The Neuroscience of Intelligence
Page 8
With respect to both a genetic basis for intelligence and the failure of early education to boost IQ, it is fair to say that Jensen’s hypotheses have not yet been refuted by another 45 years of new data. The interested reader is referred to the references at the end of this chapter for sources that delve into the Jensen controversies in greater detail (Snyderman & Rothman, 1988). Steven Pinker’s The Blank Slate is a terrific book and I highly recommend it for understanding the broader historical and philosophical context of intelligence research criticism. I also strongly recommend that any student interested in pursuing a career in intelligence research using neuroscience or other approaches read Jensen’s 1969 article. It is often cited, often misrepresented, and in my view, a classic work of psychology that still suggests important ideas and hypotheses to test with modern methods.
2.3 “Fraud” Fails to Stop Genetic Progress
Before moving on to modern advances in both quantitative and molecular genetics, we need to take one more historical side trip. Explaining this story also introduces basic strategies of quantitative genetics research. Following the 1969 article, another line of attack claimed that some of the genetic data Jensen cited to support his argument were fraudulent. These data came from identical twins reared apart as reported by Sir Cyril Burt, an eminent British psychologist in the mid-twentieth century.
The story of “fraud” begins with the undistinguished number .771. Here’s the background. Because monozygotic (MZ) twins, that is, identical twins, have 100% of their genes in common, any trait that was found in both twins was thought to have a genetic component. The more similar the twin pairs on the trait, the stronger the effect of genes. Of course, identical twins also share both the pre-natal and the post-natal environment, so the fact that identical twins may have quite similar intelligence test scores does not rule out that the similarity is due to having similar environments. Conceptually, this problem is easily addressed by comparing the similarity of a trait between pairs of identical twins who have 100% of their DNA in common to pairs of fraternal twins – that is, dizygotic (DZ) twins. Fraternal twins share most of their early environment but only 50% of their DNA, so any similarities should not be as strong in the fraternal twins as they are in the identical twins.
Indeed, this is the undisputed case in many studies of intelligence that collectively report average correlations for identical twins as about .80 and .60 for fraternal twins (Loehlin & Nichols, 1976). Adoption studies are even more powerful and compelling because they separate genetic and environmental influences more clearly than comparisons of identical and fraternal twins reared together. The Denmark Adoption Studies, for example, shifted the debate about the etiology of schizophrenia decidedly toward a genetic component because adopted children who had a biological parent with schizophrenia grew up with a higher risk of having schizophrenia than other adopted children with biological parents who were not schizophrenic. David Rosenthal was one of the principle investigators for the Denmark studies and I worked in his laboratory at The National Institute of Mental Health (NIMH) in my first job after graduate school. He once told me that although these studies did not elucidate much about schizophrenia other than a genetic component of some kind was involved, the beauty of the adoption study design was its simplicity. Basically, only two simple numbers counted. Anybody can see a higher rate of schizophrenia in the one group compared to the other, so in this case, it was hard to deny some role for genetics (although some anti-genetic critics certainly tried).
There are a relatively small number of well-done adoption studies of intelligence. These studies are quite difficult and complex to do because so many variables are difficult to control (e.g., age at adoption, age at intelligence testing, indexing similarities of environments in a quantitative way, the rate of participant dropouts from the study, no random assignment to environments). Nonetheless, results consistently report higher intelligence test score correlations between adopted children and their biological parents compared to correlations with adoptive parents. In fact, the correlations with adoptive parents are very low and even near zero (Petrill & Deater-Deckard, 2004), especially as children grow older (see Hunt, 2011, pp. 230–231 for an excellent summary), another observation hard to explain for critics who argue against genetic effects on intelligence. Interestingly, one recent adoption twin study reports higher IQs in adopted children compared to their non-adopted siblings, suggesting that enriched educational opportunity in the adoptive home leads to an increase of about 3–4 IQ points in early adulthood (Kendler et al., 2015). The study is noteworthy for the large sample of sibling pairs and a replication in a large sample of half-siblings. These samples were identified in Sweden, a country that registers such information systematically. This study suggests a small effect for the environment of the adoptive home and this finding does not contradict or impugn the heritability data in any way. The heritability studies always demonstrate that some environmental effects must be at work. Nonetheless, some caution is warranted because the IQ measure consisted of only four subtests used by the Swedish military. As noted in Chapter 1, all IQ scores are estimates of an underlying construct and small differences between groups are difficult to attribute to any causation.
An even more powerful design combines adoption and twins. Think about studying identical twins adopted away from their biological parents in early life and each one raised separately in a different family and exposed to different everyday environments, one twin often not even knowing of the existence of the other. Are identical twins reared apart still very similar to each other on things like intelligence test scores?
This brings us back to .771. In mid-twentieth-century Britain, Sir Cyril Burt did the first major studies of intelligence in identical twins adopted away from biological parents and reared in separate adoptive families. Over a number of years, Burt gave intelligence tests to pairs of identical twins who had been reared apart, an extremely rare group quite difficult to find and enter into a research study. He first mentioned a correlation of .77 in 15 pairs of identical twins reared apart (Burt, 1943), suggesting a strong genetic component to intelligence. Subsequently, he added six twin pairs and reported a correlation of .771 (Burt, 1955). His third report included 53 pairs with a correlation in the identical twins reared apart of .771 (Burt, 1966).
The three reports had different sample sizes ranging from 15 to 53 but each new, larger sample showed the same correlation of .771 (.77 in the first report). Burt’s results were a key element of Jensen’s 1969 argument. Critics of Jensen’s genetic view revisited Burt’s publications looking for possible flaws and .771 caught their attention. They argued that the same correlation value to three decimal places based on different sample sizes was statistically improbable. They concluded that Burt surely committed scientific fraud, and this example is still cited today to undermine the idea that genes are important for intelligence. In the wake of the fraud charges, Jensen, who knew Burt (who died in 1971), examined Burt’s original data files and found a number of serious concerns that he reported in detail (Jensen, 1974). Jensen was willing to exclude Burt’s data from his argument, but still maintained that other data supported a role for genetic influences on intelligence. Most independent investigations of Burt’s data doubt the claim of intentional fraud (Mackintosh, 1995). We may never know for sure, but the main point is this.
Subsequent twin studies done by different investigators around the world with large samples arrive at an average value for the correlation of intelligence scores among identical twins raised apart of .75 (Plomin & Petrill, 1997). Burt’s value was .77. For comparison, based on 19 studies ranging in sample sizes between 26 and 1,300 identical twin pairs, the average value for identical twins raised together is about .86 (see Loehlin & Nichols, 1976, table 4.10, p. 39). These values compare to the fraternal twin data (Loehlin & Nichols, 1976) that show average correlations for intelligence of about .60 based on pair sample sizes of 26–864. The overall story from twin and adoption studies has been apparent for
some time (Bouchard & McGue, 1981; Loehlin, 1989; Pedersen et al., 1992) and is nicely summarized in Figure 2.1 (Plomin & Petrill, 1997). In subsequent research, however, a rather dramatic new insight has emerged that further informs the pattern in Figure 2.1. We now know that age at time of intelligence testing makes quite a difference for heritability estimates. This is discussed in section 2.4.
Figure 2.1 Intelligence variance accounted for by genetics based on family, twin, and adoption data. T, reared together; A, reared apart; MZ, identical twins; DZ, fraternal twins; sib, sibling; PO, parent–offspring.
Reprinted with permission, Plomin and Petrill (1997).
Thus, the .771 “fraud” ends with recognition of overwhelming data from independent researchers that are fully consistent with Burt’s analyses, flawed as they may have been. Any single study, or any one researcher, can be flawed, but the basic conclusion that genes play an important role in intelligence is consistently supported by data from numerous studies of twins, adoptees, and adopted twins. This is an excellent example of looking at the weight of evidence (recall my three laws from the Preface: no story is simple; no one study is definitive; it takes many years to sort out conflicting and inconsistent findings and establish a weight of evidence). Many issues about the role of genetics in intelligence remain unresolved. For example, cross-sectional and historical data show that average IQ scores have consistently increased by about three points every decade around the globe. This is known as the Flynn Effect (Flynn, 2013; Trahan et al., 2014). Some critiques of the genetic role in intelligence argue that such an increase cannot be attributed to the slow pace of genetic evolution; they are correct. The increase, however, may not be a g effect (te Nijenhuis & van der Flier, 2013) and the causes are unknown, but its existence alone does not disprove a major role for genetic influences on intelligence.The weight of evidence summarized in this chapter leaves no reasonable doubt. Only extreme ideologues are still in denial. Data from more recent twin studies, described later in this chapter, expand the basic genetic findings to a new level of focus for neuroscience with the addition of DNA assessments. But let us not ignore that genetic studies also highlight a role for non-genetic factors with some surprising empirical observations.
2.4 Quantitative Genetic Findings also Support a Role for Environmental Factors
While the atmosphere surrounding intelligence research was still toxic and Burt’s data were under attack, a group of researchers at the University of Minnesota led by Professor Thomas Bouchard embarked on a new project to identify a large sample of identical twins reared apart. Ultimately, 21 years of searching (1979–2000) yielded 139 twin pairs from around the world who participated in the project. Some twins had no contact with each other until they were reunited in Minnesota, where all the twins completed an elaborate battery of tests for about 50 hours over a week. This included tests of intelligence, personality, attitudes, values, and many physical characteristics.
Genetic components were found for several personality traits like extroversion, and, surprisingly, even for some attitudes and values. However, these identical twins reared apart were most similar on intelligence scores with a correlation of .70 (Bouchard, 1998, 2009). When correlations are computed in identical twins reared apart, the correlation is also one way to estimate hereditability, so a correlation of .70 indicates that 70% of the variance in intelligence is due to genetic factors and 30% is not. Although this result from a large, careful study did not end all skepticism about a role for genetics, it started to temper many critics who were suspicious of Burt’s results and more inclined toward yet-to-be identified environmental factors. Like the impact of the Denmark Adoption Studies of schizophrenia on psychiatry, the Minnesota study began to shift the tide toward a renewed objective interest in genetic contributions to intelligence.
All of the twin and adoption studies of intelligence that demonstrate an important role for genes also are consistent in showing that genes do not account for 100% of the variance. So, an important consequence of the genetic studies is demonstrating that non-genetic factors must be involved in some way. Prior to the current interest in epigenetics and gene/environment interactions, there were attempts to apportion the contributions of genetic and non-genetic environmental factors. The most common view was about 50–50. However, there is considerable variability among studies regarding this proportion and there’s an interesting factor that accounts for much of this variability. The factor is the age when twins are tested (Haworth et al., 2010; McGue et al., 1993).
Based on cross-sectional data, in young twins 4–6 years old, the heritability of intelligence estimate is about 40%, and the heritability rises to a high of about 85% when the twins are older adults. In other words, the genetic influences on intelligence variance actually increase with age and environmental influences decrease. Note that cross-sectional means that different twin pairs participated in different studies at different times. Suppose we followed the same twins and retested them periodically as they got older. Would we see the same trend in such longitudinal data? The answer is yes. In a large Dutch twin study (Posthuma et al., 2003b), the same identical twins were given mental test batteries repeatedly over time to assess general intelligence. The heritability estimate of general intelligence was 26% at age 5, 39% at age 7, 54% at age 10, 64% at age 12, and starting at age 18 the estimate grew to over 80%. The increases could be due to several factors including more genes “turning on” with increasing age or gene–environment interactions. A detailed discussion of heritability estimation and genetic modeling is beyond the intention of this book, but see Hunt (2011, chapter 8) for a detailed presentation.
Here we are focused on an overview of genetic studies that provide a rationale for neuroscience approaches. Nonetheless, I want to present data that illustrate important findings about non-genetic factors that come from quantitative genetic studies. Until now, I have discussed environmental factors as a single category. One common quantitative genetic model divides environment into two categories: shared and non-shared factors. Shared environment is what it sounds like. Twins and siblings grow up in the same family, live in the same neighborhood, and attend the same schools. They have many shared general experiences that may influence intelligence. There are also many experiences unique to each person such as different friends, different classes and teachers. These unique influences are the non-shared environment.
In these models, genetic influences, shared and non-shared environmental factors together account for 100% of the variance in any characteristic like intelligence differences among people. The amount of variance attributed to each component can be distinguished and estimated statistically by comparing similarities of intelligence scores for identical twins, fraternal twins, and siblings, with samples from each group reared together and reared apart. Differences in intelligence test score correlations in these groups are used to estimate how much variance each of the three components contribute (Plomin & Petrill, 1997). Although this basic three-component model does not incorporate gene/environment interactions, it has provided important observations.
Let’s consider additional data from the Dutch twin study described at the end of the last section. Figure 2.2 shows the influence of genetics and both shared and non-shared environment on intelligence scores for people of different ages. The black part of the bar shows the genetic influence we have noted at the end of the last section. The white part of the bar shows shared environment, and the gray part shows non-shared environment influence. Shared-environment influences, the white bars, peak at age 5 and then decrease to virtually zero by age 16. Non-shared environment, the gray bars, has generally greater influence in the early years, but some non-shared influence continues through at least age 50. Note the sources of non-shared influences likely change over time.
Figure 2.2 Genetic, shared, and non-shared environment influences on intelligence variance for different age groups.
Reprinted with permission from Hunt (2011), based on Posthuma et al. (2003b).
<
br /> To recap this key piece of the genetic story, the heritability of general intelligence increases with age to about 80% by the end of teenage years and the effects of shared environment on intelligence decrease to near zero a bit earlier. These findings are extraordinary and among the most powerful and important in all of psychology. They are difficult to explain if you are convinced that genes are unimportant for intelligence. They also give pause to the idea that enriching childhood family experiences, as pleasant as they may be for many good reasons, has a lasting effect on the development of intelligence. However, these data also show that both shared and non-shared environment have effects at different developmental stages and in different amounts, especially before age 18. They are just not as strong as was once believed, but they demonstrate clearly that genes alone are not the whole story. All genetic researchers know that genes always express their function within an environmental context that may influence expression in many ways. Specific sources of these non-genetic effects on intelligence are not yet determined, just like specific genes are not yet identified, although general factors like schooling have some influence (Ceci, 1991; Ceci & Williams, 1997; Tommasi et al., 2015). As noted, part of the complexity is that environmental factors like social–economic status (SES) often are confounded with genetic factors for intelligence because intelligence plays a role in income and other factors that determine SES. This is discussed further in Textbox 2.1.
Nonetheless, by the beginning of the twenty-first century, the ascendance of the genetic view for intelligence was mirrored in three “laws” of quantitative genetics (Turkheimer, 2000): “First Law: All human behavioral traits are heritable. Second Law: The effect of being raised in the same family is smaller than the effect of the genes. Third Law: A substantial portion of the variation in complex human behavioral traits is not accounted for by the effects of genes or families.” Dr. Turkheimer discusses the implications of these laws in the context of new challenges for explaining how genetic and environmental influences may work. Recently, Plomin and Deary (2015) offered their own version of three laws: all traits show significant genetic influence; no traits are 100% heritable; heritability is caused by many genes of small effect. This chapter is not the place to discuss these “laws” in detail. I cite them here to emphasize the sea change in thinking about the role genes play in complex traits, virtually all of which have high heritability estimates according to a comprehensive meta-analysis that included almost every twin and adoption study ever conducted (Polderman et al., 2015). High heritability is a primary reason that neuroscience research on intelligence is expanding so quickly.