IQ and fertility: what registers and surveys count

0 The short answer

Higher IQ is not simply linked to fewer children: the answer depends on which sex was studied, how children were counted, and at what age. In Swedish registers of 779,146 men born 1951 to 1967, higher conscription scores went with more children, while US survey studies and a 2018 systematic review of 17 datasets, which include women, found small negative links (a weighted correlation of minus 0.11) that were stronger for women. Whether any of this changes the average IQ of a population is a separate question, and every published estimate rests on assumptions the data cannot check.

1 What Does "Do Smarter People Have Fewer Children?" Actually Ask?

The question bundles at least four different measurements, and a study can answer yes to one and no to another. The first is completed family size, the number of children a person has by the end of the reproductive years. The second is timing, usually the age at the first birth. The third is childlessness, the share of people who never have a child. The fourth is the sex studied, because the fertility window, the social pressures and the data available differ between men and women. A claim of the form "smart people have fewer kids" quietly picks one combination of these and states it as if it were all four.

The research question is old. In the 2019 Swedish register study published in Proceedings of the Royal Society B, Kolk and Barclay note that Francis Galton, Karl Pearson and Ronald Fisher all examined differential fertility, and that most researchers on the topic were concerned about what they called "dysgenic" population effects, meaning a falling average ability through differences in reproduction. That word is a hypothesis about a population, not a finding about a person, and this page treats it that way: it reports what each study measured, attributes every projection to the authors who made it, and gives the published critiques beside it. It takes no position on policy and says nothing about what any reader should do with their own family plans.

Two separate questions are also easy to merge. The first is descriptive: in a given group of people born in a given period, how are test scores related to the number and timing of children? The second is a projection: if that relationship persisted and test scores were partly heritable, how would the average score of the next generation differ? The first question can be answered by counting. The second needs a heritability estimate, data for both sexes, an assumption about the environment, and an assumption about how long the pattern lasts. The sections below keep them apart.

The test scores in these studies are also not interchangeable. The Swedish and Norwegian studies use military conscription batteries given to young men, converted to a nine point stanine scale. The US studies use tests given in adolescence or in high school. The page on what an IQ score measures explains why a composite score blends several abilities, and the page on the g factor explains why such composites tend to correlate with one another. For reading the correlations that follow, the page on the standard deviation of 15 is the right background.

2 How Do Researchers Measure the Link, and What Can Go Wrong?

Four design choices decide most of what a fertility study can say: who is in the sample, when fertility is counted, how ability is scored, and whether women are included. Each is a source of disagreement between studies, and each is documented in the papers reviewed here.

Who is in the sample matters because the people least likely to appear in surveys can differ in both ability and family outcomes. Kolk and Barclay argue that older studies, which relied on surveys or school classes, would often miss people with low scores, especially when samples were drawn from older children in secondary education. Their data come from Swedish administrative registers covering the complete population, including institutionalized men, and children are linked through birth records. They report that births known to the authorities represent over 99 percent of all births.

When fertility is counted matters because the same person looks different at 30 and at 45. In the Swedish data, men with lower scores had their first child earlier. Kolk and Barclay report that the lowest scoring category had a mean age of 27.6 at the first child and the highest scoring category 31. When they measured fertility before age 30, the gradient was completely reversed, and men with low scores had twice as many children as men in the highest categories. They conclude that data until at least age 45 are needed, and that studies of men in their early 30s or younger risk severe bias. This is a methodological point about timing, not a result about any group.

How ability is scored matters because tests differ in content, norming and ceiling. The Swedish test was normalized by the military each year, so scores are relative within a birth cohort, and the authors state that there can be no increase or decline in IQ scores over time within their data. A fertility study using a conscription score therefore cannot say anything about the Flynn effect. The page on how IQ scores are normed explains what a relative score means.

Whether women are included matters most for the population question. Conscription historically covered men, so the register studies cannot assess the female gradient. Both the Swedish and the Norwegian authors say so plainly, and the Swedish authors write that they cannot say anything certain about the population level effect because they lack data on cognitive ability and fertility for women. The US survey studies include both sexes but rest on smaller samples and on self-reported or survey-collected fertility. The two kinds of evidence therefore answer overlapping but different questions, and the headline "smart people have fewer kids" often mixes them.

A final caution concerns confounding. Education, family income, marriage and family background all relate to both test scores and fertility. A brother comparison, which the Swedish study uses, removes everything the brothers share, including parental education, income, neighborhood and, on average, half of their genes. The Norwegian authors take a different position: they chose not to stratify their analyses on education, warning that doing so risks collider bias and unobserved confounding, and they describe their own study as descriptive. Neither approach turns an association into a cause, and the verbs on this page follow that limit.

3 What Did the Swedish Register Study of 779,146 Men Find?

Among all Swedish men born from 1951 to 1967, higher conscription scores went with more children, mainly because men with the lowest scores were far more often childless. The study by Kolk and Barclay (2019) defines the population as all men born in Sweden in those years who were alive until at least age 45, a total of 779,146. Register data give monthly event histories of births from 1968 to 2012. Nearly all of the fertility data were measured at or after age 50, which the authors say leaves a virtually complete count and misses less than 1 percent of births.

The measure of ability came from the Swedish Enlistment Battery, a one to two day examination that every man had to attend before conscription, typically at ages 18 to 20. The battery included subtests of logical, spatial, verbal and technical ability, each scored on a stanine scale, then summed and converted to a stanine with a mean of 5 and a standard deviation of 2. The authors present results translated into IQ scores on a Wechsler scale. About 3 percent of the men did not take the test: roughly 2 percent attended but were not administered it, probably because of disabilities, and about 1 percent did not attend. The untested group had lower education and lower fertility, and the 2 percent who attended without testing had a mean of 1.0 child.

The descriptive results are simple to state. The overall mean was 1.80 children. The lowest scoring category had 1.41, and the categories above the median had between 1.87 and 1.89. Above the median the authors found no large differences in average fertility, so the gradient is steep at the bottom and flat at the top. Over 40 percent of the men had two children, and the lower fertility at the bottom came mainly from a large share of childless men together with a small proportion of men with two or three children. In the cohorts studied, 20 percent of men were childless at the latest age of measurement, and the lowest scoring group had approximately twice the relative probability of childlessness of the median group.

The regression results sharpen the picture. In the brother comparison models, which use only variation between full brothers, men in the lowest category (below 76) had 0.56 fewer children than men at the median, and men in the highest category (above 126) had 0.09 more. Men with scores of 81 to 89 had 0.12 fewer children than the median, and men with scores of 111 to 119 had 0.06 more. When the authors treated the score as continuous, a one point increase on the stanine scale, which they note equals 0.5 standard deviations, was associated with 0.036 more children in the full population and 0.074 more in the brother comparison. By our arithmetic that is about 0.07 and 0.15 more children per standard deviation, against a mean of 1.80, or about 4 and 8 percent of the mean.

Three further results matter for interpretation. First, the positive gradient appeared within each level of attained education, so it was not explained by the fact that higher scorers have more schooling. Second, the relationship was consistent across the birth cohorts studied, with a slightly stronger positive gradient for the earliest cohorts. Third, higher fertility among higher scoring men came from more children with a single partner, because having children with more than one woman was more common among men with lower scores. The authors conclude that the relationship is positive in their data, which they describe as inconsistent with a large literature predicting dysgenic deterioration, and they add that they cannot speak to women.

A follow-up by the same authors, published in Intelligence in 2021, asked whether income and marriage account for the pattern. It used Swedish administrative data on 18 birth cohorts of men, with tax records and sibling comparisons. Cognitive ability, education, income, marriage and fertility were all positively associated with each other. Income explained only part of the positive gradient. Much of the association was explained by marriage, but a positive association remained among both ever-married and never-married men, and both low income and low cognitive ability were strong predictors of childlessness and low fertility. The result persisted in the subsample of brothers. The page on how IQ relates to income covers the income side, and the sibling page on IQ and relationships covers partnering.

4 Does a Norwegian Register Study of 953,692 Men Show the Same Pattern?

Yes for men: Norwegian registers covering birth cohorts from 1950 to 1981 show a stable positive link between conscription scores and lifetime fertility, again driven mostly by childlessness at the bottom of the score range. The study by Bratsberg and Rogeberg, published in Biology Letters in 2023, analyzes 953,692 Norwegian-born males and 1,706,438 registered births. Conscription coverage rose from 72 percent to 86 percent across the 1950 to 1960 cohorts and exceeded 90 percent for all but two of the later cohorts. The ability score combines three timed tests, one of vocabulary, one of arithmetic and one of matrix-style figures, summed into a stanine.

The authors compare the average score of fathers, weighted by how many children each had, with the unweighted average score of all men. For the 22 cohorts observed to age 50 (born 1950 to 1971), which comprise 572,029 scored potential fathers and 1,104,594 children, the two means were 5.045 and 4.982. The difference of 0.064 stanines is, in the authors' words, equivalent to IQ point increases of approximately 0.5 over one generation, or about 0.17 IQ point per decade. They stress that this is only a summary indicator of the fertility gradient for men. It should not be viewed as an estimate of generational IQ change, which would also depend on the gradient for women and on heritability.

The timing pattern echoes the Swedish one. Low scoring men had the highest fertility at young ages, while higher scoring men began later but ended with higher fertility, and this held across every cohort even though overall male fertility, measured at age 40, fell from 1.9 to 1.5 as the average age of fathers rose by 3 years. The overall positive relationship was driven mainly by a high rate of childlessness in the lowest scoring group, while low scoring men were more likely to progress to additional children at higher parities. Childlessness rates almost doubled across the cohorts, with larger absolute changes among low scoring men.

The authors list two limitations. Conscription covered men, so the female gradient and the net effect on a population remain unclear, and they note that other studies report stronger negative gradients for women. And the study is descriptive: it does not identify the drivers of the patterns, and the authors say that credible causal inference would need an instrument or a reform that affects education or ability alone. They tie the gradient to the strong fertility gradients for education and earnings in Norway and Sweden, which fits the Swedish follow-up on income and marriage.

5 What Do US Survey and Cohort Studies Find, and Why Do Results Differ by Sex?

US studies that follow people to the end of childbearing report small negative links between test scores and number of children, larger in women than in men, and they point to education and the timing of the first birth as the main routes. Three studies carry most of the weight, and each has a different design.

The Wisconsin Longitudinal Study began with Wisconsin high school graduates of 1957. In Retherford and Sewell's 1988 analysis, which covered over 9,000 graduates, the IQ selection differential for the complete cohort was estimated at eight tenths of an IQ point decline per generation. That figure is the generational change in mean IQ that would follow, hypothetically, if each child had the IQ of the mean of its parents. The authors estimate that the contribution of females to the decline was almost five times that of males. They call the eight tenths an upper bound on the decline in mean genotypic IQ, and their educated guess, based on heritability findings, is that the generational change in mean genotypic IQ is about one third of a point.

In 2010 Meisenberg used the National Longitudinal Survey of Youth 1979 in The Reproduction of Intelligence. Intelligence measured in 1980 was related to the number of children reported in 2004, when respondents were aged 39 to 47. The abstract lists partial correlations, controlling for age, for four groups defined by sex and by the paper's racial categories, and all four were negative. This page reports only the sex contrast in the largest subsample: minus .156 for women and minus .069 for men. The coefficient was also larger for women than for men in the second racial category the abstract lists. The page does not compare racial categories, because the differences depend on how the sample was drawn and on variables the survey did not measure. The paper reports that the link runs mainly through the g factor, is mediated in part by education and income and to a lesser extent by gender attitudes, and that the selection differential for the total sample of 7,344 respondents was minus 1.63 points.

The third study, Woodley of Menie and colleagues (2019) in Twin Research and Human Genetics, returned to the Wisconsin sample with 5,629 people (2,617 men and 3,012 women), IQ measured in 1957 and fertility measured in 2011, when respondents were in their early 70s. Their path model had the age at first birth carry the negative effect of cognitive ability on the number of children. The path from the latent cognitive ability factor to the age at first birth was .25 for men and .46 for women, the only path that differed substantially by sex. The authors read this as higher ability being a risk factor for delaying the first birth, which shortens the fertility window, especially for women, and not as a direct reduction in the number of offspring.

The table has a simple reading and a cautious one. The simple reading is that every study in this section that reported the link separately by sex found a larger negative link for women than for men. The cautious reading is that the two kinds of evidence differ in design, not only in sex: the register studies have population coverage but no women, while the surveys include both sexes but with smaller samples, survey-collected fertility, and test scores taken in high school or adolescence. Comparisons of the sexes inside one design are the more informative ones, and the Wisconsin and NLSY79 results are of that kind. Differences between men and women in average test scores are a separate question, covered on the page on average IQ by gender.

6 What Does the 2018 Systematic Review of 17 Datasets Conclude?

The 2018 review by Reeve, Heeney and Woodley of Menie summarized 17 datasets and found a weighted correlation of minus 0.11 between general cognitive ability and number of offspring, stronger for women than for men and stronger in later decades, while warning that the literature is too limited for a full psychometric meta-analysis. The paper, published in Personality and Individual Differences, is titled a systematic review, and its abstract states that limitations of the literature currently prohibit a psychometric meta-analysis. It nevertheless provides a quantitative summary, using a random effects model on the 17 unique datasets that passed its inclusion criteria. Later papers, including the 2019 Wisconsin study above, refer to it as a meta-analysis, which is why both terms circulate.

A correlation of minus 0.11 is small. By our arithmetic, it means that about 1.2 percent of the variance in number of offspring is shared with the ability score (0.11 squared is 0.0121). The review reports a sex effect, with stronger correlations among females than males, and it reports that the correlation appears to have strengthened through the 20th century and into the early 21st. Its abstract also reports differences by population group, which depend on how each dataset was sampled and which are left to the paper here. The review's discussion draws strong inferences from the trend, and the Norwegian authors, who cite it as the background to their register study, note that it describes substantial consequences if the trend continued. They then present register evidence that does not show that pattern for men.

The review sits in an uneasy relationship with the register studies, and the reason is design. The Swedish authors summarize the earlier literature as showing no clear gradient for cohorts born in the first half of the 20th century, then a small to moderate negative gradient for cohorts born in the second half, with a steeper gradient for women, and with low ability linked most commonly either to childlessness or to large families. They add that this earlier research was, with very few exceptions, based on surveys or samples of school classes. A review can only be as complete as the datasets it pools. The register studies are not pooled into it, and they cover men only, so the two bodies of evidence are complementary and neither cancels the other.

7 What Do Polygenic Scores for Educational Attainment Show About Fertility?

Studies in Iceland and the United States found that people with higher polygenic scores for educational attainment had slightly fewer children, mostly because they had them later, but these are mostly scores for education, not for IQ, and only one of the studies, in Wisconsin, also had measured IQ. A polygenic score adds up the many small genetic variants that large genome-wide studies have linked to a trait, and it explains only part of the differences between people. The page on what heritability means for IQ explains why a polygenic score for intelligence or education is much weaker than the public often imagines.

The largest study is Kong and colleagues (2017) in the Proceedings of the National Academy of Sciences. The authors studied the reproductive history of 109,120 Icelanders and found that an educational attainment score was associated with delayed reproduction and fewer children overall. The effect was stronger for women, and it remained highly significant after adjusting for each person's own educational attainment. Based on 129,808 Icelanders born between 1910 and 1990, they estimate that the average score has been declining at about 0.010 standard units per decade, and they add that, because the score captures only a fraction of the underlying genetic component, that component could be declining two to three times faster. Their significance statement says the rate is small per generation but marked on an evolutionary timescale. Bratsberg and Rogeberg summarize the Icelandic result as implying about a 0.3 IQ point decline per decade, which is their reading, not a figure in the Kong abstract.

A US study used a different sample. Beauchamp (2016) built polygenic scores for several traits in individuals of European ancestry born between 1931 and 1953 in the Health and Retirement Study, a representative study of the US population. The results imply that natural selection has been slowly favoring lower educational attainment in both females and males, at a rate of about minus 1.5 months of education per generation. The author notes that this pales next to the increases in educational attainment observed in contemporary times, and that the results cannot be projected over more than one generation.

Two other studies complicate a simple reading. Conley and colleagues (2016) examined US data across the 20th century and reported that changing patterns of the number of children ever born by phenotype were not matched by shifts in genotype and fertility relationships over time. Their conclusion was that the trends provide no evidence that social sorting is becoming increasingly genetic or that dysgenic dynamics have accelerated. And the 2019 Wisconsin study by Woodley of Menie and colleagues found that the effect of the polygenic score on fertility was indirect: a latent cognitive ability factor, built from educational attainment and IQ, wholly mediated it, with the age at first birth carrying the negative effect. They estimate a decline in the genetic component of IQ of minus .052 points per decade from the score alone (95 percent interval minus .005 to minus .094), and they call that a substantial underestimate. After a correction for missing heritability they give minus .208 points per decade (interval minus .020 to minus .383) using a heritability of 0.40, and minus .424 (interval minus .041 to minus .766) using 0.80. The intervals nearly reach zero at their upper ends.

A genome-wide study of reproduction itself adds context. The Nature Genetics analysis of Barban and colleagues (2016) covered both sexes, with 251,151 individuals for age at first birth and 343,072 for number of children ever born, and identified 12 independent genome-wide significant loci. That work shows that reproductive behavior has measurable genetic correlates, which is a different claim from saying that intelligence is selected for or against.

Two limits apply to these studies. The Swedish authors point out that it was not possible to isolate cognitive ability net of education in the recent genomic studies, with a partial exception for the Icelandic one, and that they were conducted in contexts where the link between socioeconomic success and fertility may differ from contemporary Sweden. And a decline in an average score describes a population over decades, not the ability of anyone born in those decades.

8 What Does "Dysgenic" Mean, and What Do the Authors and Their Critics Say?

"Dysgenic" names a hypothesis that differences in reproduction lower the average of a heritable trait; the published estimates are small per generation, depend on assumed heritability and an unchanged environment, and have been challenged on several grounds. The hypothesis needs three things to be true together: a negative link between the trait and reproduction in both sexes, a heritable trait, and an environment that does not change enough to offset the shift. A negative correlation in one sex, or in one cohort, does not establish all three.

The table lists the figures that authors have published, with the source of each stated. Each row comes from a paper that was read, in full or, where the Sources section says so, as an abstract.

Several published critiques bear on these figures. The Swedish authors report that their results are inconsistent with the large literature predicting dysgenic deterioration, and the Norwegian authors conclude that their data show no dysgenic male fertility in those countries over the period, while saying the net effect remains unclear without women. Conley and colleagues found no evidence that dysgenic dynamics accelerated in the United States. Meisenberg himself calls the heritability estimate debatable. Retherford and Sewell labeled their own figure an upper bound.

A further line of evidence concerns the trend in average scores. Bratsberg and Rogeberg (2018) used Norwegian conscription data for birth cohorts 1962 to 1991 and showed that the rise in scores, its turning point and the later decline can all be recovered from variation within families. Their analysis controls for all factors shared by siblings and finds no evidence for prominent causal hypotheses of the decline that involve genes and environmental factors varying between, but not within, families. The abstract states a general conclusion about between-family explanations, and this page does not extend it to differential fertility beyond that. The detail belongs to the page on the Flynn effect and its reversal and the page on IQ trends across generations, which this page links without repeating.

9 How Large Could a Population Effect Be? The Arithmetic Behind a Per Generation Figure

A population effect is the fertility weighted gap in scores multiplied by the share of that gap that is inherited, so it is a product of assumptions, and the sign of the fertility weighted gap changes with the sex and country that supply the data. Meisenberg sets out the formula. The selection differential, S, is the average score of parents weighted by their number of children, minus the average score of everyone. The response to selection, R, equals S times the narrow sense heritability, h squared. The rest of this section checks the published figures with that formula. Anything marked as our arithmetic is our calculation, not a claim by the authors.

Take Meisenberg's NLSY79 numbers. The total sample selection differential was minus 1.63 points, and he assumed a heritability of 0.5. Our check gives minus 1.63 times 0.5, which is minus 0.815, matching his "approximately minus 0.8". With a 28 year generation, one century is about 3.57 generations, and minus 0.8 times 3.57 is about minus 2.9, matching his figure of about 2.9 points per century. Because R is S multiplied by h squared, the answer moves in direct proportion to the heritability assumed. The two corrected estimates of Woodley and colleagues, minus .208 and minus .424 points per decade, differ by roughly a factor of two by our arithmetic, and the only difference in their method is the heritability of 0.40 or 0.80. Meisenberg notes that the Flynn effect produced environmental gains on the order of 10 points per generation across the 20th century, and he writes that this environmental effect was at least ten times greater than the decline predicted from genetic selection.

Now take the Norwegian numbers. The 0.064 stanine gap is, by our check, 0.064 divided by the stanine standard deviation of 2, times 15, which is about 0.48 IQ points, in line with the authors' approximately 0.5. The authors apply no heritability to that gap, and neither does this page: the gap is for men only, and carrying a heritability assumption from one study to data from another country is the kind of step that produces unreliable forecasts. The point is that the sign of the weighted gap depends on whose fertility is counted.

The formula leaves out at least six things. It leaves out women in the register studies, so the net selection is unknown. It leaves out the environment, since R is a genetic response only, and Meisenberg's figure assumes no migration and an unchanged environment. It leaves out changes over time: the Norwegian gradient was stable across cohorts, while the 2018 review reports a strengthening negative correlation. It leaves out assortative mating, which the page on IQ and relationships covers. It leaves out the fact that a published selection differential describes one cohort observed to one age. And it leaves out measurement error, because a difference of a point or less per generation is far smaller than the uncertainty around any one person's score, a point the page on reliability and validity develops.

10 The Evidence by Design: What Each Study Can and Cannot Show

None of the studies on this page randomized anything, so the right verbs are "is associated with", "went with" and "was found among", and the table shows what each design can carry. The register studies have population coverage but only men. The surveys include both sexes but are smaller. The genetic studies mostly measure a score for education, not for IQ. The review pools datasets of mixed quality.

Three design points follow. First, a brother comparison removes what brothers share, but it cannot remove differences between brothers, and it does not randomize ability. Second, a polygenic score for education is a different object from an IQ score, and Kolk and Barclay note that the genomic studies could not isolate ability net of education, with a partial exception. Third, any projection to a population multiplies an association by assumptions, as the arithmetic section showed.

A short checklist makes headlines easier to read. Ask which sex was studied, because a register of men cannot speak for women and a survey of both can. Ask at what age children were counted, because a count before 30 can reverse the sign in men. Ask whether the claim concerns number of children, timing or childlessness. Ask whether the figure was observed or projected, and what heritability and environment were assumed. And ask who was in the sample, because people who are missing from a survey can differ from those who stay in it. The sibling pages on poverty and IQ and environmental toxins and IQ apply the same design-first reading to other questions.

11 What This Does Not Say About You

Population patterns between test scores and number of children do not describe your plans, your family, your partner or what a score means for you. A weighted correlation of minus 0.11 leaves most of the variation in family size unexplained. A gap of 0.56 children between the lowest group and the median in one Swedish brother comparison is an average over hundreds of thousands of men and says nothing about any one of them. Timing, health, partnership, income, housing and personal values all appear in these studies, and none is captured by a single test score.

An ACIS report gives a Full Scale IQ and six index scores, with percentiles and a 95 percent confidence interval, for the person who takes the test. It is online and unsupervised, it is not a clinical or diagnostic instrument, it is not for hiring, school accommodations or admission to high IQ societies, and it is offered in English only. The technical manual documents how scores are built and what their limits are. If you are curious about your own profile for its own sake, the page on how to choose a test and read the result covers the options, the Full Scale IQ test page describes the longest form, and the page on the CHC model explains the framework behind the six indices.

12 What the Evidence Supports, Stated Narrowly

The verified evidence supports a short list of statements, each tied to a design, and it does not support a single answer for all men and women or a forecast of any population's average score.

• In Swedish men born 1951 to 1967, higher conscription scores went with more children. The register covered 779,146 men. The gradient was steep at the bottom and flat above the median, and it held within each level of education and in brother comparisons.
• In Norwegian men born 1950 to 1981, the positive link was stable across cohorts. It was driven mostly by childlessness among the lowest scoring men, and the authors call their study descriptive.
• In US survey and cohort studies, small negative links appeared, larger in women than in men. The review of 17 datasets gave a weighted correlation of minus 0.11, and the larger link for women appeared in each US study and in the review that reported results by sex.
• Timing carries much of the pattern. Low scoring Swedish men had their first child earlier, and counting children before age 30 reversed the sign. In the Wisconsin data the path from ability to age at first birth was .25 for men and .46 for women.
• Polygenic scores for educational attainment went with later and fewer children in Iceland and the United States. The effect was stronger for women in Iceland, and the scores capture only part of the genetic component and mostly index education, not IQ.
• Published dysgenic estimates are small per generation and assumption dependent. Meisenberg's figure of about 0.8 points per generation assumes a heritability of 0.5 and no migration or environmental change, and critiques from register data, genetic studies and within-family analyses are on record.
• No study here gives the net effect on a population. The register studies lack women, and the surveys rely on assumptions.

What the evidence does not support is any statement that people choose fewer children because of their ability, any statement about a group of people, or any advice on family decisions. None of these follow from the designs.

The measurement standards set the same limit. The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014) open with Standard 1.0, which calls for a clear statement of each intended score interpretation for a specified use and appropriate validity evidence for each intended interpretation. The APA Ethical Principles of Psychologists and Code of Conduct, Standard 9.02(a), says psychologists administer, score, interpret or use assessments in a manner and for purposes that are appropriate in light of the research on or evidence of their usefulness and proper application. A score from one test, taken once, offers no evidence for forecasting the number or timing of anyone's children, and a population pattern is not evidence about any one person.

13 Sources Behind This Page

These references separate population registers, survey cohorts, a systematic review and genetic studies, so that a result from one design is not mistaken for another. Correlations, differentials and sample sizes keep the context of the papers that report them. Where only an abstract was read, the text says so. Arithmetic marked as ours, such as converting the Norwegian stanine gap to IQ points, is not attributed to any author.

• Kolk M and Barclay K. Cognitive ability and fertility among Swedish men born 1951 to 1967: evidence from military conscription registers. Proceedings of the Royal Society B: Biological Sciences, 2019, volume 286, issue 1902, article 20190359.
• Kolk M and Barclay K. Do income and marriage mediate the relationship between cognitive ability and fertility? Data from Swedish taxation and conscriptions registers for men born 1951 to 1967. Intelligence, 2021, volume 84, article 101514 (abstract only).
• Bratsberg B and Rogeberg O. Stability and change in male fertility patterns by cognitive ability across 32 birth cohorts. Biology Letters, 2023, volume 19, issue 6.
• Bratsberg B and Rogeberg O. Flynn effect and its reversal are both environmentally caused. Proceedings of the National Academy of Sciences, 2018, volume 115, issue 26, pages 6674 to 6678 (abstract only).
• Reeve C, Heeney M and Woodley of Menie M. A systematic review of the state of literature relating parental general cognitive ability and number of offspring. Personality and Individual Differences, 2018, volume 134, pages 107 to 118 (abstract only).
• Retherford R and Sewell W. Intelligence and family size reconsidered. Biodemography and Social Biology, 1988, volume 35, issues 1 to 2, pages 1 to 40 (abstract only).
• Meisenberg G. The reproduction of intelligence. Intelligence, 2010, volume 38, issue 2, pages 220 to 230.
• Woodley of Menie M, Rindermann H, Pallesen J and Sarraf M. How intelligence affects fertility 30 years on: Retherford and Sewell revisited with polygenic scores and numbers of grandchildren. Twin Research and Human Genetics, 2019, volume 22, issue 3, pages 147 to 153.
• Kong A, Frigge M, Thorleifsson G, Stefansson H, Young A, Zink F and others. Selection against variants in the genome associated with educational attainment. Proceedings of the National Academy of Sciences, 2017, volume 114, issue 5, pages E727 to E732 (abstract only).
• Beauchamp J. Genetic evidence for natural selection in humans in the contemporary United States. Proceedings of the National Academy of Sciences, 2016, volume 113, issue 28, pages 7774 to 7779 (abstract only).
• Conley D, Laidley T, Belsky D, Fletcher J, Boardman J and Domingue B. Assortative mating and differential fertility by phenotype and genotype across the 20th century. Proceedings of the National Academy of Sciences, 2016, volume 113, issue 24, pages 6647 to 6652 (abstract only).
• Barban N, Jansen R, de Vlaming R, Vaez A, Mandemakers J, Tropf F and others. Genome-wide analysis identifies 12 loci influencing human reproductive behavior. Nature Genetics, 2016, volume 48, issue 12, pages 1462 to 1472 (abstract only).

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014) and the APA Ethical Principles of Psychologists and Code of Conduct, Standard 9.02(a), provide the interpretive framework: evidence must support the use of a score, and limits must be stated. Citing them does not imply that their authors endorse ACIS or any other product. Publication records were checked against Crossref on October 6, 2026.

14 Frequently Asked Questions