FIVE

Testing for Talent

No man is sure he does not need to climb.
It is not human to feel safely placed.
“A girl can't go on laughing all the time.”

—William Empson, “Reflection
from Anita Loos,” 1940

A Tall Story

IMAGINE THAT YOU LIVE in a society where most of the important decisions are taken and the most important rewards enjoyed by people who are tall. Actually, you don't have to imagine very hard, because there's a good deal of evidence that modern industrialized societies do grant substantial economic power and privileges to tall people (more on that evidence below). But suppose height were even more central to economic and social life than it now is. Suppose, for instance, that every time you filled in a bureaucratic form you had to declare your height, and there were separate toilets and changing rooms for shorter and taller people. Suppose that endless daily rituals kept underlining the distinction for you: hosts deciding seating plans at dinner always insisted on alternating short people and tall people; clothes shops devoted separate sections to short people and to tall people, sporting significantly different colors and styles; and special magazines appeared with titles like Tall Tales and Short and Sweet. And suppose you discovered that there were large discrepancies in the representation of short and tall people in positions of power and influence: that, say, among the chief executives of major companies, only one person in thirty was of below-average height. What evidence might it take to persuade you that this was a reasonable, fair, and economically efficient state of affairs?

One possibility is that someone might offer you evidence of ability, based on various psychometric tests. Suppose it were clear that, on the basis of the tests currently available, tall people performed on average better than short people from very early in their schooling. The apparent discrimination against short people in employment and earnings might just reflect the fact that short people were, on average, less talented—at least as measured by the tests. Would that be enough to justify the outcome? Clearly not. It would depend on at least two further factors: whether the tests were actually identifying abilities relevant to the people's later performance in employment, and whether the degree of discrepancy in test performance was large enough to explain the later discrepancy in economic rewards.

Imagine, then, that as you inquired into these tests that children take in school, you discovered that in addition to some tests of arithmetic and verbal comprehension, there were also tests of athletic ability, such as performance on the high jump and scores in basketball. What was presented as an “ability score” was just some aggregate of performance on a group of these different tests, some of which just seem to be ways of favoring tall people from the start. Don't worry, you might be told reassuringly, statisticians have established that these tests are justified because they have a very high power in predicting individuals' subsequent success. At this point the shade of Franz Kafka might begin to show some restlessness: the economic outcomes are justified because they are predicted by the tests, and now you're told that the tests are justified because they predict the economic outcomes? This clearly won't do.

In fact, it wouldn't do even if you could be convinced that the tests were both positive indicators of real underlying talent and positive predictors of economic outcomes. Two factors (in this case, talent and economic outcomes) are not necessarily related just because both of them happen to be correlated with some third factor (test scores). If you're doubtful about this, consider a similar faulty inference: how fast you drive may increase your chances of arriving early at your destination and also increase your chances of having an accident, but having an accident does not increase your chances of arriving early at your destination. In other words, correlation is not transitive: factor A (arriving early) can be positively correlated with B (driving fast), and B can be positively correlated with C (having an accident), without A's being positively correlated with C. That is because the aspects of B that are correlated with A are not the same as those of its aspects that are correlated with C.

To avoid this circular reasoning, someone who wanted to use the predictive power of psychometric tests to defend the economic rewards enjoyed by tall people as being a just return on talent would need to argue that the features of psychometric tests that enable them to reflect talent are exactly those features that help to predict economic outcomes. Because talent and economic outcomes are both complex, multidimensional variables, that is not a straightforward exercise. It would require ruling out two possible causes of the correlation between test scores and economic outcomes.

We could call the first of those causes irrelevant inclusion: test scores might be influenced partly by talent but also partly by some other unrelated characteristics, which are themselves unjustifiably rewarded in economic outcomes (such as the ability to play basketball or to do the high jump). There could indeed be a correlation between test scores and economic outcomes, but one that had nothing to do with talent. We might call the second possibility unjustified exclusion: test scores reflect only some aspects of talent, and the aspects they fail to measure might also be unjustifiably ignored by the process that determines economic rewards. The correlation between test scores and economic outcomes could show that both were reflecting an incomplete and inadequate measure of talent. In both cases the argument would need to be based on detailed examination of the tests, not just on the correlation of the test scores with the outcomes.

As it happens, researchers have done exactly that. As I hinted at the outset, the idea that tall people gain a more than equal share of economic power and rewards is not a fiction. It's been known for around a century that tall people earn more and are more represented in high-status occupations. The social psychologist E. B. Gowin wrote in 1915 that business executives tended to be taller than “average men”; he also compared the height of people with different status in the same profession, noting that bishops tended to be taller than small-town preachers, sales managers were taller than salesmen, and so on.¹ The economists Anne Case and Christina Paxson have shown that the correlation between height and economic success remains very strong today: for instance, they report that “an increase in U.S. men's heights from the 25th to the 75th percentile of the height distribution—an increase of 4 inches—is associated with an increase in earnings of 9.2 percent”; and they report similar findings for women.² The explanation they propose is that “the height premium in earnings is largely due to the positive association between height and cognitive ability, and it is cognitive ability rather than height that is rewarded in the labor market.”³

To prove their point, as Case and Paxson know, it's not enough to demonstrate that test scores are correlated with height and height with labormarket outcomes. Though they don't express it this way, they also need to ensure that the correlation with test scores is not due to either irrelevant inclusion or unjustified exclusion. For instance, they show that height in children at ages 5 and 10 is strongly correlated with scores on a range of tests of different cognitive skills, ranging from figure drawing to linguistic skills and mathematical ability.⁴ The probable reason is that height and cognitive skills are both the outcome of factors enhancing the growth and development of young children.⁵ This makes irrelevant inclusion unlikely: if the correlation of scores with height were due to some extraneous factor unrelated to cognitive ability (like jumping or basketball in our hypothetical example), it's unlikely the extraneous factor would be present to the same degree in all of these different tests. Unjustified exclusion is harder to rule out for certain: maybe some important talents that are more strongly present in short people are being overlooked by these tests. But in the absence of plausible theories about what these talents might be, the wider the range of tests that are used, the less likely it is that such talents are being overlooked. All in all, therefore, it does seem as though the correlation between test scores and height does reflect a genuine causal relationship between talent and height.⁶ But is that causal relationship important enough to account for the fact that tall people receive, on average, so much more generous economic rewards?

The answer is no. Interestingly, Case and Paxson don't find that the effect of height on salaries is due entirely to the causal relationship between height and talent. Using data from the United States, they find that controlling for test scores reduces the return on height for men in the labor market but does not eliminate it completely.⁷ The effect of height on earnings falls to about half of its previously estimated level when test scores are taken into account, meaning that fully half of the effect of height on salaries remains unexplained by cognitive ability. In the data for women, the measured return on height is reduced: it remains positive but no longer statistically significant (in their data from the United Kingdom, this is true for both women and men). It's clear that a substantial part of the measured return to height in the labor market is really due to underlying cognitive ability. But it seems likely that height also continues to convey some additional advantage. Whether that's because of a prejudice in favor of height or because height correlates with other characteristics that employers value besides cognitive ability, or because (perhaps unconsciously) employers use height as a more reliable signal of talent than it actually is, this evidence cannot tell us.⁸ And in case you were wondering, the difference in earnings between women and men has nothing to do with the difference in their heights. Whether you control for height or not, women face the same salary disadvantage compared to equivalently talented men.

Short and tall men. Tall men earn significantly more than short ones. © Anna Peisl / Corbis.

Gender, Talent, and Rewards

Let's return, then, to the issue of gender, which poses a similar puzzle about the difference in economic rewards. The period since 1900, and especially since the Second World War, has seen industrialized countries, as well as many developing countries, remove formal barriers to women's participation in almost all areas of employment. These countries have also revoked or outlawed many barriers that were based on explicit custom and practice. This is a remarkable experiment, representing the first large-scale attempt in the history of the human species to remove obstacles to the division of labor between men and women and to allow members of both sexes to perform (almost) any kind of work they can persuade someone else to pay them to do. During prehistory, men and women appear to have occupied largely different roles (though, because it was prehistory, we have no written evidence and have to make inferences on the basis of other kinds of evidence, as we saw in chapter 4). Throughout much of recorded history they occupied largely different roles because barriers to their doing otherwise were enforced by those groups that held economic and political power, which consisted overwhelmingly, if not quite universally, of men. Most merchant guilds in medieval Europe, for instance, refused entry to women or permitted it only to the widows of deceased members and then only under fairly restrictive conditions.⁹ The landscape was not uniform, and few of the barriers were absolute: examples of women entering apparently male occupations can be found in many countries, but the formal and informal obstacles they faced were often daunting.¹⁰ Many of these barriers persisted long after mechanization had neutralized any advantage that men had once possessed in physical strength.

Since the removal of such explicit barriers, women have begun to occupy many of the roles formerly occupied only by men. This period coincided with technological developments such as the invention of household labor-saving devices¹¹ and the contraceptive pill, which reduced significantly the costs to women of participating fully in the labor market and pursuing the studies necessary to make that participation possible.¹² It's also not a coincidence that this change came soon after women gained the vote, which in most of the industrialized countries occurred after the First and Second World Wars, the first conflicts in which women's contribution in previously male occupations such as munitions manufacturing had proved essential to the war effort.¹³ (Famously neutral Switzerland was the last republic in the Western world to grant women the vote in national elections, in 1971.) Changes in technology, changes in the law,¹⁴ and changes in attitudes all reinforced one another, with changes in attitudes given particular visibility by such signals as women wearing trousers and smoking cigarettes (dubbed “torches of freedom” by the American Tobacco Company in 1929).¹⁵ Women could not only manufacture chemical weapons every bit as well as men; they could inhale them, too.

Whatever the exact configuration of causes, by the standards of previous historical periods, the resulting pace of change has been spectacular. In 2008, just under 60 percent of US women of working age were in the labor force. Although still lower than the 73 percent participation rate of men, this figure was a dramatic change from 43 percent in 1970 and just over 30 percent in the late 1940s.¹⁶ Of women with children above the age of five, 77 percent were in the labor force by 2008.¹⁷ Not only have women entered the workforce in historically unprecedented numbers, but they have entered hitherto male-dominated professions, from accountancy to zoology, and in many of them have come to equal or outnumber men in the space of a few decades. In 2009, women represented 51.4 percent of what the US Bureau of Labor Statistics calls “management, professional and related occupations.” In the United States women also significantly outnumber men in enrollment in higher education, which is the main port of entry into interesting and remunerative work for most of the population. In short, an unprecedented tide of talented and motivated women swept into the masculine economy, producing an incalculable, unrepeatable, but surely vast increase in activity, output, and innovation in many areas of social and economic life.

And yet that tide did not sweep everywhere, and it did not lift all boats. The very speed of this dramatic change throws into sharper relief those areas where change has been slow or nonexistent. Three things remain highly puzzling. First, some occupations continue to see low proportions of women even though their formal barriers are apparently no higher than in occupations where women have attained parity. Women represent only 32 percent of lawyers, 25 percent of architects, 20 percent of computer programmers, 15 percent of taxi drivers and chauffeurs, 7 percent of civil engineers, 2.2 percent of electricians, and 1.3 percent of airline pilots. Only 32 percent of physicians and surgeons are women, even though women make up nearly three-quarters of all health care practitioners.¹⁸ There are some differences between the United States and other rich countries (notably because overall female participation in the workforce is lower in countries such as Italy, Japan, and Germany), but broadly similar patterns are observed across the industrialized world.

Second, women's salaries continue to be lower than men's even within occupations. The overall ratio of US women's earnings to those of men was 81 percent in 2010, and in some professions it was much lower (77 percent among lawyers, for example, and 58 percent among personal financial advisers).¹⁹ Third (and related), across a broad range of economic activities, many of the most prestigious and highly remunerated positions continue to have startlingly low rates of representation of women. In 2010, women made up only 15.7 percent of board members and just 2.4 percent of chief executive officers of Fortune 500 corporations.²⁰ There's some evidence that when women are appointed to leadership positions, these tend to be more precarious ones (this phenomenon has been christened the “glass cliff”).²¹ Even within a specific area of activity, such as the restaurant and catering industry, women make up more than 70 percent of waiters and waitresses and 41.5 percent of cooks but only 20.7 percent of chefs and head cooks.²² What is going on?

The story of height might make you think I'm going to suggest that the explanation might be differences in talent. But the point of the analogy with height was to show that using measures of talent to explain differences in rewards requires that these measures satisfy a number of rather particular conditions. Outside a small number of occupations for which physical strength remains important, there's simply no evidence at all that the underrepresentation of women in certain occupations, their lower salaries, or their underrepresentation in particularly high-status or well-rewarded positions has anything at all to do with differences in talent between women and men.

We can start by looking at psychometric tests of cognitive ability, or skill. These do not test a single type of skill but many, usually through the administration of a number of component tests that form part of an overall package or battery. There are some differences between men and women, on average, in their performance on these different tests of skill: women tend to perform better on average than men in tests of verbal ability, and men tend to perform better on average than women in tests of some (though not all) types of visual and spatial skills.²³ These average differences are, however, usually small compared to the variation between individuals of either gender, and some of them are found in certain environments but not in others.²⁴ There has been vocal and justified criticism of the specious precision of some of the tests, which are known to be influenced by contextual factors such as the information given to subjects before the tests are administered.²⁵ In particular, the relative performance of women has been shown to be subject to what is known as “stereotype threat”: when primed to be aware of stereotypes of poor female performance, women perform less well than when primed with neutral messages.²⁶ These considerations mean that a lot of uncertainty surrounds the evidence on gender differences in performance on these component tests, but for the moment it seems reasonably likely that at least some such differences are real.

Psychologists speak of g, a measure of general intelligence, as a persistent common component of all reputable psychometric tests; it is also highly heritable and highly correlated with economic performance.²⁷ This heritability does not mean, incidentally, that environmental influences on g are unimportant. On the contrary, average IQ scores have risen rapidly over time in most countries around the world, by around three IQ points per decade. This increase is known as the Flynn effect, after the researcher who first documented it, and it is entirely implausible that it could be due to genetic change.²⁸ It therefore indicates that the skills captured by IQ tests are strongly influenced by environmental factors that have been changing over time, though to date there is no hard evidence as to what these might be.²⁹

There is a large and contentious literature debating whether g shows any significant average difference between women and men. Some studies find no difference, some find a difference in favor of men (equivalent to between one and five points on a standard IQ scale), and a much smaller number find a difference in favor of women.³⁰ The male advantage, where it is found, is equivalent to only about a decade's worth of the Flynn effect and could therefore be due entirely to differences in the learning environments for boys and girls, if these are as great as differences in the learning environment from one decade to the next. Within this margin of variation, such findings are simply irrelevant to the question of whether differences in talent sufficiently explain men's and women's differential performance in the labor market. We would not learn anything from knowing that the average difference was “really” zero, or 3 percent in favor of men, or 1 percent in favor of women.

The reason is very simple. Precisely because there are differences on average in men's and women's performance on the component tests, the common factor of talent that such tests appear, statistically, to reveal, will be sensitive to the choice of component tests that make up the overall battery, and in particular whether there are more of them that tend to favor women's performance or more that favor men's. In other words, the fact that all batteries of tests can be used to derive a measure of g should not make us think that the measures of g they produce are identical. On the contrary, specific estimates of g that particular tests produce can rank a group of individuals very differently according to the types of task that those individuals happen to be good at.³¹ As the psychologist Earl Hunt has written, “You can find a measure of general intelligence that is g and verbally loaded, and produce an advantage for females, or produce a measure of general intelligence that is g and loaded on spatial-visual reasoning, and find an advantage for males.”³²

As we saw with height, the correlation of some measure of g with economic rewards can serve as a justification for those economic rewards only if the strength of that correlation is not exaggerated either by irrelevant inclusion of some extraneous characteristic that happens to favor one sex's test performance or by the unjustified exclusion of some genuine component of talent that is also underrewarded by economic performance. Most reputable general psychometric tests on which comparisons are based do not suffer from blatant examples of irrelevant inclusion (as they would if, for example, the ability to remember football scores were one of the component tests). But unjustified exclusion is another matter. A test battery that is verbally loaded could be considered to be unjustifiably excluding a number of spatial-visual tests, while a test battery that is visually-spatially loaded could be considered to be unjustifiably excluding a number of verbal tests. Both kinds of battery may exclude other dimensions of genuine talent along which men and women may differ. Observing the correlation of the resulting g measure with economic rewards could not tell you whether those rewards “really” reflected talent.

Notice how the argument with respect to gender is quite different from the argument with respect to height. The test batteries do not consist of one group of tests on which tall people do better on average and another group on which short people do better: tall people do better, on average, on all of them. So while you can argue whether the tests exaggerate the cognitive disadvantages of short people, and whether those disadvantages are inherited or acquired, it's not easy to claim that they arbitrarily exclude equally reasonable component tests on which short people would perform better. And different weightings of the component tests might affect the size of the performance difference between short people and tall people, but it would not affect the fact that such a difference existed. This shows, though, how challenging it would be to demonstrate that differences in talent were the real explanation for gender differences in economic rewards. The fact that the exercise has been done, carefully and broadly convincingly, for height only underlines the fact that it has not been done for gender.

Unlike in the case of height, when you have some tests on which men do better and others on which women do better, you can't use the “average” effect across the two types of test to justify economic rewards unless you have separately found a way to justify the weighting of the different components in the average. This isn't some pedantic technical objection: on the contrary, we face this kind of difficulty all the time in making comparisons between complex, multidimensional options. You can't choose between a fun but insecure job and a secure but dull one by ranking each job according to an index that gives arbitrary weights to security and to fun: you really have to decide how much security and fun are important to you. You can't settle an argument about whether it's better to live in the city or the country— even if you thought that question had an objective answer—just by comparing places to live on an index that includes an arbitrary mix of dimensions that tend to favor cities (like quality of nightclubs) with others that tend to favor the country (like the extent of fields and woodland).³³ Even if each dimension captures something we all agree is important, how to weigh the dimensions against each other can never be decided by a purely technical procedure.

It's uncontroversial that there are many aspects of talent that the psychometric tests aiming to measure intelligence cannot capture. That's not because no one has considered the problem but because the very basis of psychometric testing (getting subjects to sit down with a pen and paper or a keyboard and mouse) is like capturing a moving scene in a pencil drawing: the medium intrinsically misses some important features of what is going on. For instance, it doesn't capture reactions between people, a sensitivity to which is one of the most important talents that anybody can exercise but which intelligence tests are by their nature ill-equipped to assess. Nor, usually, does it capture spatial orientation, the ability to find your way around, which is related to certain types of visual spatial ability but distinct from them (though various kinds of computer simulation make it easier to test this than it used to be).³⁴

Another important talent that tests usually fail to assess is the ability to form accurate memories of a scene and to use them to inform our dealings with other people. One of the most famous case histories in psychology concerns a patient known as H.M., who, because of the removal of part of his brain to control severe epileptic seizures, was unable to form short-term memories and as a result was entirely incapable of functioning socially in a normal way. Yet H.M. scored high on several standard intelligence tests.³⁵ This is not a criticism of intelligence tests: it's a criticism of people who think that intelligence tests tell you all you need to know about the value of someone's economic contribution to society. Similarly, given what we know about the multidimensional nature of talent and the accumulated evidence that men and women tend to cluster differently along different dimensions of that talent, if we were to discover that some purported measure of general intelligence conclusively showed either sex to be superior to the other, that would tell us a lot about the implicit weighting of the measure and little about the men and women it was supposed to be measuring.

Personality and Talent

Talent is not just about cognitive skills but also about certain other types of ability, such as hard work and organization. Might gender differences in these noncognitive abilities be more systematic than differences in cognitive skills? A reasonable consensus in psychology has identified five major personality traits (the “Big Five”) that are generally heritable and reasonably stable over time (though less so than cognitive skills).³⁶ They are also found fairly consistently across cultures, though not entirely so (for instance, richer countries display larger variations in reported personality traits).³⁷ These five traits are openness to experience, conscientiousness, agreeableness, emotional stability, and extraversion. Unlike cognitive skills, they are measured by tests that depend on honest self-reporting, which is why they are much more rarely used for recruitment purposes than are IQ tests: people applying for jobs in marketing know better than to give answers indicating low scores on the extraversion scale. This may also make them more dependent than IQ tests on norms about the kinds of personality it is desirable to have. When it's considered cool to be open to experience, you can bet many more people will give answers that claim to be just that. (This doesn't mean that the answers are simply false: being open to experience really is easier when people around you encourage you in being so.) Indeed, the accuracy of these tests may even depend on the kinds of personalities who take them (conscientious people may be much more cautious in making claims about their conscientiousness). We need, therefore, to be careful before we grant too much authority to any particular personality test.

All the same, psychologists have found gender differences in some of the Big Five traits. Economists have found that such traits also have a tendency to predict labor-market performance, although the correlations are all smaller than those of IQ tests, are less uniformly corroborated across different studies, and are not always either statistically significant or economically very important.³⁸ Conscientiousness is the trait most strongly (and positively) associated with labor-market outcomes (as Woody Allen is credited with observing, “Eighty percent of success is showing up”), but it is only about half as predictive as IQ. Most (though not all) studies find that women score higher on conscientiousness.³⁹ Emotional stability, on which men score consistently higher on average, is also positively associated with labormarket outcomes, but to a lesser degree.⁴⁰ Agreeableness is negatively associated with labor-market outcomes, but mainly for men. Openness to experience and extraversion are only weakly associated with higher performance, and there are few consistent findings across the many different studies.⁴¹ Furthermore, the most careful recent study to examine the possible causal channels by which personality might affect labor-market outcomes argues that much of the effect happens through its influence on educational achievement.⁴² There is certainly evidence that personality affects educational achievement,⁴³ but the gender differences in labor-market performance that we are interested in are large even when educational achievement is taken into account.

Overall, although it seems a reasonable guess that gender differences in noncognitive abilities could have a stronger association with talent than do measures of IQ, none of the studies conducted so far suggests that they actually explain more than a small fraction of the difference in labor-market outcomes. This is partly because the association between personality and labor-market performance is typically weaker and less consistent across studies than it is for IQ; it is also because the gender differences in personality traits are not systematically favorable to men. Men score higher on emotional stability but typically lower on conscientiousness; they are also penalized for agreeableness to an extent that women are not. These differences therefore tend to cancel out in terms of overall average effects. One recent study concludes that “only 3 to 4 percent of the gender gap is explained by differences in personality including differences in traits and trait returns,” and no statistical study has made a credible claim that personality differences explain most of the gap.⁴⁴ Overall, then, average differences in talent between men and women do not seem any more promising an explanation of the gender gap in earnings when personality differences are taken into account than when we consider only cognitive skills.

Are Men More Extreme?

It has sometimes been claimed that even if mean scores on psychometric tests show very small differences, if any, between men and women, men's scores show a tendency to higher variance, which would explain why men might be more highly represented at the top extreme of the distribution of economic rewards (as well as at the bottom). This tendency to higher variance has also been given a genetic explanation, on the grounds that for psychological traits influenced by genes on the X chromosome, the fact that men carry only one X chromosome, while women carry two, means that men will be more vulnerable to the effect of unusual alleles (which in women would more commonly be offset by the “normal” allele on the other chromosome).⁴⁵

There is indeed evidence for several of the building blocks of this argument. The X chromosome appears to carry a particularly large number of genes that are involved in cognitive development (a fact that incidentally tends to support Geoffrey Miller's hypothesis that sexual selection by women was an important factor in human cognitive evolution, since the X chromosome also carries a particularly high frequency of genes related to sex and reproduction).⁴⁶ Measures of g tend to show somewhat higher variance for men, and small differences in variance can translate into large differences in representation at the extremes of the distribution of test scores.⁴⁷ Scores on tests of more specialized abilities can show even greater differences in representation at the extremes (although, intriguingly, some such discrepancies have declined strikingly over time, suggesting a strong influence of socialization).⁴⁸ There are, for example, many more male than female students in the extreme upper tail of the distribution of scores on various advanced mathematical tests, though there remains much controversy about the likely explanation for the gap.⁴⁹

Nevertheless, there's a serious flaw in the argument that differences in variance are likely to account for differences in representation of men and women at the upper extreme of the distribution of economic outcomes. The flaw lies in the implicit assumption that these traits, as measured by test scores, translate directly into economic outcomes, so that more of the trait leads on average to a higher outcome, however strongly the trait is present. That may be true at the lower end of the distribution, since single genetic mutations can interfere in important ways in normal cognitive development, and the incidence of mental disability is around 30 percent higher in men than in women.⁵⁰ But within the normal range as well as at the upper end of the distribution, traits determine economic outcomes, to the extent that they do, in the context of complex interactions between the bearer of the trait and other individuals. People who possess single traits to an extreme degree do not necessarily enjoy extreme rewards as a result. Think, for instance, of a trait like talkativeness. Talkative people tend to be economically more successful than very taciturn people. But this doesn't mean that extremely talkative people are extremely successful: on the contrary, unstoppable windbags tend to annoy others to a degree that can seriously hinder their professional prospects. So an argument that men are more likely to have extreme traits, even if true, doesn't tell us anything about whether these traits are the reason for their presence at the extreme of the distribution of economic rewards.

In fact, the kinds of ability that psychometric tests can uncover are very rarely associated unambiguously with success in any field (other than the field of taking psychometric tests). For example, visual and spatial ability, on which women tend to score less well on average than men, is certainly relevant for driving skills. Yet most large-scale statistical studies find that women are better drivers, on average, than men.⁵¹ Visual and spatial skills are important, certainly, but so are risk assessment and courtesy to other road users. Even if individuals who score highly on visual and spatial skills might drive better, on the whole, than those who score less highly, the effect might not go all the way: those with extremely high visual and spatial skills might overestimate their ability to handle the risks posed by other drivers with lower levels of skill or become impatient at having to share the road with them.

Similar points, incidentally, can be made about the association between underlying neurophysiological influences on certain skills and the skills themselves: for example, increases in testosterone appear to enhance spatial reasoning abilities in women and in men with low testosterone levels. But they decrease spatial reasoning abilities in men with normal or high testosterone levels, a finding that will surprise no one who has spent time navigating kitchen space in the company of teenage boys.⁵² As it happens, men have higher average testosterone levels than women and score higher on certain visual-spatial tests, and the former phenomenon may well be causally responsible for the latter. But because the effect of testosterone on skills does not continue to increase beyond a certain level, if men and women had the same average testosterone levels while men had higher variance, men's overall scores on visual-spatial tests would tend to be lower, not higher, than those of women. This illustrates a general point: the theory that men are more extreme because they have only one X chromosome appeals to a simple model in which genes have a continuously increasing influence on physical traits like testosterone, physical traits have a continuously increasing influence on behavioral traits, and behavioral traits have a continuously increasing influence on economic outcomes. The model is seductively easy to think about, but in lots of ways it doesn't fit the evidence.

Overall, therefore, the claim that the major differences in economic rewards of men and women can be explained by differences in talent—either on average or specifically at the upper extreme of the distribution of talent— is entirely unconvincing. We know that talent can produce higher test scores, and we know that higher test scores are correlated with economic rewards. But outside a small number of fairly idiosyncratic occupations, we don't know that the particular tests on which men perform better than women uncover the particular talents that make people do their jobs better overall.⁵³ Better visual-spatial skills might help you negotiate the water cooler with more dexterity, or maybe do better on the golf course, but these differences are a slim basis for an overall theory of what makes people in a modern economy do their jobs well.

It's not impossible that someone might come up with such a theory in the future, a theory that could really explain why those aspects of talent on which the test scores favor men are also the aspects that count for most in a modern economy. So none of what I've written here should be interpreted as ruling out the possibility of a talent-based explanation. But no one, to my knowledge, has come up with one yet. In the absence of such an account, particular tests of talent cannot be justified on the grounds that the scores they yield are correlated with economic rewards when those economic rewards are in turn justified because of their correlation with individuals' scores.

If it's not talent that explains the discrepancy in economic rewards between women and men (and if, as we saw, it's not height either), what is the explanation? Chapter 6 looks at a different possibility: instead of differences in talent, could the explanation be that men and women have different tastes?