Do IQ Tests Really Work? How Accurate and Valid They Are

Yes, within limits. A properly normed IQ test given one-to-one by a trained examiner measures one thing consistently, general reasoning ability compared with other people of the same age, and that measure predicts school results, years of education and, more loosely, job performance. It does not capture all of intelligence, a single score carries about 5 points of error either side, and an online test with no norm sample does not count as an IQ test. This guide shows how accurate IQ tests are, what the evidence says about whether they are valid, where they go wrong and how to judge any test you are shown.

• Accuracy: the standard error of measurement on a full-scale score is commonly 2 to 3 points, so read a reported score as a band of about 5 points either side (an IQ score is a range, not a number).
• Stability: adult scores keep their rank order over time, with a correlation of about .76 at age 20 across a five-year gap (Breit and colleagues, 2024); childhood scores move more.
• Prediction: intelligence at age 11 correlated .81, at the level of underlying traits, with national exam results at 16 in more than 70,000 English children (Deary and colleagues, 2007).
• Limits: creativity, motivation and practical judgement sit outside the score, and anxiety, practice and out-of-date norms each move it by a few points.
• Online tests: a number printed without a norm sample is not an IQ and cannot be compared with a clinical report.

What does it mean for an IQ test to work?

A test works if it passes four checks, and each has its own technical name. Reliability asks whether results repeat; validity asks whether the test measures what it claims to measure. A scale that reads 5 kilograms heavy every time is reliable but not valid, so a test needs both. The table sets out the four checks.

Our guide to how IQ tests work explains how a norm sample turns raw answers into a score, and are IQ tests reliable measures of intelligence and what makes an IQ test reliable go deeper on reliability.

How accurate is an IQ test?

The accuracy of one score is stated as the standard error of measurement: the typical distance between the score reported and the score the person would average over many sittings. It follows from the test’s reliability: the standard error equals the standard deviation times the square root of one minus the reliability. The table applies that formula on the usual scale, where the standard deviation is 15.

Full-scale scores on well-made batteries sit near the bottom of that table, which is why manuals report 2 to 3 points and why a reported 100 is best read as 95 to 105. Index and subtest scores rest on fewer items, so they are less reliable than the total. The practical rule is that a gap of a few points between two scores is not a real difference, and that comparing bands is safer than comparing numbers.

Accuracy also means stability over time. A meta-analysis of 205 longitudinal studies found a rank-order stability of about .76 for adults at age 20 over a five-year gap (Breit and colleagues, 2024), so most people keep their place but individuals do move, and childhood scores are less stable than adult ones (see what 11,119 children show). Our guides to how long an IQ test stays valid and the most accurate age to be tested cover the timing.

Do IQ tests predict anything that matters?

The cleanest large-scale evidence is educational. In a study of more than 70,000 English children, intelligence measured at age 11 correlated 0.81 with a general achievement factor built from national exam results in 25 subjects at age 16, and general intelligence contributed to success in all 25 (Deary and colleagues, 2007). That figure is a correlation between underlying traits, so it overstates what one raw score can tell you about one pupil, but the direction is not in doubt.

A meta-analysis of longitudinal studies found correlations of .56 with educational attainment, .43 with occupational level and .20 with income (Strenze, 2007). Squared, those figures say how much of the spread in an outcome goes with the score: about 31% for education, 18% for occupation and 4% for income. Tests predict tendencies in groups, not the future of one person. For more, see does IQ predict school success and does IQ predict income.

For job performance the picture is more disputed. The figure most often quoted for general mental ability, .51, comes from a 1998 review of 85 years of selection research (Schmidt and Hunter, 1998). A 2022 re-analysis in the Journal of Applied Psychology argued that the corrections for restriction of range used in earlier meta-analyses had overestimated the validity of many selection procedures, and it lowered the mean estimates by .10 to .20 points while most procedures that ranked high stayed high (Sackett and colleagues, 2022). Ability tests are still useful in hiring, but weaker predictors than the old headline implied; our guide to IQ tests for job interviews covers the hiring side.

What do IQ tests not measure?

A full-scale score summarises performance on tasks of reasoning, memory and speed. It says little about creativity (see IQ and creativity), emotional skills (IQ versus EQ), persistence (grit versus IQ) or practical judgement. Theories of multiple intelligences argue for a wider definition, and is intelligence limited to IQ sets out the debate. That makes the score narrow, not invalid: a thermometer that cannot report the humidity still works as a thermometer.

Where do IQ tests go wrong?

Five things most often move a score without moving the ability behind it, and each is worth a few points rather than a whole band:

• Anxiety: studies of test anxiety find a relationship near r = -.20 with test performance, about three points on the usual scale (test anxiety and IQ scores).
• Practice: retaking a cognitive test raises scores by about .26 standard deviations, roughly 4 points, more with coaching or when the same form is reused (Hausknecht and colleagues, 2007).
• Old norms: raw performance has risen about 0.28 points a year (Pietschnig and Voracek, 2015), so a test normed decades ago flatters today’s test taker.
• Language and culture: bias in the narrow statistical sense is usually small, but familiarity with the content can still help (cultural bias in IQ tests).
• Over-reading one number: a single score is a range, not a point, and it is a snapshot of one sitting.

Our guide to the factors that affect IQ test results ranks these by size and timescale.

Do online IQ tests work?

Mostly not as IQ tests. A clinical IQ test is individually given, timed under standard conditions and scored against a recent norm sample of thousands of people. Many online tests have none of that: nobody checks the conditions, there may be no published norm sample, and the number printed is a guess at where you would rank. A short online test can be a fair way to try puzzle formats (see our guide to good online IQ tests), but treat the result as a rough indication, not a measurement.

To compare an online number with a clinical report, check four things: whether a norm sample is described, which standard deviation the scale uses, whether a margin of error is printed and whether the timing was standardised. Our pages on the official IQ test and professional IQ tests show what a clinical result looks like.

How should you read an IQ result?

• Find out which test and which edition produced it, and when its norms were collected.
• Check the scale: most tests use a standard deviation of 15, some use 16 or 24 (standard deviation 15 versus 16).
• Read the number as a band of about 5 points either side, and use the label bands rather than single points (how to read an IQ test report).
• Translate it into a percentile with the IQ percentile calculator and compare it with other evidence such as school work.

Quick answers about whether IQ tests work

• Are IQ tests accurate? A well-made, individually given test is accurate to about 5 points either side of a reported score; it is far less accurate as a verdict on a person.
• Are IQ tests reliable? Yes: full-scale reliability is high, which is what a standard error of 2 to 3 points means.
• Are IQ tests valid? They predict school results and, more loosely, work outcomes, but they do not measure all of intelligence.
• Can an IQ test be wrong? Yes: anxiety, illness, practice and old norms can each move a score by a few points.
• Are online IQ tests accurate? Without a norm sample and standard conditions they cannot be compared with a clinical score.
• Do IQ tests measure intelligence? They measure general reasoning ability, which is part of intelligence. See the pros and cons of IQ testing for the full balance.

Sources

• Deary, I.J., Strand, S., Smith, P. and Fernandes, C. (2007): Intelligence and educational achievement, Intelligence, 35(1), 13-21.
• Strenze, T. (2007): Intelligence and socioeconomic success: a meta-analytic review of longitudinal research, Intelligence, 35(5), 401-426.
• Breit, M., Scherrer, V., Tucker-Drob, E.M. and Preckel, F. (2024): The stability of cognitive abilities: a meta-analytic review of longitudinal studies, Psychological Bulletin, 150(4), 399-439.
• Schmidt, F.L. and Hunter, J.E. (1998): The validity and utility of selection methods in personnel psychology, Psychological Bulletin, 124(2), 262-274.
• Sackett, P.R., Zhang, C., Berry, C.M. and 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.
• Hausknecht, J.P., Halpert, J.A., Di Paolo, N.T. and Moriarty Gerrard, M.O. (2007): Retesting in selection: a meta-analysis of coaching and practice effects for tests of cognitive ability, Journal of Applied Psychology, 92(2), 373-385.
• Pietschnig, J. and 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.
• Reliability and standard-error figures in the second table are computed from the formula shown; the 2 to 3 point standard error is the range described in our explainer on IQ score margin of error.