How Stable Are Intelligence-Test Scores Across the Lifespan?

What does a lifespan stability correlation measure?

A lifespan stability correlation measures whether people who score relatively higher than others at one testing occasion also tend to score relatively higher when tested again. The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies calls this rank-order stability: the association between people’s relative scores across two occasions. It summarizes how scores move together in a sample. It does not mean that a given percentage of people received exactly the same score twice, or say whether the group’s average increased or decreased. Those are separate questions. A raw score is a person’s performance under a test’s scoring rules, before interpretation against a reference group. Imagine two people who take the same test at two ages. At the first occasion, one scores higher; at the second, both score higher, but the first still scores higher than the other. Their raw scores changed while their ordering stayed the same. A high rank-order correlation can describe that pattern. Mean-level change asks whether the group average shifted; rank-order stability asks whether relative positions were associated. Neither answers the other question. Individual-level change is different again: it asks how a particular person’s score changed between occasions. A correlation is calculated across people, so it does not report the size or direction of change for each person. It also cannot identify what caused a score to change. The meta-analysis separates mean-level change, individual-level change, and rank-order stability for this reason. It synthesizes 205 longitudinal studies, 87,408 participants, and 1,288 test-retest correlations, then models an average pattern across studies. That estimate is not a forecast for one examinee. A coefficient belongs to the scores, ages, interval, and sample that produced it. When a source calls scores “stable,” ask which kind of stability it measured. A rank-order correlation supports a claim about the persistence of relative differences in that study’s sample. It leaves group-average movement and any one person’s change to be examined separately. This distinction helps readers interpret lifespan evidence without turning a sample-level association into a claim that scores—or people—remain unchanged.

Sources: The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies

What is the broad lifespan pattern from adolescence to old age?

A lifespan stability correlation is an association between people’s scores at two occasions. In the 2024 meta-analysis, the modeled average rank-order stability for participants around age 20 retested five years later was ρ = .76. The review also found high stability from late adolescence through late adulthood. That is a broad pattern across studies, not a coefficient that predicts what will happen to every person or test. “The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies” combined 205 longitudinal studies involving 87,408 participants. The researchers analyzed 1,288 test–retest correlations between observed, or manifest, cognitive ability scores. A test–retest correlation summarizes how scores on one occasion are associated with scores on another. Rank-order stability asks whether people who scored relatively higher than others at one occasion also tended to score relatively higher at the next. The coefficient does not mean that 76 percent of participants kept exactly the same score. The age pattern was not flat. The meta-analysis modeled stability as relatively low in preschool years, rising rapidly during childhood, then remaining consistently high from late adolescence through late adulthood. It does not mean that the same people were followed continuously from childhood to old age in one representative sample. The curve combines evidence gathered under different study conditions, so its age estimates are a synthesis rather than a personal timetable. The interval between test occasions also mattered. Stability declined as the retest interval grew longer, while the modeled decline flattened from about five years onward. The age-20, five-year estimate is a reference point chosen from the model, not a universal lifespan constant. A correlation from a short interval and one from decades apart answer different questions, even if both are called “stability.” The review supports substantial persistence of relative differences in adulthood while allowing variation by interval, measure, and sample. The authors examined different cognitive abilities rather than treating “intelligence” as an identical measure in every study. At the modeled age of 20 and five-year interval, estimated stability varied by ability, roughly from .65 to .80. General intelligence was among the more stable abilities, but many specific abilities were also substantially stable. This shows why one headline coefficient can hide differences in what tests assess. It does not establish that a browser quiz measures the same constructs or has the same stability. One methodological distinction affects the numbers. The 1,288 correlations were based on manifest scores: scores as measured, with measurement error included. The meta-analysis also reports analyses adjusted for test reliability, often called disattenuated estimates. This adjustment aims to estimate the association between underlying scores after accounting statistically for imperfect measurement. It changes the question and depends on reliability estimates; it is not simply a more accurate version of the observed correlation for every practical purpose. The age pattern remained similar after adjustment, but observed and adjusted coefficients should not be treated as the same statistic. The synthesis has limits. Studies differed in tests, abilities, ages, intervals, and participant groups, and the authors report substantial between-study heterogeneity. A modeled mean summarizes varied evidence but cannot erase differences among its parts. The .76 estimate informs us about the broad evidence base at a specified age and interval; it is not a guaranteed retest result, individual forecast, or proof that scores remain unchanged. Read the meta-analysis as evidence for a strong adult pattern in relative ordering, alongside dependence on age, elapsed time, and measured ability. It offers breadth across many longitudinal studies; the next question is how that pooled pattern compares with unusually long follow-ups of the same historical test.

Sources: The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies

What do the Scottish follow-ups show across six to eight decades?

In the 1932 Scottish Mental Survey follow-up, the observed age-11-to-77 correlation was .63; in a separate Lothian Birth Cohort of 1921 analysis, the age-11-to-90 correlation was .54. Both studies also report higher estimates after correcting for the restricted range of childhood scores in their follow-up samples: .73 and .67, respectively. These are correlations between people’s scores in the study samples, not probabilities that an individual’s score stayed the same. The 1932 survey tested 87,498 Scottish schoolchildren born in 1921 and attending school on June 1. In “The Stability of Individual Differences in Mental Ability from Childhood to Old Age: Follow-up of the 1932 Scottish Mental Survey,” the researchers report retesting 101 people at age 77, 66 years after the original assessment. They re-administered the Moray House Test, a group test that had been used in the survey. The study also reports a change in the group’s average raw score. For those 101 participants, the mean rose from 43.3 at age 11 to 54.2 at age 77. A raw score is the number or value earned under a particular test’s scoring rules; it is not automatically an age-adjusted score or a modern IQ point. People can earn different scores later while retaining some of their relative ordering: those who scored higher than peers at the first occasion tended, on average, to score higher at the second. Why are the corrected correlations larger? A correlation can be reduced when the retested group covers a narrower spread of initial scores than the population from which it came. The researchers therefore estimated what the association might have been with a broader range of childhood ability represented. This range-restriction correction raises .63 to .73. It is an estimate, not a newly observed correlation among every child in 1932, and depends on assumptions about how selection narrowed the range. The observed result and corrected estimate should be shown separately, not blended into a single supposedly exact value. Neither number is an individual probability or a forecast for a new test. A correction does not repair missing follow-up data or reveal the unobserved scores of people who did not return; it estimates an association under a specified statistical adjustment. The age-90 result comes from a different follow-up, not from the same 101 age-77 participants returning at 90. “The Stability of Intelligence from Age 11 to Age 90 Years: The Lothian Birth Cohort of 1921” reports the same Moray House Test administered to 106 people at ages 11 and 90. Its abstract gives an observed correlation of .54 and a range-restriction-corrected estimate of .67. These numbers support the direct answer, but the different samples and follow-up designs matter as much as their size. The two sets of figures should not be treated as points on a curve from .63 at 77 to .54 at 90. They come from distinct follow-up analyses, with different participant sets and ages, and the age-90 study does not retest the age-77 sample. The lower observed coefficient at 90 therefore does not, by itself, demonstrate that stability weakened between those ages. Nor does the difference between either observed value and its corrected counterpart tell us how much any one participant changed. Taken together, the Scottish studies show that childhood and late-life scores can remain substantially associated across six to eight decades, even while average raw performance differs across occasions. The same-test follow-up is unusually long, but its small returning subsets cannot automatically represent every original child or an individual reader. Corrected values are estimates under assumptions; observed values describe the retested samples. Read each with its sample, ages, measure, and correction attached.

Sources: The Stability of Individual Differences in Mental Ability from Childhood to Old Age: Follow-up of the 1932 Scottish Mental Survey; The Stability of Intelligence from Age 11 to Age 90 Years: The Lothian Birth Cohort of 1921

How should long follow-ups be compared fairly?

A fair comparison starts with the design behind each coefficient: participants’ ages at first testing, retest interval, who returned, whether the instrument and outcome stayed the same, and whether the value is observed or adjusted. The 2024 meta-analysis, Scottish follow-ups, and adult twin study all examine persistence in cognitive scores, but across different ages, samples, time spans, and measures. “The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies” combines 205 studies, 87,408 participants, and 1,288 test-retest correlations. Its estimate of .76 is modeled for a reference point, age 20 and a five-year interval. It summarizes patterns across many studies; it is not one cohort followed for five years or an individual forecast. The Scottish studies offer unusually long follow-up with the Moray House Test at age 11 and again in old age. “The Stability of Individual Differences in Mental Ability from Childhood to Old Age: Follow-up of the 1932 Scottish Mental Survey” reports 101 retested participants from an original survey of 87,498 Scottish schoolchildren. The observed age-11-to-77 correlation was .63; adjusted for restricted ability range in the returning sample, it was .73. A separate study, “The Stability of Intelligence From Age 11 to Age 90 Years,” reports .54 observed and .67 adjusted among 106 Lothian Birth Cohort participants tested at 11 and 90. The age-77 and age-90 results come from distinct follow-ups. Range correction addresses the possibility that returnees cover a narrower range of childhood scores than the original cohort. A narrower range can reduce an observed correlation. The correction estimates an association over a broader range under assumptions; it does not recover missing scores or measure everyone from the original survey. Observed and adjusted figures answer related but different questions, so readers should not simply choose the larger value. “Stability and change in intelligence from age 11 to ages 70, 79, and 87” reports raw correlations of .67 from age 11 to 70, .66 to 79, and .51 to 87 in Lothian cohorts. The variation shows that coefficients can differ by endpoint and returning sample, even in related cohorts. The separate age-90 estimate cannot prove stability declined from 79 to 90: this is not a controlled comparison of an unchanged sample at each age. “Genes Determine Stability and the Environment Determines Change in Cognitive Ability During 35 Years of Adulthood” contributes a different comparison. It followed 7,232 male twins tested with the Armed Forces Qualification Test in early adulthood, then retested a subset of 1,237 in late middle age. The correlation was .74. This describes adult stability over 35 years in a specific cohort using another instrument. The paper’s twin-model analysis of causes is a separate inference dependent on design and assumptions; the correlation alone does not identify causes. These values are not a leaderboard. The .76 is modeled across studies at a specified age and interval; .63 and .54 are observed childhood-to-old-age associations in selected follow-ups; .73 and .67 are adjusted estimates; and .74 comes from an adult twin cohort over 35 years. Compare what each design contributes: breadth, very long same-test follow-up, variation across endpoints, or adult evidence from another population. They cannot be ranked as if age, interval, measure, retention, and correction were held constant.

Sources: The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies; The Stability of Individual Differences in Mental Ability from Childhood to Old Age: Follow-up of the 1932 Scottish Mental Survey; The Stability of Intelligence from Age 11 to Age 90 Years: The Lothian Birth Cohort of 1921; Stability and change in intelligence from age 11 to ages 70, 79, and 87: the Lothian Birth Cohorts of 1921 and 1936; Genes Determine Stability and the Environment Determines Change in Cognitive Ability During 35 Years of Adulthood

Does stability mean that aging decline is absent?

No. A correlation can show that people who scored relatively higher at one age tended to score relatively higher later; it cannot show that scores, group averages, or every person’s performance stayed fixed. In the Lothian Birth Cohorts analysis, researchers compared Moray House Test performance first recorded at age 11 with later scores: age 70 for the 1936 cohort, and ages 79 and 87 for the 1921 cohort. The raw correlations were .67, .66, and .51. In the older cohort, the age-11 score did not predict how much a participant’s score changed between ages 79 and 87. Early performance was associated with later score level, but did not explain the amount of change during that later interval. This concerns these cohorts and analyses; it is not evidence that aging causes no decline, or a forecast for an individual. (“Stability and change in intelligence from age 11 to ages 70, 79, and 87: the Lothian Birth Cohorts of 1921 and 1936.”) Keep three questions apart: Did the average score move? Did a particular person’s score change? Did people retain roughly the same relative ordering? The first is mean-level change; the second is individual-level change; the third is rank-order stability. A high correlation answers the third. In a hypothetical group, everyone’s raw score could rise by the same amount while their ordering remains unchanged. Conversely, the average might barely move while some people improve and others decline, shifting positions. Neither pattern can be read from one correlation alone. In Lothian, childhood-to-old-age association and change within old age were examined separately. The broader evidence leaves room for change. “The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies” synthesized 205 studies, 87,408 participants, and 1,288 correlations. It reports that measured abilities differ in stability patterns. Its discussion distinguishes effortful-processing abilities, such as reasoning and processing speed, from comprehension knowledge, and describes different average trajectories across adulthood. These patterns do not mean every person follows the average path, or that change in one ability means the same change in another. A test samples particular tasks; its total score can combine abilities with different trajectories. Stable overall ordering can coexist with component changes, while a change in one component need not reorder everyone’s total scores. Substantial rank-order stability and real age-related change can both be true. Do not translate persistence of rank into biological permanence, or use a correlation to claim that an individual did not decline. Ask which score was compared, at what ages, in which retained sample, and whether the result concerns average movement, personal change, or relative ordering. In Lothian, the age-11 score tracked later level but did not account for the amount of change from 79 to 87. Stability describes a relationship among scores in a defined study; decline or improvement must be assessed with the relevant measure and outcome.

Sources: Stability and change in intelligence from age 11 to ages 70, 79, and 87: the Lothian Birth Cohorts of 1921 and 1936; The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies

What does a correlation leave unproved about one person or one test?

A correlation between scores at two ages cannot tell whether a particular person’s change is larger than expected measurement error, why that person changed, or whether a score should be used to make a diagnosis. It describes how scores vary together across the people included in a study. Even a strong association can coexist with movement in the average score or meaningful changes for some individuals. To decide whether one person changed reliably, a reader needs evidence about the specific test’s precision and a method for interpreting change across occasions; the correlation alone supplies neither.

A raw score is the number or pattern of responses recorded under a test’s scoring rules. It is a description of performance on that form, not yet a statement about how that performance compares with other people. A normed interpretation compares a score with a defined reference group using the test’s scoring system. Without a suitable reference group and a documented conversion, a raw result cannot be translated responsibly into a population rank or a conventional IQ scale. Likewise, a correlation reported for one instrument does not provide the conversion rules for another.

Validity concerns the evidence and theory supporting an interpretation of scores for a proposed use. The Standards for Educational and Psychological Testing describe validity in those terms: support attaches to a particular interpretation and use, rather than automatically to a test label or any number that resembles a familiar scale. A lifespan study can contribute evidence about the persistence of relative scores on the measures it studied. It cannot, by that fact alone, establish that a different test measures the same construct, that its score conversion is sound, or that the result answers a clinical or educational question. Those claims require evidence tied to the other instrument, the examinee population, administration, scoring, and intended decision.

That boundary matters for Test IQ Free. The publication describes an adaptive 50-question reasoning test covering pattern, number, and spatial reasoning. Its displayed estimated IQ range comes from a provisional, unvalidated conversion with no validated adult reference population. The range is neither a confidence interval nor an established comparison with a population. The long-term studies discussed in this article did not evaluate this test or its conversion, so their correlations cannot validate that range or establish what it means for an individual. The site’s guide presents the result as provisional and directs readers to methodology information; that product description is not independent validation evidence.

The optional $9 report adds explanation and practice structure around the same provisional result, not validation or accuracy. Answers and domain feedback describe performance during that session; the estimated range carries a stronger interpretation that the current evidence does not support.

A practical rule follows: use stability research to understand the pattern reported for the study’s own participants, test, ages, and interval. For a personal score, separately ask how the test was administered, what its raw result represents, what reference group supports any comparison, and what evidence supports the use being proposed. The Standards for Educational and Psychological Testing emphasize evidence for intended score interpretations; the 2024 meta-analysis estimates associations across other studies and measures. Neither licenses transferring those findings to an unrelated online conversion. For a consequential decision, seek a qualified assessment whose evidence and interpretation fit that decision.

Sources: Standards for Educational and Psychological Testing; Intelligence and IQ Test Guide | Test IQ Free; The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies

Why can two sound stability estimates differ?

Two stability estimates can both be sound and still differ because they summarize different people, ages, intervals, tests, and adjustments. A coefficient is not a free-standing property of intelligence; it is an association between particular scores. Comparison starts with the study design and number construction, not with which result looks larger.

Elapsed time matters. “The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies” synthesized 205 studies and 1,288 correlations, finding that stability declined as the retest interval lengthened, with the modeled decline flattening from about five years onward. That broad pattern does not require every pair of studies to follow the same curve: the synthesis spans different ages and measures. A five-year adult estimate is not directly comparable to an age-11-to-90 correlation. The age-90 Moray House follow-up, for example, tested 106 people at ages 11 and 90; its unusually long interval is part of what the result describes.

The participants retained matter too. “Stability of IQ from Age 11 to Age 77 Years: The Scottish Mental Survey 1932–1998” retested 101 adults from a much larger school survey. Their observed correlation was .63. Returning participants may cover a narrower range of childhood scores than the original population; restricted range can weaken an observed correlation. The paper reports .73 after correcting for range restriction. The age-90 study likewise reports .54 observed and .67 corrected. These adjusted values estimate a broader-range association under assumptions; they are not additional observed participants and do not automatically replace the observed values. Selective return can also change who is represented, so the follow-up group is not a miniature of everyone originally tested.

Measurement consistency is another limit. A test with more score noise generally yields a weaker observed retest association than a more reliable measure. The 2024 meta-analysis examined reliability-adjusted analyses, but reliability data covered a less comprehensive, potentially unrepresentative subset. An adjusted estimate therefore depends on available reliability information and its assumptions; label it separately from the manifest-score correlation calculated from scores as measured. “The Stability of Intelligence from Age 11 to Age 90 Years” and the age-77 Scottish study used the Moray House Test at both occasions. The meta-analysis combines many instruments and abilities. Different test content samples performance differently, and age may change which abilities are sampled well. Coefficients from distinct measures are not interchangeable readings of one fixed quantity.

A headline coefficient should come with its denominator in plain language: who returned, what test and scores were compared, at which ages and interval, and whether the number is observed or adjusted. Those details explain why estimates can differ without either study being wrong. They also keep an adjusted population association from being mistaken for a personal forecast or a correction from being treated as certainty.

Sources: The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies; The Stability of Individual Differences in Mental Ability from Childhood to Old Age: Follow-up of the 1932 Scottish Mental Survey; The Stability of Intelligence from Age 11 to Age 90 Years: The Lothian Birth Cohort of 1921

What is the responsible way to read a stability claim?

Read a stability coefficient as a description of one comparison, not a verdict about a person. Ask who was included, which scores were correlated, how far apart assessments were, and whether the number was observed or corrected. Separate three questions: did people retain relative ordering, did the group average move, and did a particular person change beyond measurement uncertainty? A correlation addresses the first most directly. The 2024 meta-analysis, “The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies,” pooled 1,288 test–retest correlations from 205 studies involving 87,408 participants. Its modeled estimate at age 20 over five years was .76; it found high rank-order stability from late adolescence through late adulthood. This synthesis describes a broad pattern, not a personal forecast. The Scottish follow-ups add unusually long intervals using the same Moray House Test. The 1932 survey began with 87,498 children, but its age-77 analysis included 101 returnees. Their age-11-to-77 correlation was .63 observed and .73 after correction for restricted range. A separate age-11-to-90 follow-up reported .54 observed and .67 corrected among 106 older participants. Corrected figures estimate an association with a broader score range under assumptions; they are not observations of everyone in the original surveys. Mean raw scores also differed between ages 11 and 77 in the first follow-up. Average movement can coexist with correlation: scores may change while relative order persists. Historical raw scores do not translate directly into modern IQ-point change. Another PubMed abstract reports raw age-11-to-70, -79, and -87 correlations of .67, .66, and .51 in a Lothian cohort. Childhood score did not predict the amount of change from ages 79 to 87 in that analysis. An association over decades does not settle who changed within a later interval. Each result belongs to its cohort, retained participants, test, and analysis. Do not line these numbers up as a contest. The meta-analysis averages across designs; the Scottish results track selected historic cohorts on one test; an adult twin study follows a specific male-twin cohort over 35 years. First-test age, interval, measure, sample range, retention, and correction differ, so their coefficients are not directly comparable. Write those design details beside a coefficient, then read raw-score averages separately. Ask whether a claim concerns relative ordering, average movement, or one person’s change. The Standards for Educational and Psychological Testing frame validity as evidence supporting a proposed score interpretation and use. Correlations from other instruments cannot provide that evidence for a new test. That boundary matters for Test IQ Free. Its adaptive 50-question test covers pattern, number, and spatial reasoning; its estimated range uses an unvalidated conversion with no validated adult reference population. The range is neither a confidence interval nor an established population comparison. Better representative long-term research using comparable measures could refine lifespan estimates, but would not validate an unrelated browser score. Evidence supports substantial adult rank-order stability in studied samples, with limits on what it establishes about an individual.

Sources: The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies; The Stability of Individual Differences in Mental Ability from Childhood to Old Age: Follow-up of the 1932 Scottish Mental Survey; Stability and change in intelligence from age 11 to ages 70, 79, and 87: the Lothian Birth Cohorts of 1921 and 1936; Standards for Educational and Psychological Testing

What can you responsibly do next?

For low-stakes curiosity, take a browser reasoning test as practice and review its worked answers. Treat answers and domain feedback as performance on those questions in that session; they can help you choose a reasoning operation to practice. Test IQ Free offers an adaptive 50-question test across pattern, number, and spatial reasoning. Its displayed IQ range is a provisional estimate from an unvalidated conversion with no validated adult reference population. It is not a confidence interval or established comparison with other adults, so the free test is enough to try the questions without treating its range as personal standing. The optional $9 report adds explanation and a structured practice plan, but uses the same unvalidated conversion. Payment does not make the estimate more accurate or show that practice raises general intelligence. If a result will affect a consequential educational, clinical, or other decision, seek a qualified professional and an assessment selected for that purpose. The “Standards for Educational and Psychological Testing” frames validity as evidence supporting a proposed score interpretation and use; a browser result alone cannot supply that evidence. Practice freely for curiosity; seek qualified interpretation when the decision requires it.

Sources: Intelligence and IQ Test Guide | Test IQ Free; Standards for Educational and Psychological Testing

Questions readers ask