Average IQ by gender: what the means, the spread and the tails actually show

0 The short answer

In the largest representative samples of children, males and females have almost the same average IQ, and the differences that recur sit in two other places: the spread of scores, which is wider for boys in most but not all samples, and a handful of specific abilities. In 178,599 English 11 year olds the mean gap in general cognitive ability was 0.01 of a standard deviation, while the variance of boys' scores was 16 percent larger than the variance of girls' scores (a variance ratio of 1.16). Studies of adults disagree about whether a small mean gap exists, and the answer moves with the test battery, the sample and the statistical model. No group statistic describes any one person.

1 What does average IQ by gender actually measure?

An average IQ by gender is the mean score of two groups on one shared scale, and a 2007 consensus statement reports that the most widely used tests were designed so that the overall mean does not favor either group. An IQ is a score compared with a reference group, set to a mean of 100 and a standard deviation of 15, as the pages on how IQ scores are normed and the standard deviation of 15 explain. Asking for the average IQ of women and of men therefore asks whether two groups, scored on one ruler, sit at different points on it. The overview of average IQ, including its section on sex and other groups gives the general answer. This page gives the evidence behind it.

Halpern, Benbow, Geary, Gur, Hyde and Gernsbacher wrote in a 2007 consensus statement that the most commonly used intelligence tests have been designed so that there are no overall sex differences in IQ scores. In their account, items that show an advantage for either sex are discarded during test construction or balanced with items that favor the other. They concluded that standardized intelligence tests cannot settle whether there is a smarter sex. Such a design has a consequence. If the overall mean is close to equal partly by construction, a finding of no difference in the average is not independent proof that women and men are identical on every ability. Irwing made a related point about one test in a 2012 paper. He noted that the WAIS-III manual documents expert review and differential item functioning analyses used to eliminate gender bias, argued that such procedures can also remove items that are not biased, and concluded that there are some grounds to think the test may underestimate sex differences in cognitive abilities. The adult standardization samples in the tables below still show differences in g in both directions.

Reynolds, Hajovsky and Caemmerer set out the mechanism in a 2022 review in Intelligence. Differences in specific abilities often cancel out when scores are combined. If a battery leans heavily on abilities that favor one sex, a composite difference appears, and a composite should not be confused with g, the general factor that the page on the g factor defines. The same review adds that mean differences in g or IQ from comprehensive batteries given to large, representative samples do not consistently favor either sex. Those two points organize everything below: the total score is the place where differences are most likely to cancel, so the spread of scores and the individual domains carry more of the information.

Four terms recur on this page. First, the studies record sex as male or female. This page uses "gender" in its title because that is the word people search, and "sex" for what the studies measured. None of the data sets discussed here measure gender identity, so nothing below speaks to it. Second, the standardized mean difference, written d, is the difference between the male and female means divided by the average within-sex standard deviation, so it is expressed in standard deviation units. Hyde defined it that way in her 2005 review and labeled effects of 0.10 or less as close to zero, 0.11 to 0.35 as small, 0.36 to 0.65 as moderate and 0.66 to 1.00 as large. Multiplying d by 15 converts it to IQ points, which is our arithmetic and not a figure any of the authors report. Third, the variance ratio is the male variance divided by the female variance, so a value above 1.00 means that male scores are more spread out. Fourth, "the tails" means the extreme high and low ends of the distribution, where small differences in spread change the counts most.

Several neighboring questions have their own pages and are only linked here: whether people can estimate their own IQ, including the sex gap in self-estimates, the research on humor and IQ, the page on spatial intelligence and mental rotation, where the male advantage on rotation is covered, and the page on how IQ tests are checked for bias. School achievement gaps, causes, and group comparisons by race or ethnicity are outside this page.

2 What do the whole-population surveys show?

The largest and most representative samples, one from 1932 and several from the 2000s, show nearly equal means and, in most comparisons, more boys in both tails. Deary, Thorpe, Wilson, Starr and Whalley described the cognitive ability distribution of more than 80,000 children, almost everyone born in Scotland in 1921, who were tested at age 11 in 1932, in a 2003 paper in Intelligence. They found no significant mean difference between boys and girls and a highly significant difference between their standard deviations, with boys over-represented at both the low and the high extremes. The authors noted that these were the first findings of this kind from a whole population. The page on average IQ in the UK treats the survey as a national cohort. Here only the comparison by sex matters.

Johnson, Carothers and Deary then analyzed two population-wide Scottish surveys of 11 year olds in a 2008 paper and showed that the distribution of general intelligence is not a clean bell curve. It had less variability in the higher range than in the lower range, and although the mean was 100 the modal score was about 105. Even above the modal level, males showed more variability than females. The authors concluded that, although present at the high end, the difference in variability did not appear to account for sex differences in high-level achievement.

Two national samples of school pupils from the 2000s, one from the UK and one from England, confirm the pattern. Strand, Deary and Smith analyzed Cognitive Abilities Test scores for more than 320,000 UK pupils aged 11 to 12, tested between September 2001 and August 2003, in a 2006 paper. Girls' mean verbal reasoning score was 2.2 standard score points higher than boys', their nonverbal score was 0.3 points higher, and boys' quantitative score was 0.7 points higher. For all three tests the standard deviation was substantially greater among boys, and boys were over-represented at both the top and the bottom extremes, with the exception of the top 10 percent in verbal reasoning. Calvin, Fernandes, Smith, Visscher and Deary then studied 178,599 pupils, 89,545 girls and 89,054 boys, who attended English state schools, represented 93 percent of the UK's local education authorities and completed the third edition of the same test in 2004, in a 2010 paper. The cognitive rows of their Table 3 are reproduced below with the original values.

Read the last column first. The general factor differs by 0.01 of a standard deviation, a figure the authors describe as statistically significant in a sample this large but of negligible size, and the boys' variance is 16 percent larger. The paper reports boys 22 percent more variable on quantitative reasoning, 14 percent on nonverbal reasoning and 12 percent on verbal reasoning. A gap of 0.01 standard deviations is about 0.15 IQ points, which is our arithmetic and far below the one point step of a score report, while a variance ratio of 1.16 changes how many children fall in the extremes, which is the subject of a later section.

Two designs add something the national samples cannot. Arden and Plomin followed British children at ages 2, 3, 4, 7, 9 and 10, with samples from more than 10,000 down to more than 2,000, and reported in a 2006 paper that boys had greater variance on a general ability factor at every age except 2. The section on age returns to their mean differences. Deary, Irwing, Der and Bates compared 1,292 pairs of opposite-sex full siblings from the US National Longitudinal Survey of Youth 1979, using the ASVAB and the AFQT, in a 2007 paper. Males had only a marginal advantage in mean g, less than 7 percent of a standard deviation, but substantially greater variance, and the top 2 percent of AFQT scores held almost twice as many males as females. The page on ASVAB and IQ explains the AFQT itself.

3 The Studies Side by Side

Set side by side in two tables, the national samples of children agree that means are within a fraction of a standard deviation and that boys are more variable, while the adult studies disagree about the sign of the mean and the spread finding is not universal. The first table lists whole-population samples, national samples and cohorts, most of them of children. The second lists adult standardization samples and syntheses of many studies. The rows are not pooled, because the instruments, ages and statistical models differ, and a pooled figure would hide the disagreement that is the main finding for adults.

The adult standardization samples and syntheses follow in the second table.

Three features of the tables matter most. The first three rows of the first table are whole-population or national samples of 11 year olds, and in every one the mean difference is small and the male spread is larger. The adult rows come from standardization samples and from syntheses of many smaller studies, and their means disagree in sign. The spread finding is also not universal: Irwing and Lynn's synthesis of university students found women more variable on one version of the Progressive Matrices, Irwing's WAIS-III analysis found no significant variance difference on g, and the Romanian samples showed no consistent pattern. A fair summary has to carry all three. The studies are linked where they are first discussed, and the Sources section lists every one.

4 Why do studies of adults disagree about the mean?

Adult studies disagree about the sign and size of a small mean difference, and researchers on different sides of the dispute point to test composition, sample selection and the statistical model as places where a few points can be added or removed. Three bodies of work pull in different directions.

The first reports a male advantage that appears in adolescence. Lynn and Irwing meta-analyzed 57 studies of the Standard and Advanced Progressive Matrices in general population samples in a 2004 paper. They found no difference among children aged 6 to 14 and higher male means from age 15 through old age, with an adult male advantage of 0.33 d, which they equate to 5 IQ points. A separate meta-analysis of 15 studies of the Colored Progressive Matrices, a version for younger children, found boys ahead by 0.21 d, or 3.2 IQ points, at ages 5 to 11. Irwing and Lynn then meta-analyzed 22 studies of university students in a 2005 paper and found males ahead by 0.22 to 0.33 d, equivalent to 3.3 and 5.0 IQ points. Their result on variability cut against the usual expectation. In the 8 studies of the Standard Progressive Matrices that reported standard deviations, females showed significantly greater variability (F = 1.20), and in 10 studies of the Advanced Progressive Matrices the difference was not significant.

That university student meta-analysis drew a published criticism. Blinkhorn criticized it in Nature in 2005. In their 2006 reply, Irwing and Lynn described the criticism as directed at their university student result, an average advantage of 4.6 IQ points for men on the Progressive Matrices, a figure that sits between the 3.3 and 5.0 reported in the 2005 abstract, and said that Blinkhorn maintains the adult sex difference is negligible. They answered that his criticism did not place their result in the context of several other studies showing an adult male advantage of around 4 to 6 IQ points. Blinkhorn responded in the same issue. This page reports Blinkhorn's position only as the authors' reply describes it, because his own text could not be read for this page, and it does not summarize his response. The exchange documents that the adult question is disputed between researchers.

In a 2012 paper Irwing analyzed the US standardization sample of the WAIS-III, 2,450 people aged 16 to 89, and found a difference favoring men in g of 0.19 to 0.22 d. The variance ratios from the same analysis were 1.04 for g, which was not significant, 1.39 for working memory, which meant significantly greater male variability, and 0.65 for processing speed, which meant significantly greater female variability.

The second body of work reports parity or a female advantage. Flynn and Rossi-Casé analyzed Raven's Progressive Matrices data of high quality from five advanced nations in a 2011 paper and found that females matched males both below and above the age of 14. They argued that evidence against parity at mature ages rests on suspect samples: at ages 15 to 18 more males than females are school dropouts, and at ages 18 to 24 a female deficit among university students may reflect a gap between IQ and academic achievement. Colom and colleagues applied three methods of testing whether sex differences load on g to cognitive batteries given to 10,475 adults and concluded in a 2000 paper that the sex difference in g was negligible. Keith, Reynolds, Patel and Ridley used the Woodcock-Johnson III standardization sample from ages 6 to 59 and reported in a 2008 paper that the general factor showed fairly consistent, statistically significant differences favoring females in adulthood, a result they called inconsistent with the developmental theory that predicts a male advantage.

The third is a synthesis across test types. Waschl and Burns pooled 98 studies of inductive reasoning, with 96,957 adults, in a 2020 paper and reported an overall standardized difference of 0.13, a range from minus 0.54 to 0.68, and significant variation in size and sometimes direction across types of test. The pattern is of a small and unstable difference, not a fixed one.

Four sources of instability recur in the literature. The first is battery composition. Johnson and Bouchard reported in a 2007 analysis that general intelligence works as an all-purpose problem solving ability that masks sex differences in more specialized abilities: at the level of specific tests, the differences were greater in residual scores, with g removed, than in full scores. A battery with more speed, spatial or numerical content therefore moves a composite in a different direction from a battery with more verbal content.

The second is how individual subtests behave. van der Sluis and colleagues analyzed the Dutch WAIS-III in a twin register sample, not a standardization sample, of 294 women and 228 men aged 18 to 46 in a 2006 paper. They found that the Information subtest proved to be biased in favor of males and that the sex differences in the first-order factors were not attributable to g. Pezzuti, Tommasi, Saggino, Dawe and Lauriola reported in a 2020 paper that in the Italian WAIS-IV standardization the intercepts of Information, Arithmetic and Comprehension were not equal for men and women, so latent mean differences in verbal comprehension and working memory could be biased. In plain terms, women and men with the same underlying ability did not receive the same expected score on those subtests, and a few such subtests can move a composite.

The third is who is sampled. A difference found among university students, school dropouts or volunteers describes those groups, and a later section returns to this. The fourth is the statistical model. Irwing wrote in the 2012 paper that the confused state of the debate is perhaps attributable to selection biases, dependence on method, the quality of tests and what they measure, the need to establish measurement invariance and the fact that g is not normally distributed, and that no study, including his own, is immune from all of these difficulties.

The practical reading follows from the numbers. The published adult estimates on this page run from a female advantage in g in one standardization study, through no difference, to about 3 to 5 IQ points favoring men (0.19 to 0.33 d multiplied by 15, our arithmetic, while the authors of the Progressive Matrices meta-analyses cite around 4 to 6 points in their reply). Those 3 to 5 points are one fifth to one third of a standard deviation, small beside a standard deviation of 15, and across these studies the sign differs with the battery, the sample and the model. The right question to ask of any adult headline is not which side is correct but which instrument, which sample and which model produced the number.

5 Which cognitive abilities differ by gender?

A 2022 review of studies of children and adolescents found the most consistent differences in specific abilities, with females ahead on latent processing speed and males ahead on latent visual processing, and with offsetting differences that mostly cancel in the total. The table sorts the evidence by the Cattell-Horn-Carroll broad ability, using the abbreviations that the CHC model defines, and names the ACIS index that samples each domain. The Reynolds, Hajovsky and Caemmerer review is the main source for the pattern. It summarizes seven studies of latent abilities in children and adolescents, five of them based on the norming data of prominent batteries, and the adult rows come from the standardization studies named in each cell. One labeling point matters: CHC places quantitative reasoning under fluid reasoning, so the table keeps quantitative reasoning apart from quantitative knowledge.

The pattern in the English sample shows why the total hides so much. The general factor differed by 0.01 d, yet with g removed the verbal residual factor favored girls by 0.26 and the quantitative residual factor favored boys by 0.28, while the nonverbal residual factor differed by only 0.02. That is the pattern Johnson and Bouchard described: larger differences appear once the general factor is taken out. Reynolds and colleagues note that a battery weighted toward abilities that favor one sex produces a composite difference. Applied here, a battery weighted toward processing speed would tilt a total toward women and one weighted toward spatial and quantitative tasks toward men, and a battery that includes both would show partly offsetting differences.

Task format also matters. Reynolds and colleagues note that the paper and pencil Coding task requires drawing symbols, whereas newer digital administrations require only tapping, and they suggest that comparing the two could show how much of the female advantage reflects graphomotor speed rather than cognitive processing speed. The page on processing speed as an IQ domain and the page on working memory tests for adults describe those two domains.

Reynolds and colleagues add that the similarities hypothesis is sometimes read to mean that no important sex differences exist, a reading they reject. Both statements stand together with the small means above: similar totals and real differences in particular abilities are compatible.

6 How does a larger male spread change the tails?

A variance ratio near 1.1 or 1.2 is a modest difference in spread that still changes the number of people at the extremes. The variance ratio is the male variance divided by the female variance. In the English sample it was 1.16 for g and 1.12 to 1.22 across the three tests. Hedges and Nowell analyzed six national probability samples in a 1995 paper in Science and found that average sex differences were small and stable over time, that male scores consistently had larger variance, and that males typically outnumbered females substantially among high scorers, except in reading comprehension, perceptual speed and associative memory. Halpern and colleagues, summarizing those data, report variance ratios that differ by 3 to 20 percent. Hyde's 2014 review tabulates the ratios in the Hedges and Nowell data at 1.03 to 1.16 for verbal tests, 1.05 to 1.20 for mathematics and 1.27 for spatial tests.

The pattern differs by domain, and a recent synthesis shows how. Giofrè and colleagues meta-analyzed 571 effect sizes from 75 studies of the Wechsler Intelligence Scales for Children in a 2024 paper. Males were more variable in several domains, including visual processing and crystallized intelligence, females were more variable in processing speed, and short-term verbal memory showed no difference. The differences in variance ratios across domains were linearly related to the size of the mean differences, so the sex with the advantage at the mean was often the more variable sex. Not every sample agrees. Irwing and Lynn found women more variable on one version of the Progressive Matrices among university students, Irwing found no significant variance difference on g in the WAIS-III standardization sample, and Iliescu and colleagues, in the Romanian study listed in the tables above, found no consistent pattern in six measures.

What does a ratio like this do to the counts? The table applies the normal curve, with women at a mean of 100 and a standard deviation of 15, which is our arithmetic and not a figure from any study. The cutoffs of 130 and 70 are two standard deviations from the mean, the points that the page on how rare a given IQ is works with, and each tail holds about 228 of every 10,000 women.

Three readings follow. With equal means and a ratio of 1.16, about 58 percent of the people above 130 and about 58 percent of those below 70 would be men, assuming equal numbers of women and men, against 50 percent when the spreads are equal. The mean and the spread must be read together, because a mean difference moves the two tails in opposite directions: a 3.3 point male advantage with no difference in spread produces 1.65 men per woman above 130 and 0.58 below 70. And the ratio grows with distance from the mean. At 145 and at 55, three standard deviations out, the same variance ratio of 1.16 gives 1.98 men per woman, although fewer than 14 in every 10,000 women lie that far out, so the extreme tails involve very few people in absolute terms.

The last row of the table shows that the arithmetic runs in both directions. It applies the ratio in favor of females, 1.20, that Irwing and Lynn reported for students on the Standard Progressive Matrices, and it gives 0.62 men per woman in each tail, or about 38 percent of each tail, which reverses the ratios above.

The method reproduces published results. Hyde's 2014 review reports a calculation by Hedges and Friedman for d = 0.01 and a variance ratio of 1.05 that gives 111 males for every 100 females above the 95th percentile and 133 above the 99.9th percentile, and the same normal model returns those figures when the cutoffs are set at the pooled 95th and 99.9th percentiles. Two limits apply. Real score distributions are not exactly normal: Johnson, Carothers and Deary found less variability in the higher range of the Scottish distribution than in the lower range, so the table is an illustration and not a prediction for any test. And a cutoff on a score is a statistical threshold, not a diagnosis or a label for a person.

Observed tail ratios among selected students are not constants either. Wai, Cacchio, Putallaz and Makel reported male to female ability ratios in the right tail (the top 5 percent) from more than 1.6 million seventh graders who took the SAT or ACT across 30 years, from 1981 to 2010, in a 2010 paper. Male to female ratios in mathematical reasoning were substantially lower than 30 years earlier but stable over the last 20 years and still favored males, while more females scored in the extreme right tail of verbal reasoning and writing. On what spread can explain, Johnson, Carothers and Deary concluded that the sex difference in variability did not appear to account for sex differences in high-level achievement, and Hyde argued in her 2014 review that variance ratios of this size cannot explain why only 18 percent of undergraduate engineering degrees went to women at the time she wrote.

7 How much do the two distributions overlap?

For total-score differences of the sizes reported here, the two distributions overlap heavily, so a group mean says little about which of two people scores higher. Two simple measures show it. One is the chance that a randomly chosen member of the higher-scoring group outscores a randomly chosen member of the lower-scoring group. The other is the share of the lower-scoring group that scores above the higher-scoring group's average. Both follow from d when the two curves are normal with equal spreads, which is our arithmetic, and the table applies them to figures reported earlier on this page.

Read the rows against intuition. At d = 0.22 the member of the lower-scoring group wins about 44 of every 100 pairings, and 41 percent of that group scores above the other group's average. Only at the two latent factor figures, d = 0.97 for visual processing and d = 1.30 for processing speed in one standardization sample, does one group outscore the other in 75 to 82 of every 100 pairings. The page on IQ score versus percentile explains how a percentile places one score within a reference group, which is the question a personal score answers.

The same pattern holds beyond IQ. Zell, Krizan and Teeter pooled 106 meta-analyses and 386 effects across psychological domains in a 2015 paper and found an average absolute difference of d = 0.21, with 46 percent of effects small and 39 percent very small. Hyde's 2005 review had found 78 percent of 124 effect sizes in the close-to-zero or small range. The general factor figures on this page, 0.01 to 0.33, run from far below that average to somewhat above it, while the largest latent factor figures, 0.97 for visual processing and 1.30 for processing speed, sit well above it. Reynolds and colleagues add a caution in the other direction: an effect that looks small for an individual can still matter at the level of a society, or in the tails, where small differences accumulate. Both statements are true at once, and which one applies depends on whether the question is about one person or about counts in a population.

8 Do the findings change with age?

They do, because the mean gap, the spread and the specific abilities do not all appear at the same age, and the sample that is available also changes with age. Arden and Plomin's British children show the sequence most clearly. Girls had a mean advantage on the general ability factor from age 2 to age 7, and at ages 2, 3 and 4 girls were over-represented in the high tail and boys in the low tail. By age 10 the boys had a higher mean, greater variance and an over-representation in the high tail. In the Scottish and English national samples of 11 year olds the mean differences were close to zero, and Calvin and colleagues described age 11 as an age before substantive gender-related selection bias occurred.

For adults, the dispute described above turns on a developmental theory in which a male advantage appears from age 15. Lynn and Irwing described their Progressive Matrices results as supporting it, and Irwing described his WAIS-III results the same way. Keith and colleagues found their Woodcock-Johnson III results inconsistent with it, and Reynolds and colleagues describe the evidence for the model as mixed, with age as a potential moderator. Age also changes who is in the sample, so a comparison of university students aged 20 with a general population sample compares two different selections, whatever the instrument. Hyde's 2005 abstract states the general point: gender differences can vary substantially in magnitude at different ages and depend on the context in which measurement occurs.

Two other pages on this site approach age from different directions: the page on how age norms work and the page on how IQ changes with age, which separates the stability of rank from changes in raw performance. This page makes no claim about age beyond what the studies above report.

9 Why can an online site not report an average IQ by gender?

A gender average computed from the people who happen to take an online test is not a population average, because who takes the test and who finishes it can each differ by sex, and each moves the means. Dykiert, Gale and Deary showed how sample restriction alone can create an apparent gap. They used the 1970 British Cohort Study, a nationally representative sample, and measured IQ at age 10, when the male advantage was small (d = 0.081, favoring boys). They then kept only the participants who returned at later waves. Returners were more often female and had higher IQ scores at age 10. At age 30, with 28 percent attrition, the male advantage grew by 15 percent, and at age 26, with 43 percent attrition, it grew by 48 percent. Our arithmetic turns the 48 percent increase into a gap of about 1.2 points becoming about 1.8 points (15 points times the d value, before and after). The authors concluded that part of the male advantage reported by some researchers might come from greater male variance combined with sample restriction, a possibility they said still needed to be tested with a proper battery.

That cohort at least recorded who dropped out. A website usually cannot see who never began a test or who stopped halfway, and it cannot see how the people who did finish differ from the population. Reynolds and colleagues state the general rule: selected measures and samples provide selected results. Hyde reports in her 2005 review that Hedges and Nowell argued the usual method of meta-analysis, which aggregates many small convenience samples, should be augmented or replaced by large probability samples where possible, as in ability testing, which is why the national surveys come first on this page. Flynn and Rossi-Casé made a similar point about adult samples, and the same logic applies to any group that is selected before it is tested, such as recruits, which the page on average IQ by military branch examines. This page uses no data collected by ACIS or by any other test website. The page on whether online IQ tests are accurate covers the wider question of how far an unsupervised online score can be trusted.

A reader who meets a headline about men, women and IQ can test it with six questions.

1. Who was tested and how were they chosen? A whole-population survey, a norming sample and a volunteer sample answer different questions.
2. Which instrument was used, and what mix of abilities does it sample? A total moves with its ingredients, as the domain table showed.
3. How big was the sample, and was it analyzed before or after exclusions and dropouts?
4. Is the gap stated as d, in points or only as higher or lower? Multiply d by 15 to read it in IQ points.
5. Is the spread reported, as a variance ratio or a standard deviation, and what happens in the tails?
6. Is the claim about a total score or about a specific ability, and has the test been checked for functioning differently by sex?

A claim that cannot answer those questions is a number without a measurement behind it.

10 What does a personal report show that a group average cannot?

A personal report places one person against a reference group on separate abilities, which is the level at which the research above finds real differences and the level at which a group average has nothing to say. We sell a paid, self-administered online assessment, so this section is a commercial disclosure and the rest of the page should be checked against it. ACIS has 20 subtests in six domains of the Cattell-Horn-Carroll model, reported as a Full Scale IQ and six primary indices: verbal comprehension (VCI), fluid reasoning (FRI), quantitative reasoning (QRI), visual spatial ability (VSI), working memory (WMI) and processing speed (PSI). Indices and the Full Scale IQ use the standard scale with a mean of 100 and a standard deviation of 15, subtest scores run from 1 to 19 with a mean of 10 and a standard deviation of 3, and the report gives percentiles and a 95 percent confidence interval. The adult norms cover ages 16 to 90.

The six indices line up with the domains in the table above. Processing speed is sampled by the Symbol Search subtest and the Coding subtest, visual spatial ability by Visual Puzzles, Layer Rotation and Spatial Comprehension, and working memory by Digit Span, Alphanumeric Sequencing and Visual Sequence. Because the research shows that a total moves with the mix of subtests, the six indices are the part of a report to read first. Only the Full Scale form covers all six domains. The technical manual documents the forms and should be read before relying on any score, and the page on the Full Scale IQ test explains what the longest form covers.

Prices and terms were read on October 6, 2026 and can change. All forms are a one time payment with no subscription, and breaks are allowed. Quick costs 15 dollars, covers 6 subtests in 3 domains and takes about 45 minutes. Optimized costs 30 dollars, covers 13 subtests in 5 domains and takes about 110 minutes. Full Scale costs 50 dollars, covers all 20 subtests in six domains and takes about 175 minutes. A free trial needs no card, a 5 day quality guarantee applies, and 30 days are allowed to complete a form.

The limits are as plain as the features. The assessment is online and unsupervised, it is not a clinical or diagnostic instrument, it is not meant for hiring, school accommodations or admission to high IQ societies, and it is offered in English only. A report makes no claim that any group difference in this literature applies to the person who took it. It shows where that person stands on each ability, which is a different question from the one a gender average answers.

11 What the Evidence Supports, Stated Narrowly

The evidence supports a short list of statements about groups, each tied to the samples and instruments that produced it, and nothing that ranks one sex above the other or predicts an individual. The statements below are limited to the studies reviewed on this page.

1. In the largest national samples of children, the mean difference in general cognitive ability is close to zero. It was 0.01 d among 178,599 English 11 year olds and not significant among more than 80,000 Scottish children, and a US design of 1,292 sibling pairs found a male advantage under 7 percent of a standard deviation.
2. Male scores were more variable in most of those samples. The English variance ratios ran from 1.12 to 1.22, and published ratios by domain run from 1.03 to 1.27. The size depends on the domain, one synthesis found females more variable in processing speed, one meta-analysis of university students found women more variable on one test, and some samples show no clear pattern.
3. Mean and spread must be read together. In our arithmetic, equal means with a variance ratio of 1.16 give about 58 percent men in each tail, and a mean difference pushes the two tails in opposite directions.
4. Adult means are unsettled. Published estimates run from a female advantage in general ability in one standardization study to about 3 to 5 IQ points favoring men, depending on the battery, the sample and the model, and the Progressive Matrices meta-analyses behind the male estimates drew a published criticism and a reply, so the question is a live dispute between researchers.
5. The most replicated differences are specific. Females lead on latent processing speed and males lead on latent visual processing in the studies reviewed in 2022, while fluid reasoning and working memory show negligible or inconsistent differences and comprehension-knowledge is mixed.
6. A volunteer or online sample cannot answer the question. In one cohort, restricting the sample to the participants who returned raised the apparent male advantage by 15 to 48 percent.
7. The causes are outside this page. Halpern and colleagues wrote that the reasons why males are often more variable remain elusive, and nothing here adjudicates between biological and social explanations.

The evidence does not support the claims that either sex is smarter, that a larger spread means anything about a particular man or woman, that a stated number applies to a given person, or that sex is a sound basis for selecting people for anything.

12 What This Does Not Say About You

None of these findings predicts your score or anyone else's from sex alone, and none supports treating a person differently because of it. The within-sex standard deviation is 15 points, while the largest adult mean differences in general ability in the sources are 3 to 5 points. At the 5 point figure, a randomly chosen member of the lower-scoring group still outscores a randomly chosen member of the other group in about 41 of every 100 pairings. At every figure on this page the two groups share most of their range, and the overlap table shows it. A group mean describes a distribution, never a member of it.

Several lessons follow for a reader. If you want to know where you stand, measure it with an instrument built for that purpose and read the result against its reference group, with its interval, as the page on reliability and validity explains. Do not use a gender average as a starting guess for your own score or as a limit on what you can do. If a score surprises you, remember that every individual score is a band and not a point, and that a few points of difference between two groups is small beside the uncertainty around a single result. And if you are deciding whether a test is fair, ask the question the research asks: whether particular subtests function differently by sex, not whether the totals differ.

Interpretation standards say the same. The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014) define validity as the degree to which evidence and theory support the interpretations of test scores for proposed uses of tests. A difference between group means is not evidence for any use of sex to interpret one person's score. The Standards also state that group differences in testing outcomes do not in themselves indicate that a testing application is biased or unfair, and they note that most testing professionals agree such differences should trigger heightened scrutiny for possible sources of bias, which is why the subtest-level findings on Information and Arithmetic matter. The American Psychological Association's Ethical Principles of Psychologists and Code of Conduct point the same way. Standard 9.02 asks psychologists to use instruments whose validity and reliability have been established for the population tested, Standard 9.05 asks test developers to use appropriate psychometric procedures for the reduction or elimination of bias, Standard 9.06 asks psychologists to consider the purpose of an assessment and the characteristics of the person assessed and to state the limits of their interpretations, and Standard 3.01 prohibits unfair discrimination based on gender.

13 Sources Behind This Page

Every figure on this page is traceable to one of the following, and each is linked at the point where it is used. Conversions of d to IQ points, the overlap shares and the counts in the tail table are our arithmetic on published figures and are labeled as such. The full text was read for Calvin and colleagues (2010), Reynolds and colleagues (2022), Irwing (2012), Hyde (2005 and 2014), Halpern and colleagues (2007) and Jensen and Reynolds (1983). For the three Nature items on the criticism of the university student meta-analysis only the PubMed records were read, so Blinkhorn's two items are cited as records without their text. For Iliescu and colleagues (2016) the abstract content and the authors' conclusion about variance were read only in secondary sources. For the other papers only the abstract could be read, and the page states no figure from those papers that the abstract does not give. All were checked on October 6, 2026.

• Deary I J, Thorpe G, Wilson V, Starr J M and Whalley L J. Population sex differences in IQ at age 11: the Scottish mental survey 1932. Intelligence, 2003, volume 31, issue 6, pages 533 to 542. Johnson W, Carothers A and Deary I J. Sex differences in variability in general intelligence: A new look at the old question. Perspectives on Psychological Science, 2008, volume 3, issue 6, pages 518 to 531.
• Strand S, Deary I J and Smith P. Sex differences in Cognitive Abilities Test scores: A UK national picture. British Journal of Educational Psychology, 2006, volume 76, issue 3, pages 463 to 480. Calvin C M, Fernandes C, Smith P, Visscher P M and Deary I J. Sex, intelligence and educational achievement in a national cohort of over 175,000 11-year-old schoolchildren in England. Intelligence, 2010, volume 38, issue 4, pages 424 to 432.
• Arden R and Plomin R. Sex differences in variance of intelligence across childhood. Personality and Individual Differences, 2006, volume 41, issue 1, pages 39 to 48. Deary I J, Irwing P, Der G and Bates T. Brother-sister differences in the g factor in intelligence: Analysis of full, opposite-sex siblings from the NLSY1979. Intelligence, 2007, volume 35, issue 5, pages 451 to 456.
• Dykiert D, Gale C and Deary I J. Are apparent sex differences in mean IQ scores created in part by sample restriction and increased male variance? Intelligence, 2009, volume 37, issue 1, pages 42 to 47. Hedges L V and Nowell A. Sex differences in mental test scores, variability, and numbers of high-scoring individuals. Science, 1995, volume 269, issue 5220, pages 41 to 45.
• Halpern D F, Benbow C P, Geary D C, Gur R C, Hyde J S and Gernsbacher M A. The science of sex differences in science and mathematics. Psychological Science in the Public Interest, 2007, volume 8, issue 1, pages 1 to 51. Wai J, Cacchio M, Putallaz M and Makel M C. Sex differences in the right tail of cognitive abilities: A 30 year examination. Intelligence, 2010, volume 38, issue 4, pages 412 to 423.
• Hyde J S. The gender similarities hypothesis. American Psychologist, 2005, volume 60, issue 6, pages 581 to 592. Hyde J S. Gender similarities and differences. Annual Review of Psychology, 2014, volume 65, pages 373 to 398.
• Zell E, Krizan Z and Teeter S R. Evaluating gender similarities and differences using metasynthesis. American Psychologist, 2015, volume 70, issue 1, pages 10 to 20. Reynolds M R, Hajovsky D B and Caemmerer J M. The sexes do not differ in general intelligence, but they do in some specifics. Intelligence, 2022, volume 92, article 101651.
• Johnson W and Bouchard T J. Sex differences in mental abilities: g masks the dimensions on which they lie. Intelligence, 2007, volume 35, issue 1, pages 23 to 39. Keith T Z, Reynolds M R, Patel P G and Ridley K P. Sex differences in latent cognitive abilities ages 6 to 59: Evidence from the Woodcock-Johnson III tests of cognitive abilities. Intelligence, 2008, volume 36, issue 6, pages 502 to 525.
• Irwing P. Sex differences in g: An analysis of the US standardization sample of the WAIS-III. Personality and Individual Differences, 2012, volume 53, issue 2, pages 126 to 131. Lynn R and Irwing P. Sex differences on the progressive matrices: A meta-analysis. Intelligence, 2004, volume 32, issue 5, pages 481 to 498.
• Irwing P and Lynn R. Sex differences in means and variability on the progressive matrices in university students: A meta-analysis. British Journal of Psychology, 2005, volume 96, issue 4, pages 505 to 524. Blinkhorn S. Intelligence: a gender bender. Nature, 2005, volume 438, issue 7064, pages 31 to 32. Irwing P and Lynn R. Is there a sex difference in IQ scores? Nature, 2006, volume 442, issue 7098, page E1. Blinkhorn S. Is there a sex difference in IQ scores? (Reply). Nature, 2006, volume 442, issue 7098, pages E1 to E2. Flynn J R and Rossi-Casé L. Modern women match men on Raven's Progressive Matrices. Personality and Individual Differences, 2011, volume 50, issue 6, pages 799 to 803.
• Colom R, Juan-Espinosa M, Abad F and García L F. Negligible sex differences in general intelligence. Intelligence, 2000, volume 28, issue 1, pages 57 to 68. Waschl N and Burns N. Sex differences in inductive reasoning: A research synthesis using meta-analytic techniques. Personality and Individual Differences, 2020, volume 164, article 109959.
• Iliescu D, Ilie A, Ispas D, Dobrean A and Clinciu A. Sex differences in intelligence: A multi-measure approach using nationally representative samples from Romania. Intelligence, 2016, volume 58, pages 54 to 61. Giofrè D, Toffalini E, Perugini A, Esposito L, Amoretti G and Geary D C. Sex differences in cognition: A meta-analysis of variance ratios in the Wechsler Intelligence Scales for Children. Personality and Individual Differences, 2024, volume 229, article 112776.
• van der Sluis S, Posthuma D, Dolan C, de Geus E, Colom R and Boomsma D. Sex differences on the Dutch WAIS-III. Intelligence, 2006, volume 34, issue 3, pages 273 to 289. Pezzuti L, Tommasi M, Saggino A, Dawe J and Lauriola M. Gender differences and measurement bias in the assessment of adult intelligence: Evidence from the Italian WAIS-IV and WAIS-R standardizations. Intelligence, 2020, volume 79, article 101436.
• Jensen A R and Reynolds C R. Sex differences on the WISC-R. Personality and Individual Differences, 1983, volume 4, issue 2, pages 223 to 226.
• American Educational Research Association, American Psychological Association and National Council on Measurement in Education. Standards for Educational and Psychological Testing, 2014 edition, chapters 1 and 3, read in the published edition PDF on October 6, 2026.
• American Psychological Association. Ethical Principles of Psychologists and Code of Conduct, adopted August 21, 2002, effective June 1, 2003, with amendments effective January 1, 2017, Standards 3.01, 9.02, 9.05 and 9.06, read in the APA pamphlet PDF on October 6, 2026.

14 Frequently Asked Questions