A colleague says a decision in a meeting was unfair. You understand why she would see it that way, although you do not feel her anger and may disagree with her conclusion. A questionnaire might ask whether you generally try to see other people’s points of view. A photograph test might ask you to choose a word for someone’s expression. Neither result alone tells us whether you understood this colleague accurately. Measuring cognitive empathy requires specifying whose mind is being inferred, what information is available, and what would count as a correct understanding.
Measuring cognitive empathy means assessing a person’s capacity or tendency to represent another person’s perspective, emotion or intention while preserving the distinction between self and other. It is related to, but not identical with, affective empathy, which concerns sharing or responding emotionally to another’s state. It also overlaps with theory of mind, the broader ability to attribute mental states, including beliefs and knowledge that are not emotional. This Methods & Measures guide compares self-report, standardized tasks and interaction-based accuracy measures, with attention to language, culture and the target person’s own account.
Decide which part of empathy the score should represent
Empathy is not a single observable act. Mark Davis’s Interpersonal Reactivity Index separated perspective taking from empathic concern, personal distress and imaginative identification. The perspective-taking scale asks about a habitual tendency to consider another person’s point of view. Empathic concern asks about other-oriented feeling. The subscales can be related without being interchangeable. A person may understand a rival’s motives and feel little sympathy, or feel distress for someone while misunderstanding what that person actually wants.
Lawrence and colleagues’ work on the Empathy Quotient identified cognitive empathy, emotional reactivity and social-skill factors in their psychometric analyses. That structure underscores why a total score can conceal different profiles. Yet a factor label does not make every item a pure process measure. A self-report item about social understanding requires introspection, willingness to answer candidly and a common interpretation of “understanding.” A scale may be useful for group patterns while remaining too indirect for diagnosing an individual’s real-time understanding of a particular person.
The Institute uses cognitive empathy here for other-oriented understanding, especially of perspectives and feelings, rather than for every successful social inference. Theory-of-mind tasks may involve knowing that someone holds a false belief about an object without requiring an empathic stance. An emotion-recognition task may identify a conventional facial label without revealing the person’s reasons, needs or account of the situation. Good measurement states the overlap and the limit rather than treating all social-cognition tasks as equivalent tests of empathy.
A precise claim also distinguishes tendency from ability. Someone may be capable of considering another view but choose not to do so in a rushed argument. A questionnaire may capture how often they say they try; a performance task may show what they can infer when instructed. Neither alone proves what happens in an unobserved conversation. Motivation, attention, relationship and incentives affect whether ability is used.
What self-report can and cannot establish
Perspective-taking items are economical and can cover many situations. They are useful when a study asks whether people who describe themselves as taking others’ viewpoints also report certain attitudes or behavior. Davis’s multidimensional design helps researchers avoid mistaking personal distress for other-oriented understanding. But a respondent is both the person being measured and the judge of their own skill. Confidence in being empathic need not equal accuracy.
Social desirability is an obvious concern. In many settings, saying “I consider all sides” sounds admirable. A person might answer aspirationally or infer the expected response. Conversely, someone with high standards for understanding others may rate themselves modestly. Item wording and reference group matter: “often” could mean several times a day in one household and once a week in another. Comparing groups without checking response styles, translation and item meaning can produce a spurious difference.
Self-report can measure intention or self-concept even when it is not an accuracy test. That distinction should be preserved in titles and conclusions. If a questionnaire score predicts a later willingness to ask questions, it may have practical value. It still does not show the participant correctly inferred what a particular stranger believed. An intervention that raises self-reported empathy may have changed language, norms or awareness rather than social accuracy. A stronger design combines self-report with observed behavior and an independent criterion.
Scale scores also have limits at the individual level. Internal consistency indicates that items covary, not that they are unbiased or diagnostic. Test-retest stability can show a recurring response pattern while obscuring situational change. A total score can mix cognitive and affective items. Researchers should report subscales, reliability in the actual sample and the construct each subscale is intended to represent. They should not translate one number into a moral ranking of people.
Standardized tasks sample particular cues
The revised Reading the Mind in the Eyes Test asks participants to select a mental-state word that best matches a photographed eye region. Its fixed stimuli and answers make scoring straightforward and allow comparison across participants. Yet an eye photograph is not a full person in context. Correct answers depend on reading visual cues, knowing the offered words and using assumptions about expression. A high score may indicate skill with that task without demonstrating accurate understanding of someone’s lived situation.
The vocabulary demand is especially relevant when comparing languages, ages or educational groups. A participant who cannot distinguish two English mental-state words may be scored as poor at social inference even if they would understand a person’s situation in their own language. The task’s original development can support its use for a defined question, but validity must be reconsidered when stimuli, words or populations change. Translation is not only replacing labels; it requires checking whether expressions and mental-state categories carry comparable meanings.
A film-based task such as Dziobek and colleagues’ Movie for the Assessment of Social Cognition adds moving social context and questions about characters’ mental states. That may resemble everyday inference more closely than cropped photographs, but it also adds memory, language comprehension, attention to plot and cultural familiarity with the interaction. A missed answer may reflect one of these demands. A richer stimulus is not automatically a purer measure. Researchers need suitable control questions and should examine which errors the task actually captures.
Standardized tasks are helpful for testing specific predictions. A researcher might compare accuracy when only facial cues are available with accuracy when the person’s words and situation are available. If performance changes, the added context matters. But neither format should be called the complete measure of cognitive empathy. Real interaction includes feedback, the chance to ask, and the possibility that the target revises their account. Static tests usually remove those features for control.
Accuracy requires a defensible target
To know whether someone understands another person’s perspective accurately, a study needs a criterion. One option is to ask the target person what they were thinking or feeling at a particular moment, then compare an observer’s inference with that account. This is closer to interpersonal accuracy than matching a consensus label for a photograph. Yet the target’s self-report is not infallible. People may be uncertain, revise their interpretation or conceal information. The criterion is an informed reference point, not direct access to a mind.
Another option is to test understanding of a known informational asymmetry. For example, the observer sees one object while the target cannot. If the observer predicts the target’s choice using only what the target knows, the task probes self-other distinction. Todd and Tamir’s review describes how people’s own knowledge can remain an anchor when they infer others’ minds. A task can reveal this egocentric bias, but performance also depends on inhibition, memory for who saw what, and motivation to use the target’s viewpoint.
Perspective taking and perspective getting should be distinguished. Eyal, Steffel and Epley found that trying to imagine another person’s perspective did not reliably improve accuracy in their studied contexts, whereas obtaining information from that person could. This finding challenges a common assumption that effortful imagination is the same as accurate understanding. It does not mean imagination is useless for every relational goal. It means an accuracy measure should reward correspondence with the target, and a practical intervention may need conversation rather than solitary inference.
Interaction-based studies can record questions asked, paraphrases offered and whether the target feels understood. Each outcome addresses a different construct. Asking questions may show curiosity; correct answers may show knowledge; a target’s feeling of being understood may reflect tone, trust and relationship history as well as cognitive accuracy. A complete study can use several outcomes and avoid substituting one for another. A person may infer correctly while communicating poorly, or communicate warmly while making an incorrect assumption.
Keep affective empathy visible but separate
Understanding another’s anger is not the same as feeling anger. Affective responses can sometimes motivate attention to a person’s situation, but they can also create personal distress that pulls attention toward oneself. Davis’s subscales make this distinction operational. A measure of cognitive empathy should not classify a person as incapable merely because they report less emotional contagion. Conversely, feeling moved by a story does not prove a correct inference about its narrator.
Emotion-recognition tasks occupy a middle ground. They may require cognitive labeling of an affective cue, but the measured skill is narrower than understanding what caused the feeling or what the person needs. A smile can express pleasure, politeness, tension or irony. The best answer on a standardized test reflects the task’s scoring convention. In natural interaction, asking the person may be more informative than confidently decoding a face. Researchers should state whether their target is cue classification, inference from context or responsive conversation.
It is tempting to aggregate all these measures into a single empathy index. That may be useful only if the researcher has a theory of why the components should combine and evidence that the composite predicts something meaningful. Averaging a self-reported tendency, a photograph score and an affective reaction can hide disagreement among methods. Divergence can be informative: it may reveal that someone feels concern but relies on inaccurate assumptions, or is perceptive on a test but seldom chooses to engage.
Common confounds in comparison studies
Language ability affects both questionnaire comprehension and task responses. A researcher should measure or accommodate vocabulary and reading demands when comparing groups that differ in language exposure. Visual perception affects performance on photographic stimuli. Working memory and attention affect a film task. General response speed matters if a task uses time limits. These are not minor nuisances when they differ systematically between groups; they can change what a score means.
Culture and context shape the available cues. A facial expression or conversational convention may be familiar to one participant and unfamiliar to another. An apparent deficit can arise when the test’s assumed norm is narrow. This does not mean no cross-cultural comparison is possible. It means the stimuli, translation and criterion should be examined in the populations studied, with enough room for more than one culturally intelligible route to understanding.
Target selection matters too. Inferring the perspective of a close friend is different from inferring that of a stranger. A target who shares a participant’s background may be easier to understand, but familiarity can also invite projection. Todd and Tamir’s account of egocentric mentalizing explains why one’s own experience often serves as a starting point, while also emphasizing conditions that amplify or attenuate that bias. A good design samples targets and situations rather than assuming one vignette represents general ability.
Repeated exposure can create practice effects. Once a participant learns the vocabulary or style of an eyes test, a higher retest score may reflect familiarity rather than increased understanding. If an intervention claims to improve cognitive empathy, it should use suitable control groups and preferably alternative test forms or novel targets. Outcome assessors should be unaware of condition where possible. The intervention’s content should not simply teach answers to the post-test items.
Interpret scores at the level supported
A transparent report names the construct, instrument, target population, language and scoring rule. It states whether the result concerns self-reported perspective-taking, accuracy on a specific social-cognition task, or correspondence with a target person’s own account. It reports uncertainty and recognizes that a modest average group difference may coexist with substantial overlap among individuals. A task score is not a diagnosis, a character judgment or a direct map of how someone behaves with a particular person.
Evidence from multiple methods can strengthen a claim when the methods make different errors. If a person says they try to understand others, accurately infers several targets’ perspectives, and asks clarifying questions during interaction, the convergence is more persuasive than any one measure. But convergence should not be forced. A weak correlation may reveal measurement noise or a real distinction between intention, ability and use. The next study should test that distinction rather than silently choosing the most flattering score.
For an intervention, preregistering the main outcome and timing helps prevent a selective account. A program may increase knowledge of empathy language immediately but not improve later interaction. Another may improve willingness to ask questions without changing photograph-test accuracy. Both results can be valuable if described precisely. Claims about lasting relationships or professional care need outcomes in those settings, measured after enough time for change to matter.
The Institute’s measurement framework is synthesis of the cited construct and task literature. It is not a validated scale. Its practical instruction is to separate what the observer says about themselves, what they choose on a standardized test, what they infer about a real target, and how they respond when the target can correct them. Those are linked facets of social understanding, not duplicate readings of a hidden meter.
What the evidence does not show
No single questionnaire, eye-region photograph or film task proves that a person understands others accurately in every situation. Self-report measures tendencies and self-concepts under particular wording; task scores depend on stimuli, vocabulary, attention and scoring conventions. The target person’s account is essential for many accuracy questions but is itself a report, not perfect access to private experience.
The cited studies do not show that cognitive empathy and affective empathy always rise and fall together. They do not show that deliberately imagining another person’s perspective is equivalent to learning what that person thinks. Nor does a theory-of-mind task that tests beliefs automatically measure empathic concern. A low score should not be turned into a diagnosis, a moral defect or a universal account of someone’s relationships.
The review of egocentric mentalizing helps specify one confound: self-information can bias inference about others. It is not a validation study for every cognitive-empathy instrument. Claims about the accuracy of a particular test require studies of that test, its scoring and its intended population.
Return to the colleague
To test whether you understood your colleague, a researcher could ask what she meant by unfair, what information she had, and whether your account matches hers. A general questionnaire might predict that you would try. An eyes test might show skill with certain visual cues. Neither replaces the conversation. Your emotional reaction also matters to the relationship, but it is a different outcome from accurately representing her view.
Measuring cognitive empathy well means allowing social understanding to be both fallible and improvable. The strongest design asks a narrow question, chooses a criterion suited to it, and checks the language, context and motivation that could produce the same score for different reasons. The result is a bounded account of how someone understands another mind, not a single label for the whole person.
Research references
- Davis, M. H. (1983). Measuring individual differences in empathy: Evidence for a multidimensional approach. Journal of Personality and Social Psychology, 44(1), 113–126. View source
- Lawrence, E. J., Shaw, P., Baker, D., Baron-Cohen, S., & David, A. S. (2004). Measuring empathy: Reliability and validity of the Empathy Quotient. Psychological Medicine, 34(5), 911–924. View source
- Baron-Cohen, S., Wheelwright, S., Hill, J., Raste, Y., & Plumb, I. (2001). The “Reading the Mind in the Eyes” Test revised version: A study with normal adults, and adults with Asperger syndrome or high-functioning autism. Journal of Child Psychology and Psychiatry, 42(2), 241–251. View source
- Dziobek, I., et al. (2006). Introducing MASC: A movie for the assessment of social cognition. Journal of Autism and Developmental Disorders, 36(5), 623–636. View source
- Eyal, T., Steffel, M., & Epley, N. (2018). Perspective mistaking: Accurately understanding the mind of another requires getting perspective, not taking perspective. Journal of Personality and Social Psychology, 114(4), 547–571. View source
- Todd, A. R., & Tamir, D. I. (2024). Factors that amplify and attenuate egocentric mentalizing. Nature Reviews Psychology, 3, 164–180. View source