Average IQ by birth month and the relative age effect

0 The Short Answer

No birth month produces higher intelligence. What the research consistently finds is a relative age effect: within a single school cohort, the oldest children outperform the youngest by a margin large enough to matter, because they are up to eleven months older when the test is given. Which months are advantaged depends on where a country sets its school entry cutoff, so the answer changes across borders while the mechanism stays the same.

The size of the effect at its peak is striking. In England, where the academic year runs from 1 September to 31 August, Crawford, Dearden and Greaves reported at the Institute for Fiscal Studies that the gap between August born and September born children at age 5 reached 79 percent of a standard deviation on the national assessment and 65 percent on an independently administered cognitive battery. That is a difference of roughly 10 to 12 points if expressed on a scale with a standard deviation of 15, which is the conversion this site uses throughout and which is set out on the page on the 15 point standard deviation.

The rest of the finding is what makes it useful rather than alarming. The gap shrinks steadily as children age and the relative age difference becomes a smaller fraction of their lives. By the end of compulsory schooling it is around 12 percent of a standard deviation. In adulthood, measured on employment, wages, health and reported happiness, the same authors found essentially nothing. And when the same children are assessed at the same age rather than on the same date, most of the difference disappears, which tells you what the effect was made of.

1 Three Different Questions Hiding Behind One

The phrase smartest birth month bundles together three claims with three separate evidence bases, and almost all confusion about this topic comes from answering one while thinking about another. Separating them first makes the rest short.

The first is the relative age effect. Within a school year group, children born just after the entry cutoff are the oldest and children born just before the next cutoff are the youngest. When everyone sits the same test on the same day, the oldest have had up to eleven more months of development and schooling. This is a real, large, repeatedly measured effect, and it is the answer to the question people are usually asking without realising it.

The second is the season of birth effect: the claim that something about the time of year a person is conceived or born, through daylight, temperature, maternal nutrition or infection exposure, alters development. This is a different claim with a different mechanism, and the evidence for it in relation to intelligence is weak. It is not weak in relation to every outcome, which is why a section below covers the one seasonal finding in psychiatry that has survived meta-analysis.

The third is a measurement question that people rarely state: does an intelligence test, as opposed to a school exam, show this pattern at all? The answer depends entirely on how the test is scored. A raw score compared against a grade cohort will show it. A score converted against an age matched reference band will show much less of it. That distinction is the practical payoff of this page.

That argument is worth holding on to, because it is also the reason this page does not publish a table of months ranked by score. Any such table would be a table about one country's calendar, presented as a fact about human beings.

2 The Mechanism, Stated Precisely

A single school entry cutoff date creates a cohort in which the oldest member is roughly 20 percent older than the youngest, and that ratio is the whole engine. Bedard and Dhuey made this point directly in the Quarterly Journal of Economics in 2006, in a paper that established the international scope of the finding.

Consider a class of five year olds. The oldest child has lived about 72 months and the youngest about 61. Eleven months is a sixth of the younger child's entire life. At age 16 the same eleven months is a twentieth of a life, and by age 40 it is a rounding error. That arithmetic predicts the shape of every result on this page: the effect is largest in the earliest years of school and decays monotonically as the denominator grows.

Four distinct things can produce the observed gap, and they are almost perfectly correlated with one another, which is why untangling them took decades. A child born late in the academic year is younger on the day of the test. That child also started school younger. That child may have received fewer terms of schooling before the test, depending on local admissions policy. And that child is the youngest relative to classmates, which changes how teachers, peers and the child rate their ability. Those are four different causal stories that produce identical raw data.

The distinction matters because the remedies differ. If the gap is age at test, adjusting scores for age fixes it. If it is age of starting school, letting parents defer entry fixes it. If it is length of schooling, aligning admission dates fixes it. And if it is relative age, nothing fixes it, because somebody must always be the youngest in the room. A later section reports which of the four the evidence actually supports.

One more feature of the mechanism deserves stating. It is not confined to test scores. The same eleven months shows up in who gets picked for selective streams, who gets identified as needing support, and who gets chosen for competitive activities where physical maturity matters, a pattern documented in the material on measured ability and athletic selection. Wherever people are compared against age peers and a threshold is applied, relative age has somewhere to act.

3 The International Evidence, Organised by Cutoff

The finding replicates across countries, and the table below is arranged by study, country and entry cutoff rather than by month, because the cutoff is what determines which months are affected. Reading it any other way reproduces the error this page exists to correct.

Bedard and Dhuey's headline range, 4 to 12 percentiles in grade four falling to 2 to 9 percentiles in grade eight, is the cleanest summary of the phenomenon available. It is a range rather than a number because the size differs by country, which is itself informative: the effect is not a constant of human development, it is a product of institutional design.

Sprietsma's result at age 15 is the one that most tempts overinterpretation in the other direction. Finding a significant relative age effect in about half of 16 countries and regions is not evidence that the effect is unreal in the other half. It is evidence that by age 15 the effect has decayed to a size that a PISA sized sample can detect in some systems and not others. Absence of detection and demonstrated absence are different claims, and collapsing them is the most common error in reading this literature.

Note also what changes and what does not across the table. The cutoff date changes. The affected months change. The direction never does. In every system studied, being relatively older within the cohort is associated with better measured performance, which is exactly what a mechanical age at test explanation predicts and what a seasonal biology explanation does not.

4 England Is the Best Documented Case

England has a single national cutoff, universal testing at fixed ages and an administrative dataset covering every state school pupil, which makes it the place where the effect has been measured in the most detail. The main source is the Institute for Fiscal Studies report When You Are Born Matters: Evidence for England, by Claire Crawford, Lorraine Dearden and Ellen Greaves, drawing on a programme of work that includes the earlier 2010 working paper by Crawford, Dearden and Meghir.

The academic year runs from 1 September to 31 August, so September born children are the oldest in their year group and August born children the youngest. The datasets are the National Pupil Database, which covers all state funded schools since January 2002, and the Millennium Cohort Study, which has followed approximately 18,500 children born across the United Kingdom between September 2000 and January 2002.

The attainment figures, all comparing August births to September births, run as follows. At age 5, on the teacher assessed Foundation Stage Profile, the gap is 79 percent of a standard deviation. At age 7, on Key Stage 1, it is 56 percent. At age 16 it is 12 percent of a standard deviation, expressed as a 6.4 percentage point difference in the probability of achieving five GCSEs at grades A star to C. The gradient is not confined to the two extreme months: January born pupils are 2.8 percentage points less likely to clear that bar than September born pupils, and May born pupils 4.4 percentage points less likely.

The consequences extend past exam results. August born pupils are 5.4 percentage points more likely to be recorded as having mild special educational needs at age 11. They are around 2 percentage points less likely to enter university at 18 or 19, around 2.3 percentage points less likely to attend a Russell Group institution if they do go, and around 1 percentage point less likely to graduate with a degree.

Two further findings from the same report belong here because they show the effect operating on people rather than on spreadsheets. Among pupils who scored at the expected level in externally marked Key Stage 2 English, those born in August were about 4 percentage points more likely to be under assessed by their teacher than September born pupils, roughly a third more often, with a 2 percentage point difference in maths. And August born children rated their own academic competence significantly lower than September born children at ages 8 and 14, with prior attainment explaining about 60 percent of that difference at 8 and 85 percent at 14.

5 Untangling Four Explanations That Produce Identical Data

The most useful thing in the English work is not the size of the gap but the decomposition of it, because that is what tells you whether the difference is about ability at all. The four candidate causes are perfectly or almost perfectly correlated in ordinary data, so separating them required exploiting variation that most countries do not have.

The variation came from three sources. Local authorities in England differ in their admissions policies: some have a single entry point, others admit children in September, January and April, which breaks the link between birth month and length of schooling. Survey data are collected on interview dates rather than fixed national test dates, which breaks the link between birth month and age at assessment. And year groups differ in their age composition, which creates variation in relative age that is independent of absolute age.

Read the first two rows together and the conclusion is hard to avoid. Between them, age at test and relative age account for roughly three quarters of the gap at age 7 and essentially all of it at age 11. The two policy levers people most often propose, deferring school entry and equalising the amount of schooling before tests, account for less than a quarter at age 7 and almost nothing by age 11.

The authors state their own caveat, and it should be repeated rather than buried. Age at test and relative age remain highly correlated even in their data, so a model that cannot fully separate them will tend to split the difference between them and may attribute too much to one. Their Key Stage 2 result, in which relative age overtakes age at test, contradicts their own earlier work, and they say so. What is not in doubt is the pair together against the other two.

The evidence from other countries points the same way. As summarised in the same report, Fredriksson and Öckert found for Sweden that increasing school starting age by one year raised grade point average at 16 by 0.2 standard deviations while relative age accounted for only 6 percent of the test score difference at that age. Smith found for Canada that a one year increase in age at test was associated with a 5.8 percentage point reduction in the likelihood of repeating grade 3, against 0.6 percentage points for a one year increase in age at entry. Datar found for the United States that older entrants gained 0.12 standard deviations more than the youngest over a two year period.

6 The Test That Settles It: Same Age Instead of Same Date

If the gap is mostly about when the assessment happened rather than about the child, then measuring the same children at the same age instead of on the same date should make most of it disappear. That comparison has been run, and it does.

The Millennium Cohort Study collects its assessments through household interviews spread across the year, which means two children born eleven months apart can be assessed at the same age rather than on the same date. Crawford, Dearden and Greaves exploited exactly that. On socio-emotional development at age 5, the difference between August and September born children was 17 percent of a standard deviation when the comparison was made at the same point in time. Assessed at the same age, it was 3 percent of a standard deviation, and not significantly different from zero.

That is one outcome rather than all of them, and the cognitive gaps do not vanish as completely, because relative age keeps operating even when absolute age is held constant. But the direction of the result is unambiguous, and it generalises: the more of the age difference you remove from the comparison, the less of the birth month difference remains.

The report's authors then took the argument to its practical conclusion. Using Key Stage 2 English results from 2008 to 2009, they found that August born pupils scored about 7 points lower than September born pupils on the raw scale. They showed that raising the threshold by 3 points for September births, by 2 for October and November, by 1 for December and January, leaving it unchanged for February and March, and lowering it by 1, 2 and 3 points respectively through the spring and summer eliminates the month of birth difference in the proportion of pupils reaching the expected level.

That correction is age norming, applied to a school exam. It is the same operation that converts a raw score into an index score on a properly constructed intelligence test, and the fact that it works on Key Stage results is the clearest available demonstration of what the relative age effect is made of. The machinery behind it is described on the page on how norms are built.

The authors are explicit about the interpretation, and their sentence is the honest bottom line for this whole section: these differences arise from the organisation of the educational system, and there is nothing fundamentally different about August born children. Somebody has to be the youngest in each academic cohort, and in England it happens to be them.

7 What Happens by Adulthood

The effect that dominates a five year old's test scores is close to invisible in a thirty year old's life, and that is where the honest answer to the original question lives. Two independent bodies of evidence say the same thing.

The first is the English adult data. Crawford, Dearden and Greaves used the Labour Force Survey, covering individuals aged 25 to 64 who were born in England between 1948 and 1987 and first interviewed between 2002 and 2011, plus Understanding Society as a check. Those born at the end of the academic year were no more or less likely to be in work than those born at the start, although slightly more likely to be unemployed. They did not earn more or less per hour or per week. They were, subjectively, no healthier and no happier. In the most recent birth cohorts the hourly wage difference actually ran in favour of August births, which the authors attribute to earlier labour market entry.

The second is Norwegian. Black, Devereux and Salvanes reported in the Review of Economics and Statistics in 2011, at volume 93, issue 2, pages 455 to 467, that starting school younger produced a short run earnings advantage that had essentially disappeared by age 30. The working paper version is open. Their explanation is mechanical rather than cognitive: starting later reduces potential labour market experience at any given age for a given level of education, and that matters less as careers lengthen.

The residual that does persist is the one that runs through selection rather than through ability. If an examination taken at 16 determines who continues into further and higher education, then a 6.4 percentage point gap at that threshold converts into a durable difference in credentials regardless of what happens to the underlying scores afterwards. That is why the university entry figures survive when the wage figures do not, and it is the argument developed further in the section on selection below and in the analysis of institutional sorting on the page on measured ability by university.

8 The Study That Tested Intelligence Outside a School

Almost every result above comes from tests administered inside schools, which is exactly the setting where age at test and relative age are hardest to separate. One study measured cognitive ability outside that setting entirely, and its numbers are the most directly relevant on this page.

Black, Devereux and Salvanes used the population registers of Norway together with the cognitive test administered at military conscription, taken by men at about 18 to 20 years of age. The composite is built from three subtests covering arithmetic, word similarities and figures, reported in stanine units with a mean of 5 and a standard deviation of 2, and it is available for about 84 percent of the relevant male population. The main specifications run on roughly 247,000 observations, and the design includes family fixed effects so that brothers can be compared with brothers.

Two features make the design unusual. The Norwegian rule that children start school in the calendar year they turn 7 creates a sharp discontinuity around 1 January, which gives an instrument for actual school starting age. And because the conscription test is taken outside school, there is variation in the mapping between birth month and the year the test happens, which lets age at test be separated from school starting age rather than absorbed into it.

The results are precise and small. Being one year older at the time of the test raised the score by about 0.22 stanine, which the authors describe as roughly one tenth of a standard deviation. Starting school one year later lowered the score by about 0.06 stanine. Combining the two, which is what an in school test would estimate, gives 0.16, or about 8 percent of a standard deviation. On a scale with a standard deviation of 15, one tenth of a standard deviation is about 1.5 points and 8 percent is about 1.2 points, which is our conversion of their reported figures rather than a number they printed.

Set that against the 79 percent of a standard deviation observed in five year olds and the trajectory is complete. What begins as an eleven point difference at school entry has decayed by age 18 to something on the order of one or two points, measured on a test taken outside school, in a design large enough to detect far less than that. The same authors note that in school estimates from Sweden, where Fredriksson and Öckert found a school starting age effect of about 20 percent of a standard deviation on ninth grade grade point average, are larger precisely because in school tests cannot make this separation.

9 Season of Birth Is a Separate Claim With Separate Evidence

There is a real seasonal literature in medicine, and confusing it with the relative age effect is the second most common error in this area. The two are distinguishable by a simple test: a relative age effect moves when a country moves its school cutoff, and a seasonal biological effect does not.

The best documented seasonal finding concerns schizophrenia rather than intelligence. Davies, Welham, Chant, Torrey and McGrath published a systematic review and meta-analysis in Schizophrenia Bulletin, volume 29, issue 3, pages 587 to 593, in 2003, pooling eight studies covering 126,196 patients against 86,605,807 general population births across 27 Northern Hemisphere sites. They found a significant excess of winter and spring births, with a pooled odds ratio of 1.07 and a 95 percent confidence interval from 1.05 to 1.08. The population attributable risk was 3.3 percent, and the strength of the effect correlated with latitude.

Three things about that result are instructive for the present question. It is genuinely seasonal rather than institutional, because it tracks hemisphere and latitude rather than school calendars. It is small: an odds ratio of 1.07 in an uncommon condition. And it took a pooled sample of more than 86 million births to establish with confidence, which is a useful calibration for how large a study needs to be before a seasonal claim about anything is credible.

No comparable body of evidence exists for season of birth and measured intelligence. Where seasonal patterns in test scores have been reported, the difficulty is always the same: in any single country, season of birth is confounded with position in the academic year, and separating them requires either comparing across countries with different cutoffs or exploiting a cutoff change. The cross country comparison, as the Institute for Fiscal Studies work points out, is exactly what shows that the pattern follows the calendar of the school system rather than the calendar of the sun.

The other seasonal correlate worth naming is not biological at all. Birth months are not uniformly distributed, and the distribution differs by maternal age, education and income in ways that vary between countries and over time. Any raw comparison of outcomes by month in a single dataset therefore carries a composition difference as well as an age difference, which is one more reason that a table of average scores by month would be misleading even if somebody produced one.

10 Why a Properly Normed Test Removes Most of This

The relative age effect is a scoring problem before it is anything else, and the fix is the ordinary machinery of a normed test. This is the point at which the answer stops being about school calendars and starts being about measurement.

A school examination produces a raw score, compares it against a fixed threshold or against everyone in the same grade, and reports the result. Every month of age advantage inside that grade goes straight into the number. An intelligence test does something different. Raw performance is converted against a reference distribution for the test taker's own age band, so a score of 100 means average for people of that age rather than average across everyone who sat the test. The whole point of that conversion is to strip out exactly the variation that the relative age effect exploits.

The demonstration is in the English data already cited. When the same authors applied an age adjustment to Key Stage 2 English results, adding points to the threshold for autumn births and subtracting them for summer births, the month of birth difference in the proportion reaching the expected level was eliminated. The test did not change. The scoring did.

Two honest qualifications belong here. First, age norming removes the age at test component, which is the single largest driver, but it does not remove the relative age component, which is a real accumulated difference produced by years in a peer group. A child who has spent five years as the youngest in the room and has come to believe they are less able carries that into an age normed test as well. Second, an age band is finite. ACIS norms against age bands across a range from 16 to 90, and within any band there is residual variation the conversion cannot see.

The relevant ACIS figures, from the published technical manual, put the size of the remaining problem in context. The standard error of measurement is 2.40 points for the Verbal Comprehension index, 2.48 for Fluid Reasoning, 3.39 for Visual Spatial, 4.02 for Quantitative Reasoning, 4.12 for Working Memory and 5.22 for Processing Speed, with 1.61 for the General Ability Index built from 15 subtests. A 95 percent confidence interval spans roughly 1.96 standard errors either side of the observed score, so even the tightest composite reports a band about 6 points wide. A residual birth timing effect of the size Black, Devereux and Salvanes estimated at age 18, on the order of one to two points, sits well inside every one of those intervals. How those intervals are constructed is covered on the reliability and validity page.

ACIS is a self-administered online assessment for adults aged 16 to 90, not a clinical or diagnostic instrument, and nothing here is educational or clinical advice. A parent worried about a summer born child's school placement needs the school and, where appropriate, an educational psychologist, not a website.

11 Where the Effect Does Real Damage: Selection Thresholds

The reason a shrinking, artefactual effect still matters is that education applies thresholds, and a threshold converts a small difference in score into a large difference in what happens next. This is the part of the finding with practical consequences, and it survives the conclusion that the underlying ability difference is minimal.

Take the English numbers as the worked example. A 6.4 percentage point gap in the probability of achieving five GCSEs at grades A star to C, at the exact bar that gates further and higher education, is not a small thing for the individuals sitting on either side of it. Neither is a 5.4 percentage point difference in being recorded as having mild special educational needs at age 11, or a 2.3 percentage point difference in attending a Russell Group institution among those who reach university at all.

The same logic operates wherever selection happens against age peers. Ability grouping inside schools, entry to selective institutions, identification for enrichment programmes and team selection in youth sport all apply a threshold to a comparison among people who differ in age by up to a year. In each case the relatively older are overrepresented above the line, and the mechanism does not require anyone to be biased. It requires only that the comparison be made among a cohort defined by a cutoff date.

The consequences are not confined to attainment. The same report found that children born at the end of the academic year rate their own academic competence lower at ages 8 and 14, are more likely to hold an external locus of control, and are more likely to engage in risky behaviours at younger ages when compared at the same age. Prior attainment explained about 60 percent of the difference in ability beliefs at age 8 and 85 percent at 14, which leaves a real remainder that scoring adjustments would not touch.

That remainder is the reason relative age is worth understanding rather than dismissing. The measurement artefact can be corrected by arithmetic. The years a child spends believing they are the slow one in the room cannot, and that belief is one of the routes through which a scoring convention becomes a life outcome. Related material on how self perception and measured ability come apart is on the page on self estimated intelligence.

12 What the Evidence Actually Supports

The defensible answer to the question people are asking has four parts, and none of them is a month.

First, within a school cohort, the relatively oldest outperform the relatively youngest by a large margin in the early years. Bedard and Dhuey put it at 4 to 12 percentiles in grade four across OECD countries. The English data put it at 79 percent of a standard deviation at age 5. This is not in dispute and it replicates wherever it has been looked for.

Second, the effect is a function of the entry cutoff, not the month. Which months are advantaged moves when the cutoff moves, which is why no country independent table of scores by month can be constructed and why anyone offering one has confused an institution with a fact about people.

Third, the effect shrinks with age and is small or absent by adulthood. It falls from 79 percent of a standard deviation at 5 to 56 percent at 7 to 12 percent at 16 in England. At about 18, measured outside school on a quarter of a million Norwegian conscripts, the age at test effect is roughly a tenth of a standard deviation. On adult employment, wages, health and happiness, the English evidence finds essentially nothing.

Fourth, most of what remains is an artefact of the measurement occasion rather than a difference in underlying ability. Age at test and relative age together account for about three quarters of the gap at age 7 and effectively all of it at age 11, and adjusting scores for age eliminates the month of birth difference in the proportion of pupils reaching the expected standard. That is a strong claim and it rests on a decomposition the authors themselves qualified, so the correct confidence level is high rather than certain.

What follows for a reader is straightforward. If the question is whether your birth month set a ceiling on your ability, the answer is no, and the studies that look most like a yes are measuring how old you were on exam day. If the question is whether it affected what you were allowed to do at 11 and 16, the answer in some systems is yes, through thresholds rather than through capacity. And if the question is what your current ability profile looks like, that is a measurement question with a measurement answer, addressed by an age normed instrument rather than by a calendar. The relationship between a score and its position in the population is set out on the page on scores and percentiles, and what the bands mean is on the page on interpreting a result.

13 Sources Behind This Page

Every figure above traces to one of the following, and each is linked at the point where it is used. Conversions from standard deviation units to a 15 point scale are our arithmetic on published figures and are labelled as such in the text.

• Bedard K and Dhuey E. The Persistence of Early Childhood Maturity: International Evidence of Long-Run Age Effects. The Quarterly Journal of Economics, 2006, volume 121, issue 4, pages 1437 to 1472.
• Crawford C, Dearden L and Greaves E. When You Are Born Matters: Evidence for England. Institute for Fiscal Studies Report R80, 2013, ISBN 978-1-909463-08-0. Source for the Foundation Stage Profile, Key Stage, special educational needs, university entry, self assessment, decomposition and adult outcome figures.
• Crawford C, Dearden L and Meghir C. When you are born matters: the impact of date of birth on educational outcomes in England. Institute for Fiscal Studies Working Paper W10/06, 2010.
• Sprietsma M. Effect of relative age in the first grade of primary school on long-term scholastic results: international comparative evidence using PISA 2003. Education Economics, 2010, volume 18, issue 1, pages 1 to 32. Sixteen countries and regions.
• Black SE, Devereux PJ and Salvanes KG. Too Young to Leave the Nest? The Effects of School Starting Age. Review of Economics and Statistics, 2011, volume 93, issue 2, pages 455 to 467; working paper version National Bureau of Economic Research 13969.
• Davies G, Welham J, Chant D, Torrey EF and McGrath J. A Systematic Review and Meta-analysis of Northern Hemisphere Season of Birth Studies in Schizophrenia. Schizophrenia Bulletin, 2003, volume 29, issue 3, pages 587 to 593.
• Fredriksson and Öckert on Sweden, Smith on Canada and Datar on the United States are cited as summarised within the Institute for Fiscal Studies report above, which reviews them directly. Their figures are attributed here through that route rather than from the original papers.
• ACIS reliability, standard error of measurement and index structure figures are from the published ACIS technical manual. The manual documents the adult English speaking reference frame.

Two things this page deliberately does not contain. There is no table of average scores by calendar month, because such a table would encode one country's school cutoff as though it were a property of people born in that month. And there is no estimate of a smartest month, because the literature does not support one and the mechanism it does support predicts that the answer would change at every border.

The professional framework governing interpretation is explicit. The Standards for Educational and Psychological Testing (2014), published jointly by the American Educational Research Association, the American Psychological Association and the National Council on Measurement in Education, require that score interpretations be supported by evidence for the specific use proposed, that reliability and measurement error be reported alongside every score, and that the composition and limits of the reference sample be disclosed. APA standards on test use and the International Test Commission guidelines on test use make the same requirements. Under those standards, an age normed score describes a person's standing relative to their own age band on a set of tasks, which is precisely why the effect described on this page is a scoring artefact rather than a finding about human beings. Researchers who need verified administrations with participant level export can use the research workspace or administer the assessment to participants through private links.

14 Frequently Asked Questions