IQ tests are highly accurate at ranking people on the narrow skills they sample—full-scale scores are reliable to within roughly 3 to 5 points on retest—but they are not a complete measure of intelligence, and their accuracy drops sharply when you try to read a single score as a verdict on a person. That distinction between statistical precision and interpretive overreach is where most public confusion about IQ accuracy comes from.
The reliability of a modern IQ test is genuinely impressive. The validity of using it to predict school performance, job performance, or health outcomes is real but bounded. And the accuracy of treating a score as a fixed, innate number stamped on a brain is poor. This explainer separates those three claims, because they have different evidence bases and different error bars.
What an IQ score actually computes
An IQ score is not a count of correct answers. It is a position in a distribution, derived from a norming sample, expressed on a scale with a mean of 100 and a standard deviation typically set to 15 for the major adult tests. A score of 115 means the tester placed about one standard deviation above the average of the reference group, not that the person answered 115 units of intelligence.
The mechanism has three layers. First, a battery of subtests samples specific abilities—vocabulary, matrix reasoning, working memory, processing speed, and others depending on the instrument. Second, raw scores on those subtests are converted to scaled scores using norms collected from a large, stratified sample. Third, the scaled scores are combined into index scores and a full-scale score, usually through factor-analytic weighting that reflects how the subtests cluster together.
That clustering is the load-bearing idea. Modern tests are built on a hierarchical model in which a general factor (g) sits above broad domains such as verbal comprehension and fluid reasoning, which sit above narrow tasks. The full-scale score is an estimate of that general factor. The index scores are estimates of the broad domains. A person's full-scale score can mask substantial unevenness—someone with strong verbal skills and weak processing speed can land at 105 while their two index scores differ by 20 points or more.
Norms are the other mechanism most people miss. A raw score earns its IQ meaning only relative to the reference sample. Tests are restandardized periodically because populations drift—average performance on many reasoning tasks has risen over the past century, a phenomenon called the Flynn effect, for which norms must be refreshed. A score derived from stale norms will systematically overstate standing, which is why the age of the norming sample matters as much as the test itself. [CIT2021, authoritative test publisher technical manual]
Quantified reliability and validity
Reliability and validity are different questions. Reliability asks whether a test gives consistent answers. Validity asks whether it measures what it claims and predicts what it should. An IQ test can be highly reliable and still overinterpreted.
On reliability, the full-scale score is the strongest component. Internal consistency and test-retest coefficients for full-scale IQ on major adult instruments commonly fall in the range of about 0.90 to 0.97, depending on the instrument and the population studied. Converted to practical terms, that implies a standard error of measurement of roughly 3 to 5 points. This is why competent assessors report a confidence interval around a score—often something like 100 to 110—rather than a single number presented as exact.
On validity, the correlations are moderate to strong for specific outcomes and weaker for broad claims. Full-scale IQ tends to correlate with academic achievement at roughly 0.5 to 0.7 in many studies, with job performance around 0.3 to 0.5, and with income more weakly still. These are group-level relationships. They tell you something about the predictive power of the score across many people, not what any one person will do.
The counterintuitive detail most thin explainers skip is that the subtest profile is often the least trustworthy part of a report. Subtest scores have lower reliability than the full-scale score, and the gap between two index scores needs to be fairly large—often 10 to 15 points or more, depending on the instrument—before it is statistically meaningful rather than noise. Assessors who interpret every 5-point subtest dip are reading a signal the test cannot deliver. The full-scale score is the most reliable number on the page; the fine detail around it is the least.
A comparison table of what IQ predicts
| Outcome | Typical correlation with full-scale IQ | How to read it |
|---|---|---|
| School achievement | Roughly 0.5 to 0.7 | Moderate to strong; IQ is one predictor among several |
| Job performance | Roughly 0.3 to 0.5 | Moderate; stronger for complex roles than routine ones |
| Income | Roughly 0.2 to 0.4 | Modest; many other factors dominate |
| Health and longevity | Modest, positive | Real association, causal pathways still debated |
| Creativity | Weak to moderate | Open-ended tasks reward traits IQ does not sample well |
| Personality or character | Near zero | IQ is not designed to measure these |
Read this table as a map of signal strength, not a hierarchy of importance. A modest correlation can be statistically solid and still be nearly useless for predicting any individual person.
Where accuracy breaks down: edge cases and limits
Accuracy degrades in specific, predictable places. The first is the population tested. Most major IQ tests were normed on general samples, and their behavior at the extremes—very low or very high scores—is less precise than near the middle. Extreme scores are also where ceiling and floor effects bite, compressing differences between people at the tails.
The second is the testing condition. People take tests under testing conditions: standardized instructions, timed subtests, a quiet room, an unfamiliar examiner. Fatigue, anxiety, illness, medication, sleep loss, and language background can all move scores by meaningful amounts. A person tested while sick is not getting a clean reading of their stable ability.
The third is the construct itself. IQ tests sample reasoning, vocabulary, working memory, and processing speed. They do not sample sustained effort over long projects, social judgment, creativity, musical or athletic skill, persistence, or practical knowledge outside the test's item bank.
The fourth is the group-level gap problem. There are well-documented average score differences between demographic groups, and the mainstream interpretation in the field is that environmental and test-construction factors account for the gap. The point for accuracy is narrower: group averages say very little about any individual within a group, and treating an average as a prediction about a person is a use the test was never built to support.
The fifth is timing. A single test at a single age is a snapshot. Stability of IQ scores increases with age, but a score in childhood is a less reliable predictor of adult scores than a score taken in late adolescence or adulthood. The test measures current functioning against a norm group; it does not measure a fixed ceiling.
Red flags when reading any IQ report:
The report presents one exact score with no confidence interval. Subtest or index differences are interpreted as fixed traits with no error band. The score is used to predict life outcomes for a person rather than describe current performance. The norms are old or unstated, so no one can judge the reference group. The testing conditions—time limits, language, health—are not documented. A single score is used to decide something high-stakes with no other evidence.
Common misconceptions that inflate the accuracy of IQ tests
The biggest misconception is that an IQ score is a fixed number. It is a measurement with a confidence interval, taken at a point in time, against a particular norm group. Repeating the test months later can produce a different score, most of which is measurement noise rather than a change in the person.
The second is that the full-scale score and the subtest profile are equally trustworthy. They are not. The full-scale score carries the error band of the sum of many items, while individual subtests and index gaps carry more noise. That is why the most defensible reading of a report is usually the full-scale score plus a judgment about whether the overall picture is consistent.
The third is that a high IQ means broad competence and a low IQ means limited capacity. IQ predicts academic-style performance fairly well but says little about practical judgment, character, or specific talents outside the test's item types.
The fourth is the idea that a score difference between two people of a few points means one is meaningfully "smarter." With a standard error of measurement of 3 to 5 points, a 3-point gap is well within noise. A meaningful difference in a practical sense needs to be larger than the error band, and even then it is a statement about measured performance, not an essence.
The fifth is that "IQ tests are biased and therefore useless." The more defensible position is that IQ tests measure what they measure with real reliability and real predictive power, and that this predictive power is narrow and must be interpreted by someone who understands its error bands. Dismissing the test entirely and worshipping the score are both errors. The science supports a middle reading: a trustworthy instrument for a limited question.
Practical steps for interpreting an IQ score
Start by finding the confidence interval. Every credible report should state the error band around the full-scale score. If the number is 105 with a range of 100 to 110, treat the range as the result and the single number as shorthand. Do not compare scores across two people unless the gap exceeds roughly 8 to 10 points, because smaller gaps are within combined measurement error.
Next, check the norming date. Ask when the norms were collected and whether the test has been updated since. A score is only as current as its reference group. If the norms are old, the score may overstate standing, and the report should say so.
Then examine the index scores for consistency. If the full-scale score and the index scores tell a coherent story, the full-scale number is the most defensible summary. If they diverge substantially, the report should explain why and the reader should treat the full-scale score with caution rather than as the whole truth.
Finally, treat the score as one input among several. Use it alongside grades, work samples, structured interviews, and observed performance when making decisions about a person. The test's real strength is in efficiently estimating a narrow set of cognitive skills; its real weakness is any decision that requires more than that estimate.
How accurate are IQ tests, in one honest answer
The full-scale IQ score is one of the more reliable measurements in psychology, with test-retest precision typically within about 3 to 5 points and moderate-to-strong correlations with academic and job outcomes at the group level. Its accuracy as a description of current cognitive performance against a defined norm group is high. Its accuracy as a verdict on a person's fixed intelligence, potential, or worth is low, because the test is not designed for that claim and its error bands, narrow item types, and norming conditions all limit what a single number can mean.
The practical rule is that IQ tests are accurate about what they measure and inaccurate about what they do not. A confidence interval, a current norm group, and a clear-eyed sense of the test's narrow item sample will get you most of the way toward reading a score correctly. Treating a 3-point difference as meaningful, an old norm as current, or a single subtest as a trait is where accuracy quietly leaks away.