Mars and Venus
In 1992 a book told its readers that men are from Mars and women are from Venus. John Gray’s Men Are from Mars, Women Are from Venus became a worldwide best seller (the author’s website claimed more than 30 million copies in 2005), and its premise was that the sexes differ so deeply in how they think, feel, and handle stress that they might as well be from different planets. The book was a success because it explained something people had felt in their own homes. The question for a scientist is a different one: how far apart are men and women, measured in the same way, on one trait after another?
The first great attempt to answer it was a book by the psychologists Eleanor Maccoby and Carol Jacklin in 1974, The Psychology of Sex Differences, which sifted through a research literature of well over a thousand studies. They concluded that four differences were well established: in verbal ability, in visual-spatial ability, in mathematical ability, and in aggression. Many popular accounts remembered only the list of differences. The book itself had also knocked down a number of popular beliefs, such as that girls are more “social” or suggestible, or lack achievement motivation. Fifty years on, with hundreds of meta-analyses, the picture is clearer, and more interesting than either side expected.
Course 9 (Lesson 1) has already described one summary: the psychologist Janet Hyde’s 2005 count of 124 effects from 46 meta-analyses, of which 78 percent were small or close to zero. A 2015 review by Ethan Zell and colleagues, of 106 meta-analyses (more than 20,000 studies and 12 million participants), found the same: an average gap of 0.21 standard deviations, and 85 percent of the effects small or very small. We will not repeat it. We will ask the next questions: which differences are big, which are small but matter, which have changed over time, and what changes them.
A ruler for differences
To compare gaps we use d, the distance between the averages measured in standard deviations (see Lesson 1). It helps to have landmarks. The largest differences between men and women are physical. In Hyde’s tally the biggest were how fast and how far people throw a ball (d of about 2). Adult height in the United States, in a national health survey of 2015 to 2018, is 175.3 centimeters for men and 161.3 for women, d of about 1.9. Even at that size the groups overlap: pick a man and a woman at random and the man is taller about 91 times in 100, so in about one pairing in eleven the woman is taller. Almost every psychological difference is far smaller, and most of them lie between a coin toss and a 60 percent chance for the higher-scoring group. Explore a ruler of real gaps below.
A ruler of real differences
Each bar is the gap between men’s and women’s averages from a large study or meta-analysis. Blue means men average higher, red women. Click a bar to see what the gap means for a random pair, and for guessing someone’s sex from a score.
Values are rounded from the studies named in each note and apply to the groups studied (often American or European samples). Direction and size can change with age, test, and country.
On a trait where the gap between averages is d = 0.5 (a medium difference), a woman and a man are picked at random. About how often does the man score higher?
The spinning shape
One of the largest cognitive differences is a spatial skill, mental rotation: picturing how a three-dimensional object would look if it were turned. In 1971 the psychologists Roger Shepard and Jacqueline Metzler showed people pairs of drawings of block figures, each built of ten cubes joined into an arm with three right-angled bends, and asked whether the two were the same shape turned around. The time it took to answer rose in a straight line with the angle between them, as if people really were turning the shape in their heads. (Shepard later recalled that the idea began with a dream of rotating objects.) A paper-and-pencil version, the Vandenberg and Kuse test of 1978, made the task easy to give to large groups, and it quickly turned up a difference between the sexes.
The difference is real and replicated. Across many studies men score higher on average, by about 0.5 to 0.7 standard deviations, depending on the test; in one review of 70 studies it was 0.57. That is among the biggest psychological gaps in the research and still leaves much overlap, with a man ahead in about two pairs out of three. It is larger when the test is timed, and it grows as the time limit shrinks. Strategy matters too: when people talk through how they solved the problems, men more often use a shortcut that rules out wrong answers by their overall shape without rotating anything. And it appears early, but how early is disputed. A review of 303 comparisons involving more than 30,000 children found a small male advantage emerging in childhood and growing to a moderate one in adolescence, and some studies of infants a few months old report that boys do better, while one of 208 six-month-olds found no difference.
Can it be trained? Yes, for both sexes. A 2013 review of 217 training studies found that practice improved spatial skills by about half a standard deviation, and improved men’s and women’s scores almost equally (0.54 and 0.53). Because both gained, the average gap did not shrink in these studies. The most striking exception is small. In a 2007 experiment at the University of Toronto, students with little experience of video games played either an action game or a puzzle game for ten hours over four weeks. Women gained more than men from the action game, and the gap in mental rotation was reduced. The study was small, about twenty students. A spatial-skills course begun in 1993 at Michigan Tech for first-year engineering students with weak spatial skills was later aimed especially at women, on the same idea.
What about different cultures? Here the studies disagree, and it is better to say so. In 2011 the economists Moshe Hoffman, Uri Gneezy, and John List timed nearly 1,300 villagers in northern India on a four-piece jigsaw. Among the Karbi, who trace descent through fathers, men finished about a third faster than women; among the neighboring Khasi, who trace descent and inherit property through women, the gap was not significant. Education explained about a third of the difference, and critics noted that a single puzzle and two societies are thin evidence. In a larger internet survey of more than 200,000 people from 53 nations, men beat women on mental rotation in every nation, and the gap was larger in richer and more equal ones. In a navigation game played by more than half a million people in 57 countries, the size of the gap went with how unequal the country was, though the players were volunteers. A meta-analysis of 266 navigation studies found men ahead by about 0.35 standard deviations overall, and by 0.15 in children under 13. The results do not support “fixed by biology” or “just culture.”
A popular claim holds that men turn the map to face the way they are going and women turn their heads. We could not find a study that tested it, and we would not present it as science. What studies do find is that, when giving directions, men tend to use more distances and compass directions and women more landmarks, and that this difference can vanish: in one experiment, drawing people’s attention to compass directions with extra map legends made men and women use them equally.
Voles, hunters, and a long story
An evolutionary explanation is most convincing when it could have failed. The best-known test is a pair of vole species. Meadow voles are polygynous: males range over a wide area looking for mates. Pine voles are monogamous, and the sexes stay near home. In 1990 Steven Gaulin and colleagues reported that male meadow voles ranged farther than females, did better in laboratory spatial tests, and had a larger hippocampus relative to their brains; in the monogamous pine vole, none of those differences appeared. (The animals are voles, not the moles or mice sometimes quoted.) Later work was less kind. A comparison of 11 species found no support for the range hypothesis, and a 2024 analysis of 21 species, including human peoples who live by foraging, again found little evidence that differences in navigation follow differences in range.
For humans the story is mixed. Hadza men in Tanzania, tracked with GPS for more than 2,000 person-days, walked farther and explored more land than women, while one study of 73 Tsimane people in Bolivia found no sex difference in dead reckoning, the sense of one’s direction after wandering. A second theory, that women evolved better memory for where things are, has partial support: a meta-analysis of 36 studies found women ahead on remembering object locations in most conditions, but the result depended on the objects, and masculine objects favored men. Voles show that the idea can be tested, and in one species pair it passed. Whether human spatial skill was shaped this way is, so far, untested.
How much does it matter? Spatial ability in adolescence predicts who earns advanced degrees in science and engineering later: in a study that followed 400,000 American students for 11 years, it marked out those who later earned advanced degrees in science and engineering. Yet one recent American study that followed 690 children from birth to age 26 found that the sex difference in spatial skill explained only a small part of the gap in who chooses STEM majors, and no one has a good estimate for engineering or physics.
Words and numbers
The other famous differences in the 1974 list have moved. Language first. In the United States, girls outscore boys in reading and writing, and the gap grows through school: in national test data covering 3.9 million students, it reached 0.32 standard deviations in reading and 0.55 in writing by grade 12. The OECD’s PISA test of 15-year-olds in 2018 found girls ahead in reading by 30 points on average across rich countries, a gap of roughly a third of a standard deviation by our estimate. School grades show it too: across 369 samples girls were ahead overall by 0.23, by 0.37 in language courses, and by a very small 0.07 in math. The popular claim that boys are four times as likely to be dyslexic does not hold up: school referrals do over-identify boys, but studies of whole populations find about 1.5 to 3 boys for each girl. Stuttering is different: in adults the ratio is four to one or more, probably because more girls recover.
Now math. In the best-known test of the claim, Janet Hyde and colleagues analyzed the scores of about seven million American students in grades 2 to 11 and found gaps between boys and girls of between −0.02 and +0.06 standard deviations, essentially zero at each grade. A review of 242 studies found an overall gap of 0.05; a study of 493,495 students in 69 nations found the gap on the TIMSS test to be −0.01 and on PISA to be 0.11. Differences in math attitudes and anxiety were larger than those in scores: girls reported more anxiety, boys more confidence. These results do not mean the gap is zero everywhere. When researchers followed all 2.65 million first and second graders in France, they found no gap at school entry and a boys’ advantage of about 0.20 after one year. It grew with schooling, not with age. Where the gap shows up depends on the country, the test, and the year.
In mainland China the pattern is similar, with its own twists. In the 2018 PISA results for the four Chinese provinces and cities that took part (Beijing, Shanghai, Jiangsu, and Zhejiang, home to about 13 percent of the mainland population), girls led boys in reading by 13 points while boys led in math by 11 and in science by 12; in Hong Kong girls led in all three, including math by 6 points; in Singapore boys led in math by 4. A good deal of the national debate is about whether the equality of a society changes the gaps, and the evidence is contested. One influential 2008 analysis of PISA data said the math gap disappears in the most gender-equal countries. A larger analysis of 69 nations found that composite measures of equality mostly did not predict the math gap, though some specific measures did, such as women’s share of research jobs. A third, of 75 nations, found no consistent link.
The very top
Average gaps in math may be near zero, but at the top of the scale the numbers look different. In the 1970s and 1980s, among young Americans who scored 700 or more on the SAT math test at age 12 or 13, boys outnumbered girls about 13 to 1. By 2005 the ratio had fallen to about 2.8 to 1 in the same talent search, and a 2018 study of two million top students puts it at about 2.5 to 1 at the very highest level, 1 in 10,000. Why does the tail look so different from the middle? One reason is that, in many data sets, boys’ scores are more spread out than girls’. Statisticians measure this with a variance ratio, which for math in large samples runs from about 1.05 to 1.2: the boys’ variance is 5 to 20 percent larger. Havelock Ellis proposed this “variability hypothesis” in the 1890s, and it has been argued over ever since. In January 2005, at a conference on the science workforce, the president of Harvard, Larry Summers, suggested that differences in variability might be one reason for the shortage of women among the very top scientists, and a national argument followed.
Try the arithmetic below. The point is not to settle the argument, but to see how little it takes to change the ratio far out, and how little it changes in the middle.
Why the tails look different
Two bell curves, women’s and men’s. Change the gap between the averages and how spread out men’s scores are, then move the cut-off and read how many men there are for every woman above it.
A model of two normal curves; real score distributions have heavier tails and ceilings. The same ratios apply at the bottom of the scale, with women the minority there.
There are two cautions. The same spread puts more boys at the bottom, too: in reading, boys are over-represented among the weakest readers, a worry in many countries. And the variance ratio does not say why. Hyde calculated in 2014 that ratios of this size cannot account for the fact that women earn only about a fifth to a quarter of engineering degrees in the United States, and the ratio is not constant across countries: in PISA data from 2003 it ran from 1.24 in Canada and Iceland to 1.00 in the Netherlands and 0.95 in Indonesia, where girls were the more variable. Brain studies show the same pattern: in a pooled analysis of 16,683 brain scans from people aged 1 to 90, men’s measures were more variable in all the subcortical volumes. Whether the cause is biology, culture, or both is unsettled.
The international mathematics Olympiad shows what changes at the top. Girls were 9.2 percent of contestants from 2000 to 2021, and women won about 3 percent of the gold medals through 2021. But the share varies by country and decade: China’s team had three girls in nineteen places in the 1980s (China first competed in 1985) and one in sixty in 1999 to 2008, Serbia’s 13 in 60, and the United States fielded no woman at all for its first 23 competitions. The first woman to win the Fields Medal, the highest honor in mathematics, was Maryam Mirzakhani of Iran in 2014; she died in 2017 at the age of 40, and she is said to have worked by drawing on large sheets of paper and writing formulas around the drawings. A second, Maryna Viazovska, won in 2022, and a third, Hong Wang, in 2026; she was born in Guilin, studied at Peking University, and received her doctorate from MIT.
A threat that faded
If stereotypes shape performance, the stereotype itself might cost points. In a famous 1995 study the psychologists Claude Steele and Joshua Aronson found that Black students scored lower on a hard verbal test when it was presented as measuring ability, and in 1999 Steven Spencer, Steele, and Diane Quinn reported that women did worse on a math test when told it showed gender differences. This “stereotype threat” became one of the best-known ideas in social psychology. Later work cut it down to size. A 2015 meta-analysis of studies of girls found an average effect of −0.22 and clear signs of publication bias; a 2019 analysis of tests taken in real conditions put it at about −0.14; and a 2018 pre-registered study of more than 2,000 Dutch students aged 13 and 14, designed in advance to give an unbiased answer, found no effect. The best summary today: a small, fragile effect in published experiments that does not hold up in the best-designed tests, and not an explanation for the gaps.
Another view
Are “small average gaps” really small? The psychologist Marco Del Giudice and colleagues argue that looking at one trait at a time hides the picture. In a 2012 analysis of 10,261 American adults on a personality questionnaire, the gaps were small one by one (an average of 0.26), but combined into a single overall distance they were large: the multivariate distance, D, was 1.49, which they estimated could tell men from women about three times in four, and 2.71 (about nine in ten) when measurement error was removed. A 2020 replication found D of about 2.1. The same logic applies to interests: Rong Su and colleagues, pooling inventories filled in by 503,188 people, found that men preferred working with things and women with people at d = 0.93, a large gap, and 1.11 for engineering. Hyde and others reply that D is built to find the biggest difference and says nothing about what it means, and that many such traits are partly shaped by culture and change over time. A 2022 attempt at reconciliation by Alice Eagly and William Revelle concluded that the two views organize the same data in different ways. There is also a second debate, about whether differences in personality and in interests grow or shrink as societies grow richer and more equal; different data sets disagree, and one large study (Breda and colleagues, 2020) found that the stereotype that math is “not for girls” was stronger in more developed countries, which may explain why some gaps widen there. What the sides share is more important than what divides them: single traits mostly differ little; a handful differ a lot; and none of this tells you what a given man or woman will be like.
Closer to home
In the United States women earn about 58 percent of all bachelor’s degrees, 66 percent of those in the biological sciences, 41 percent in mathematics and statistics, 24 percent in engineering, and 23 percent in computer science, in 2021–22. The computer-science figure was 37 percent in 1983–84, fell to under 18 percent by 2010–11, and has since risen again; in 2021–22 there were more women graduates (about 24,500) than in 1984–85 (about 14,400). A change that large in a few decades is not a change in brains. Interests matter, too: Su and colleagues found a gap of 1.11 standard deviations in interest in engineering, and interests, like the numbers of women in a field, change with the culture and with the way fields are taught and advertised.
In China, women’s share of university students rose from about 26 percent in 1985 to about 53 percent in 2021 (one study’s estimate). The share of women among academicians of the Chinese Academy of Sciences was about 6 percent in 2019, and about 5 percent for the Academy of Engineering. Women’s record on the national college entrance exam is more complicated: an economics study (Cai and colleagues, 2019) found that women do worse on the high-stakes 高考 than on lower-stakes mock exams, and decline more after a bad earlier result, which points to pressure and not raw ability. The slogan 妇女能顶半边天 (“women hold up half the sky”) is attached to the Maoist drive to bring women into the workforce, though we could not trace who first said it or when. The “Iron Girls” of the 1960s, led by Guo Fenglian at Dazhai in Shanxi from 1963, when she was 17, became its public image. By the International Labour Organization’s modeled estimates, the share of Chinese women aged 15 and over in the labor force has fallen from 73 percent in 1990 to about 59 percent in 2025, while the American figure has stayed between about 55 and 59 percent.
What to keep
- The size of a gap, not its existence, is the useful question. Height and throwing are large; most psychological gaps are small and leave the groups overlapping a great deal.
- Mental rotation is one of the largest cognitive gaps (about 0.5 to 0.7). It responds to practice in both sexes, and its relation to culture is disputed.
- The gaps in reading, writing, and math have changed with school, country, and decade. The math gap is near zero on average, while the gap at the very top has narrowed from about 13 to 1 to under 3 to 1.
- Greater spread among boys can magnify small differences far out, but it does not explain who studies engineering. Famous claims such as stereotype threat have shrunk when retested, and both sides of the “small or large” argument agree that averages say little about one person.
Think it over
Your notes are saved in this browser.
Sources and further reading
- John Gray, Men Are from Mars, Women Are from Venus, HarperCollins, 1992; Eleanor Maccoby and Carol Jacklin, The Psychology of Sex Differences, Stanford University Press, 1974.
- Janet Hyde, “The gender similarities hypothesis,” American Psychologist 60 (2005); Janet Hyde, “Gender similarities and differences,” Annual Review of Psychology 65 (2014); Zell, Krizan, and Teeter, “Evaluating gender similarities and differences using metasynthesis,” American Psychologist 70 (2015); Alice Eagly and William Revelle, “Understanding the magnitude of psychological differences between women and men requires seeing the forest and the trees,” Perspectives on Psychological Science 17 (2022).
- Roger Shepard and Jacqueline Metzler, “Mental rotation of three-dimensional objects,” Science 171 (1971); Voyer, Voyer, and Bryden, Psychological Bulletin 117 (1995); Lauer, Yhang, and Lourenco, Psychological Bulletin 145 (2019); Voyer, Psychonomic Bulletin & Review (2011) on time limits; Mary Hegarty, Psychonomic Bulletin & Review (2018).
- David Uttal and colleagues, “The malleability of spatial skills,” Psychological Bulletin (2013); Jing Feng, Ian Spence, and Jay Pratt, “Playing an action video game reduces gender differences in spatial cognition,” Psychological Science 18 (2007); Sheryl Sorby and Beverly Baartmans, Journal of Engineering Education 89 (2000).
- Moshe Hoffman, Uri Gneezy, and John List, “Nurture affects gender differences in spatial abilities,” PNAS (2011); Richard Lippa, Marcia Collaer, and Michael Peters, Archives of Sexual Behavior (2010); Antoine Coutrot and colleagues, Current Biology 28 (2018); Nazareth, Huang, Voyer, and Newcombe, Psychonomic Bulletin & Review (2019); Wan, Newcombe, and Fitzhugh, Cognitive Processing 14 (2013).
- Jacobs, Gaulin, Sherry, and Hoffman, “Evolution of spatial cognition,” PNAS 87 (1990); Clint and colleagues, Quarterly Review of Biology (2012); Hults and colleagues, Royal Society Open Science (2024); Wood and colleagues on Hadza GPS tracking, Nature Human Behaviour 5 (2021); Voyer, Postma, Brake, and Imperato-McGinley, Psychonomic Bulletin & Review (2007); Jonathan Wai, David Lubinski, and Camilla Benbow, Journal of Educational Psychology 101 (2009).
- Reilly, Neumann, and Andrews, American Psychologist (2019); Voyer and Voyer, Psychological Bulletin (2014); OECD, PISA 2018 Results, Volume II; Rutter and colleagues, JAMA 291 (2004); Shaywitz and colleagues, JAMA 264 (1990); Quinn and Wagner, Journal of Learning Disabilities (2015); Yairi and Ambrose, Journal of Fluency Disorders (2013).
- Hyde and colleagues, “Gender similarities characterize math performance,” Science 321 (2008); Lindberg, Hyde, Petersen, and Linn, Psychological Bulletin (2010); Else-Quest, Hyde, and Linn, Psychological Bulletin 136 (2010); Hyde and Mertz, PNAS 106 (2009); Guiso, Monte, Sapienza, and Zingales, Science 320 (2008); Stoet and Geary, PLoS ONE 8 (2013); Martinot and colleagues, Nature (2025).
- Larry Hedges and Amy Nowell, Science 269 (1995); Machin and Pekkarinen, Science 322 (2008); Wai, Hodges, and Makel, Intelligence 67 (2018); Wierenga and colleagues, Human Brain Mapping 43 (2022); the International Mathematical Union’s list of Fields Medalists.
- Steele and Aronson, Journal of Personality and Social Psychology 69 (1995); Spencer, Steele, and Quinn, Journal of Experimental Social Psychology 35 (1999); Flore and Wicherts, Journal of School Psychology (2015); Shewach, Sackett, and Quint, Journal of Applied Psychology (2019); Flore, Mulder, and Wicherts, Comprehensive Results in Social Psychology (2018).
- Del Giudice, Booth, and Irwing, “The distance between Mars and Venus,” PLoS ONE 7 (2012); Su, Rounds, and Armstrong, “Men and things, women and people,” Psychological Bulletin 135 (2009); Breda and colleagues, PNAS 117 (2020); Stoet and Geary, Psychological Science 29 (2018).
- US National Center for Education Statistics, Digest of Education Statistics, tables 318.30 and 325.35; Cai, Lu, Pan, and Zhong, Review of Economics and Statistics (2019); Chinese Academy of Sciences and Academy of Engineering membership, National Science Review (2021); International Labour Organization modeled estimates of female labor-force participation, via Our World in Data.