Adjusted for…
Height: 1.28
Total volume: 1.35
Left accumbens
Raw: 1.23
Adjusted for…
Height: 1.23
Total volume: 1.22
Right accumbens
Raw: 1.20
Adjusted for…
Height: 1.20
Total volume: 1.20
Left amygdala
Raw: 1.35
Adjusted for…
Height: 1.35
Total volume: 1.37
Right amygdala
Raw: 1.27
Adjusted for…
Height: 1.28
Total volume: 1.27
Left caudate
Raw: 1.18
Adjusted for…
Height: 1.18
Total volume: 1.16
Right caudate
Raw: 1.19
Adjusted for…
Height: 1.19
Total volume: 1.20
Left pallidum
Raw: 1.14
Adjusted for…
Height: 1.10
Total volume: 1.09
Right pallidum
Raw: 1.19
Adjusted for…
Height: 1.16
Total volume: 1.15
Left putamen
Raw: 1.20
Adjusted for…
Height: 1.22
Total volume: 1.20
Right putamen
Raw: 1.23
Adjusted for…
Height: 1.23
Total volume: 1.23
Left thalamus
Raw: 1.22
Adjusted for…
Height: 1.18
Total volume: 1.33
Right thalamus
Raw: 1.20
Adjusted for…
Height: 1.18
Total volume: 1.30
Source: All ratios are implicitly compared to 1.0. For example, the entry of 1.22 for Total brain volume represents a male-to-female ratio of 1.22. Adapted from Ritchie, Cox, Shen et al. (2018): Tables 1 and S1. For all variance ratios, p < .001 (all but the variance ratio for the left hippocampus adjusted for height had p < 10–4). All variables are adjusted for age and ethnicity. Ratios greater than 1.0 indicate greater male variance.
*Adjustments for gray and white matter for total brain volume were not performed because of collinearity.
The remarkable aspect of the table is how little the VRs are affected by controlling for height or total brain volume, unlike the story for regional brain volumes presented earlier. The mean of the 28 differences was a trivial 0.0014. The largest of all 28 of the differences between the adjusted ratios and the raw ratio was just 0.04.
The table shows the results for only 14 subcortical regions. The full analysis in the Ritchie study included 68 regions, with measures not just of volume but for surface area and cortical thickness. Males had greater variance in all 68 regions, and those differences were significant for 64 out of the 68. The surface area variance ratio was significant in 66 of the 68 regions. The exception was the measure of cortical thickness, where women had a thicker cortex than men across almost the entire brain. The variance ratios for cortical thickness were nonsignificant with a single exception.38
Greater Male Variance in Other Biological Traits
Greater male variance is found in a wide variety of physiological traits. I won’t try to list them all, but these examples will give you a sense of the prevalence of greater male variance.
The U.S. National Health and Nutrition Examination Survey (NHANES) for 2015–16 included 10 basic physiological measures. The variance results for adults ages 20–39 (936 males, 1,017 females) are shown below: 39
Weight
Effect size (d): –0.62
Variance ratio: 1.15
Height
Effect size (d): –1.91
Variance ratio: 1.25
Body mass index (BMI)
Effect size (d): –0.02
Variance ratio: 0.77
Upper leg length
Effect size (d): –1.49
Variance ratio: 1.19
Upper arm length
Effect size (d): –1.45
Variance ratio: 1.09
Waist circumference
Effect size (d): –0.24
Variance ratio: 0.99
Saggital abdominal diameter
Effect size (d): –0.36
Variance ratio: 1.01
Pulse
Effect size (d): +0.35
Variance ratio: 1.08
Systolic blood pressure
Effect size (d): –0.68
Variance ratio: 1.23
Diastolic blood pressure
Effect size (d): –0.32
Variance ratio: 1.33
Of the 10 parameters in the NHANES data, seven show greater male variability, two show effectively equal male and female variability, and only one, BMI, shows clearly greater female variability.40 The mean variance ratio was 1.10.[41]
In 1988, the U.S. Army conducted an anthropometric survey of its uniformed personnel, taking 132 measurements of length, breadth, and circumference of various portions of the body plus a measure of overall weight using a sample of 1,774 men and 2,208 women balanced to reflect the racial/ethnic and age groups in the active-service Army.42 The measures ranged from the basic (height, weight) to the arcane (bitragion coronal arc, bispinous breadth). Of the 132, 3 percent had variance ratios of 1.0, 14 percent had variance ratios less than 1.0 (women had greater variance), and 83 percent had ratios greater than 1.0 (men had greater variance). The average VR for the 132 anthropometric measures was 1.12.
A team of British scholars analyzed grip strength across the lifespan, combining 12 British studies with 49,964 subjects. From the ages of 5 to 9, girls had slightly higher grip strength than boys (d = +.10) and slightly greater variability (VR = 0.88). Thereafter, males had higher grip strength and greater variability. By the age of 20 and for each 5-year age group through the oldest group (90–94), both the effect sizes and the VRs were at least 1.8.43
MRI was used by Canadian and U.S. scholars to measure skeletal muscle mass in 200 women and 268 men ages 18–88 and of varied adiposity.44 The results are shown below.
Total skeletal muscle (SM)
d: –2.60
VR: 1.95
SM relative to BMI
d: –1.47
VR: 0.86
Lower body SM
d: –2.10
VR: 1.54
Upper body SM
d: –2.55
VR: 2.09
The effect sizes not adjusted for BMI are so large that there was virtually no overlap between the males and females. The variance ratios were large as well. When adjusted for BMI, the effect size remained large, but women were more variable than men.[45]
This is just a sampling. The generalization seems secure: In childhood, the sex differences in variability are scattered and small. Male variability increases after puberty. As adults, greater male variability extends from the regions of the brain throughout the body—not on every parameter, but on a large majority of them.
Greater Male Variance in Sexually Selected Attributes
In 1989, psychologist David Buss conducted a study of sex differences in human mate preferences across 37 cultures worldwide. Using parental investment and sexual selection theory, he predicted the results for five target attributes: In choosing mates, males would value youth, physical attractiveness, and chastity more than females do; females would value good providers (operationally defined as “good financial prospects” and ambition/industriousness) more than men do. All of the five predictions were empirically supported, though to different degrees.46
Fourteen years later, psychologists John Archer and Mani Mehdikhani returned to Buss’s data with an additional hypothesis, based on Trivers’s theory of parental investment but also on the prevalence of men who make high parental investment. Some men behave as if they are pursuing the primordial male repro
ductive strategy of impregnating as many females as possible; others behave as if they are pursuing a strategy of attracting women through the promise of being good fathers. “If there are alternative reproductive strategies among men but not among women, we would predict greater variability among males than among females in psychological characteristics associated with sexual selection,” the authors hypothesized, and used Buss’s database to test the hypothesis. They also used meta-analyses of sex differences in physical aggression, another trait predicted to be sexually selected.47
The weighted means for effect size and variance ratio are shown below:48
Physical aggression
Effect size (d): –0.70
VR: 2.04
Good looks
Effect size (d): –0.59
VR: 0.95
Chastity
Effect size (d): –0.30
VR: 1.82
Ambition and industriousness
Effect size (d): 0.50
VR: 1.91
Good financial prospects
Effect size (d): 0.76
VR: 1.41
Age difference
Effect size (d): 2.00
VR: 2.09
“Good looks” was the exception, with a VR slightly under 1.0. Males showed substantially more variability than females on the other five, with ratios ranging from 1.41 to 2.09, consistent with their hypothesis.[49]
Greater Male Variance in Personality
The database used for the McCrae cross-national study in personality described in chapter 2, based on observations rather than self-reports, has also been analyzed for variability. In the United States, males had greater variability than females on all five factors of the Five Factor Model, with VRs 1.05 for neuroticism, 1.21 for extraversion, 1.14 for openness, 1.08 for agreeableness, and 1.20 for conscientiousness. The mean for all five was 1.13.50
This pattern applied to Anglophone and European countries generally. Excluding the United States, the mean VR for 24 other Anglophone and European countries was 1.08. Those 24 countries did not show a higher male variance on neuroticism (mean VR = 0.97). The mean VRs for the other factors were 1.08 for extraversion, 1.13 for openness, 1.11 for agreeableness, and 1.13 for conscientiousness.51
The results also differed markedly by personality factor. Males had greater variability in conscientiousness (84 percent of the countries), openness (75 percent), and agreeableness (69 percent), but a majority of countries had greater female variability in neuroticism (59 percent) and extraversion (53 percent). Overall, greater male variability in personality is not nearly as consistent as for the other topics I cover.
Greater Male Variance in Mental Tests
In the early 1990s, testing experts Larry Hedges and Amy Nowell set out to conduct a comprehensive study of sex differences in mental test scores, variability, and high-scoring individuals in the United States for all of the large and nationally representative datasets over the period from 1960 to 1992.52 Their article, published in Science in 1995, was the definitive statement of where things stood when they wrote. They presented variability ratios for 37 different mental test measures. Male variance was higher in 35 of the 37.53 The exceptions were a test of word memory and one of coding speed—and, the authors noted, “In both cases, measures of the same constructs in other surveys showed greater male variability.”54 Overall, Hedges and Nowell concluded, “These data demonstrate that in U.S. populations, the test scores of males are indeed more variable than those of females, at least for the abilities measured during the 32-year period covered by the six national surveys. Moreover, there is little indication that variance ratios are changing over time.”55
The National Assessment of Educational Progress (NAEP) now provides an even longer trendline—44 years for reading and 37 years for math. From the first test in 1971 to the most recent one in 2015, 12th-grade boys have had higher variance than girls on all 13 tests for which I have data, with ratios ranging from 1.07 to 1.20, with a mean of 1.12. The trendline is absolutely flat. And yet in all of those tests, girls outscored boys at the mean.56 On the nine math tests for which I have data, all showed greater male variance, with ratios also ranging from 1.07 to 1.20 and a mean of 1.13. The SAT shows even more consistent but quite small variance ratios. From 1996 to 2016, the variance ratio on the math test was never smaller than 1.03 and never larger than 1.09. For the reading test, variance was almost but not quite equal, ranging from 1.02 to 1.05. There has been no trend in either test.
VARIABILITY IN MENTAL TESTS DURING CHILDHOOD
I concentrate on test scores in adolescents and older because so many sex differences, physiological and mental, change during adolescence and persist thereafter. But greater male variability on mental tests emerges early as well. Psychologists Rosalind Arden and Robert Plomin explored variance in g in large British samples at ages 2, 3, 4, 7, 9, and 10. They found significantly greater variance among boys at every age except 2.57
Internationally, greater male variability in test scores was formerly thought to be inconsistent. A 1994 review by psychologist Alan Feingold found a median VR of 0.95 in tests of vocabulary (six countries), 1.01 for reading comprehension (three countries), 1.09 for math (20 countries), and 1.14 for spatial ability (nine countries).58 The variance ratios were often 1.0 or less, indicating greater female variance. But Feingold had to work with a heterogeneous set of tests from a comparatively small number of nations. Since the PISA tests began in 2000, the picture has come into clearer focus.
Thirty-nine countries plus Macao and Hong Kong participated in the 2003 PISA administration.59 The mean scores for reading and math showed the pattern you saw in chapter 3: a small male advantage in math (mean d = –0.11), a greater female advantage in verbal (d = +0.36). But when it came to variance ratios, 38 of the 41 countries showed higher male variance in the math test, with a mean VR of 1.16. The difference in girls’ and boys’ ratios was statistically significant (p < .05) for 37 of the 41 countries. For reading, despite the universal female advantage in mean scores, 40 of the 41 showed higher male variance (Indonesia had a VR of exactly 1.0), with a mean VR of 1.19. The difference in girls’ and boys’ ratios was statistically significant (p < .05) for 35 of the 41 countries.60
The most recent PISA results have not changed much since 2003, except that more countries are participating. For the 2015 math and science tests, males had higher variability in 65 out of 67 countries; for the reading, in 63 out of 67. The table below shows the average variance ratios grouped by geographic region.61
Anglosphere
Mean variance ratio
Reading: 1.15
Math: 1.14
Science: 1.18
East Asia
Mean variance ratio
Reading: 1.12
Math: 1.18
Science: 1.18
Eastern Europe
Mean variance ratio
Reading: 1.12
Math: 1.12
Science: 1.14
Latin Am./Caribbean
Mean variance ratio
Reading: 1.09
Math: 1.12
Science: 1.13
Mideast/No. Africa
Mean variance ratio
Reading: 1.17
Math: 1.16
Science: 1.14
Scandinavia
Mean variance ratio
Reading: 1.19
Math: 1.14
Science: 1.18
SE Asia
Mean variance ratio
Reading: 1.12
Math: 1.10
Science: 1.07
Western Europe
Mean variance ratio
Reading: 1.12
Math: 1.14
Science: 1.15
The table shows remarkably consistent greater male variability in regions that are culturally, socioeconomically, and educationally diverse. Furthermore, this uniformity of greater male variability existed across three tests that showed quite different effect sizes by sex. Averaged across countries, boy
s slightly outscored girls on the math test (d = –0.05) while girls fractionally outscored boys on the science test (d = +0.01). Girls substantially outscored boys on the reading test (d = +0.32). And yet the mean VRs for those countries were –1.14, –1.14, and –1.15 for reading, math, and science respectively.
How Big Does a Variance Ratio Have to Be Before It Becomes Important?
In his 1994 review of greater male variance in mental tests, Alan Feingold pronounced that a VR between 0.9 and 1.1 was “negligible,”62 and thereby established a guideline that others who discuss variance ratios have often followed. I consider this to be an important error. VRs between 0.9 and 1.1 can be socially and culturally important for any trait for which performance at the extremes has consequences.
Consider the implications for a normally distributed trait if the VR is just 1.09. Assume that male and female means are equal, so that the only source of a disparity is the greater male variance. With a VR of 1.09, we can expect males to outnumber females by 31 percent in the 99th percentile, 38 percent in the top half of the top percentile, and 57 percent in the top tenth of the top percentile. Those are noteworthy disproportions in the abstract, but even more noteworthy when we consider the consequences. The social consequences of this seemingly small disparity can be great.
In a population of 250 million adults—roughly the number of Americans ages 20 and older—the top 0.1 percent amounts to 250,000 people. Think about the 250,000 people with the nation’s highest visuospatial and math skills. They constitute some substantial proportion of the top programmers and hardware designers in Silicon Valley, the staffs of quantitative hedge funds, and the nation’s most eminent mathematicians, physicists, chemists, biologists, and engineers. Apply the same logic to other fields requiring different abilities—the 250,000 most gifted attorneys, the 250,000 most gifted managers, the 250,000 most gifted in the performing arts. America is far from having tapped the talents of all 250,000 of the most gifted in any field, but it is probable that a large proportion of the most important accomplishments in all fields are done by those who are in the top 0.1 percent, three standard deviations above the mean. Add in the accomplishments and positions of the much larger numbers of people in the entire 99th percentile—2.5 million people—and even the 31 percent disparity at that level, given a VR of only 1.09, will have a large aggregated impact.
Human Diversity Page 44