Intelligence Research Paper

Comments on intelligence have a long history. Homer’s Odyssey repeatedly refers to the hero’s superior thinking. In the 1500s the Spanish philosopher Juan Huarte de San Juan wrote extensively on the difference between people who solve problems through their memory and those who rely on imagination. Huarte also speculated that the differences were due to differences in biological capabilities, but his investigations were restricted by the limited (and often erroneous) biological theories of his time. Truly scientific theories of intelligence had to wait until the nineteenth century, when Francis Galton proposed that individual differences in mental capacity were reflections of individual differences in the efficiency of the central nervous system. This lead Galton to try to develop tests of ‘nervous functioning,’ such as the speed with which a person could make a motor movement in response to a signal, and then to try to relate performance on these tests to performance in complex activities, such as success in college courses. At the time this effort was not seen as a success, and it was abandoned. Subsequently more sophisticated statistical analyses have shown that this conclusion may be premature. Some of our modern theories of intelligence can be traced to Galton’s ideas. Shortly after Galton’s proposal Alfred Binet, a French scientist, was asked to develop a test that could be used to identify children who were likely to fail in the French public school system. Binet took a pragmatic approach. He assumed that (a) mental competence grows as a child ages and (b) that children who are ahead or behind normal development at one age are likely to be similarly ahead or behind at a later age. Accordingly, Binet developed an intelligence test that was actually a ‘battery’ of subtests, in which children were asked to perform a variety of tasks that were ‘typical of their age.’ For instance, an average four-year-old child was supposed to be able to repeat back three digits: an eight-year-old should repeat back five digits.

The most widely used modern versions of intelligence tests, such as the Stanford–Binet test (derived from Binet’s original work by Lewis Terman, a professor at Stanford University) and the Wechsler adult and child intelligence scales, retain Binet’s emphasis on a battery of tasks, rather than relying on a single task to measure intelligence. Tests that are used to evaluate candidates for jobs and academic programs also use the battery approach. The Scholastic Assessment Test (SAT), widely used in the USA for college admission, for instance, consists of separate batteries evaluating linguistic and mathematical skills. So does the Armed Services Vocational Aptitude Battery (ASVAB), which is used as a selection device for the United States Armed Services.

There has been concern that tests such as these, which assume that all examinees have had certain experiences, are simply evaluating a person’s exposure to different cultures. This is unlikely, at least for the range of countries represented by modern industrialized society, for testing seems to serve the same purpose in European, North American, and industrialized Asian nations. However there have been numerous attempts to develop tests that used the same type of problem (but at different levels of difficulty) rather than to take the test-battery approach. The most successful of these tests require the examinees to find patterns in observations, and especially in geometric patterns.

2. One Intelligence Or Many?

As different tests were developed a surprising finding emerged. Although tests of vocabulary, paragraph comprehension, arithmetic problem-solving, or pattern recognition involve different processes on their face, in practice there are moderate to high correlations between the various types of tests and subtests. The correlation coefficients, r, usually lie in the range 0.6 to 0.9. This suggests that there is a single individual difference, ‘general intelligence,’ that determines a person’s mental competence almost independently of the way in which that competence is expressed. While this turns out to be a reasonable approximation for many purposes, research in the 1990s demonstrated a more complex picture. The following facts seem to be a more accurate statement of the situation:

(a) In the range from normal to high mental competence there appear to be three factors of intelligence. One is Fluid Intelligence (Gf ), which can be thought of as the ability to see new patterns in data and develop new solutions to novel problems. The second is Crystallized Intelligence (Gc), which reflects the ability to apply previously learned information to the current problem. The third is Visual–Spatial Intelligence, which reflects the ability to manipulate visual images and see visual relationships.

(b) Measures of fluid and crystallized intelligence are correlated (r about 0.6) in most studies, suggesting but not proving that they may be partly a reflection of a general intelligence (g) factor. An alternative explanation for the correlation is that the Gc and Gf depend upon different combinations of elementary abilities, such as pattern recognition, short term memory, and ability to maintain attention. At the present time we do not know which of these interpretations is correct.

(c) Visual–spatial reasoning has a much lower correlation with Gc and Gf than the two intelligences have with each other. This indicates that visual— spatial reasoning is a somewhat different ability than intelligence, as indicated by Gc and Gf.

(d) The above remarks apply in the normal–above normal-gifted range. Test scores are much more highly correlated at the lower end of the scale than at the upper end. Put a slightly different way, unusually high scores in one type of test (say, Gc) may be associated with a fairly wide range of scores on a different test (say, Gf ). However unusually low scores on one type of test are usually associated with low scores on another type of test. Put somewhat crudely, the case for general intelligence is weak, but the case for general stupidity is strong.

There are a few exceptions to statement (d). Turner’s syndrome is a genetic disease that affects women. Turner’s syndrome patients have low visual–spatial intelligence test scores, but score in the normal range on tests requiring verbal reasoning. People who have suffered damage to specific areas of the brain may lose mental competence in one area, such as memory, without necessarily losing competence in other areas. (People who suffer widespread brain damage, not surprisingly, show widespread deterioration of functioning.) These exceptions aside, statements (a) to (d) above are a reasonable summary of the distributions of test scores in normal populations.

3. Do Test Scores Matter?

A person’s score on an intelligence or aptitude test is of little difference in itself. It becomes interesting only if test scores predict performance in some socially important arena, such as schools or the workplace. There has been considerable controversy over whether the tests (a) predict nothing at all, (b) predict performance in academic settings only, or (c) predict performance in both academic and workplace settings. The facts are fairly well known. The disagreement is over what to make of them.

In university settings the correlation between test scores and grades is in the range 0.3 to 0.4. The correlations are somewhat higher for students going through technical training, such as Armed Services technical training schools, at least partly because these students represent a wider range of abilities than university students. These correlations do not allow for the fact that performance scores can only be obtained for those people who have a high enough test score to be enrolled or employed in the first place. When statistical adjustments are made to allow for this selection effect the correlation rises to about 0.5. Moving outside of academics, similar scores are obtained in industrial settings. For instance, one study found a correlation of 0.38 between scores on an SATtype test given to management trainees and the level of management position that the trainees achieved 20 years later. Similar correlations have been observed in many studies relating test scores to industrial performance, although the range of obtained correlations is generally higher in industrial than academic studies. This may be because it is difficult to obtain accurate measures of personal performance in many industrial settings.

Mathematically, a correlation of 0.35 indicates that about 10 percent in the variance of one score is predictable from knowledge of another. In this case, it means that 10 percent of the variance of academic or industrial performance would be predictable from knowledge of a cognitive test score. The higher 0.5 estimate raises the variance effect to 25 percent. Both of these values are some distance from perfect prediction. However, ‘variance’ is a statistical term whose meaning does not easily translate into the everyday concept of ‘variation.’

There are three other ways of looking at the situation. One is to consider what sorts of correlations are typically found between different aspects of human variation. For instance, the correlation between adult height and weight (within sexes) is about 0.5, not far from the estimated population value for test-performance relations.

A second way to look at the issue is to ask whether there are alternative predictors that do as well as tests of cognitive competence. The evidence here is striking. The typical personality test has a correlation of about 0.20 with indices of workplace performance, just slightly more than one-half of the value for a cognitive test. However, it is important to note that the personality and cognitive tests appear to predict different aspects of workplace performance, for prediction from a combination of cognitive and personality tests is higher than the prediction from a cognitive test alone. If just one measure of prediction of performance is to be used, a cognitive test will be a better predictor than a personality test. However, the ideal selection system would use both types of predictor.

A third way to try to evaluate the importance of intelligence is to ask how much workplace or academic performance is improved by using a test as a screening device. The answer to this question depends upon the correlation between test scores and performance and the extent to which the academic or workplace institution can select candidates. If all applicants are to be accepted, then there is no point in using any test. On the other hand, if only 10 percent of the applicants are to be accepted, then a test that has a correlation of 0.3 with performance can be used to select a workforce (or student body) that, on the average, will outperform an unselected group by more than 150 percent. This shows that the use of cognitive tests for selection is most justified when (a) the selection is rigorous and (b) there is a substantial difference between the benefits of good and poor performance.

Finally, there are interesting observations about the distributions of test scores in the population. It is well established that people who have high test scores as adolescents or young adults will, on average, tend to get better jobs and do better economically. It is even true that people with high test scores tend to live longer than those with low test scores! However, it is very difficult to establish cause and effect from such studies, because education, health, and economic success are themselves positively correlated.

4. What Causes Intelligence?

When we ask ‘what causes intelligence?’ it is important to distinguish between two different ways that this question can be asked. At the individual level, we can ask what characteristics of a person cause them to be or not to be intelligent. At the population level, we can ask what measurable demographic variables are associated with intelligent performance. We look first at the individual level and then the population level.

4.1 The Individual Level

At the individual level, it becomes important to distinguish between physical and social causes of cognitive competence, and between fluid and crystallized intelligence (Gc and Gf ). People who perform well on tests of fluid intelligence and reasoning are, to a very large extent, people who are adept at keeping track of several things at once, and who can manage fairly large amounts of information. This ability is known as an ability to manage ‘working memory’ for information relevant to the problem currently being worked on. It is contrasted with ‘long-term memory,’ which refers to memory for information about how the world works and information about biographic events that happened some time ago. Working memory, in turn, appears to be related to the functioning of the frontal and prefrontal areas of the brain. There is now a good deal of evidence suggesting (but not yet proving) that individual differences in the ability to activate information in these areas of the brain is related to scores on tests of fluid intelligence. Just what these relations are, and how they relate to individual differences in neural functioning in other parts of the brain, is an important topic in research on intelligence. While individual differences in fluid intelligence are related to individual differences in brain functioning, it is important not to exaggerate the relationship. Statistical analyses suggest that individual differences in brain functioning can account for only a part of the wide individual differences in the ability to deal with ‘new and unusual problems,’ that is, fluid intelligence. By default, the remaining influences must be social (e.g., education, early training that might set a particular style for problem solving), but as of this research paper psychologists have been unable to determine what these environmental influences are.

The situation is quite different for tests of crystallized intelligence, that is, the ability to recognize and apply previously acquired solutions to new situations. In part this ability appears to depend upon generalized pattern-recognition abilities in the brain. However, it is also extremely responsive to education and training. This is seen most clearly by studies of expertise, individual differences in solving problems within a particular domain. Study after study has shown that specialized problem-solving depends upon the acquisition of schematic forms of reasoning appropriate to the domain at hand. Furthermore, these schema are very largely acquired by extensive practice. This has been shown in domains as far apart as chess, physics, and economics. Fluid intelligence appears to be one determinant of how efficiently information can be acquired during the schema-acquisition period. (This may partially account for the correlation between measures of fluid and crystallized intelligence.) However, knowledge builds on knowledge, so soon the expert has an advantage over the novice in learning how to learn within a particular field. Whether or not there is a generalized ability to learn to learn, regardless of the field being studied, is an open question. If there is, this ability is probably quite closely related to fluid intelligence.

As would be expected, any influence that causes a deterioration of brain structure is likely to have a deleterious effect on intelligence. In modern industrialized society alcoholism is undoubtedly the biggest single effect; repeated studies have shown a negative correlation between excessive alcohol use and intelligence test performance. Once again, whether or not this is causal is hard to determine, as excessive alcohol use could either produce or be produced by low intelligence. However, the effect can be trans generational. A major cause of mild mental retardation in children is excessive alcohol use by the mother during pregnancy. Other negative influences on intelligence include prolonged malnutrition (primarily in the developing nations) and exposure to atmospheric lead.

Aging has a paradoxical effect on intelligence. Fluid intelligence test scores drop. This is not surprising, as there is considerable evidence for an age-related drop in performance in tasks involving working memory. However there are very large individual differences in the extent of the drop. If we compare the fluid intelligence test scores of otherwise comparable individuals in their 20s and 60s we find that the top scores of the older group are only slightly below the top scores of the younger examinees, but the lowest scores in the older group are considerably below the lowest scores in the younger examinees.

Crystallized intelligence, in the sense of scores on such things as vocabulary and knowledge tests, may rise slightly during the adult working years, and declines only slowly until people reach their 70s or beyond. Because we live in a specialized society this may actually underestimate the abilities of older people. People become quite competent in those things that they practice, so during their working years people may become ‘cognitive specialists’ in the tasks that they encounter every day. Perhaps for this reason, general cognitive tests, such as the Department of Labor’s General Aptitude Test Battery (GATB), seem to underestimate the out-of-laboratory performance of older workers. Of course, these statements apply only so long as the workplace and social environments remain the same.

4.2 Population Level Variables

Another way to answer the question ‘what causes intelligence?’ is to look at the correlations between intelligence test scores and various other measures of individual differences. This sort of analysis can be quite informative, although it only indirectly provides an indication of the mechanisms that produce cognitive competence at an individual level.

One of the oldest questions in the study of intelligence is whether intelligence is inherited or acquired through experience. The answer, not surprisingly, is that it is partly inherited and partly acquired. Very many studies have been conducted in an attempt to estimate the percentage of variance in a population that is due to genetics. This is called the heritability coefficient. Conceptually, the clearest experiment is to examine the correlation between monozygotic (MZ) ‘identical’ twins raised apart to the correlation between MZ twins raised together. Such situations are rare but they do occur, usually in cases of adoption immediately after birth. Correlations as high as 0.8 have been reported between MZ twins raised apart. By contrast, the test scores of dizygotic (DZ) ‘nonidentical’ twins raised apart is usually in the range 0.4 to 0.5. If one believed that the environments for each of the adoptees were chosen randomly this data is consistent with a heritability coefficient of 0.8. However there are a variety of reasons to believe that these correlations are inflated to some unknown degree by common environmental factors (e.g., sharing a common uterine environment prior to birth). Other studies of correlations between family members of varying degrees of relation (e.g., parent–child, siblings, grandparent– grandchild), using statistical techniques too complicated to be easily explained, have lead geneticists to conclude that the heritability coefficient is somewhere in the range 0.4 to 0.6. It is important to realize that this value applies to a modern industrialized society, where the difference between the well-to-do and the poor in terms of nutrition, disease, and schooling are not large. In a society in which there were very wide differences in living conditions and educational opportunity we would expect environmental effects on intelligence to be large, and hence the heritability coefficient would decrease.

Notwithstanding these qualifications, one fact is clear. Within the range of environments found in a modern postindustrialized or industrialized society, the single most important influence on intelligence is heritability. It accounts for slightly over one-half of the variance in intelligence test scores, on a population basis. This does not mean that you can say that one-half of an individual’s mental competence is due to genetics: the variance referred to applies only to the population as a whole. Given advances in understanding the human genome it may be possible, in the future, to associate specific genes with intelligence. When this is done it will be possible to discuss genetic effects on an individual basis. As of the year 2001, such research was in its infancy.

What in the environment determines intelligence? Three effects have been noted. One is schooling: people who receive formal instruction (and especially people who are literate and read a good deal) do better on intelligence tests. Another is the physical environment: certain diseases, environmental hazards, and pathogens can cause reductions in intelligence. The most notable are alcoholism, exposure to atmospheric lead, and any disease or physical insult that harms the brain. The last influence is not inevitable. Some brain injuries lead to a drop in intelligence, others do not.

A third, and particularly puzzling, environmental influence is the ‘cohort effect’ phenomenon. In industrialized nations the population average on intelligence tests rose throughout the twentieth century. (We do not have comprehensive data for nonindustrialized nations.) The rise was particularly marked at the low end: on a percentage basis there are fewer people with very low scores (IQ below 80) than there used to be. The cause for this rise is unknown. However, it is clearly an environmental effect, for the change is far too rapid to be due to genetic changes. Understanding genome patterns for intelligence (see above) and understanding the reason for the cohort effect are perhaps the most interesting questions in research on intelligence as we progress through the twenty-first century.

5. Who Has Intelligence? Demographic Variables

The question of whether or not different demographic groups differ in intelligence is one of the most hotly debated, socially contentious issues in the social sciences. It turns out that the facts are relatively clear. The reasons for those facts, however, are far from clear.

Male–female differences present a fairly straightforward case. While absolute equality cannot be proven, there are no marked differences between men and women in either fluid or crystallized intelligence. It is possible that women exhibit a slight advantage on some verbal tasks, but the size of the effect is not large. Men, on average, do considerably better than women on visual–spatial reasoning tasks that involve visualization or perceptual judgments of motion. This is important, because such judgments are required in tasks that involve spatial orientation, such as walking around a building and then trying to imagine what it would look like from above. The difference between population means on such tasks is about 0.5 standard deviation units. In less statistical terms this means that about 70 percent of the male population scores above the median score for the female population. Because visual–spatial reasoning has been shown to be correlated with achievement in mathematics, it has been suggested that the female difference in visual–spatial reasoning is one of the reasons that women are not proportionately represented in jobs that depend on mathematical reasoning. While this is a reasonable conjecture, with some evidence to support it, social factors are also responsible for this difference in occupational achievement.

There are racial and ethnic differences in intelligence test scores. This statement is largely based on studies in the USA, although a few studies from Europe (where there is less ethnic diversity) show generally the same pattern. In tests carried out in the late 1990s, ‘African-Americans’ scored about 0.8 standard deviation units below the ‘White’ mean (and it should be noted that this latter social category includes many disparate groups), while ‘Asian-Americans’ scored approximately 0.3 units above the ‘White’ mean. People identified as ‘Hispanic-Americans’ (who may be of any race) scored between the ‘African-American’ and ‘White’ mean, and there were differences between particular groups of ‘Hispanic-Americans’. Furthermore, there was no evidence that the tests were unfair to particular groups, in the sense that the tests predicted performance more accurately in one group than another. The correlations between test scores and academic or job performance (and other relevant statistics) were substantially the same across all groups. This indicates that the cognitive skills that are (a) tapped by the test and (b) required in academics and the workplace are not equally distributed across racial and ethnic groups in the USA.

These statistics do not provide a reason for the differences. Environmental effects are clearly involved, for greater opportunity for social equality has been accompanied by a lessening of the difference between groups. For instance, the ‘African-American’ vs. ‘White’ difference was about 1.2 standard deviation units prior to the end of legal segregation and separate, usually inferior, educational opportunities for African-Americans. This clearly indicates a social influence. However the rate at which the gap is closing has slowed since about 1980. This and other evidence has led some individuals to claim that ethnic differences are partly due to genetic differences between groups. The claim has been strongly challenged by other researchers. The controversy is impossible to resolve at this time, largely because genetic differences between ethnic groups are confounded with social differences in economic status, living conditions, family structures, and a host of other environmental variables. To make the situation even more confusing, it is logically possible that differences between groups A and B are genetic, while the differences between A and C are environmental. Perhaps someday advances in molecular biology will make it possible to trace genetic influences on intelligence at the single-gene level. It is also possible that in the future a better understanding will be obtained of the way special environmental factors, such as childrearing practices or schooling, influence intelligence. Until this research is done any causal ‘explanation’ of racial and ethnic differences in cognitive competence will be highly speculative.

6. Final Comment

A century of modern research on intelligence has led to a great many interesting conclusions. Perhaps the most important is that individual differences in cognitive competence are the best single predictors of success in industrialized and information-age societies. This does not mean that intelligence is the only thing that is important, but it does mean that it is important. The changes in population levels of intelligence throughout the twentieth century are not yet understood. When they are, though, they may well lead to ways of improving human cognitive potential in the twenty-first century.

Bibliography:

1. Hunt E 1995 Will We Be Smart Enough? A Cognitive Analysis of the Coming Workforce. Russell Sage, New York
2. Mackintosh N J 1998 IQ and Human Intelligence. Oxford University Press, Oxford, UK
3. Sternberg R J (ed.) 1994 Encyclopedia of Intelligence. Macmillan, New York
4. Sternberg R J, Detterman D K (eds.) 1986 What is Intelligence? Ablex, Norwood, NJ