Empathic Accuracy — How To Actually Understand What Someone Feels

The term empathic accuracy was coined and systematically studied by William Ickes, a social psychologist at the University of Texas at Arlington, beginning in the late 1980s. His central contribution was methodological before it was theoretical: he figured out how to measure something that had previously been treated as unmeasurable.

Prior to Ickes's work, empathy research was almost entirely self-report based. Researchers would give people surveys asking how empathetic they are, or how well they understood someone after a brief interaction. The problem is obvious in retrospect: self-assessed empathy and actual accuracy are substantially uncorrelated. People who are confident they understood someone often didn't. People who were uncertain sometimes did. The feeling of understanding and the fact of understanding are different things.

Ickes developed what he called the "dyadic interaction paradigm." Two strangers (or couples, or friends, depending on the study) interact for several minutes while being videotaped. Then each person watches the recording alone, pausing it at any moment when they were aware of having a specific thought or feeling. They record what they were thinking or feeling at that moment. Then the recordings are compared: what did Person A think Person B was thinking or feeling at moment X, versus what did Person B report actually thinking or feeling at moment X? The resulting score is an index of empathic accuracy — how often the inference matched the report.

The average scores in these studies consistently land between 20 and 35 percent for strangers, rising to 50 to 60 percent in long-term intimate relationships. That rise sounds like progress. It is. But it also means that even in close partnerships, people misread their partners' inner states roughly 40 to 50 percent of the time. And critically, confidence in one's accuracy doesn't track with actual accuracy. People are frequently wrong with high confidence.

Ickes also documented systematic patterns in the errors. People aren't wrong randomly. They're wrong in predictable directions: they overattribute their own emotional reactions to others (projection), they fill gaps in their reading with culturally scripted emotional sequences (if someone looks like this, they must feel that), and they interpret ambiguous cues in ways that serve their own relational interests or anxieties.

The distinction between projection and attunement is the conceptual heart of empathic accuracy research, and it maps onto a much older psychoanalytic distinction that is worth retrieving.

In psychoanalytic usage, projection refers to the attribution of one's own internal states — feelings, desires, conflicts — to an external person or object. The projector doesn't experience themselves as projecting; they experience themselves as perceiving. The fear or anger or desire feels as if it belongs to the other person. This is different from accurate perception of what the other person actually feels.

The everyday form of this is not pathological, it's just cognitively routine. When you try to understand what another person is feeling, you have to build a model of their inner state. The fastest raw material for that model is your own inner state in what seems like a comparable situation. Your brain uses you as the sample. It says: if I were in that situation, I'd feel X; therefore, they probably feel X. This is efficient and often approximately right, but it breaks down systematically in two conditions: when the other person's experience differs significantly from what yours would be in that situation (due to different history, different values, different physiological baseline, different relationship to the situation), and when your own emotional activation is high enough to dominate the signal.

Attunement is the alternative — orienting toward the other person as the primary source of information about their inner state, rather than using yourself as the primary model. This is what therapists who demonstrate high empathic accuracy actually do differently. Research comparing high-accuracy and low-accuracy therapists by Arthur Bohart and Leslie Greenberg shows that high-accuracy therapists ask more questions early, revise their hypotheses more readily when disconfirming information arrives, attend closely to nonverbal and paralinguistic cues (pace, tone, pausing, word choice) alongside content, and resist premature closure on their reading — they stay in a state of productive uncertainty rather than moving quickly to a fixed interpretation.

This has a direct parallel in how scientists distinguish hypothesis-testing from hypothesis-generating modes. Low empathic accuracy often looks like hypothesis-testing: you form an early model of what the person feels and then process subsequent information through it. High empathic accuracy looks more like hypothesis-generating: you hold your interpretation loosely, keep generating new possibilities as you take in more information, and treat the other person as the authority on their own inner state.

Empathic accuracy varies substantially across individuals, and the sources of that variation are now reasonably well-mapped.

The facets of personality most predictive of high empathic accuracy are not what common sense would suggest. Agreeableness — being warm, cooperative, and interpersonally oriented — shows only weak correlations with empathic accuracy. The person who most wants to understand others is not necessarily the person who most accurately does. What predicts accuracy more robustly is openness to experience (the willingness to entertain complexity and revise models) and low neuroticism in threatening social contexts (the capacity to remain cognitively open rather than defensively closed when interpersonal information is challenging).

Attachment style also predicts empathic accuracy, but in a counterintuitive pattern documented by Jeffrey Simpson and colleagues. Securely attached people show high empathic accuracy across the board. Anxiously attached people show high empathic accuracy for negative emotional content and low accuracy for positive content — they're attuned to threat but miss warmth. Avoidantly attached people show a motivated inaccuracy pattern: when the interaction becomes emotionally intense, their empathic accuracy decreases, apparently because accurate reading of emotional distress would require an approach response that conflicts with their habitual distancing. They unconsciously stop reading accurately when accuracy would require closeness.

This motivated inaccuracy finding is important because it suggests that low empathic accuracy is not always a failure of attention or skill. Sometimes it's a defense. People are sometimes inaccurate because accuracy would be uncomfortable — it would require seeing something they'd rather not see, or feeling something they'd rather not feel, or doing something they'd rather not do. This means that improving empathic accuracy is not only a skill-building project. It's also an emotional regulation project and, for some patterns, a relational healing project.

Gender differences in empathic accuracy have been studied extensively, with mixed results. Women show higher self-reported empathy than men in virtually every study. They show higher empathic accuracy in some studies but not consistently across conditions. The meta-analytic picture, reviewed by Ickes and his colleagues, suggests that actual accuracy differences between men and women are small and context-dependent, while the confidence differential is larger: women are more likely to hold their empathic readings loosely and check them, while men are more likely to hold early readings with higher confidence — which can translate into lower accuracy precisely because overconfident readings get revised less.

Understanding what degrades empathic accuracy is as practically useful as understanding what improves it.

Cognitive load. When people are cognitively busy — distracted, multitasking, under time pressure — their empathic accuracy drops significantly. The capacity to read another person accurately requires attentional resources, and those resources are finite. This has direct implications for how consequential conversations are conducted. A conversation about something emotionally significant, held while also checking a phone, managing another task, or rushing, is a conversation in which accuracy is structurally compromised.

Emotional activation. When your own emotional arousal is high, your reading of others becomes less accurate. The amygdala activation associated with anger, fear, or high stress narrows the perceptual field and biases interpretation toward threat-consistent readings. Paul Ekman's research on "hot cognition" shows that emotionally activated people both over-read threat signals and miss non-threat signals in others' expressions. This is precisely the context — a heated argument, a high-stakes confrontation — in which accurate reading of the other person is most urgently needed and most difficult to achieve.

Stereotype salience. When category membership is salient — race, gender, political affiliation, class — empathic accuracy toward outgroup members drops, and drops substantially. The work of Susan Fiske and colleagues on "social neuroscience of prejudice" documents that when outgroup members are categorized rather than individuated, the neural circuits associated with mentalizing (thinking about others as subjects with inner lives) show reduced activation. You can't accurately read the inner life of someone you're not experiencing as a full subject.

The illusion of understanding in close relationships. Counterintuitively, familiarity can decrease empathic accuracy for specific emotions. The "closeness-communication bias," documented by Boaz Keysar and colleagues, shows that people are more confident and less accurate when communicating with close others than with strangers. The mechanism appears to be that familiarity creates an illusion of transparency — I know this person so well that I know what they mean — which short-circuits the careful attention that strangers require. The result is that long-term partners sometimes read each other less accurately than strangers read each other, for the same reason that you stop seeing the furniture in a room you've walked through a thousand times.

Perspective-getting rather than perspective-taking. Nicholas Epley and colleagues have documented that simply imagining yourself in another person's situation — the standard advice for improving empathy — does not reliably improve empathic accuracy. This is because imaginative perspective-taking draws on your own experience as the primary source, replicating the projection problem. What does improve accuracy is perspective-getting: actually asking the other person how they feel, what they're thinking, what the experience is like for them. This seems obvious, but it is practiced far less frequently than imagining, because asking feels presumptuous or vulnerable, while imagining feels like sufficient effort.

Slowing interpretation. The practical analog of holding hypotheses loosely is slowing the moment at which your reading of someone hardens into certainty. High-accuracy empathizers show a consistent pattern in Ickes's research: they generate multiple possible interpretations of an ambiguous cue before settling on one, and they settle later — they tolerate uncertainty about what someone feels for longer before resolving it. Low-accuracy empathizers resolve faster and generate fewer alternatives before resolving. The exercise, in practice, is to notice when you've formed a reading of what someone feels, and deliberately ask: what else could be going on? Not to be indecisive, but to stay open long enough to receive more information.

Attending to nonverbal channels with disciplined attention. Research on which cues most reliably signal accurate emotional content consistently shows that vocal prosody — the pace, rhythm, volume, and tone of how someone speaks — is a more reliable indicator of emotional state than facial expressions, which are more subject to social display rules and deliberate management. Paul Ekman's work on microexpressions (involuntary facial movements lasting 1/25 to 1/5 of a second that reveal emotional states even when the person is trying to conceal them) is well-known, but the practical take from the broader research is less about exotic skill and more about basic attentional discipline: slow down and notice how someone is saying something, not just what they are saying.

Accuracy-motivation. Ickes's research shows that empathic accuracy can be increased simply by increasing the motivation to be accurate. In studies where participants are told that accuracy is important and will be measured, their scores increase compared to uninstructed conditions. The practical equivalent is the deliberate commitment to accuracy in high-stakes conversations — explicitly reminding yourself that your reading might be wrong and that getting it right matters. This sounds simple because it is. It's also underused.

The consequences of empathic accuracy and inaccuracy scale with power. The more power a person has to affect others' lives, the more their empathic accuracy — or inaccuracy — matters.

Research by Dacher Keltner at Berkeley has documented what he calls the "power paradox": as people acquire power, their empathic accuracy tends to decrease. The causal mechanism involves multiple pathways. Power reduces the need to attend carefully to others (because your outcomes depend less on accurately reading them), increases the salience of your own emotional state, and is associated with greater abstraction in thinking about other people. Keltner's work, and related research by Michael Kraus, shows that lower-class individuals show higher empathic accuracy than upper-class individuals across a range of tasks — apparently because their outcomes depend more directly on accurately reading the more powerful people around them.

The implication is structural: the people who make the largest decisions about the largest numbers of people tend to be, by virtue of their position, operating with degraded empathic accuracy. This is not inevitable — it describes average tendencies, not fixed outcomes — but it means that institutional design needs to actively counteract this tendency rather than assuming that wise leaders will simply be accurate.

In judicial contexts, empathic accuracy failures are documented in sentencing research. Judges who do not individuate defendants — who process them as category instances rather than as subjects — show bias patterns that can be traced to inaccurate empathic readings. The "defendant is a stranger" effect, in which judges show lower empathic accuracy toward people dissimilar to themselves, tracks with longer sentences for minority defendants in multiple studies.

In policy contexts, the consequences of low empathic accuracy are less tractable because they're less visible. Programs designed with projection rather than genuine attunement tend to address the problems that the designers imagine the recipients have, rather than the problems they actually have. The history of poverty policy, urban planning, mental health care, and international development is partly a history of well-intentioned interventions built on systematically inaccurate readings of the inner lives of the people they were meant to serve.

If empathic accuracy is a trainable capacity — and the research says it is — then it is also a civilizational resource. Not metaphorically. Literally.

Most of what we call political and social conflict involves people making decisions about each other based on models of each other that are wrong. Not evil models, necessarily. Just wrong. The conservative voter and the progressive voter typically have profound misunderstandings of what motivates the other. The misunderstanding is not symmetrical — research by Jonathan Haidt and colleagues shows that both sides misconstrue the other's values, but the pattern of misconstrual differs. But the aggregate effect is that large numbers of people are making consequential decisions — about whom to elect, which policies to support, which groups deserve resources — based on projected models of the other side's inner life that are substantially inaccurate.

If empathic accuracy were higher — not perfect, not unlimited, but meaningfully higher than the current baseline — the category of "those people" would begin to dissolve. Because "those people" is a low-empathic-accuracy category. It's a category you can maintain only by reading the inner lives of its members poorly. Once you read them accurately, they stop being those people. They become specific people with specific inner lives that you have specific, accurate information about. And it is very hard to sustain generalized contempt for people whose specific inner lives you have accurate information about.

This is not naive. It doesn't dissolve genuine conflicts of interest. But it removes a very large layer of manufactured conflict that is sustained by nothing more than mutual projection. The problems underneath — of resources, of power, of competing values — are real but negotiable. The projected layer — the stories we carry about what the other side wants and feels and believes — is not real, and it forecloses negotiation before it begins.

Law 1 says we are human. Empathic accuracy is the practice of acting on that claim at the level of one actual person in front of you. It is the practice of insisting that your model of another person be built on what they actually feel rather than on what you'd feel in their place. That insistence, practiced at scale, changes who can be seen as human. And who can be seen as human changes everything.

The Accuracy Check: After any significant conversation in which you formed a read on what someone was feeling, check it. Ask directly: "I thought you seemed [frustrated / worried / excited]. Did I get that right?" Track how often you're accurate versus off-base, and in which direction you tend to err.

The Three Alternatives: When you catch yourself confident about what someone is feeling, stop and generate two alternative readings before acting on your first one. This is not to create endless uncertainty but to widen the aperture enough to receive disconfirming information if it's there.

The Nonverbal Audit: In your next important conversation, spend the first two minutes attending specifically to how the person is speaking — pace, hesitation, volume, tone — rather than only the content of what they're saying. Notice what additional information this channel provides. Notice when it diverges from the verbal content.

The Direct Ask: Replace "I imagine you must be feeling..." with "How are you actually feeling about this?" Practice treating the other person as the authority on their own inner state rather than a subject of your imagination. Notice how often the answer surprises you.

The Projection Inventory: Choose a person you feel you understand well — a close partner, family member, or colleague. Write down what you believe they are currently most stressed about, what they want most from you, and what they feel most proud of. Then ask them directly about each. Compare. Use the gaps not to judge your accuracy but to identify your projection tendencies.

Empathic accuracy is not about achieving perfect understanding. It is about choosing accuracy over convenience, revision over certainty, and the other person's actual inner life over the one you've constructed for them. That choice, scaled, is the practice of recognizing other people as real. And the world organized around that recognition looks very different from the one we have.