Are smart people more empathetic? What the meta-analyses actually measured

0 The short answer

Smarter people score somewhat higher on tests that ask them to read other people's thoughts and feelings, but the link is small, and it says much less about whether they care. Across the meta-analyses, correlations between intelligence and performance tests of emotion recognition, mental state reading and interpersonal sensitivity run from 0.19 to 0.24, which leaves about 94 to 96 percent of the differences in those scores unexplained by IQ (our arithmetic). The evidence on caring, meaning affective empathy and prosocial behavior, comes mostly from questionnaires, and how empathic people say they are barely tracks how well they perform on mindreading tasks. These are group statistics, and they describe no one in particular.

1 What Do Cognitive Empathy, Affective Empathy and Emotional Intelligence Mean?

Empathy is not one ability, and the question of whether smart people have more of it has no single answer until the word is split into the parts that researchers measure. In their survey of the neuroscience of the field, Zaki and Ochsner describe two empathic sub-processes: sharing the internal states of others, and explicitly considering those states. The first is usually called affective empathy, because the observer feels something in response to the other person. The second is usually called cognitive empathy, or mindreading, because the observer works out what the other person thinks or feels without necessarily feeling it. Murphy and Lilienfeld define cognitive empathy as the ability to accurately recognize or infer the thoughts and feelings of others, which is the sense used in most of the studies below.

Two neighbors need separating from both. Prosocial behavior is what a person does, such as helping or sharing, and it can occur with or without strong feeling. Emotional intelligence is a family of constructs that includes the ability to perceive, use, understand and manage emotions. The meta-analysis by Olderbak, Semmler and Doebler organizes ability emotional intelligence into four branches: perceiving emotion, facilitating thought using emotion, understanding emotions and managing emotions. The page on emotional intelligence versus IQ covers that construct and its trait versions in full, so this page uses it only where it shares evidence with empathy.

The second distinction matters even more, because the same word is attached to three different kinds of instrument.

• Performance tests have correct answers. A person looks at a photograph, a video or a vignette and chooses what the other person is feeling or thinking, and the score is the number of correct choices. The Reading the Mind in the Eyes test, in which a person determines mental states from photographs of pairs of eyes, is the most used example, and its format is described in Peterson and Miller.
• Questionnaires ask people to rate themselves. The Interpersonal Reactivity Index, used in the prosocial behavior study discussed later, has scales for perspective taking and for empathic concern, among others. A questionnaire score is a belief about oneself, not a performance.
• Behavior is observed or reported conduct, such as helping a stranger, and it is the hardest of the three to measure well.

Intelligence tests are performance tests too. They have correct answers, and they are scored against norms, as the page on what IQ measures and the page on the CHC model explain. That shared format is the first reason a performance test of empathy might correlate with IQ more than a questionnaire would, and later sections test that reasoning against the data. The popular label "empath" is not a construct in any of the studies reviewed here, and it is left aside. For the other extreme of the question, the page on psychopathy, narcissism and IQ covers empathy deficits in those traits, and the page on autism and IQ covers social cognition in autism, so neither is repeated here.

2 What Do the Studies Show, Side by Side?

Read by measure, the evidence splits cleanly: performance tests of understanding others correlate modestly with IQ, questionnaires of empathy correlate with those tests hardly at all, and the evidence on caring is a single set of self-reports from one sample. The table sorts the main studies by design, sample and what was measured. Sample sizes are stated before each conclusion, as the studies report them, and the sections that follow open each row.

Three features of the table shape everything below. The correlations between intelligence and performance tests are small, about 0.19 to 0.25, except for one corrected value and one latent value that come from different methods. The results depend on how each side was measured, as the Wendt row shows most sharply. And the rows that speak to caring, as distinct from reading, are the few at the bottom, so the page treats them with correspondingly more caution.

3 What Did the Meta-Analysis of the Eyes Test Find?

The best-known meta-analysis found a small positive correlation of 0.24 between intelligence and the Reading the Mind in the Eyes test, with verbal and performance abilities contributing equally. The paper, Eyes and IQ by Baker, Peterson, Pulos and Kirkland, appeared in Intelligence in 2014. Its abstract notes that the test had been used as a measure of mental state understanding in over 250 studies while its relation to intelligence was seldom considered. The analysis of 77 effect sizes with 3,583 participants revealed a small positive correlation, r = 0.24, with no difference between verbal and performance abilities. The authors conclude that intelligence plays a significant role in performance on the test.

An earlier small study pointed the same way. Peterson and Miller gave a two-subtest Wechsler Abbreviated Scale of Intelligence, a face-processing task and the Eyes test to 42 college students. Their abstract reports that verbal IQ contributed significantly to the variance in Eyes test performance while the face perception measure did not, a result they called surprising, since the test had been conceptualized as relatively free of general cognitive abilities. With 42 people the study is a signal and not an estimate, which is why the meta-analysis matters more.

How large is 0.24? Squared, it is 0.0576, so IQ accounts for about 6 percent of the differences in Eyes test scores and the other 94 percent lies elsewhere (our arithmetic). A concrete illustration, also our arithmetic and assuming a straight-line relation with normally distributed scores, helps: a person two standard deviations above the mean on IQ, which is a score of 130 on the scale explained on the page about why the standard deviation is 15, would be expected to land about 0.48 standard deviations above the mean on the Eyes test. That is roughly the 68th percentile, well above average and nowhere near the top. Individuals scatter widely around that expectation in both directions.

The second large synthesis changes how the first should be read. Kittel, Olderbak and Wilhelm pooled 119 effect sizes from 61 studies with 8,611 persons in Assessment. They found acceptable internal consistency (alpha of 0.73) and reported that the test related more strongly to emotion perception (r = 0.33, ρ = 0.48) than to alternative theory of mind measures (r = 0.29, ρ = 0.39). It related weakly to moderately to vocabulary (r = 0.25, ρ = 0.32), to cognitive empathy (r = 0.14, ρ = 0.20) and to affective empathy (r = 0.13, ρ = 0.19). The abstract reports both a raw correlation r and a value labeled ρ; in meta-analysis a ρ is conventionally a correlation corrected for the unreliability of the measures. The authors conclude that the test operates as an emotion perception measure more than as a theory of mind measure, which challenges how its results are interpreted.

Put together, the two syntheses say that a popular test of reading minds from eyes shares about as much with a vocabulary measure (0.25) as with IQ (0.24), shares less with measures of cognitive and affective empathy (0.14 and 0.13), and may be mostly a test of perceiving emotion. That is a finding about one instrument. The next section asks whether other performance tests agree.

4 Do Other Tests of Reading Emotions and People Show the Same Link?

Yes, two further meta-analyses of different performance tests land at the same small value, 0.19, which suggests that the Eyes test is not a special case. The first, by Schlegel, Palese, Mast, Rammsayer, Hall and Murphy, examined emotion recognition ability, the ability to recognize others' emotions from nonverbal cues, which is measured with performance-based tests. Their abstract reports 471 effect sizes from 133 samples and a significant mean effect size of r = 0.19, controlled for nesting within samples. Different intelligence types (crystallized, fluid, spatial, memory, information processing speed and efficiency) yielded similar effect sizes, whereas academic achievement measures such as SAT scores were unrelated to emotion recognition ability.

Two moderators in that abstract deserve attention. Effect sizes were higher for tests that present facial, vocal and bodily cues at once than for tests using static pictures, and higher for tests with higher reliability and more emotions. Effect sizes also increased with the mean age of the sample. The reliability finding is a reminder that a noisy test caps the correlation it can show, so a small observed value does not mean a small true link; it also does not license reading the observed value as larger than it is. The authors conclude that emotion recognition ability is a sensory-cognitive ability that is distinct from, yet related to, intelligence. A separate five-study validation of one such test, with a total of 1,284 participants, reported that scores correlated highly with other performance tests and with cognitive intelligence, and also with higher self-reported empathy, openness and neuroticism and with better social adjustment. That last result is a useful counterweight to the questionnaire findings below, because it shows some tie between a performance test and a self-rating, though the abstract gives no magnitude.

The second synthesis, by Murphy and Hall in Intelligence, covers interpersonal sensitivity, defined as accurate judgment or recall of others' behavior or appearance. It pooled 38 independent samples with 2,988 participants and reported a highly significant small-to-medium effect for intelligence measures to correlate with decoding accuracy, r = 0.19. The moderators it examined included whether the judgment concerned emotions or intended meanings, whether the channel was audio alone or audio with video, and the gender of the targets. The same value as the emotion recognition meta-analysis, reached from different studies and a different set of tests, makes the pattern harder to dismiss as an artifact of one instrument.

Interpersonal sensitivity has a wider research literature on its social correlates. A meta-analysis of 215 independent studies by Hall, Andrzejewski and Yopchick concluded that higher interpersonal sensitivity was generally associated with favorable or adaptive psychosocial functioning, with significant mean correlations for 27 of 40 categories of psychosocial variables. That paper does not isolate intelligence from those correlates, and it is cited here only to show that the sensitivity literature is large. A later section shows that the social benefits of mindreading measures were questioned once general cognitive ability was held constant.

Read together, the three performance-test syntheses give a consistent picture at the level of the average person. Cognitive empathy, measured as the number of correct judgments about what others feel or think, rises a little with intelligence, in the 0.19 to 0.24 band, and the link is similar across several kinds of intelligence. By the same arithmetic as before, 0.19 squared is 0.036, so IQ would account for about 4 percent of the differences, and a person at 130 would be expected near the 65th percentile on such a test. Those are expectations for groups, not predictions for any one person, a point the section on what these studies do not say returns to.

5 How Does Ability Emotional Intelligence Relate to Fluid and Crystallized Intelligence?

Among the four branches of ability emotional intelligence, the branch that involves understanding emotions correlates most strongly with intelligence, and it does so about equally with fluid and crystallized ability. The meta-analysis by Olderbak, Semmler and Doebler in Emotion Review pooled 352 effect sizes with a total of 15,333 participants. Its abstract reports that for each branch, the strength of relations with fluid intelligence (Gf) and crystallized intelligence (Gc) was equivalent. Understanding emotions had the strongest relation with Gf and Gc combined (ρ = 0.43, 81 effect sizes, 11,524 participants). The other three were weaker: facilitating thought using emotion (ρ = 0.19, 51 effect sizes, 7,254 participants), managing emotions (ρ = 0.20, 74 effect sizes, 11,359 participants) and perceiving emotion (ρ = 0.20, 79 effect sizes, 9,636 participants). For perceiving emotion, relations were also moderated by stimulus type.

Two cautions apply. The values are marked ρ, which, as above, conventionally denotes a corrected correlation, so they are not directly comparable with the raw r of 0.19 from the emotion recognition meta-analysis, though the perceiving branch landing at 0.20 is consistent with it. And these are ability tests of emotional intelligence, which have right and wrong answers by design. They are not the self-rated trait scales that most online quizzes and many workplace programs use, and the page on emotional intelligence versus IQ explains why that difference matters for what the scores mean.

What the pattern adds to the question is one outlier. The branch closest to emotion recognition, perceiving, sits where the other recognition meta-analyses do, at about 0.20, and so do facilitating thought and managing emotions, at 0.19 and 0.20. Only understanding emotions, the branch that concerns reasoning about what emotions mean and where they lead, runs more than twice as high. This is our observation on published values, not a claim the authors make, and it fits a simple reading in which the more an emotional task resembles reasoning about a problem, the more it overlaps with a reasoning test. None of the four branches is a measure of caring, and no study in this table tests whether the part of empathy that concerns caring sits near or far from IQ.

A reasonable question is whether any of this means that emotional intelligence is just intelligence under another name. The meta-analytic values above say that ability emotional intelligence overlaps with intelligence by amounts that vary from small to moderate depending on the branch, not that the two are the same. How much of the overlap is shared knowledge, shared reasoning or shared test-taking skill cannot be read from a correlation, and the next section shows how contested that reading is.

6 Why Do Performance Tests of Empathy Overlap With IQ?

The most direct test of the question found that mindreading tasks track general cognitive ability far more closely than they track people's own descriptions of their mindreading, which suggests that the tasks measure partly what an intelligence test measures. The study, Mindreading measures misread? by Wendt, Zimmermann, Spitzer and Müller, appeared in Psychological Assessment in 2024. It was preregistered and involved 700 participants from the US general population. The authors tested two rival interpretations: that mindreading tasks may capture general cognitive ability, and that mindreading self-reports may capture perceived rather than actual mindreading ability. They used structural equation modeling to compare the observed associations with what each interpretation predicts.

The abstract reports three correlations between latent variables. Mindreading tasks showed a negligible correlation with mindreading self-reports (0.05) and a large one with general cognitive ability (0.85). Mindreading self-reports, in contrast, were specifically associated with perceived performance in the mindreading tasks (0.29). Squared, 0.85 is 0.72, so by our arithmetic about 72 percent of the variance in the latent task factor is shared with the latent general ability factor, a much larger share than the 4 to 6 percent seen in the meta-analyses of single observed tests. A plausible reason, which is our reading of how latent models work and not a claim in the abstract, is that a latent correlation is estimated free of measurement error, while an observed correlation between one imperfect test and one imperfect IQ score is held down by the noise in both. The two sets of numbers would then describe the same relation at different levels of purification, though the abstract does not say so.

The authors also tested what the measures predict. Neither mindreading self-reports nor task-based measures showed positive unique associations with psychosocial functioning when controlling for general cognitive ability and a general positive self-evaluation. Instead, negative unique associations emerged for both methods, although the effect was not robust for tasks. They conclude that the results cast doubt on the validity of commonly used mindreading measures and support the rival interpretations. This is a strong claim from one study in one country, and it should be read as a serious challenge and not as a settled verdict.

A published commentary by Conway, Long, Sevi, Catmur and Bird in Psychological Assessment in 2025 takes up the argument from the other side. It criticizes how Wendt and colleagues categorized the tasks and raises deeper questions: what exactly mental states are, how mindreading can be measured accurately when the correct answer lacks ground truth, and what contributions general cognitive ability and specific experience with the kinds of minds being read make to individual differences. The commentary does not dispute that general ability matters; it argues that without a psychological theory of how people read other minds, the measures cannot be judged properly. The debate is therefore about interpretation, and the correlation itself is not in dispute.

The wider measurement picture supports caution. A systematic review by Yeung, Apperly and Devine of 273 studies identified 75 measures used to investigate individual differences in adults' theory of mind and examined their sensitivity, reliability and validity. It reports that ceiling effects were prevalent, that there was limited evidence to establish the reliability or validity of these measures because psychometric properties were rarely reported, and that interrelations among measures were inconsistent. In plain terms, many mindreading tests in adults are not yet built to separate people well, and the ones that do separate people may be doing so partly through vocabulary and reasoning, which is what the Eyes test meta-analyses suggested.

Three readings of the overlap remain open. The first is that the tasks are partly intelligence tests with social content. The second is that mindreading is a real social skill that is itself partly cognitive, so some overlap is genuine and expected. The third is that the tasks are simply imperfect measures and the overlap reflects what all imperfect tests share. The data in these studies cannot fully separate them. What they do establish is that a high score on a performance test of mindreading is, to a measurable degree, a high score on a cognitive ability, and that is relevant to anyone who reads it as a measure of kindness.

7 Does Saying You Are Empathic Predict How Well You Read Others?

Questionnaire measures of empathy are nearly unrelated to performance on tasks of the same ability, with a pooled overlap of about 1 percent in the largest meta-analysis. The paper, Are self-report cognitive empathy ratings valid proxies for cognitive empathy ability? by Murphy and Lilienfeld, appeared in Psychological Assessment in 2019. Based on 85 studies and a total of 14,327 participants, its abstract reports that self-report cognitive empathy scores account for only approximately 1 percent of the variance in behavioral cognitive empathy assessments. The relation was not significantly different from that shown by affective empathy scale scores, and effect sizes were not moderated by self-report empathy domain, gender composition, the type of stimulus, child versus adult samples, or normative versus clinical or forensic samples. Publication bias did not markedly affect them. The authors raise serious concerns about using self-report scores as proxies for ability.

A later set of studies reached the same place from a different angle. Sunahara and colleagues analyzed data from three studies with a total of 2,376 participants and distinguished the tendency to take the perspective of others and the tendency to catch the emotions of others. They tested those against tasks of emotion recognition, theory of mind, empathic accuracy and affect sharing. Their abstract reports little evidence of an association between any of the self-reported empathy measures and either social-cognitive ability or affect sharing. It also reports that alexithymia, the difficulty of recognizing and understanding one's own emotions, was more related to social-cognitive abilities than the tendency to take perspective or to vicariously experience emotions. The Wendt study adds the same result from a third side, since its 0.05 between tasks and self-reports points the same way as the 1 percent of variance in the meta-analysis, which as a correlation is about 0.1.

This result matters for the headline question for a reason that is easy to miss. Most people use the word empathy to mean something they can feel in themselves, and that is what a questionnaire records. Performance tests record something else, and the two barely move together. So a finding that smarter people do better on a performance test of mindreading cannot be carried over to a claim that they would describe themselves as more empathic, nor that they would be. No single study here shows that. We did not find a meta-analysis that pools the correlation between IQ and questionnaire empathy, and the gap is recorded in the ledger for this page. One study with both in the same sample is discussed in the next section.

An inference, labeled as ours, shows why the gap matters. If questionnaire scores share about 1 percent of variance with task scores, which is a correlation of about 0.1, and task scores correlate with IQ at about 0.24, then the path from IQ to questionnaire empathy that runs through task ability would carry about 0.02 (0.1 times 0.24, our arithmetic). Questionnaires can of course relate to IQ by other routes, such as verbal fluency in answering items, and the next study found one such link. The point of the sketch is only that the strongest evidence for an IQ link applies to the measure that people with questionnaires in hand are least likely to be using.

8 Is There Evidence That Smarter People Behave More Kindly?

The one study here that put intelligence, empathy and helping behavior in the same sample found a positive link with self-reported helping, but it is a single cross-sectional sample of university students, and its measure of behavior was a questionnaire. The paper, Why are smarter individuals more prosocial? by Guo, Sun, Cai, Zhang and Song, appeared in Intelligence in 2019. Its abstract describes 518 undergraduate students (254 women, mean age 19.79) who completed the Chinese version of the Raven's Standard Progressive Matrices, a Self-Report Altruism Scale that distinguishes the recipient of help, the Interpersonal Reactivity Index and the internalization subscale of a moral identity scale.

The abstract reports that fluid intelligence was significantly correlated with self-reported prosocial behavior, and that moral identity, perspective taking and empathic concern could account for the positive association between intelligence and prosocial behavior. The mediation effects of moral identity and empathy were consistent across gender. Perspective taking is the cognitive side of empathy and empathic concern is the affective side, so both appear in the account. The abstract does not give the size of the correlations, and the full text was not available to this page, so no coefficient from the paper is quoted here. Readers who see specific coefficients attributed to it in other summaries should check them against the paper.

Several limits follow from the design, and they change which verbs are safe. A cross-sectional survey measures everything once. Words such as mediate and account for are statistical terms for how the associations fit a model, and they do not show that intelligence leads to empathy or that empathy leads to helping. The altruism measure was self-reported, so it captures how participants describe their conduct, which can drift toward how they would like to see themselves. The sample was university undergraduates in one country, a group with a narrower spread of ability than the general population, and narrower spread usually shrinks correlations in a way that is separate from any real effect. And the questionnaire measures of empathy are exactly the kind that the previous section found to track task performance only weakly.

For adjacent questions, other pages on this site cover the nearby territory. The page on IQ and personality covers agreeableness, the personality trait closest to kindness. The page on IQ and crime covers conduct at the antisocial end of the range. The page on IQ and relationships covers how partners match. None of them reverses the point here: a modest positive link between ability and self-reported helping is plausible and has some support in one sample, and a firm claim that smarter people are kinder has no support at the level of pooled evidence.

9 Are Gifted People More or Less Empathetic?

The best pooled evidence on gifted people covers emotional intelligence, not empathy as such, and it found no systematic difference between gifted and non-gifted students. The 2026 paper in Intelligence by Ferrándiz, Ferrando-Prieto, Sánchez-Meca, Pons-Parra and colleagues is a preregistered systematic review and meta-analysis of comparative observational studies conducted since 1990. According to its abstract, 684 records were screened, 61 full texts were assessed and 29 studies met the inclusion criteria. The authors synthesized group differences in 28 dimension-specific random-effects meta-analyses, estimating standardized mean differences with 95 percent confidence intervals and prediction intervals, and classified emotional intelligence outcomes by theoretical framework (ability versus trait-mixed) and by measurement modality (performance-based versus self-report).

Across dimensions, the pooled effects provided no evidence of systematic differences between gifted and non-gifted students. Adaptability was the only dimension favoring gifted students, and its prediction interval included negative values, which the authors read as limited generalizability and context-dependent effects. They conclude that the evidence does not support a generalized emotional intelligence advantage associated with giftedness and that dimension-level synthesis is important. Because giftedness is commonly defined by high test scores, the review is a test of what high ability brings with it in this domain. The finding fits the performance-test results above, where the link with IQ is real but small.

A neighboring question is whether gifted people are more prone to social and emotional difficulty, which is sometimes presented as the other face of the empathy story. A systematic review by Tasca and colleagues in Child Psychiatry and Human Development covered 19 studies, 17 of them with children or adolescents and 12 with a non-gifted control group. Socio-emotional problems such as withdrawal appeared in 3 of 8 studies, internalizing disorders such as anxiety in 5 of 9 and externalizing disorders such as hyperactivity in 3 of 5. The authors concluded that no univocal conclusion on the relationship between intellectual giftedness and socio-emotional or behavioral problems could be drawn, mainly because the studies differed in the age of participants, the informants and the instruments. They suggest that age and the gap between verbal and non-verbal ability should be considered critical factors.

Taken together, the two reviews undercut both versions of the stereotype, that gifted people are unusually empathic and that they are unusually cold or awkward. Neither version has solid pooled support in the studies reviewed. The reviews concern identified gifted students, mostly children, and may not carry over to adults who score high on a test, so the pages on gifted adults, on being twice exceptional and on the signs of high intelligence cover those groups, and the sibling page on high-IQ quirks audits the viral lists of traits attributed to smart people.

10 Do Sex, Age, Culture and Language Change the Picture?

Any comparison of IQ and performance tests of empathy has to hold sex, age and language constant, and the largest study of the Eyes test shows why. Greenberg and colleagues tested sex and age differences with the English version of the Eyes test in adolescents and adults across 57 countries, using one discovery dataset (305,726 participants) and three validation datasets (642, 5,284 and 1,087 participants). Their abstract reports a replicable on-average female advantage in performance, that performance increases through adolescence and shallowly declines across adulthood, that the female advantage is evident across the lifespan, and that it is significant in 36 of the 57 countries. They also report that empathizing-systemizing D-scores predicted performance above and beyond sex, and, in exploratory country-level analyses, that the female advantage was negatively linked to prosperity and autonomy and positively linked to collectivism.

Three implications follow for the IQ question. The first is sex. Women score higher on average on this test, while the page on average IQ by gender shows that average IQ differs little between women and men, so a study that does not account for sex can blur or distort a link between IQ and the test. Whether the studies in the meta-analyses held sex constant varies, and their abstracts do not say. The second is age. Both Greenberg and the emotion recognition meta-analysis find that performance depends on age, which means a sample's age mix can move a correlation. The third is language. The Eyes test asks people to choose among descriptor words, and in 57 countries it was given in English. Kittel and colleagues found that the test relates to vocabulary at 0.25, which makes the language of the test and the language ability of the person plausible sources of overlap with verbal intelligence, though that is an inference and not a result of the Greenberg paper.

A fourth issue is the design itself. Every study on this page is observational, and nearly all are cross-sectional or meta-analyses of cross-sectional correlations. None can say whether intelligence helps people read others, whether growing up reading others builds vocabulary and reasoning, or whether schooling, family and culture shape both. The correlations describe how two scores go together across people at a given time. They do not settle which comes first, and the section on what the evidence supports keeps that distinction in its wording.

11 What Do These Studies Not Say About You, or About Someone You Know?

None of these findings can tell you how empathic you are, how well you read people or whether you care, because they describe how scores move together across large groups, and an individual can sit anywhere around the trend. The numbers make that concrete. At a correlation of 0.24, about 94 percent of the differences in Eyes test scores are not explained by IQ, and at 0.19 about 96 percent are not (our arithmetic). A person at an IQ of 130, about the 98th percentile on the scale explained on the page about percentiles and IQ scores, would be expected to score near the 68th percentile on the Eyes test and near the 65th on an emotion recognition test, with wide scatter around either expectation. Many people at 130 score below the mean on these tests, and many people at 100 score far above it.

The same logic cuts the other way. A person who is kind, perceptive and good with people may have any IQ, and a person who struggles in social situations may have any IQ too. The studies do not support reading social difficulty as a sign of low or high intelligence. Where social or emotional difficulties are a concern, they call for conversation with a qualified clinician, and this page is not a clinical resource. Two things it does not claim are worth stating plainly: it makes no statement about any diagnosis, and it does not describe any reader's character.

Three misreadings are common enough to name. The first is turning a modest correlation into a trait of the whole group, as in smart people are more empathic or smart people are cold. The pooled evidence supports neither, and the gifted reviews above found no systematic difference in emotional intelligence. The second is treating a questionnaire score as a measure of ability. A person who rates themselves as highly empathic has reported a belief about themselves, and the largest meta-analysis found that such ratings share about 1 percent of variance with task performance. The third is treating a task score as a measure of kindness. A task such as the Eyes test measures how accurately a person labels expressions in photographs, which several analyses suggest depends partly on vocabulary and reasoning. Neither a task score nor a questionnaire score is a measure of how a person treats others.

A final point concerns this site. ACIS measures cognitive abilities. Its report gives a Full Scale IQ and six index scores for verbal comprehension, fluid reasoning, quantitative reasoning, visual spatial ability, working memory and processing speed, and none of them is empathy. The ACIS report cannot tell you how empathic you are, and a high ACIS score should not be read as a statement about how you treat people.

12 What the Evidence Supports, Stated Narrowly

The evidence supports a small positive correlation between intelligence and performance on tests of reading others' emotions and mental states, and it supports little else with confidence. Sorted by strength, the statements that the studies on this page can bear are as follows.

1. Performance tests of understanding others correlate modestly with intelligence. Three meta-analyses reach 0.24 for the Eyes test (77 effect sizes, 3,583 participants), 0.19 for emotion recognition (471 effect sizes from 133 samples) and 0.19 for interpersonal sensitivity (38 samples, 2,988 participants).
2. The overlap is not specific to one kind of intelligence or one kind of test. The Eyes test meta-analysis found no difference between verbal and performance abilities, and the emotion recognition meta-analysis found similar effects across crystallized, fluid, spatial, memory and speed measures.
3. Ability emotional intelligence correlates with fluid and crystallized intelligence at 0.19 to 0.20 for three of its branches and at 0.43 for the branch of understanding emotions, with equal strength for the two kinds of intelligence (352 effect sizes, 15,333 participants).
4. The overlap may be partly a feature of the tasks. The Eyes test relates to vocabulary at 0.25 and to emotion perception at 0.33, and one preregistered study of 700 adults found a latent correlation of 0.85 between mindreading tasks and general cognitive ability. That interpretation is contested in a published commentary and is not settled.
5. Questionnaires of empathy barely track tasks. The pooled overlap is about 1 percent of variance (85 studies, 14,327 participants), and a later analysis of 2,376 participants found little evidence of an association.
6. On caring, the evidence is thin. One cross-sectional sample of 518 students linked fluid intelligence to self-reported helping through empathy and moral identity, and no pooled estimate of IQ and questionnaire empathy was found. For giftedness, a meta-analysis of 29 studies found no systematic difference in emotional intelligence.

The statements the evidence does not support are as important. It does not show that high intelligence makes people more empathic, that empathy raises intelligence, that smart people care more or less than others, or that a test of reading eyes measures kindness. It does not provide a basis for estimating anyone's empathy from an IQ score or the reverse.

The Standards for Educational and Psychological Testing and the American Psychological Association guidance say why those limits matter for reading any score. The Standards for Educational and Psychological Testing, published jointly by AERA, APA and NCME in 2014, define validity as the degree to which evidence and theory support the interpretations of test scores for proposed uses. On that definition, evidence that an Eyes test score correlates 0.24 with IQ supports the interpretation that the score reflects some cognitive ability. It does not support the separate interpretation that the score measures empathy as caring, and each of those interpretations needs its own evidence. The page on reliability and validity explains that distinction, and the page on how IQ scores are normed explains what a norm-referenced score does and does not say. The APA Guidelines for Psychological Assessment and Evaluation, approved by the APA Council of Representatives in March 2020, state in Guideline 7 that individual performance on psychological tests is only one piece of an assessment and should be integrated with information from multiple sources. A single score on any task, including the ones discussed here, is not a character assessment.

A last note on this site. ACIS is an online, unsupervised cognitive assessment in English. It is not a clinical or diagnostic instrument, and it is not for hiring, school accommodations or admission to high IQ societies. The Full Scale form covers all 20 subtests for a one time payment of 50 dollars, as read on the ACIS home page on October 20, 2026, takes about 175 minutes with breaks allowed, and comes with a free trial that needs no card. Its report gives each score with a 95 percent confidence interval, and the technical manual documents the instrument. A reader curious about how their reasoning, vocabulary and working memory compare with adult norms can use it. A reader curious about their empathy needs a different kind of measure and, for any practical purpose, a qualified professional.

13 Sources Behind This Page

Every figure above is traceable to one of the following, and each is linked at the point where it is used. Sample sizes, correlations and counts of studies are quoted as the papers report them in their abstracts, and any conversion into variance explained, percentiles or products of correlations is labeled as our arithmetic. DOIs were checked against the Crossref record. The full texts of several papers were not available, so only abstract-level results are cited.

• Baker C A, Peterson E, Pulos S and Kirkland R A. Eyes and IQ: A meta-analysis of the relationship between intelligence and "Reading the Mind in the Eyes". Intelligence, 2014, volume 44, pages 78 to 92.
• Kittel A F D, Olderbak S and Wilhelm O. Sty in the Mind's Eye: A meta-analytic investigation of the nomological network and internal consistency of the "Reading the Mind in the Eyes" test. Assessment, 2022, volume 29, issue 5, pages 872 to 895.
• Peterson E and Miller S L. The Eyes Test as a Measure of Individual Differences: How much of the Variance Reflects Verbal IQ? Frontiers in Psychology, 2012, volume 3, article 220.
• Schlegel K, Palese T, Mast M S, Rammsayer T H, Hall J A and Murphy N A. A meta-analysis of the relationship between emotion recognition ability and intelligence. Cognition and Emotion, 2020, volume 34, issue 2, pages 329 to 351; and Schlegel K, Fontaine J R J and Scherer K R. The Nomological Network of Emotion Recognition Ability. European Journal of Psychological Assessment, 2019, volume 35, issue 3, pages 352 to 363.
• Murphy N A and Hall J A. Intelligence and interpersonal sensitivity: A meta-analysis. Intelligence, 2011, volume 39, issue 1, pages 54 to 63; and Hall J A, Andrzejewski S A and Yopchick J D. Psychosocial correlates of interpersonal sensitivity: A meta-analysis. Journal of Nonverbal Behavior, 2009, volume 33, issue 3, pages 149 to 180.
• Olderbak S, Semmler M and Doebler P. Four-branch model of ability emotional intelligence with fluid and crystallized intelligence: A meta-analysis of relations. Emotion Review, 2019, volume 11, issue 2, pages 166 to 183.
• Wendt L P, Zimmermann J, Spitzer C and Müller S. Mindreading measures misread? A multimethod investigation into the validity of self-report and task-based approaches. Psychological Assessment, 2024, volume 36, issue 5, pages 365 to 378; and Conway J R, Long E L, Sevi L, Catmur C and Bird G. Theoretical limitations on mindreading measures: Commentary on Wendt et al. (2024). Psychological Assessment, 2025, volume 37, issue 3, pages 129 to 132.
• Murphy B A and Lilienfeld S O. Are self-report cognitive empathy ratings valid proxies for cognitive empathy ability? Negligible meta-analytic relations with behavioral task performance. Psychological Assessment, 2019, volume 31, issue 8, pages 1062 to 1072.
• Sunahara C S and colleagues. Revisiting the association between self-reported empathy and behavioral assessments of social cognition. Journal of Experimental Psychology: General, 2022, volume 151, issue 12, pages 3304 to 3322.
• Guo Q, Sun P, Cai M, Zhang X and Song K. Why are smarter individuals more prosocial? A study on the mediating roles of empathy and moral identity. Intelligence, 2019, volume 75, pages 1 to 8.
• Ferrándiz C, Ferrando-Prieto M, Sánchez-Meca J, Pons-Parra R and colleagues. Emotional intelligence in gifted and non-gifted individuals: A systematic review with meta-analysis. Intelligence, 2026, volume 117, article 102037.
• Tasca I, Guidi M, Turriziani P, Mento G and Tarantino V. Behavioral and socio-emotional disorders in intellectual giftedness: A systematic review. Child Psychiatry and Human Development, 2024, volume 55, issue 3, pages 768 to 789.
• Greenberg D M, Warrier V, Abu-Akel A and colleagues. Sex and age differences in "theory of mind" across 57 countries using the English version of the "Reading the Mind in the Eyes" test. Proceedings of the National Academy of Sciences, 2023, volume 120, issue 1.
• Zaki J and Ochsner K N. The neuroscience of empathy: progress, pitfalls and promise. Nature Neuroscience, 2012, volume 15, issue 5, pages 675 to 680.
• Yeung E, Apperly I and Devine R. Measures of individual differences in adult theory of mind: A systematic review. Neuroscience and Biobehavioral Reviews, 2024, volume 157, article 105481.
• American Educational Research Association, American Psychological Association and National Council on Measurement in Education. Standards for Educational and Psychological Testing. Washington, DC: AERA, 2014, read October 20, 2026; and American Psychological Association, APA Guidelines for Psychological Assessment and Evaluation, approved by the APA Council of Representatives in March 2020, read October 20, 2026.

14 Frequently Asked Questions