Evidence status: mixed results “Empathy” includes emotional sharing, concern, perspective understanding and accurate inference, and the measures are not equivalent. A 2019 meta-analysis found that self-reported cognitive empathy explained about one per cent of performance variance on behavioural tasks. Other work finds weak or conditional links among empathic feeling, facial mimicry and recognition accuracy. This is a version 0.1 research draft, with evidence reviewed to August 2026.
How Everyday Psychology Takes Shape · Season Two: How Do We Read Other People? · Article 12
“I am a very empathetic person” can mean several things. The speaker may easily catch another person’s distress, care deeply, imagine different perspectives, or accurately know what another person thinks. Ordinary language places these meanings together. Research cannot assume they are one capacity.
A person can care intensely and misunderstand what help is wanted. Someone can infer an emotion accurately and choose not to help. Feeling similar to a target is not the same as understanding the target’s particular experience. Unless the target is specified, “increase empathy” becomes a morally attractive programme that is difficult to test.
The final article of this season is not an argument against the moral value of empathy. It proposes a stricter relation. Care makes understanding worth pursuing. Accuracy tells us whether the understanding is right. Responsiveness and correction determine whether mistakes can be repaired. These achievements need one another, but they do not arrive automatically as a package.
How many things does psychology call empathy?
Judith Hall and Rachel Schwartz examined hundreds of published studies and concluded that empathy had been used so promiscuously, with such diverse conceptual and operational definitions, that cumulative science was threatened. Common distinctions include affective empathy or experience sharing; empathic concern; cognitive empathy, perspective understanding or mentalising; and empathic accuracy—how correctly one person infers another’s thoughts and feelings.
Questionnaires ask whether people consider themselves good at understanding others or inclined to take another perspective. Behavioural tasks present faces, stories or recordings and compare inferences with the target’s report or another task criterion. One relies on self-knowledge and the other on situated performance. Sharing a label does not guarantee close agreement.
The empathic-accuracy tradition associated with William Ickes often asks perceivers to watch recordings of natural interactions, infer a target’s thoughts or feelings at particular moments, and compare those inferences with content supplied by the target. This is closer to daily life than one static expression, but it still has limits. A target’s report can be incomplete, similarity coding requires rules, and perceiver accuracy depends on how clearly the target is willing and able to express.
Accuracy is not a fixed radar possessed by the judge alone. It is formed by perceiver, target, relationship and information conditions.
Why self-rated understanding barely predicts task performance
In 2019, Brett Murphy and Scott Lilienfeld pooled 85 studies with a total of 14,327 participants, comparing self-reported cognitive empathy with behavioural task performance. Self-report accounted for only about one per cent of performance variance, and the association was not meaningfully different from the association involving self-reported affective empathy. The authors concluded that using self-report as a proxy for cognitive-empathy ability presents serious problems.
This does not make the questionnaires empty. They may measure identity, willingness to attempt perspective taking, or the importance a person assigns to empathy. The error is to call the score an ability and infer successful understanding from it.
A 2021 meta-analysis of facial mimicry gives another reason for restraint. Mimicry showed a weak positive association with empathy dispositions but no significant overall association with facial emotion-recognition ability. Effects varied with task conditions. Bodily following of another face may participate in social coordination without serving as a single mechanism for accurate mind-reading.
Together, these findings reject a linear model in which stronger feeling guarantees better understanding. Intense sharing can redirect attention towards one’s own distress. “I would be devastated in that position” may project a reaction over someone whose experience is different.
Even behavioural tasks must be interpreted carefully. Recognising an emotion label from a standard face, inferring a speaker’s changing affect in a video and understanding a partner’s unspoken concern are related but non-identical performances.
Accuracy does not automatically create benevolence
Empathic accuracy is often treated as a moral ability, but an accurate model can support help, negotiation, sales or manipulation. A person who knows that someone fears losing a relationship can offer reassurance or exploit the fear. Accuracy means that the model is closer to the target. It does not determine the purpose for which the model is used.
Conversely, care can harm when it is inaccurate. Arranging help that was not wanted, presuming that a form of life must be pitiable, or continuing after the person says “that is not what I need” can remove agency in the name of empathy.
We do not have to wait for perfect understanding before acting. Emergencies demand help under incomplete information. A better requirement is revisability: begin with low-risk support, observe the response and allow refusal. Part of the ethical quality of empathy lies in its capacity to hear “you have misunderstood me”.
Care and accuracy can also operate at different levels. A general concern for another group may be morally productive while offering little knowledge of one particular member. The concern should motivate contact, not authorise assumptions.
Neurodiversity makes a one-way ability model especially problematic
Empathy language is often used to describe supposed deficits in autistic people or other groups, although different measures do not capture one unified capacity. Social misunderstanding can also be mutual: people with different communicative and experiential structures may find one another difficult to predict. Treating the majority group’s display conventions as the only standard records relational mismatch as a defect located in one party.
This series does not provide clinical diagnosis, and no single task should be used to decide whether an individual “has empathy”. Performance on gaze or standard-expression tasks cannot represent all caring, moral conduct or understanding within familiar relationships.
The same precision helps with AI. A system that generates warm, timely language can be experienced by users as empathic. That shows that language and timing have created an effective social signal; it does not establish human-like emotional experience inside the system. Evaluation should ask specifically whether the tool identified a need, provided useful support, acknowledged uncertainty and accepted correction, instead of awarding the global trait “empathetic”.
A more reliable daily structure
Understanding another person can be divided provisionally into four movements. First, notice that the person may have an experience unlike your own. Second, label the explanation formed from face, behaviour and context as a hypothesis. Third, obtain information through questions, listening and observation over time. Fourth, revise action in response to feedback.
These movements will not always be sequential and should not become a conversational script. Their value is to separate achievements that are often merged: I care; I inferred; I checked; I changed. A relationship can continue with gaps, but the gaps should be visible when consequences are high.
“I do not know” must remain available. Sometimes the other person does not wish to disclose. Sometimes the perceiver lacks background. Sometimes the two use emotion words differently. Uncertainty is not a shameful failure of empathy. Pretended certainty closes correction.
A person who repeatedly corrects us is also providing evidence about the reliability of our method. Respect means updating rather than treating each correction as an exception to an otherwise perfect self-image.
Understanding is not possession of another person’s mind
This season began with the spotlight effect and transparency, moved through faces, behavioural slices, attribution, conversation, emotion and text, and returned to one shared structure: we do not directly enter another person’s experience. We use cues, form provisional models and revise them in interaction.
The other person is not a passive ground truth. They choose to express, conceal, clarify and change. Understanding is therefore not a one-way copy of an inner state. It is a relational achievement that can succeed or fail, and its accuracy depends on the information conditions both people form.
The version 0.1 conclusion is that more empathic feeling or a higher self-rating cannot guarantee accurate understanding. Care still matters because it influences whether we attend and accept obligations to help. Accuracy requires appropriate evidence. Respect requires giving the person room to correct and refuse. Mature empathy is not increasing confidence that we can see inside other people. It is keeping our understanding revisable without reducing our concern.
Primary research sources
- Hall, J. A., & Schwartz, R. (2019), “Empathy Present and Future”.
- Murphy, B. A., & Lilienfeld, S. O. (2019), “Are Self-Report Cognitive Empathy Ratings Valid Proxies for Cognitive Empathy Ability?”.
- Ickes, W. (1993), “Empathic Accuracy”.
- Holland, A. C., O’Connell, G., & Dziobek, I. (2021), “Facial Mimicry, Empathy, and Emotion Recognition”.
Series navigation: How Everyday Psychology Takes Shape — Season Two overview
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.