How Stable Are Adults’ Cognitive Ability Scores Over Five Years?

How stable are adult cognitive ability scores over five years?

On average, adults tend to preserve their relative ordering on cognitive measures over five years. That finding does not establish that any one person’s score stayed unchanged. In “The stability of cognitive abilities: A meta-analytic review of longitudinal studies,” published in 2024, researchers combined 205 longitudinal studies, covering 87,408 participants and 1,288 test–retest correlations. For a modeled age of 20 and a five-year interval, the mean observed rank-order stability was ρ = .76. Rank-order stability means how well people preserve their relative positions across two measurement occasions. If people who scored higher than others at the first assessment generally also scored higher at the second, their ordering was relatively stable. The statistic summarizes that pattern across a group. It does not say that each person received the same score twice, or that everyone changed by a similar amount. Some individuals can move up or down while the group’s overall ordering remains fairly similar. The value .76 is a correlation, not 76 percent of ability retained. It is not an expected change in IQ points, a percentile shift, or a confidence interval for one person. Correlation describes how two sets of scores vary together; it does not specify the size or direction of an individual’s difference. The meta-analysis reports a modeled average drawn from studies that used varied cognitive measures and samples, not one common test administered to a single representative group of adults. The age-20 figure is a model estimate, not a personal forecast for every adult. The study also reports that rank-order stability is consistently high from late adolescence through late adulthood, while declining as the retest interval lengthens; that decline flattens from about five years onward. These are broad patterns in the pooled research. They do not make five years a universal expiration date for a score, establish that an individual’s ability remained constant, or show how to interpret a particular pair of results. To make that judgment, the test, score type, conditions, and purpose of the comparison matter. The estimate answers a population question about relative ordering; it cannot by itself settle what changed for one person.

Sources: The stability of cognitive abilities: A meta-analytic review of longitudinal studies

What does “stable” mean when people compare two scores?

A score can be stable in one statistical sense and change in another. The word does not name a single result: it may refer to whether a group average moves, whether one person's measured level changes, whether people's relative ordering is preserved, or whether a person's pattern across abilities remains similar. These questions require different comparisons. A finding about one cannot automatically answer the others. The distinction is practical, not merely semantic.

Mean-level change asks whether the average score for a group rises, falls, or stays about the same between occasions. It describes the group center, not what happened to each member. Individual-level change asks whether a particular person's measured score differs over time. That difference is observable, but by itself it does not explain its cause or establish a lasting change in ability. Measurement precision and the conditions of each assessment matter to that interpretation.

Differential continuity, also called rank-order stability, asks whether people keep roughly the same relative positions compared with one another. The 2024 meta-analysis, “The stability of cognitive abilities: A meta-analytic review of longitudinal studies,” reports this kind of stability: across 205 longitudinal studies, the authors modeled a mean rank-order correlation of ρ = .76 for age 20 and a five-year interval. The quantity summarizes how scores across people at one occasion relate to scores across people at another. It does not say that a particular person retained 76% of an ability or changed by a specific number of points.

Ipsative continuity concerns the configuration of one person's scores across measured abilities: whether their relative strengths and weaknesses form a similar profile at both occasions. Consider an illustration, not a research finding: on an earlier assessment, a person’s verbal and spatial scores may be close, while their later profile shows verbal performance higher than spatial performance. Their overall composite could remain similar even as the within-person pattern differs. Conversely, a changed composite can coexist with a similar pattern across domains. The composite and profile answer different questions.

The methods article “Temporal stability and change in manifest intelligence scores: Four complementary analytic approaches” sets out these four approaches and demonstrates them with 114 gifted students assessed six months apart. That sample and interval illustrate how analysts can distinguish questions; they do not supply an estimate for adults over five years. For a five-year adult comparison, the meta-analysis supports a population-level statement about relative ordering. It does not establish group mean change, an individual's measured change, or profile continuity. Before calling two results stable, specify which of these meanings matters to the decision.

Sources: The stability of cognitive abilities: A meta-analytic review of longitudinal studies; Temporal stability and change in manifest intelligence scores: Four complementary analytic approaches

Why does a correlation of .76 not tell you how many IQ points changed?

Correlation describes how scores co-vary across people; it does not describe the size or direction of any one person's score difference. In the 2024 meta-analysis *The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies*, ρ = .76 is the modeled average rank-order correlation at age 20 over a five-year interval. It summarizes how well people tended to retain their relative positions across two occasions in the studies included. It is not a statement that 76% of anyone’s ability remained, nor a prediction that a person will gain or lose a particular number of IQ points. The estimate draws on many measures, samples, and longitudinal studies, rather than one adult cohort taking one common test. Its level of analysis matters: a pooled relationship across people cannot supply an individual change score.

Consider a simple illustration, not a study result. Suppose everyone in a group scored higher by roughly the same amount on a second occasion. Their scores could still have a similar ordering, even though the group’s measured level shifted. Conversely, one person might move relative to others while the overall correlation remains high; the coefficient alone would not show who moved or by how much. Correlation reflects both the pattern of change and the spread of scores between people. It therefore cannot be read as a personal before-and-after ruler. The methods paper *Temporal Stability and Change in Manifest Intelligence Scores: Four Complementary Analytic Approaches* makes the underlying distinction explicit: group mean change, individual-level change, relative ordering, and the shape of a person's profile are different questions. Its examples use gifted students assessed six months apart, so they clarify the concepts rather than provide a five-year estimate for adults.

The meta-analysis also reports a reliability-adjusted estimate separately from its observed correlations. For the same modeled age and interval, the abstract gives .86 for disattenuated correlations; that is a research adjustment for test reliability, not a corrected result for a particular reader. The observed .76 and adjusted .86 answer related but distinct analytic questions. Neither is an IQ-point difference, percentile movement, confidence interval, or percentage of ability retained. Converting either coefficient into those quantities would require evidence and a method for the exact instrument, score scale, and population; the meta-analysis does not provide a personal conversion. If two reports show different scores, first describe the observed difference in the score type actually reported. Whether that difference reflects lasting change requires an instrument-specific analysis of precision and comparability, not this population correlation. A high group correlation can coexist with uncertainty about a person’s score trajectory, especially when the measures or conditions are not demonstrably comparable.

Sources: The stability of cognitive abilities: A meta-analytic review of longitudinal studies; Temporal stability and change in manifest intelligence scores: Four complementary analytic approaches

How do age and the five-year interval shape the average?

Age and elapsed time both shape the average pattern, but they do not turn it into a personal timetable. The meta-analysis “The stability of cognitive abilities: A meta-analytic review of longitudinal studies” combined 205 longitudinal studies, involving 87,408 participants and 1,288 test–retest correlations among observed scores. Its modeled estimate for age 20 and a five-year interval was ρ = .76. That figure is the mean association between people’s scores across occasions under that model, not a forecast for every person aged 20 or a universal value for adults. The evidence base spans different samples, measures, and ages; the headline estimate is a synthesis, not one cohort following all adults on a single instrument. The age pattern was not a straight line. The review describes low rank-order stability in preschool children, rapid increases during childhood, and consistently high stability from late adolescence through late adulthood. In other words, people’s relative positions on the measured abilities tend to be more alike across occasions in the older age ranges studied than in early childhood. The article’s age-20 estimate is one modeled point on that broader curve. It should not be read as the exact expected correlation for every adult, much less as evidence that an individual’s measured level cannot change. The authors report that the same general age pattern remained after adjusting correlations for test reliability. This makes the developmental pattern less dependent on that particular adjustment, but it does not remove all measurement limits or convert a group model into a personal conclusion. Elapsed time matters too. Across the studies, rank-order stability declined as the interval between assessments grew. The modeled decline flattened from approximately five years onward. “Flattened” means that the average additional decrease became less pronounced in the model; it does not mean scores stopped changing at year five, or that a score becomes stale on a five-year anniversary. Nor does the curve establish a required retest schedule. A five-year interval is a useful point for describing the research pattern, while the correct timing for a new assessment depends on its purpose and on the instrument and person involved. The practical distinction is between a pooled pattern and an individual comparison. The meta-analysis can describe how strongly scores tend to preserve relative ordering across a broad body of studies at modeled ages and intervals. It cannot tell whether one person’s two scores reflect lasting change, ordinary imprecision, different testing conditions, or another influence. Even the review’s reliability-adjusted results remain statistical summaries, not an individual correction. So five years is neither proof of continuity nor an expiration date: it is context for interpreting what a particular assessment can support.

Sources: The stability of cognitive abilities: A meta-analytic review of longitudinal studies

Why can retaking a test blur the meaning of a five-year change?

A score taken five years later reflects more than the passage of time. It may also reflect what a person learned from the first assessment: the format, the operations required, or particular items. An unchanged or higher retest score cannot, by itself, show that underlying ability was unchanged or improved. A lower score does not isolate age-related change either. The comparison combines performance on both occasions with retesting effects and the properties of the measures used. “Estimating Retest Effects in Longitudinal Assessments of Cognitive Functioning in Adults Between 18 and 60 Years of Age” examined adults aged 18 to 58 who completed the same cognitive tests at intervals ranging from less than a week to 35 years. Because intervals varied, the researchers could estimate age effects and prior-assessment effects in the same analyses. They reported that positive retest effects remained detectable until seven or more years after the first assessment in the measures studied. Estimated age effects were negative while prior-assessment effects were positive. This matters for a five-year comparison: a retest contribution may still be present even when assessments are years apart. It does not establish a seven-year rule for other instruments or tell how much retest influence belongs in one person’s score. “Retest Effects in Cognitive Ability Tests: A Meta-Analysis” supports that caution. Its accessible abstract reports 174 samples from 122 studies, 786 test outcomes, and 153,185 participants. Across this varied evidence base, the analysis found significant average retest effects; its moderator analyses found that operation and content, form equivalence, interval, and participant age affected their size. The pooled result is not a forecast for every test taker. The variation is the practical finding: retesting on a different form, or a task requiring different operations, need not have the same effect as repeating a similar assessment. Several processes can contribute to a gain. A person may remember material, use a strategy learned during the first sitting, or be more familiar with how the test works. “Scoring Higher the Second Time Around: Meta-Analyses of Practice Effects in Neuropsychological Assessment” describes these as possible sources of practice effects. It reports that alternate forms, participant age, clinical diagnosis, and the interval between tests were associated with mean changes in many cases. That review concerns neuropsychological tests, so it helps explain mechanisms and moderators; it does not supply a correction for general intelligence scores. Record retest history before interpreting a difference: note the test and form, approximate date, and prior exposure to its tasks. These details do not settle whether an individual changed, but help a qualified assessor judge whether the comparison is meaningful. Neither the adult study nor the meta-analyses provide a universal five-year adjustment. Treat prior exposure as part of the evidence about the comparison, not an afterthought once the new score is visible.

Sources: Estimating retest effects in longitudinal assessments of cognitive functioning in adults between 18 and 60 years of age; Retest effects in cognitive ability tests: A meta-analysis; Scoring Higher the Second Time Around: Meta-Analyses of Practice Effects in Neuropsychological Assessment

Do all cognitive abilities and test formats show the same stability?

No single stability or retest figure applies to every cognitive ability or test format. Working-memory evidence is a bounded comparison: it shows why task design matters, but cannot be substituted for a general intelligence score or predict one person’s five-year result. The 2018 meta-analysis “Retest effects in working memory capacity tests: A meta-analysis” combined 234 effect sizes from 95 samples in 68 studies, involving healthy participants ages 12 to 70 who completed working-memory tests repeatedly. For the first-to-second administration, the weighted average retest effect was g = 0.28. Effects differed by task paradigm and were moderated by the interval between tests. Repeated performance on these tasks can shift on average, with the size depending partly on how and when people are retested. This is not an IQ-point change, a five-year estimate, or a forecast for one individual. The age range includes adults, but this is not an adult-only estimate. A broader comparison comes from “Retest effects in cognitive ability tests: A meta-analysis.” Its accessible abstract summarizes 174 samples from 122 studies, with 786 outcomes and 153,185 participants. It reports significant average retest effects and no further score gains after the third administration in the pooled evidence. Effect size varied with operation and content, test-form equivalence, interval, and participant age. This reinforces the task-specific point, while remaining an average across studies rather than a rule for every person. These retest findings address average score shifts under repeat testing; they do not replace evidence about whether adults preserve relative ordering over time. The 2024 “The stability of cognitive abilities: A meta-analytic review of longitudinal studies” estimated rank-order stability across varied ages, measures, and intervals. That estimate and a working-memory retest effect answer different questions: one concerns how similar people’s relative positions remain, the other average performance change after repeating a task. Neither says every ability or format is equally stable or gives an individual change threshold. A composite score can conceal movement within its parts. Imagine a battery combining verbal, quantitative, and spatial tasks. Two administrations could yield a similar composite while the domain pattern differs; one domain could change while the composite moves little. This is an illustration, not a reported study result. If domain mix, item format, or scoring changes, the difference may not show a clean personal trend. A shared test label alone does not establish that assessments measured abilities in the same way. Begin a five-year comparison by asking what each assessment measured: the same broad construct, domains, or a narrower capacity such as working memory? Then check whether forms and scoring are documented as comparable. A meta-analytic average gives population context, but cannot settle the meaning of one person’s pair of scores. Stability and retest effects depend partly on ability and format; “IQ” or “cognitive ability” alone is not enough to make two results directly comparable.

Sources: Retest effects in working memory capacity tests: A meta-analysis; Retest effects in cognitive ability tests: A meta-analysis; The stability of cognitive abilities: A meta-analytic review of longitudinal studies

What can make a person's retest score rise, fall, or stay level?

A two-score difference combines the measure’s precision and comparability with the person’s performance on both occasions; by itself, it cannot identify which influence produced the difference. Every test score is an estimate rather than a perfectly precise reading. A raw score is the number of items or points earned before conversion. A norm-referenced score locates performance against a specified reference group. A confidence interval is an instrument-specific range expressing uncertainty around a score, when the test provides a defensible method for calculating one. These quantities answer different questions, so a change in one cannot automatically be translated into a change in another. Several influences can contribute to a higher, lower, or unchanged retest score: the difficulty or comparability of the forms, administration conditions, familiarity with the format or content, attention on the day, health or other context, and actual change in the ability being measured. The two results alone do not show which influence mattered or how much. In particular, an increase in raw points may reflect answering more items correctly without establishing that a norm-referenced score rose; the conversion and reference group matter. A similar total can also conceal different patterns across domains, depending on what the test reports. “Scoring Higher the Second Time Around: Meta-Analyses of Practice Effects in Neuropsychological Assessment” reviewed nearly 1,600 effect sizes concerning mean performance changes on tests used to assess core neuropsychological functions. It defines practice effects as score increases that can arise from remembering specific items, learning strategies, or becoming more sophisticated at taking a test. The review reports that alternate forms, participants’ ages, clinical diagnoses, and the retest interval were associated with the size of mean change in many cases. This evidence shows why retest context can matter, but it concerns neuropsychological tests and pooled mean changes; it does not supply a correction for every adult intelligence test or explain one person’s difference. The 2024 meta-analysis, “The Stability of Cognitive Abilities,” likewise reports observed test–retest correlations separately from estimates adjusted for test reliability. That distinction matters: an across-person correlation describes how scores relate across occasions, while reliability adjustment is a research estimate, not a personal error bar. Neither statistic identifies the source of one individual’s movement. Before interpreting a rise, fall, or flat result, record the exact instrument, score type, forms, conditions, and prior exposure, then consult that instrument’s documentation. Without its evidence for precision and comparison, the responsible description is simply that the observed scores differ—or do not—not that underlying ability definitively changed.

Sources: Scoring Higher the Second Time Around: Meta-Analyses of Practice Effects in Neuropsychological Assessment; The stability of cognitive abilities: A meta-analytic review of longitudinal studies

What makes two assessments fair to compare?

A five-year comparison is fair only when both results can reasonably be read on the same measurement scale for the same question. Start with the construct: did both assessments measure general reasoning, or does one emphasize a narrower ability such as verbal knowledge or working memory? A shared label such as “IQ” does not establish that two instruments sample abilities in the same way. “The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies” includes varied abilities and tests; its modeled five-year correlation summarizes relative ordering across that mixed evidence base. It cannot establish equivalence between two specific tests or editions. Next check the instrument and form. Record the full test name and edition, and whether the same form or a documented alternate form was used. Then identify the score type. A raw score is the number of points or responses credited before conversion. A norm-referenced score compares performance with a defined reference group using a specified conversion. Raw totals are not directly comparable when tests use different items or scoring rules, and the same number on two reports does not guarantee the same meaning. “Temporal Stability and Change in Manifest Intelligence Scores: Four Complementary Analytic Approaches” emphasizes evaluating the stability of a specific instrument’s scores when they may inform consequential decisions. For a normed comparison, ask which reference group supports each score and when it was established. A norm is a score distribution from a defined comparison sample; its age, composition, and match to the person and purpose matter. A score standardized against one group cannot simply inherit the meaning of another test’s norms. A published correlation cannot supply a missing conversion: it describes how scores vary together across people, not how to translate one person’s result between scales. Administration and retest history matter too. Were instructions, timing, setting, and permitted supports comparable? Was the person already familiar with the items or format? These details help an assessor judge whether a difference could reflect procedure or prior exposure rather than a change in the ability being considered. Finally, ask what evidence describes precision for this instrument and score, and whether its manual or a qualified assessor provides a method for judging individual change. A group-level correlation is not an individual threshold. The methods article demonstrates distinct analyses on 114 gifted students tested six months apart; this illustration is not an adult five-year rule. For a consequential decision, bring both reports to a qualified assessment professional and ask: “Do these scores measure a comparable construct on a comparable scale, and what evidence supports interpreting the difference?”

Sources: The stability of cognitive abilities: A meta-analytic review of longitudinal studies; Temporal stability and change in manifest intelligence scores: Four complementary analytic approaches

How should you use a five-year-old result for a real decision?

Keep a five-year-old score as dated context when the question is low stakes. If a decision about education, clinical care, disability, or another consequential matter depends on that result, ask a qualified assessment professional whether a current assessment is appropriate. The five-year research estimate describes how well groups tend to retain their relative ordering; it does not tell you whether one person's score is still accurate for a decision today. Nor does a browser quiz establish diagnosis, eligibility, or fixed potential.

For personal reflection, start by preserving the record rather than interpreting a number in isolation. Note the assessment date, test name and version if known, why it was taken, whether the result was a raw score or a converted score, and any known conditions such as timing or prior exposure. These details help you remember what the result actually described. A raw score is the number of items or points earned before conversion. A norm-referenced score places performance against a specified comparison group. Neither becomes a timeless description of a person. If the original report does not identify its comparison group or conversion, mark that uncertainty instead of supplying one from another test.

For a consequential decision, take the old report to a qualified professional and explain the decision that needs support. The professional can consider whether a new measure is needed and whether its construct, score scale, precision, reference group, administration, and any retest history fit that use. The “Stability of Cognitive Abilities” meta-analysis cannot determine those details for an individual. A population correlation is not a personal change threshold, and five years is not a universal retest schedule. A browser result cannot settle a clinical question or determine access to services.

The free Test IQ Free assessment is a different, low-stakes option: its adaptive 50-question reasoning test offers practice, worked answers, and feedback across pattern, number, and spatial reasoning. It also displays a provisional estimated IQ range, but that range comes from an unvalidated conversion with no validated adult reference population. It is not a confidence interval or an established comparison with adults. Therefore it cannot update, confirm, or replace a professional score, and two displayed ranges should not be treated as a normed five-year comparison. The optional $9 report adds explanation and a practice plan; it uses the same provisional conversion, so payment adds no validity or accuracy.

The practical rule is to match the next step to the consequence. For curiosity, keep the old result as context and use free reasoning questions for practice. For a decision that could affect care, education, or disability support, ask a qualified professional whether a current, comparable assessment is warranted. Bring the report and ask how its score was produced, what uncertainty applies, and whether it answers the decision at hand. Until those questions are resolved, describe the result as performance on a particular assessment at a particular time, not as a verdict about who you are.

Sources: The stability of cognitive abilities: A meta-analytic review of longitudinal studies

What is the next responsible step?

For curiosity, use a free reasoning test as practice and review its worked answers. If a consequential decision depends on a score, ask a qualified assessment professional whether old and new results are comparable. The five-year estimate in “The stability of cognitive abilities: A meta-analytic review of longitudinal studies” describes average preservation of relative rank; it cannot tell what a difference means for one person.

Keep the old report with its date, test, and known conditions. Write down the decision you need to make: a general point of reference or information for an educational or clinical discussion. A browser result cannot settle diagnosis, eligibility, or another high-stakes question. Ask the professional: “Are these two scores comparable for the decision I need to make, and what uncertainty or retest effects matter?” The answer depends on the assessments and their intended use, not the five-year interval alone.

For low-stakes practice, Test IQ Free offers 50 reasoning questions, worked answers, domain feedback, and a provisional estimated IQ range. The range comes from an unvalidated conversion with no validated adult reference population; it is neither a confidence interval nor an established comparison with adults. Treat it as part of the practice experience, not as a replacement for a professional result or proof that your ability changed. For a consequential decision, take the old report to a qualified professional and ask whether a new assessment is appropriate.

Sources: The stability of cognitive abilities: A meta-analytic review of longitudinal studies

Questions readers ask