How accurate are IQ tests? What IQ scores mean, what they predict and whether IQ can change

An IQ score is a relative score: it shows where you stand compared with people of the same age (the norm group). Mensa Korea explains that the average of an intelligence quotient is set at 100 by convention and that the standard deviation differs between researchers and tests (Mensa Korea, n.d.), and most major tests, including the Wechsler scales, use a mean of 100 and a standard deviation of 15 (or 16) (Trahan et al., 2014).

On a test scaled this way, about 68 in 100 people fall within one standard deviation of the mean (85 to 115 when the standard deviation is 15), and a score of 130 or more, the top-2% mark on the Wechsler scales, is more precisely 2.28% (the 97.72nd percentile) (Mensa Korea, n.d.). At the other end, 70 or below covers about 2.2% of the population (Trahan et al., 2014). The real distribution is not a perfect bell curve, though: in surveys of all 11-year-olds in Scotland, the mean was 100 but the most common score was about 105, and scores were less spread out at the top than at the bottom (Johnson et al., 2008). So these shares should be read as approximate.

The same person can score 148 on one test and 130 on another because the scales differ. Mensa Korea explains that 148 on its own scale, with a standard deviation of 24, is the same cut-off as 130 on the Wechsler scales, with a standard deviation of 15 (Mensa Korea, n.d.), and Mensa International likewise says that 132 on one test can equal 148 on another, and that some intelligence tests do not use IQ scores at all (Mensa International, n.d.).

What do intelligence tests measure? Scores on very different mental tasks tend to rise and fall together, so people who do well on one tend to do well on the others. Spearman called this shared part general intelligence (g) in 1904, and when a diverse battery of cognitive tests is given to a sample with a wide range of ability, g accounts for about 40% of the differences in scores (Plomin & Deary, 2015). How research views Gardner’s theory of multiple intelligences, which splits intelligence into several kinds, is covered in the multiple intelligences guide. Whether IQ differs by MBTI type is examined in our guide to MBTI and intelligence.

The US adult Wechsler test released in 2024 (WAIS-5) was normed between February 2023 and January 2024 on 2,020 people aged 16 to 90, a sample matched closely to the 2022 US census by age, sex, ethnicity, education and region. According to a paper that cites the publisher’s technical manual, the full-scale IQ had a mean reliability of .97 across ages, and its stability coefficient was .93 in 201 people tested twice an average of 29 days apart (Winter et al., 2024). Index reliabilities ranged from .90 to .97, but the norm sample excluded people with intellectual disabilities, psychiatric disorders and neurological conditions (Winter et al., 2025).

In Korea, the adult Wechsler test (K-WAIS-IV) was standardised in 2012 for ages 16 years 0 months to 69 years 11 months (Seo et al., 2014). Professional tests like these are given one to one by a trained examiner.

Even so, a single score is best read as a range rather than a fixed value. In 344 students tested twice for special-education eligibility an average of 2.84 years apart, the full-scale IQ had a test-retest reliability of .82, yet 25% of the students’ scores differed by 10 points or more (Watkins & Smith, 2013). In a meta-analysis of samples with IQs under 80, most full-scale IQs changed by less than 6 points over an average of 2.8 years, but 14% changed by 10 points or more (Whitaker, 2008). Test-retest reliabilities of neuropsychological test scores are generally adequate to high (r = .7 or more) (Calamia et al., 2013).

That is why professionals read scores together with measurement error. A paper citing the American Psychiatric Association’s diagnostic criteria (DSM-5-TR) explains that people with intellectual disabilities typically score at least two standard deviations below the mean, giving 65 to 75 as an example, which includes ±5 points for measurement error (Winter et al., 2024), and the criteria of the American Association on Intellectual and Developmental Disabilities (AAIDD) also stress clinical judgement that accounts for measurement error (Trahan et al., 2014). Old norms are another problem, because they inflate scores. As IQ scores rise over time (the Flynn effect), norms become outdated (Trahan et al., 2014), and in diagnosing intellectual disability, 0.3 points are subtracted from a Wechsler full-scale IQ for each year that its norms are out of date (Winter et al., 2024).

Mensa International says that membership is open only to people who score within the top 2% on an approved intelligence test that has been properly administered and supervised, that online tests cannot be used for admission, and that the online challenge on its own website is not an official IQ test (Mensa International, n.d.). Mensa Korea likewise says it has no testing partnership of any kind with the online sites that advertise free IQ tests, and that its admission test can only be taken offline (Mensa Korea, n.d.).

Unsupervised online tests also tend to give higher scores. In a meta-analysis of 49 studies (100,434 people) comparing proctored and unproctored ability tests, unproctored scores were 0.20 standard deviations higher on average, which would be about 3 points on an IQ scale with a standard deviation of 15. The gap was smaller for tests whose answers are hard to look up online, and the authors attributed the score gap to cheating (Steger et al., 2020). These studies were about ability tests in general, though, and while writing this guide we found no study that tested any particular free IQ website.

Even well-built free online research tests have limits. The International Cognitive Ability Resource (ICAR) was developed with data from 96,958 people in 199 countries who took an online survey in exchange for feedback about their personality, and its 16-item version had an internal consistency of α = .81. In 137 students at one selective private university, after correcting for reliability and restricted range, it correlated .82 and .81 with the two composite scores of another intelligence test (Shipley-2). Yet its developers say this kind of administration is less precise than traditional assessment, and the paper gives no norms for scoring an individual’s IQ (Condon & Revelle, 2014).

There are more reasons why a free test score is not the same number as an IQ from a professional test. If a test does not say which group its norms come from, its score cannot be interpreted; old norms inflate scores (Trahan et al., 2014), and the scales differ from test to test (Mensa International, n.d.). A score from a test taken for fun and an IQ from a professional test cannot be read the same way.

IQ is clearly linked with school grades. In a meta-analysis of 240 samples with 105,185 people, the correlation corrected for sampling error, measurement error and range restriction in IQ was ρ = .54, and it varied by subject, grade level and the type of test used (Roth et al., 2015). Where this was tested, conscientiousness was linked with grades largely independently of intelligence, and for university grades, once secondary-school grades were controlled for, it added as much to the prediction as intelligence did (Poropat, 2009).

The link with job performance is smaller than often claimed. The famous .51 from a 1998 review was corrected for measurement error in job ratings and for range restriction (Schmidt & Hunter, 1998; Sackett et al., 2023); a 2022 study that redid the range-restriction corrections lowered it to .31, corrected for measurement error in job ratings and, only where appropriate, for range restriction (Sackett et al., 2022), and another team’s meta-analysis of 113 studies from the 21st century, described by the same authors, reported .16 observed and .23 when corrected for measurement error in job ratings and for range restriction (Sackett et al., 2023). In the new analysis, structured interviews rather than intelligence tests came out as the selection method with the highest mean validity (Sackett et al., 2023).

IQ is also linked with education, occupation and income, but not overwhelmingly more than other factors. Intelligence measured earlier correlated .56 with later education (59 samples, 84,828 people), .43 with occupational level (45 samples, 72,290 people) and .20 with income (31 samples, 58,758 people), values corrected for unreliability and dichotomisation. The author concluded that intelligence is a powerful predictor of success but, on the whole, not an overwhelmingly better predictor than parental socioeconomic status or grades (Strenze, 2007).

It is linked with health too. In a meta-analysis of 16 studies with 1,107,022 people (22,453 deaths), cognitive test scores one standard deviation higher in childhood or youth were associated with a 24% lower risk of death over 17 to 69 years of follow-up, but in the five studies that took adult socioeconomic status into account and the six that took education into account, the link weakened by 34% and 54% respectively (Calvin et al., 2011). Why this is so has not been settled.

Emotional intelligence (EQ) was linked with grades at ρ = .20 (42,529 people) and came third in importance after intelligence and conscientiousness, with no evidence for the direction of cause (MacCann et al., 2020). IQ is linked with many outcomes, but a correlation is not a destiny. Research on perseverance (grit) is in resilience.

In a study of 11,000 twin pairs from the US, Australia, the Netherlands and the UK, the heritability of general cognitive ability rose with age, from 41% at 9 to 55% at 12 and 66% at 17 (Haworth et al., 2010). Another review sums up that heritability rises from about 20% in infancy to about 40% in adolescence and about 60% in adulthood, with some evidence that it may reach 80% in later adulthood (Plomin & Deary, 2015).

This does not mean that 66% of one person’s intelligence comes from genes; it is an estimate of how much of the differences in scores between people in that group go along with genetic differences. And no trait has a heritability of 100%, so environmental influence is always substantial (Plomin & Deary, 2015).

The researchers suggested that heritability rises with age because, as children grow up, they increasingly select, modify and create their own experiences partly in line with their genetic tendencies (Haworth et al., 2010). A meta-analysis of 11,500 twin and sibling pairs measured at least twice between 6 months and 18 years likewise found that new genetic influences predominated in early childhood, while the rise in heritability after age 8 was explained by existing genetic influences growing stronger (Briley & Tucker-Drob, 2013). Heritability also depends on the society. US studies showed a clear, moderately sized effect of family socioeconomic status on the genetic influence, but studies from Western Europe and Australia, where education and health care are more evenly available, found no such effect or the reverse (Tucker-Drob & Bates, 2016).

In a meta-analysis of 205 longitudinal studies with 87,408 people, rank-order stability over 5 years at age 20 was ρ = .76. Stability was low in preschool children, rose quickly in childhood and stayed high from late adolescence; the level needed for decisions about an individual (.80) could only be expected from age 7 onwards, and only over short intervals in children, while adults reached it even over intervals of more than 5 years (Breit et al., 2024). In 106 people who took the same test at 11 and at 90, the two scores correlated .54 (.67 when corrected for range restriction) (Deary et al., 2013). The order of people is fairly stable, then, but not perfectly so.

Average scores have changed a lot between generations. In a meta-analysis of 271 samples with almost 4 million people from 31 countries, covering 1909 to 2013, full-scale IQ rose by about 0.28 points a year, fluid intelligence more, by 0.41 points, and gains became smaller in recent decades (Pietschnig & Voracek, 2015). Another meta-analysis of 285 studies (14,031 people) comparing two tests with different norms estimated 2.31 points per decade (Trahan et al., 2014). This is called the Flynn effect; it is a change in averages between generations, not a sign that one person’s score rises every year.

Recently the rise has slowed or reversed in some countries. In conscription data on Norwegian men born from 1962 to 1991, the rise, the turning point and the later decline all appeared between brothers within the same family, and the researchers concluded that environmental factors, not changes in who the parents were, were the cause (Bratsberg & Rogeberg, 2018). A US online volunteer sample also showed a pattern consistent with a reversed Flynn effect from 2006 to 2018 (Dworak et al., 2023), and two small studies comparing the new 2024 Wechsler test (WAIS-5) with the previous edition found a Flynn effect of only 1.2 points rather than the expected 3 (Winter et al., 2024).

Education appears to raise IQ. In a meta-analysis of 142 effect sizes (42 data sets, over 600,000 people) from quasi-experimental studies that controlled for earlier intelligence, used policy changes that lengthened compulsory schooling, or used school-entry age cut-offs, each additional year of education raised IQ by about 1 to 5 points, and the effect lasted across the life span and appeared in all the broad cognitive abilities studied. The authors call education the most consistent, robust and durable method yet identified for raising intelligence (Ritchie & Tucker-Drob, 2018).

Brain training does not live up to the hype. In a meta-analysis of working memory training studies, there was no reliable improvement on far measures such as nonverbal and verbal ability when compared with a control condition that did another activity (Melby-Lervåg et al., 2016), while a meta-analysis of 20 studies of n-back training alone reported a small but significant effect on fluid intelligence tests (Au et al., 2015). In the larger review, however, once one problematic study was removed and only control groups that did another activity were counted, the effect of n-back training on nonverbal ability was negligible (g = 0.10) (Melby-Lervåg et al., 2016). For brain games and the “Mozart effect”, see how to improve concentration.

Myth 1: IQ is inborn and never changes. Heritability is an estimate about differences within a group, and no trait has a heritability of 100% (Plomin & Deary, 2015). A meta-analysis found that each extra year of education raises IQ by about 1 to 5 points (Ritchie & Tucker-Drob, 2018), and averages have risen a lot between generations (Pietschnig & Voracek, 2015). An individual’s score can also change by 10 points or more within a few years (Watkins & Smith, 2013).

Myth 2: A high IQ guarantees success. Intelligence is fairly strongly linked with education and occupation, but its correlation with income, even corrected for unreliability and dichotomisation, was about .20, and it was not an overwhelmingly better predictor than parental socioeconomic status or grades (Strenze, 2007). Recalculated, its link with job performance was .31, a value still corrected for measurement error in job ratings (Sackett et al., 2022), and conscientiousness predicted grades largely independently of intelligence (Poropat, 2009).

Myth 3: An online IQ test gave me 140, so my IQ is 140. Even Mensa does not accept online tests for admission (Mensa International, n.d.), unsupervised ability tests tend to give higher scores (Steger et al., 2020), and scales differ from test to test. It cannot be read the same way as an IQ of 140 on a professional test.

Myth 4: Koreans have the highest average IQ in the world. The country rankings that circulate online usually draw on “national IQ” datasets. A critique of these datasets argues that most of their data come from samples that do not represent whole countries, that many are small convenience samples, some dating back to the 1940s and often of children only, and that they were turned into national figures through questionable assumptions and arbitrary adjustments, so they should not be used to draw inferences about global variation in intelligence (Sear, 2022). That critique has not yet been peer-reviewed, but another paper shows that national IQs are strongly confounded with countries’ current level of development (Wicherts et al., 2010b), and a systematic review concluded that the measurement bias of Western IQ tests used in sub-Saharan Africa needs more checking (Wicherts et al., 2010a). Country rankings of IQ are not a reliable comparison.

Myth 5: Einstein’s IQ was 160. While writing this guide we could not trace the source of this often-quoted figure. An IQ is a score that comes from taking a set test; it is not something that can be worked backwards from someone’s achievements.

Q. Where can I get an IQ test? A. A professional intelligence test is a standardised test given one to one by a trained examiner, and judgements such as intellectual disability are confirmed by both clinical assessment and individual standardised intelligence testing (Winter et al., 2024). mind.io has no IQ test. If it would be your first visit, see seeing a psychiatrist for the first time.

Q. Does an IQ of 70 or below mean intellectual disability? A. A score alone does not decide it. According to a paper citing the American Psychiatric Association’s diagnostic criteria (DSM-5-TR), a diagnosis of intellectual disability requires that deficits in intellectual functions such as reasoning, problem-solving and learning be confirmed by both clinical assessment and individual standardised intelligence testing, and that there also be deficits in everyday adaptive functioning (Winter et al., 2024). Scores are read with measurement error and clinical judgement in mind (Trahan et al., 2014).

Q. What is borderline intellectual functioning? A. It is a term researchers use for an IQ range of roughly 70 to 84; one systematic review used IQ 71 to 84 as its inclusion criterion. The American Psychiatric Association’s criteria do not class it as a disorder but list it among “other conditions that may be a focus of clinical attention” (Cho et al., 2025). It should be judged by a professional with an individual test, not by a free test score.

Q. What IQ do I need to join Mensa? A. Mensa’s criterion is not a number but whether you score in the top 2% on an approved intelligence test (Mensa International, n.d.). That is 130 on the Wechsler scales, with a standard deviation of 15, and 148 on Mensa Korea’s scale, with a standard deviation of 24 (Mensa Korea, n.d.).

Q. Does a child’s IQ stay the same into adulthood? A. The younger the child, the less stable the score. Rank-order stability is low in preschool children, and the stability needed for decisions about an individual can only be expected from age 7 onwards, and in children only over short intervals (Breit et al., 2024).

Q. Is EQ (emotional intelligence) more important than IQ? A. In research on grades, emotional intelligence came third after intelligence and conscientiousness, with no evidence for the direction of cause (MacCann et al., 2020). The emotional intelligence guide covers what emotional intelligence is and how accurate EQ tests are.

References: Mensa Korea (n.d.), Frequently asked questions [in Korean], https://mensa.kr/bbs/faq.php (retrieved 28 October 2026); Trahan, L. H., Stuebing, K. K., Fletcher, J. M., & Hiscock, M. (2014), The Flynn effect: A meta-analysis, Psychological Bulletin, 140(5), 1332-1360, https://doi.org/10.1037/a0037173; Johnson, W., Carothers, A., & Deary, I. J. (2008), Sex differences in variability in general intelligence: A new look at the old question, Perspectives on Psychological Science, 3(6), 518-531, https://doi.org/10.1111/j.1745-6924.2008.00096.x; Mensa International (n.d.), Getting your IQ tested FAQs, https://mensa.org/getting-your-iq-tested-faqs/ (retrieved 28 October 2026); Plomin, R., & Deary, I. J. (2015), Genetics and intelligence differences: Five special findings, Molecular Psychiatry, 20(1), 98-108, https://doi.org/10.1038/mp.2014.105; Winter, E. L., Trudel, S. M., & Kaufman, A. S. (2024), Wait, where’s the Flynn effect on the WAIS-5?, Journal of Intelligence, 12(11), 118, https://doi.org/10.3390/jintelligence12110118; Winter, E. L., Dale, B. A., Maharjan, S., Lando, C. R., Larsen, C. C., & Courville, T. (2025), Cognitive aging revisited: A cross-sectional analysis of the WAIS-5, Journal of Intelligence, 13(7), 85, https://doi.org/10.3390/jintelligence13070085; Seo, R., Sung, K., Kim, J., Hwang, S., Park, K., Choi, J., & Hong, S. (2014), K-WAIS-IV profiles of patients with schizophrenia [in Korean], Korean Journal of Clinical Psychology, 33(4), 719-734, https://www.accesson.kr/kjcp/assets/pdf/16896/journal-33-4-719.pdf; Watkins, M. W., & Smith, L. G. (2013), Long-term stability of the Wechsler Intelligence Scale for Children—Fourth Edition, Psychological Assessment, 25(2), 477-483, https://doi.org/10.1037/a0031653; Whitaker, S. (2008), The stability of IQ in people with low intellectual ability: An analysis of the literature, Intellectual and Developmental Disabilities, 46(2), 120-128, https://doi.org/10.1352/0047-6765(2008)46[120:TSOIIP]2.0.CO;2; Calamia, M., Markon, K., & Tranel, D. (2013), The robust reliability of neuropsychological measures: Meta-analyses of test–retest correlations, The Clinical Neuropsychologist, 27(7), 1077-1105, https://doi.org/10.1080/13854046.2013.809795; Steger, D., Schroeders, U., & Gnambs, T. (2020), A meta-analysis of test scores in proctored and unproctored ability assessments, European Journal of Psychological Assessment, 36(1), 174-184, https://doi.org/10.1027/1015-5759/a000494; Condon, D. M., & Revelle, W. (2014), The international cognitive ability resource: Development and initial validation of a public-domain measure, Intelligence, 43, 52-64, https://doi.org/10.1016/j.intell.2014.01.004; Roth, B., Becker, N., Romeyke, S., Schäfer, S., Domnick, F., & Spinath, F. M. (2015), Intelligence and school grades: A meta-analysis, Intelligence, 53, 118-137, https://doi.org/10.1016/j.intell.2015.09.002; Poropat, A. E. (2009), A meta-analysis of the five-factor model of personality and academic performance, Psychological Bulletin, 135(2), 322-338, https://doi.org/10.1037/a0014996; Schmidt, F. L., & Hunter, J. E. (1998), The validity and utility of selection methods in personnel psychology: Practical and theoretical implications of 85 years of research findings, Psychological Bulletin, 124(2), 262-274, https://doi.org/10.1037/0033-2909.124.2.262; Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2023), Revisiting the design of selection systems in light of new findings regarding the validity of widely used predictors, Industrial and Organizational Psychology, 16(3), 283-300, https://doi.org/10.1017/iop.2023.24; Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022), Revisiting meta-analytic estimates of validity in personnel selection: Addressing systematic overcorrection for restriction of range, Journal of Applied Psychology, 107(11), 2040-2068, https://doi.org/10.1037/apl0000994; Strenze, T. (2007), Intelligence and socioeconomic success: A meta-analytic review of longitudinal research, Intelligence, 35(5), 401-426, https://doi.org/10.1016/j.intell.2006.09.004; Calvin, C. M., Deary, I. J., Fenton, C., Roberts, B. A., Der, G., Leckenby, N., & Batty, G. D. (2011), Intelligence in youth and all-cause-mortality: Systematic review with meta-analysis, International Journal of Epidemiology, 40(3), 626-644, https://doi.org/10.1093/ije/dyq190; MacCann, C., Jiang, Y., Brown, L. E. R., Double, K. S., Bucich, M., & Minbashian, A. (2020), Emotional intelligence predicts academic performance: A meta-analysis, Psychological Bulletin, 146(2), 150-186, https://doi.org/10.1037/bul0000219; Haworth, C. M. A., Wright, M. J., Luciano, M., Martin, N. G., de Geus, E. J. C., van Beijsterveldt, C. E. M., et al. (2010), The heritability of general cognitive ability increases linearly from childhood to young adulthood, Molecular Psychiatry, 15(11), 1112-1120, https://doi.org/10.1038/mp.2009.55; Briley, D. A., & Tucker-Drob, E. M. (2013), Explaining the increasing heritability of cognitive ability across development: A meta-analysis of longitudinal twin and adoption studies, Psychological Science, 24(9), 1704-1713, https://doi.org/10.1177/0956797613478618; Tucker-Drob, E. M., & Bates, T. C. (2016), Large cross-national differences in gene × socioeconomic status interaction on intelligence, Psychological Science, 27(2), 138-149, https://doi.org/10.1177/0956797615612727; Breit, M., Scherrer, V., Tucker-Drob, E. M., & Preckel, F. (2024), The stability of cognitive abilities: A meta-analytic review of longitudinal studies, Psychological Bulletin, 150(4), 399-439, https://doi.org/10.1037/bul0000425; Deary, I. J., Pattie, A., & Starr, J. M. (2013), The stability of intelligence from age 11 to age 90 years: The Lothian birth cohort of 1921, Psychological Science, 24(12), 2361-2368, https://doi.org/10.1177/0956797613486487; Pietschnig, J., & Voracek, M. (2015), One century of global IQ gains: A formal meta-analysis of the Flynn effect (1909–2013), Perspectives on Psychological Science, 10(3), 282-306, https://doi.org/10.1177/1745691615577701; Bratsberg, B., & Rogeberg, O. (2018), Flynn effect and its reversal are both environmentally caused, Proceedings of the National Academy of Sciences, 115(26), 6674-6678, https://doi.org/10.1073/pnas.1718793115; Dworak, E. M., Revelle, W., & Condon, D. M. (2023), Looking for Flynn effects in a recent online U.S. adult sample: Examining shifts within the SAPA Project, Intelligence, 98, 101734, https://doi.org/10.1016/j.intell.2023.101734; Ritchie, S. J., & Tucker-Drob, E. M. (2018), How much does education improve intelligence? A meta-analysis, Psychological Science, 29(8), 1358-1369, https://doi.org/10.1177/0956797618774253; Melby-Lervåg, M., Redick, T. S., & Hulme, C. (2016), Working memory training does not improve performance on measures of intelligence or other measures of “far transfer”: Evidence from a meta-analytic review, Perspectives on Psychological Science, 11(4), 512-534, https://doi.org/10.1177/1745691616635612; Au, J., Sheehan, E., Tsai, N., Duncan, G. J., Buschkuehl, M., & Jaeggi, S. M. (2015), Improving fluid intelligence with training on working memory: A meta-analysis, Psychonomic Bulletin & Review, 22(2), 366-377, https://doi.org/10.3758/s13423-014-0699-x; Sear, R. (2022), ‘National IQ’ datasets do not provide accurate, unbiased or comparable measures of cognitive ability worldwide [preprint, version 2 posted 24 August 2026], PsyArXiv, https://doi.org/10.31234/osf.io/26vfb; Wicherts, J. M., Borsboom, D., & Dolan, C. V. (2010b), Why national IQs do not support evolutionary theories of intelligence, Personality and Individual Differences, 48(2), 91-96, https://doi.org/10.1016/j.paid.2009.05.028; Wicherts, J. M., Dolan, C. V., & van der Maas, H. L. J. (2010a), A systematic literature review of the average IQ of sub-Saharan Africans, Intelligence, 38(1), 1-20, https://doi.org/10.1016/j.intell.2009.05.002; Cho, M., Ko, E., Choi, H., Han, D. G., Kim, H. J., Kim, B., & Park, M.-H. (2025), The mental health profile of borderline intellectual functioning: A systematic review, Psychiatry Investigation, 22(11), 1230-1242, https://doi.org/10.30773/pi.2025.0209