People who score higher on intelligence tests tend to have slightly fewer children, a pattern most visible among women in American and British studies. But the strongest studies, which count children only after people finish having them and compare siblings from the same families, suggest the link is small. The pattern depends heavily on the timing of parenthood and on family background, and in Scandinavia, the association runs in the opposite direction for men.
Scientists study this question because it represents a puzzle for evolutionary biology. Intelligence test scores predict more education, higher income, and higher status. In most animals and in preindustrial human societies, status and resources went along with more surviving children, but in many modern societies, the pattern has run the other way. Researchers have proposed several explanations, including more effective use of contraception, the time demands of higher education, and changing values about family size.
This field of research has a highly contentious history. Eugenics was a movement, popularized by Francis Galton in the nineteenth century, that sought to “improve” human populations by encouraging people with traits deemed desirable to have children and discouraging others, in some places through forced sterilization. Early research on intelligence and fertility grew out of eugenicists’ worry that less intelligent people were outbreeding more intelligent ones.
Some researchers use small statistical trends to argue that modern populations face a genetic decline in intelligence, while others have questioned those assumptions. Several of the strongest claims come from a small network of researchers whose backgrounds are described below, while other findings have been replicated by independent teams.
The studies vary widely in how well they measure intelligence. Military conscription tests and batteries of school tests capture intelligence more reliably, while brief vocabulary quizzes partly reflect how much schooling someone has had. Most of these studies report their findings as correlations, which are statistical measures ranging from negative one to positive one that describe how two variables move together.
A 2017 study published in Intelligence, led by Hsin-Yi Chen, offers a concrete example. In that Taiwanese sample, adults with scores below 84 averaged 3.16 children, compared with 2.29 children for those scoring above 115, and the overall correlation was negative 0.18, a small tendency for higher scores to go with smaller families. By comparison, intelligence scores in the same sample correlated 0.65 with education.
A common pitfall in this literature involves the timing of parenthood. People with higher test scores tend to spend more years in school, start their careers later, and delay having children. To avoid this age trap, scientists look at completed fertility, counting a person’s children only after that person reaches age 45 or 50. Looking at the full lifespan provides a much more accurate picture.
Population-level trends can be misleading because they group people from vastly different backgrounds together. Different financial resources, neighborhoods, and family values create confounding, a situation where an outside factor distorts the apparent relationship between two variables. Controlling for a variable uses statistical techniques to hold that outside factor constant.
One of the strongest ways to account for family background is a sibling comparison, which looks at brothers or sisters raised in the same household. Siblings also share about half their genes on average, so these comparisons partly account for genetic differences as well. While this removes shared family influences, it does not remove individual differences, such as differences in health or experiences unique to one sibling, so it still cannot prove a direct cause.
Many of these studies also analyze the same datasets, which can create the illusion of more independent confirmation than exists. Two of the studies described below drew on the National Longitudinal Survey of Youth, which covers Americans born from 1957 to 1964, two on the General Social Survey, a long-running survey of American adults, and two on the Health and Retirement Study, which covers Americans born from 1931 to 1953.
Two used Britain’s National Child Development Study, which follows everyone born in Britain during one week of March 1958, and three used the UK Biobank, a database of about 500,000 British volunteers who tend to be healthier and wealthier than average. The three Swedish studies share the same national registers, as do the two Norwegian studies.
An early American study shows how the timing problem plays out. A 2013 study published in Intelligence, led by Charlie L. Reeve, analyzed Project TALENT, a national survey of American high school students begun in 1960. The researchers measured the number of children people had five and eleven years after high school, and at those young ages, higher test scores predicted fewer children.
The authors also estimated how much average intelligence would change per generation from genetic selection alone, while cautioning that the figures might overstate the decline because many participants had not finished having children. Based on children counted five years after high school, the projected decline for women was about 1.2 IQ points per generation, but when children were counted six years later, it fell to about 0.5 points. For men, the projection dropped from about 1.5 to about 1.15 points.
A 2019 study published in Proceedings of the Royal Society B, led by Martin Kolk, showed how this age trap works. Using Swedish national registers of men born between 1951 and 1967, who took intelligence tests in their late teens as part of mandatory military conscription, the researchers found that the lowest-scoring men had about twice as many children as the highest scorers when children were counted before age 30.
When the researchers waited until the men reached age 45, the trend flipped, and higher scorers ended up slightly ahead in family size compared to their peers. Timing played a large role: among men who became fathers, the lowest scorers had their first child at about 27.6 years old on average, compared with about 31 for the highest scorers.
Other research finds the same delay in parenthood, even though higher scorers tend to mature physically earlier. A 2025 study published in Adaptive Human Behavior and Physiology, led by Jose C. Yong and co-authored by Satoshi Kanazawa, analyzed Britain’s National Child Development Study alongside the American Add Health study, which began surveying teenagers in 1994 and 1995. The researchers found that higher childhood test scores predicted an earlier onset of puberty but a later age at first child in the British data.
For example, girls with intelligence scores above 125 had their first menstrual period at 12.60 years old on average, versus 13.24 for girls scoring below 75. Counting children born by age 33, the highest-scoring women had their first child at about 27.9, compared with 21.5 for the lowest scorers, and the gap for men was about 28.5 versus 24.4. While higher intelligence initially seemed linked to later marriage, that association was not reliable in either sample after accounting for education, nutrition, and health.
Yong, a senior lecturer at James Cook University, argues that the pattern reflects a split between fixed biological traits, such as the timing of puberty, and flexible behaviors shaped by preferences. He told PsyPost that “even though intelligent individuals are reproductively endowed at the physiological level with earlier puberty (which should typically predict faster reproduction), they opt for reduced sexual activity and fewer children at the behavioral level (resulting in slower reproduction).”
He added that the link with puberty held up after the statistical adjustments, while “the overall predicted patterns for behavior still held more rather than less, particularly for number of children.” Those patterns come from British and American data, however, and the Scandinavian results for men, described below, show that a later start does not always mean fewer children in the end.
When researchers look at completed families, the link between intelligence and having children differs notably between men and women. For men, Scandinavian data show a positive trend. In Kolk’s 2019 Swedish study, brother comparisons showed that the lowest-scoring group of men, those scoring below 74 or roughly the bottom 4 percent, had about 0.56 fewer children than the median group. The highest-scoring men had about 0.09 more children against an average of about 1.8 children, with the median representing the middle point of the score distribution.
A 2021 study in Intelligence, led by Kolk, expanded on those Swedish brother comparisons. The lowest scorers were about 20 percentage points more likely than median scorers to remain childless. Ever marrying by age 45 differed by about 20 percentage points between the lowest and highest scorers. Among men who married, the lowest group had only about 0.1 fewer children, leading the researchers to conclude that lower-scoring men had fewer children primarily because they had difficulty forming stable partnerships.
A 2023 study published in the Journal of Population Economics, led by Francis Kramarz, used Swedish military conscription tests to compare thousands of twins and close siblings. The conscription process also rated non-cognitive skills through an interview with a psychologist assessing traits such as emotional stability and social maturity. Cognitive scores and these non-cognitive ratings tend to go together, correlating at about 0.4. A standard deviation is a unit that measures how spread out scores are from the average, equaling about 15 points on a standard IQ test.
When the researchers compared men with the same non-cognitive rating, each standard deviation of higher cognitive scores was linked to about 0.02 fewer children, a very slight negative link. Kolk’s studies, which did not hold those skills constant, found a positive link. When comparing twin brothers, the higher-scoring brother tended to have about 0.09 more children per standard deviation. Non-cognitive skills showed the strongest link of all, with each standard deviation associated with 0.13 to 0.17 additional children, both across the population and within twin pairs.
This positive trend for men holds true beyond Sweden. A 2023 study published in Biology Letters, led by Bernt Bratsberg, examined Norwegian men born from 1950 to 1981 across 32 birth cohorts, or groups born in the same year. They found that high-scoring men had children later but eventually had more of them than lower-scoring men. Across the cohorts, childlessness nearly doubled, with the largest increases among the lowest scorers, though lower-scoring men who became fathers were more likely than higher scorers to go on to have additional children.
The results for men do not agree across all countries. The association is practically zero in Britain’s 1958 birth cohort, analyzed in a 2014 study published in Social Science Research by Kanazawa. It is slightly negative in the United States, with correlations of about negative 0.06 to negative 0.07, in studies by Gerhard Meisenberg and Rosemary L. Hopcroft.
In a cohort of people born from 1950 to 1956 in Aberdeen, Scotland, analyzed in a 2011 study published in Intelligence led by Sophie von Stumm, higher-scoring men had about 31 percent lower odds of having any children for each standard deviation increase in childhood intelligence. Odds describe the likelihood of an event happening compared to it not happening. Intelligence was unrelated to the total number of children for either sex in the Scottish study, though the odds of being divorced rather than married at midlife were about 13 to 14 percent lower for each standard deviation increase in childhood intelligence.
The divorce finding echoes the Swedish evidence that stable partnerships play a large role in men’s family size. Why might Scandinavia differ? In their 2019 study, Kolk and Kieron Barclay pointed out that in Nordic welfare states, education and income are positively tied to men’s fertility. They argued that preferences for smaller families spread first among higher-status groups and have since diffused across society, leaving better-off people better able to afford the two or three children most people want, and they predicted that similar positive links might emerge elsewhere.
For women, American and British surveys mostly show a small negative association. A 2010 study published in Intelligence, led by Meisenberg, analyzed the National Longitudinal Survey of Youth and reported a correlation of negative 0.16 for white women and negative 0.07 for white men. His results also showed correlations of negative 0.24 for Black women and negative 0.03 for Black men, though the Black subgroups were smaller and the result for Black men was not statistically significant, meaning the researchers could not rule out that it arose by chance. A 2015 study published in Evolution and Human Behavior, led by Hopcroft, analyzed the same American survey and found correlations of negative 0.15 for women and negative 0.06 for men.
Income adds another twist. In Hopcroft’s 2015 study, higher personal income, measured after childbearing in 2010, was associated with more children for men (a correlation of 0.08) but fewer children for women (a correlation of negative 0.15). An earlier 2006 study by Hopcroft in Evolution and Human Behavior used the General Social Survey’s ten-word vocabulary test and found that while higher vocabulary scores predicted slightly fewer children, higher-income men had more biological children.
A classic 1988 study published in Social Biology, led by Robert Retherford, tracked more than 9,000 Wisconsin high school graduates from the class of 1957. The researchers estimated a selection differential of about 0.8 IQ points per generation, a figure describing what would happen if children matched their parents’ average score. Because this calculation comes before accounting for how much of intelligence is passed down genetically, it is not directly comparable to the genetic projections described later. The link was much stronger for women.
Researchers debate whether staying in school longer explains this drop in fertility for women. In a 1989 follow-up published in Intelligence, Retherford’s team indicated that once the researchers accounted for how much schooling people completed, intelligence scores predicted little about family size. Meisenberg and Reeve both reached similar conclusions, finding the association ran largely through education and income.
On the other hand, Kanazawa’s 2014 study reported the reverse: once childhood intelligence was taken into account, education no longer predicted whether women became mothers. For each standard deviation increase in childhood intelligence, a woman’s odds of becoming a mother by age 47 dropped by 21 to 25 percent, though the study found no such association for men.
Social standing may also shape the pattern. In the Scottish study led by von Stumm, occupational class measured at midlife rather than intelligence was linked to childlessness for women. Among higher-status women, childless women scored higher than mothers, but among lower-status women, mothers scored higher than childless women. The authors suggested that lower status may be a consequence of motherhood rather than a cause.
Outside Europe and North America, the Taiwan study led by Chen reported a correlation of negative 0.25 for women, while the link for men was not statistically significant. The sample’s ages ran from 35 to 90, a span covering the country’s historical fall from about six children per woman to about one. Because older participants grew up when large families were the norm and schooling was less common, differences between generations can masquerade as effects of intelligence, a problem known as cohort confounding. The authors reported the link held after adjusting for age, though they did not publish the adjusted figures.
When researchers use sibling comparisons to control for family background, the negative trend for women’s education and grades shrinks. In the 2023 Journal of Population Economics study, Kramarz compared twin sisters in Sweden. Across the general population, each extra year of schooling went with about 0.04 fewer children, and better school grades also went with fewer children. When the researchers compared twin sisters, the schooling association shrank to about 0.01 fewer children per year and was no longer statistically significant, while the association for school grades vanished almost entirely.
Because women were not conscripted, the Kramarz study had no intelligence test scores for them, only estimates based on their brothers’ scores for population comparisons, so researchers could not test women’s own scores within families. Kramarz and colleagues suggested that some families prize education while others prize large families. Families that prize education may cultivate academic skills and have fewer children, and their children may inherit both the skills and the preference.
That dynamic would make academic skills and small families go together across families, even when the higher-scoring sibling within a family has no fewer children. Consistent with this idea, fathers who had more children and less schooling than their own skills predicted tended to raise sons whose non-cognitive skills outpaced their cognitive scores.
These modest correlations underpin a more contentious claim. A network of authors frequently publishes studies projecting a genetic deterioration in intelligence in modern populations. A 2018 systematic review published in Personality and Individual Differences, led by Reeve, pooled 17 datasets to estimate the correlation between intelligence and number of children. A systematic review is a comprehensive summary of all available research on a specific topic.
Roughly half of the datasets in the 2018 review came from a specific network of researchers, while the rest were independent studies from researchers like von Stumm and Retherford. The review reported an overall correlation of negative 0.11 between cognitive ability and the number of offspring. The authors reported that newer studies tended to find stronger negative associations than older ones, though that trend rests on fewer than 20 studies and weakens noticeably when a single 1932 study is removed. This also contrasts with the Norway study, which found a stable positive pattern for men across 32 birth cohorts.
The review’s own data highlighted the importance of completed fertility. When the researchers limited the analysis to studies where people had finished having children, the correlation for men was about negative 0.03 and not statistically significant, while the correlation for women was negative 0.12. Overall, the correlation was negative 0.13 for women versus negative 0.06 for men.
To project future declines, some of these studies use a mathematical approach based on a formula called the breeder’s equation. Researchers weight each parent’s score by their number of children, compare the result with the population average, and multiply the gap by heritability. Heritability is a statistical estimate of how much variation within a population is explained by genetics.
The 2017 Taiwan study led by Chen used this method to project a drop of 1.19 IQ points per generation. Chen’s team assumed a heritability of 0.71, and Meisenberg assumed 0.5 in his 2010 study to project a drop of 0.8 points per generation in the United States. A higher assumed value produces a larger projected decline.
A 2004 study published in Intelligence, led by Richard Lynn, reported correlations of about negative 0.08 to negative 0.18 using the same General Social Survey vocabulary quiz. The authors projected a drop of 0.9 points per generation by multiplying their correlation directly by a heritability estimate. This shortcut differs from the standard formula, which accounts for how much family sizes vary. If every adult had exactly two children, even a strong correlation would leave the next generation’s average unchanged.
These projections also rest on several assumptions, including an assumed heritability value and an environment that stays the same from one generation to the next. During the twentieth century, average test scores rose by roughly three points per decade. This trend, known as the Flynn effect, far outpaced any projected genetic change. Average scores have stalled or fallen in several countries in recent decades, including Norway, Denmark, Finland, and France, leading Lynn and colleagues to point to genetic selection.
In a 2018 study published in Proceedings of the National Academy of Sciences that tested that idea, Bratsberg and colleagues compared brothers and found that the Flynn effect and its recent decline can be recovered within families. Younger brothers outscored older brothers while scores were rising, and scored lower while scores were falling. Since brothers share parents, changes in who is having children cannot account for most of the shifts, pointing instead to environmental causes. The study was designed to rule out explanations rather than to identify which environmental factors were responsible.
Studies that measure DNA directly offer another test, and they do show a small signal related to education. A polygenic score is a tool that combines the small effects of hundreds of thousands of genetic markers to predict a trait. A 2017 study published in Proceedings of the National Academy of Sciences, led by Augustine Kong, found that an educational attainment polygenic score was linked to delayed reproduction and fewer children in Iceland, an effect that was stronger for women and held after adjusting for education. For women, each standard deviation higher on the genetic score went with about 0.08 fewer children and a first birth about 0.6 years later, in a sample of about 109,000 Icelanders.
In the Iceland study, the average score declined by about 0.010 standard units per decade, where one standard unit equals one standard deviation of the score. The directly measured drop translates to about 0.038 IQ points per decade, roughly 80 times smaller than the three-point-per-decade rise in test scores during the twentieth century. The genetic score captured only about 4 percent of the differences in education between people. To reach a larger figure of 0.30 points per decade (about a tenth of the twentieth-century rise), the authors assumed that all genetic influences on education, which they put at about 30 percent of those differences, were changing the same way as the small portion the score measures.
A 2016 study in the same journal, led by Jonathan Beauchamp, analyzed the Health and Retirement Study and estimated a decline of about 1.5 months of schooling per generation in the United States, an amount he described as small compared to the gain of roughly two years of schooling per generation in recent times. Natural selection is the process by which traits become more or less common in a population based on reproductive success. A 2022 study published in Behavior Genetics, led by David Hugh-Jones, investigated this process in the UK Biobank database.
Genetic scores predicting higher education and earnings also predicted lower fertility, though the direction of natural selection reversed among older parents or after controlling for age at first birth. In other words, among people who started families later, higher education scores were linked to having more children, not fewer. The negative effect was concentrated among lower-income groups, younger parents, people with more sexual partners, and people not living with a partner.
A 2024 study in Behavior Genetics by Hugh-Jones, which used the same survey, found that genetic scores correlating with more education were selected against across three generations of Americans. The economic explanation holds that people with greater earning power give up more income when they spend time raising children, predicting the pattern should be strongest among lower-income, less educated, unmarried, and younger parents. The study found stronger effects among low-income and unmarried parents, but not among less educated or younger ones. Together, these genetic studies point to a real but small shift, far smaller than the environmental changes in test scores over the same period.
Several researchers who advance the decline argument have documented ties to organizations widely described as funding eugenics research and promoting scientific racism. Scientific racism is the use of scientific-sounding claims to argue that some racial groups are innately superior to others. Some of these authors use the term dysgenic, a word rooted in the eugenics movement, to describe a hypothesized genetic decline in intelligence.
The term has spread beyond this network. A 2023 commentary in Biology Letters, led by Rebecca Sear, responded to the independent Norwegian study, noting that its findings of a positive trend for men in fact challenge eugenicist claims. The authors argued the term treats value judgments, such as the idea that higher test scores are desirable, as science. They also noted that, in evolutionary terms, if people with lower scores had more children, scientists would describe them as better adapted to current conditions, not as a sign of decline.
Lynn, who died in 2023, was lead author of the 2004 study and a co-author on the 2017 Taiwan study, and he promoted the claim that racial differences in intelligence are genetic. Ulster University withdrew his emeritus title in 2018. He was president of the Pioneer Fund, whose role in financing scientific racism is documented in psychologist William H. Tucker’s 2002 book, The Funding of Scientific Racism: Wickliffe Draper and the Pioneer Fund, and he edited Mankind Quarterly, which Tucker described as a journal written by racists for racists.
Meisenberg was described by the New Statesman as a fellow director of the Pioneer Fund and was editor-in-chief of Mankind Quarterly. Marian Van Court, a co-author on the 2004 study, founded the Future Generations eugenics website and wrote “The Case for Eugenics in a Nutshell” for The Occidental Quarterly, which the Southern Poverty Law Center describes as a racist journal and lists as a hate group.
Michael A. Woodley of Menie, a co-author on the 2018 review, was criticized by the New Statesman and other outlets for his involvement in the London Conference on Intelligence, an invitation-only conference founded in 2014, held four times at University College London, and revealed by the London Student in January 2018. The gathering was held on campus without the university’s knowledge, and critics and student journalists labeled it a eugenics conference. Woodley of Menie organized a published defense co-signed by 14 other attendees who disputed the characterization.
Kanazawa’s record involves a different controversy. The author of the 2014 study and co-author of the 2025 Yong study was banned from non-peer-reviewed publishing for a year after an internal investigation found he had brought the London School of Economics into disrepute over a 2011 blog post claiming Black women were less physically attractive. In the 2014 study, Kanazawa based a forecast of declining population intelligence partly on the premise that boys inherit intelligence solely from their mothers through the X chromosome. Modern genetics researchers widely reject this premise, as cognitive ability is massively polygenic, meaning it involves thousands of genes spread across all human chromosomes.
A completely different branch of research asks the reverse question. These studies investigate whether having children is linked to a person’s thinking skills later in life. A 2023 study published in Social Science & Medicine, led by Yan Zhang, analyzed cross-sectional data from the UK Biobank. Cross-sectional means researchers measured people at a single point in time.
The researchers found that fathers scored better on all five cognitive tests than childless men. Mothers scored lower than childless women on fluid reasoning, numeric memory, and prospective memory, with most of that gap explained by education. However, mothers scored better on reaction time and visual memory, an advantage that grew larger after controlling for education. The researchers controlled for polygenic scores, and the results did not substantially change.
A 2017 study published in The Journals of Gerontology: Series B, led by Sanna L. Read, looked at English adults aged 50 and older for about eight years. The researchers initially found a U-shaped link where people with either no children or many children had lower cognitive scores. The U-shape faded after socioeconomic adjustment, but childlessness remained linked to slightly lower scores in both men and women, and women who had a child after 35 scored slightly higher. For women, the link between childlessness and lower cognitive scores strengthened after statistical adjustment.
A 2020 study published in Scientific Reports, led by Kaida Ning, also used the UK Biobank database. The researchers found that parents had faster reaction times by 4 to 10 milliseconds and fewer memory errors, with larger differences for fathers. The paper also suggested that having two or three children was associated with brains that appeared 0.5 to 0.7 years younger on MRI scans. Brain age is a computer-generated estimate of how old a brain looks compared to others of the same age.
A lifestyle effect is one possibility, but it is difficult to tell if parenthood changed people’s brains or if healthier, more socially active people were more likely to become parents. These patterns are also consistent with the Scandinavian finding that the lowest-scoring men are the most likely to remain childless.
Several uncertainties remain in the study of intelligence and family size. Nearly all of this research is observational, so even the strongest designs show associations rather than cause and effect. While massive datasets from Scandinavia offer the best view for men, researchers still lack equally large sibling comparison studies that include women’s own cognitive test scores. Swedish and Norwegian register studies also reflect societies with strong welfare states, and patterns might differ in nations with less social support.
Nearly all of the completed-fertility evidence describes people born before the early 1980s, so it cannot say whether the same patterns hold for people starting families today. There is also little evidence from outside Europe, North America, and Taiwan. Yong acknowledged the same limit for his own work: “perhaps we can’t be 100% sure that the results will pan out similarly in other non-Western cultures or in later generations.”
Leave a comment
You must be logged in to post a comment.