I had always taken it for granted that men and women were about equal in intelligence. This is because I grew up reading politically incorrect authors who I trusted to tell me the truth about genetic variation, and they assured me that this was the case. But Richard Lynn’s final book, Sex Differences in Intelligence, argues that men are indeed smarter. In his opening chapter, Lynn quotes a long stream of researchers making the conventional argument, including some impressive names that you wouldn’t expect to be cowed by left-wing dogma. For example,
Richard Herrnstein and Charles Murray: “The consistent story has been that men and women have nearly identical IQs.”
Arthur Jensen and Fred Johnson: “It remains a major unresolved puzzle in differential psychology and neuroscience that the large sex difference in head and brain size is not reflected by the mean IQ difference between males and females, which is virtually nil.”
Since Jensen, Herrnstein, and Murray all agreed with more conventional academic leftists, I thought there’s probably nothing there and the whole thing is not worth looking into.
But Lynn, who passed away in July, said that everyone here is wrong. Rather, he argued that girls are smarter up to about 4 years old, the sexes become about equal between 6 and 15, but then around 16 boys start to pull away, so that there is a 4 to 5 point male advantage by age 21. Since these are things that seem like they would be straightforward to confirm or disprove, I decided to look into his claims, along with the arguments made by those who hold the standard position.
What Does It Mean to Have Higher Intelligence?
The definition of “intelligence” does not come from nature. Scientists have constructed various tests designed to measure what people commonly mean when they use the term. The idea that intelligence exists in a meaningful sense comes from the finding that how well individuals do on all kinds of mental ability examinations are correlated with one another. Psychometricians therefore talk about the g factor, which is a mathematical construct that refers to the underlying ability to think abstractly and solve problems.
IQ tests generally try to get at g, although it can never be measured directly. But some tests get closer than others. In designing them, researchers look at what kinds of questions you can ask an individual that will predict their ability to do well on other kinds of questions that we would commonly think of as related to how smart someone is. For example, rote memory, which means remembering static information, has low g loading. If you know someone is pretty good at repeating numbers, it won’t tell you much about how smart they are overall. Working memory, in contrast, or the ability to transform and manipulate information in one’s head, is highly g loaded. Skill in repeating digits backwards is more g loaded than repeating them in order.
A good IQ test focuses on highly g loaded activities. For example, Raven’s Progressive Matrices are often used to measure intelligence. Those are tests where the respondent is asked to find patterns when given a series of shapes, as in the example below.
Progressive matrices correlate at about 0.8 with g, which is why they are some of the most popular kinds of IQ tests.
The most common intelligence test for adults is the Wechsler Adult Intelligence Scale (for those 6-16 years old, there is the Wechsler Intelligence Scale for Children (WISC)). It traditionally has had two main sections: Verbal and Performance, or non-verbal.
For our purposes, regarding the question of sex differences in intelligence, we can discuss it in terms of whether men or women do better on tests like Raven’s Progressive Matrices and WAIS. We may also investigate whether one group scores higher on the mathematical construct of g, which can be derived from these or other tests. There is a subtle difference between g score and IQ test results, as will be discussed below, which means the answers you get on sex and intelligence may depend on the exact question you’re asking.
Lynn’s Argument
Below, I created a chart from Table 2.1 of Lynn’s book, which shows sex differences in intelligence for 9 different studies for young people age 12-21, covering 12 different populations.
The numbers tend to be all over the place, but whenever you focus on one researcher or team looking at the same population over time, the trend is clearly towards a growing male advantage in all but one case. A similar pattern is found in a review of scores on Progressive Matrices. Lynn collected studies from 46 different populations on every inhabited continent. The chart below shows the average male-female standardized difference across different ages.
We see a male advantage of 0.13d, or 0.13 standard deviations, for 6-year-olds, rough parity until about 15, and then an increasing male advantage from 16-up. The number for 6 year-olds may be dismissed as relying on a small sample size, based on 7 studies, largely driven by results in Oman (d = .29) and the UAE (d = .23). Meanwhile, the number of estimates is larger for every other age group. The 18-19 group is based on 19 studies, and the 20-80-year-old number relies on 17.
The figure below shows the results from 33 studies of adults exploring male-female differences on Progressive Matrices. In 32 of them, males have a higher score. The only exception is a 2012 study from Argentina that had a negligible female advantage of 0.02d.
What can explain the data, other than a true male advantage? One would have to believe that all these studies, conducted over four and a half decades across every inhabited continent, are somehow biased against women. The average difference is 0.3d, and if we exclude the abnormal result from Taiwan, it’s still 0.27d.
Next, we can look at the Wechsler Adult Intelligence Scale (WAIS). Here’s what Lynn finds for tests of adults.
Here we have 40 different studies. Now there isn’t a single one with females ahead, with the average male advantage being 0.26d. Once again, some show a relatively large tilt in favor of males, others a small one, but none go in the opposite direction. Since the sex gap isn’t huge, it’s possible to find studies where the difference is negligible, and therefore jump to the conclusion that men and women have equal cognitive ability. But no difference would imply you would sometimes see women doing better.
Lynn’s studies involved some samples that were not necessarily representative, which means that they might be biased. For example, the 2019 Progressive Matrices study from Poland looks at couples, and focuses primarily on how individuals judge the intelligence of their partners. But if women tend to “date up,” that is, pair with men who are smarter than they are, we can easily see how recruiting couples can bias the data. Yet other studies seem solid, and since they all point in roughly the same direction, we don’t have to worry too much about issues like this, even though, as discussed below, we may ask broader questions about unrepresentative samples.
This gets to the broader question of whether we should trust Lynn. Before publishing this, I talked to some scholars who are seeking to independently replicate his work, and they tell me that although he could be sloppy and made many errors, the way he represented studies generally checks out. I found a lot of typos and mistakes in his tables, and tried to correct them when producing the figures above. In Table 5.5, for example, he claimed to present 42 studies, but 2 appeared to be duplicates, which is why the figure directly above has 40. But hey, the guy was 91 years old when the book was published, and we should all be grateful that he was still contributing to the search for truth at that age.
Lynn concludes that among adults, males are +4.5IQ on Progressive Matrices, +4IQ on the Wechsler test, and +3.45IQ on other tests. In addition, all of this is consistent with larger male brain size, which remains even when controlling for body size, and faster male reaction time. There are also more specific abilities in which the male advantage is larger, particularly spatial rotation, general knowledge, and mechanical skills, and a few areas where women excel, which are also discussed in the book. Nonetheless, the headline here is that men appear slightly smarter.
The Argument Against Male Advantage
This strikes me as quite conclusive. But Lynn is only one source. I asked ChatGPT about sex differences in IQ, and it sent me to the Wikipedia page on the topic and this book chapter by Halpern and Wai, in which the authors write:
Which is the smarter sex – males or females? This may seem like an easy question to answer because it would be a simple task to compare the average scores of large samples of females and males on intelligence tests. However, this obvious strategy will not work because tests of intelligence are carefully written so that there will be no average overall difference between the sexes. (Brody, 1992; Loewen, Rosser, & Katzman, 1988). Questions that favor either sex are either eliminated from the test or matched with questions that favor the other sex to the same degree. [emphasis added] Although some researchers report a small advantage for males on tests that were standardized to show no sex differences (Nyborg, 2005), most studies do not (Colom et al., 2000; Spinath, Spinath, & Plomin, 2008).
This is quite a remarkable admission from the no differences camp! In effect, IQ tests are equal because they were rigged to be equal. I first read this in Lynn’s book, but naturally suspected he might have been exaggerating. But asking ChatGPT to give me the best case for the equality hypothesis led to the exact same fact. See here for more discussion of this.
Setting that aside, Halpern and Wai cite Jensen’s The g Factor for the proposition that even when it comes to g, which hasn’t been rigged in this way, there are no significant differences. Yet Lynn in Chapter 8 of his book discusses several statistical critiques of the correlated vectors method Jensen used to calculate sex differences in g, and points to more recent methods that do show a gap. Jensen’s 1998 book is the main source that Halpern and Wai cite for the idea that there are no sex differences in intelligence.
Interestingly, I found the same reference almost everywhere I looked. Richard Haier is considered one of the world’s foremost experts on intelligence testing. In 2005, he was the lead author on a paper that declared “Comparisons of general intelligence assessed with standard measures like the WAIS show essentially no differences between men and women…” Once again, the citation was Jensen 1998.
What else is out there? On the Wikipedia page on sex differences in intelligence, I found this meta-analysis by Giofrè et al. However, it focuses only on school age children, and here even Lynn agrees there aren’t notable sex differences in intelligence. Looking at the supplementary materials, there are only a handful of studies that include individuals over 16 years old.
One key thing to realize here is that Jensen only needed to resort to complex statistical procedures because he assumed that regular IQ tests did not show a difference, since they had been designed to produce equal results. As he wrote,
But even if there are many subtests in a battery, thereby tending to average out sex biases, a simple summation of sex differences over subtests is contaminated if the method of test construction included selecting test items on any criteria involving sex differences in item responses (as was done in creating the Stanford-Binet and the Wechsler intelligence scales). IQ scores on such tests can hardly be informative about the magnitude of possible sex differences in general ability, at least in principle. The study of sex differences must depend on tests in which item selection was based exclusively on the psychometric criteria used to maximize the reliability, validity, discriminability, and unidimensionality of the subtests.
Since we have many studies from across the world showing a male advantage in Progressive Matrices and WAIS, we arguably don’t really need this analysis, at least if we care about performance rather than g.
Following Jensen, some scholars have similarly used the method of correlated vectors to analyze datasets, finding no sex differences or even a female advantage. The idea behind the method is that IQ scores don’t simply reflect g but g plus domain-specific abilities. If you want to define intelligence as g rather than IQ performance, you can’t just take IQ scores at face value. Scholars argue the theory that men have higher levels of g than women would predict that the more that a part of an IQ test correlates with g, the larger the difference we should find between the sexes.
One problem with going beyond IQ to extract true g for the purpose of making comparisons between groups, however, is that it can lead to absurd results. One paper used the correlated vectors method to argue that more educated people were not any smarter than less educated people, but only gained more knowledge and specific skills through schooling. In investigating sex differences, Colom et al. (2002) use a Spanish dataset relying on a representative sample taking the Wechsler exam. On all 14 different subtests, men had higher scores than women, with the differences ranging from 0.11 to 0.58 standardized units. Nonetheless, through the method of correlated vectors, the authors find no difference in g between the sexes.
This seems wrong. Ashton & Lee (2005) argue that this indicts the whole idea of correlated vectors as a way to get at intelligence differences between groups.
Herein lies the implausibility of the conclusion that the zero correlation between vectors indicates a zero group difference in g. If the groups are equal in g, then the pattern of very large group differences on every subtest can only be explained by extremely large differences in the non-g aspects of every subtest. Even though these differences are no doubt accounted for in considerable part — but not entirely — by a few group factors, it is nevertheless unparsimonious to suggest that although the groups differ not at all in g, they differ massively in every other aspect of mental ability.
They show that using the method of correlated vectors, one can find no difference in g between two groups even in cases where one posits a gap that is quite large.
An additional problem with the correlated vectors method is that it depends quite a bit on the assumptions that go into the model. Keith et al. (2008) find a slight female advantage in g. Yet based on which assumptions they use, their results vary widely.
Looking at the adult figures, the methods used can induce a swing of up to 5 points, and up to 10 points based on how missing data is dealt with. Since we’re looking for a theorized gap of 2-4 IQ points or so between the sexes, methodologies that can give you differences this extreme seem problematic. And although this is not relevant in Keith et al., common vectors is also extremely sensitive to which subtests are used. Again, recall that the tests were designed to find no sex difference. It is also important to note that not all methods to get at true g show no male advantage, as at least some of them do. I’m told that there are better methods available now than that used by Jensen to extract g, but I believe the issue of being sensitive to which tests are chosen remains.
One paper has argued that the apparent male advantage in adult IQ can be in part explained by sampling bias. The 1970 British Cohort Study tested children at age 10, and then the same individuals at 26 and 30. The researchers found that boys and those who had lower IQs at 10 were more likely to drop out of the study and not get tested as adults. The authors argue that attrition or lower male participation rates could explain part of Lynn’s four-point difference, though it would not be enough to explain nearly all of it.
This is similar to an argument I saw someone make on Twitter, which was that men are more likely to be at the margins of society, and in prisons, homeless, etc., so that the dumber ones would not be included in adult studies. I think this is unlikely to have a major influence on the patterns we observe. To see why, imagine that 5% of men are somehow at the margins and unable to participate in IQ studies. Let’s arbitrarily give that crowd an IQ of 80. They would drag down the male mean by about a point if average men took their place in a study instead. If you make them 70 IQs, you bring it down by 1.5, but these seem like extreme assumptions. Nonetheless, they would still leave men with at least a two-point advantage or so. Low intelligence men simply not responding to surveys as much is potentially a bigger problem, and likely biases some studies.
One issue with these kinds of arguments is that they would have to assume that any sampling bias is practically universal. The higher male advantage in IQ is found in countries all across the world. Maybe it is a universal rule that lower IQ men either don’t participate in surveys as much or are relatively more likely to drop out of them, but that has to be proved. Many countries for which we see sex differences in IQ have much lower rates of homelessness and incarceration than the US does, which is just one complication for selection theories.
The Importance of Spatial Ability
Finally, it is worth looking some aspects of spatial ability, as some kinds of questions men do particularly well on in this area are excluded from standard IQ tests, and by extension attempts to derive g from them. One 1995 meta-analysis finds the male advantage on spatial ability to be 0.37d. Here are the results broken down by subcategory and age.1
Intriguingly, we once again see the pattern of the differences being larger the older people get. Lynn notes even more massive differences in mechanical reasoning ability, with one meta-analysis showing a one standard deviation gap. I am particularly bad in this area of life, being unable to fix things when they break, etc. But luckily I get to call myself “smart” because scientists of previous eras rigged the definition of the term in the name of achieving gender parity. Still, women are better in areas like verbal fluency, visual memory and object location, spelling ability, and perceptual and processing speed.
Lynn tells an evolutionary story in which men had more of a need for intelligence as adults to keep and attract mates, provide for them, and work their way up hierarchies. Meanwhile, female sexual value was determined more by physical attractiveness, rather than the ability to outcompete others in areas like politics and the marketplace. This is pretty much evolutionary psychology 101, and although it may sound shocking to many blank slate ideologues it is pretty unremarkable in terms of scientific theory.
There doesn’t appear to be any female equivalent of spatial ability, that is, something that should obviously be more fully included in IQ tests based on common sense or statistical criteria but has been excluded due to concerns over differential outcomes.
Psychometricians and the Equality Hypothesis
I think, in total, the evidence points to the following conclusions:
In adulthood, men score about 2-4 IQ points higher than women. Selection bias might account for around 1-point of that.
This gap may be said to not reflect underlying intelligence differences, but something specific about the tests. Yet that conclusion is based on complex methods that depend on assumptions made by the researcher and have questionable real world application. I’m not an expert in these methods, but I’m skeptical of them.
All of this is despite the exclusion of some important spatial abilities from IQ tests, where the male advantage is particularly large. There are some female favored traits excluded from IQ tests, but as far as I can tell none are as g loaded and therefore theoretically as likely to influence true g, to the extent we are comfortable thinking about the concept in this way.
The debate about true g might matter to psychometricians, but there seems to be no reason it should to normal people using the common sense definition of “intelligence.” Men are better at problem solving and know more things, so can be said to be more intelligent in the collective understanding of the term even if women are just as smart in some sense that doesn’t predict performance in the real world.
The best that can be said for the position of no differences is that if you imagine intelligence as something divorced from real world performance, scholars can use certain methods to show that the two sexes are equal. Someone suggested to me that men and women might be at the same level as far as natural reasoning skills, but men are simply more interested in the non-human or non-personal aspects of the world, with a greater ability to get into things like machines, numbers, and current events. But the higher male performance on tests, plus the understanding that they exclude some measures of spatial ability precisely because men do better on them, makes me comfortable saying that men are smarter.
Jensen must have had a singular influence in turning the no difference position into dogma. If the most famous psychologist to argue that the black-white IQ gap is genetic tells you men and women are equal in intelligence, one isn’t going to suspect that he’s selectively interpreting the data to get a politically correct result, since racial differences are much more taboo. In addition to Jensen, I grew up reading scholars like Herrnstein and Murray, and they were always willing to say many heretical things on sensitive issues, so when I saw them accept the idea of no significant sex differences they had credibility, and this is why I never looked into the topic until now.
Another thing that is going on here is that so many mental assessments are conducted on schoolchildren, and data on representative samples of adults are rarer. This means that scholars are likely to miss a result that only manifests itself later in life, and one shouldn’t expect them to work extra hard to come to conclusions that could have negative career consequences. Moreover, male-female differences are small, unlike racial and class gaps, so it is easy to get many false negatives and therefore conclude that nothing is going on.
I think that the idea of the sexes being equal in intelligence has been soothing for hereditarians. It allows them to grant something to the left, to be able to say that yes, we follow the data where it leads, and you don’t have to be afraid of science, because sometimes it even tells you what you want to hear! Of course, no one can deny disproportionate male genius, but that’s supposedly balanced out with many more male idiots, and this feels like a satisfyingly egalitarian outcome with each sex getting something (it’s probably not too comforting for idiot men that many geniuses share their sex, but whatever, society runs on the tears of professional women, not loser guys). It gets tiring for psychometricians and behavioral geneticists to keep debunking every fantasy most educated people are desperate to hold on to. So they in effect say sorry, but IQ is real and genetic, there are big racial differences, poor people are poor because they’re not as smart as rich people, and there are more male geniuses. But we can give you gender equality! If a study shows a 2 or 3 point difference, it’s always tempting to round down to zero in the abstract. But if Lynn is correct on the 4-point gap, it would indicate that the average man is smarter than over 60% of women, which is pretty important. People often point to greater variation to explain more male geniuses, but if that also goes along with a higher male average, then things are even more hopeless for leftists who want to engineer equal outcomes.
Despite what many hereditarians believe, the idea that men and women are of equal intelligence appears unlikely to be true. That doesn’t hurt my feelings, because I love truth, believe in liberty, think individuals should do whatever they want, and that society should be completely indifferent to disparate outcomes between groups. When arguing with social engineers, however, higher male IQ serves as one more thing to beat them over the head with.
Update, 1/7/24: A previous version of this article stated that all measures of spatial ability were excluded on IQ tests. It has been updated to clarify that this only refers to measures that men do particularly well on, as described in Lynn, Chapter 5.
A certain fraction of the dumbest -- most foolish? -- males behave in ways that get themselves killed. From 'misuse of power tools' to 'disrespecting gang members' -- they end up in an early grave. Are the numbers such that survivorship bias could explain some of the result?
I realised what it is that I most like about Richard Hanania. Obviously he's smart and curious, but in addition, he has absolutely no shame in telling you what he thinks about anything, regardless of liberal pieties etc. I'm not being snarky, it makes his posts super valuable, I don't have to filter for BS. It's refreshing!