can be addressed by explicitly communicating preferences, checking interpretations, and allowing time to respond.
For example, one partner might express care by solving a practical problem while the other wants emotional acknowledgment first. Neither form can be declared universally correct by an empathy questionnaire. Partners can learn to name the intended response, clarify which behaviors convey comfort, and distinguish requests for support from implied criticism.
Mutual understanding also improves when each person is permitted to describe their own experience rather than having feelings inferred exclusively from appearance. This is consistent with the relational motivation behind double-empathy research, although particular communication techniques should be evaluated according to the people and outcomes involved.
A useful empirical study would examine the same couples or friendships over time, include each partner's perspective, assess the accuracy of mutual understanding, and measure whether accommodations produce meaningful improvements. Brief stranger interactions provide a narrower window than such long-term designs.
Education, Employment, and Healthcare
In classrooms and workplaces, empathy can be mistakenly equated with eye contact, facial expressiveness, rapid small talk, or spontaneous use of particular phrases. These behaviors vary across neurotypes and cultures. Evaluation should focus on the actual function required—for example, recognizing that a colleague needs assistance—and permit more than one accessible way to fulfill it.
Clear instructions, predictable turn-taking, written summaries, reduced sensory load, and direct discussion of expectations may allow empathic intentions to translate more effectively into observable cooperation. These practices are contextual accommodations; they are not evidence that a person's fundamental emotional capacity has been repaired.
Healthcare encounters add an asymmetry of professional power. NICE instructs clinicians to develop trusting, supportive, empathic relationships, use plain language, account for individual communication needs, and adapt environments. Patient responses during a rushed, overstimulating visit should not be treated as reliable tests of their concern for others. NICE adult guideline.
When an autistic adult also experiences depression, anxiety, or another mental health condition, that condition deserves appropriate individualized attention. See Autism and Depression in Adults: Recognition, Risk, and Care for autism-specific mental health assessment and support.
Should Empathy Be Trained or Treated?
The phrase empathy training is too broad to specify a valid intervention. Programs may teach recognition of emotional cues, perspective-taking strategies, interpersonal scripts, emotion labeling, response planning, or reciprocal communication. Their outcomes differ, and improvement in one trained task does not automatically transfer to compassion, relationships, or quality of life.
An evidence-based goal begins with an individual's identified difficulty and preferences. If someone wants help interpreting an indirect request, explicit language and collaborative practice may be relevant. If the problem is overwhelming emotional arousal, support for regulation and sensory accessibility may matter more than repeatedly testing facial-expression labeling.
NICE CG142 describes structured social learning options for autistic adults with identified social-interaction difficulties and emphasizes personalized assessment and accommodations. It does not establish a general need to change autistic emotional experience or endorse medications for core autism features. NICE recommendations.
Researchers should test outcomes that matter to autistic participants, including reduced misunderstanding, greater agency, safer relationships, satisfaction, and less avoidable distress. Autistic participation in choosing those endpoints improves clinical relevance and can reveal harms hidden by an exclusive focus on outward normalization.
Research Design: What Stronger Studies Need
An informative study names the component of empathy under investigation before selecting its instruments. It determines whether the claim concerns cognitive inference, affective concern, personal distress, behavioral support, empathic accuracy, or reciprocity. Its primary outcome then directly measures that concept rather than relying on an attractive but distant proxy.
Sampling should include clinical characterization, appropriate developmental and demographic information, and representation beyond convenience cohorts. Researchers should report missing data, who could not complete the task, relevant language demands, co-occurring conditions, and how group assignment was established. Differential attrition can otherwise masquerade as a psychological difference.
Psychometric work should examine content validity, measurement invariance, differential item functioning, reliability, and convergence with appropriate behavioral or ecological outcomes. Double-empathy paradigms should include the two directions of interpretation, matched conditions, opportunities for sustained interaction, and participant-reported experiences.
Analysis should report effect sizes and uncertainty, prespecified hypotheses, multiple-outcome correction where warranted, and the practical meaning of differences. A statistically detectable group mean does not justify stereotyping every person in that group.
Autistic researchers and participants can help identify what laboratory measures omit, how wording is interpreted, and whether the studied outcomes reflect everyday needs. Participatory design is methodologically valuable when accompanied by transparent measurement and sound experimental analysis.
Strength of Evidence and Remaining Uncertainty
Established: empathy is multidimensional, different instruments probe different processes, and autism research documents heterogeneous social-cognitive performance. The 2025 review provides substantial evidence that pooled effect estimates change markedly with the selected construct and method. Its affective findings make a blanket statement about absent emotional empathy scientifically indefensible. Cusson et al. (2025).
Established methodological limitation: the psychometric literature does not adequately validate many self-report empathy measures for strong cross-group comparisons. Evidence about observed average score differences must be separated from evidence about the underlying construct. Harrison et al. (2022).
Supported but context-dependent: experimental studies of autistic-to-autistic communication, mixed-neurotype rapport, and narrative empathic accuracy show that relational context and pairing can change outcomes. These studies examine particular interaction processes and cannot all be reclassified as direct tests of affective empathy. Crompton et al. (2020), Cheang et al., and Morrison et al..
Preliminary or contested: broad causal explanations based on one mechanism, a universal direction of empathic superiority in any neurotype, and claims that every double-empathy experiment measures cognitive or affective empathy. The 2026 methodological critique identifies specific unresolved measurement questions. Fellowes (2026).
Individual-level interpretation remains the decisive boundary. A clinically meaningful assessment considers the person, the situation, and the activity or support at issue. No group-level effect size is an individual diagnosis, moral judgment, or treatment prescription.
Frequently Asked Questions
Can autistic people feel empathy? Yes. Autistic people can experience concern, emotional resonance, distress, and care, with considerable individual variation. The scientific literature finds differences in particular tasks and subscales, while a general lack of affective empathy is not supported by the higher-quality subset in the 2025 meta-analysis. Cusson et al..
Why do some studies say autistic people have lower empathy? Many instruments emphasize social inference, majority-convention responses, or unidimensional questionnaire totals. Group averages on those tests may differ, but the construct being scored must be specified. The instrument itself may also perform differently across populations.
Is cognitive empathy the same as caring? Cognitive empathy concerns understanding a person's perspective or state. Caring is an other-oriented emotional or ethical response. These can be related without being identical, and a test of one cannot be substituted for a test of the other.
Does high personal distress mean high empathy? Personal distress is an emotional response to another person's suffering, but its self-oriented quality differs from empathic concern and constructive support. The two should be assessed independently.
Does the Empathy Quotient diagnose autism? No. It is a self-report instrument with important construct and cross-group validity limitations. Formal adult autism assessment integrates developmental, behavioral, functional, and clinical evidence according to established diagnostic frameworks and guidelines.
Does the double empathy problem prove that autistic and nonautistic people have equal empathy? It raises a valuable relational question and motivates interaction research. Existing studies support some context-sensitive communication findings; a 2026 review concluded that many experiments do not directly measure cognitive or affective empathy. Fellowes (2026).
Can alexithymia affect empathy questionnaire results? Yes. Difficulty identifying or describing feelings can influence self-report and aspects of emotional processing. Its contribution is variable, and studies using different outcomes do not all reach the same conclusion. Bird and colleagues (2010) and Cusson et al. (2025).
How should a friend or clinician respond to a perceived empathy difficulty? Specify the concrete misunderstanding, ask what each person intended, provide explicit communication and appropriate time, and consider sensory or emotional demands. A persistent functional or mental health concern calls for individualized, respectful assessment rather than a judgment based on stereotypes.
Conclusion
Autism and empathy research is most accurately understood as a science of measurement, interpersonal interpretation, and individual variation. Evidence for group differences is strongest in certain cognitive or unidimensional tasks, whereas average affective-empathy differences are small and not robust in the higher-quality subset of the major 2025 synthesis. Self-report instruments also face important validity problems, and experiments on cross-neurotype communication do not automatically establish results about emotional resonance.
The responsible scientific interpretation keeps each claim attached to its construct, instrument, population, and setting. This approach produces more useful findings for psychology, clinical assessment, and everyday relationships because it explains which conditions support understanding and care, and which conclusions a particular test cannot bear.