• Open Access


Why do children differ so much in their progress in the early school years? For the past half-century, environmental factors have been the prime focus, such as characteristics of schools (e.g., physical facilities, teacher training, discipline systems), neighborhoods (e.g., poverty, crime, pollution), and families (e.g., parental education, use of language, disciplinary practices). Far less attention has been given to the possibility of genetic influences on characteristics of children that affect academic learning (other than IQ) or, more intriguingly, genetic mediation of the effects of schools, neighborhoods, and families (Wooldridge, 1994). Decades of research on the nature and nurture of children's development in families have led to a consensus in developmental psychology that recognizes the importance of genetics as well as environment (Plomin, 2004). However, this fundamental issue of the interplay of nature and nurture has just begun to be addressed in relation to education (Plomin & Walker, 2003). One goal of this monograph is to consider the nature–nurture issue in the early school years in relation to individual differences in performance in reading, mathematics, and science as well as general cognitive ability, which we refer to as learning abilities and disabilities.

However, our main goal is to go beyond this rudimentary nature–nurture question to investigate three issues that have far-reaching ramifications for the field of education because they move toward the question of “how” rather than “how much.” All three issues concern the etiology of relationships between things: between the normal (learning abilities) and the abnormal (learning disabilities), between ages (7, 9, and 10 years), and between learning abilities and disabilities (English, mathematics, and science) as well as their relationship to general cognitive ability. In short, they are questions of cognitive developmental architecture. The first issue, the relationship between learning abilities and learning disabilities, requires a sample large enough to assess abnormal development in the context of normal development. The second issue, genetic and environmental influences on change and continuity, requires longitudinal data during the early school years. The third issue requires multivariate data on learning abilities and disabilities. Answers to all of these questions require a genetically sensitive design such as the twin method that capitalizes on the experiment of nature provided by identical and fraternal twins as described in Chapter II.

These requirements are met by the Twins' Early Development Study (TEDS), a large and representative sample of twins whose progress during the early school years has been assessed longitudinally at 7, 9, and 10 years (Oliver & Plomin, 2007; Trouton, Spinath, & Plomin, 2002). Some of these results have been reported previously in diverse literatures such as education, learning disabilities, and language in addition to child development (Oliver & Plomin, 2007). As is necessarily the case with longitudinal projects, these papers were published as data collection progressed during the course of the 10-year TEDS project, often at one age using different samples, models, and analytic strategies. Our goal here was to examine genetic and environmental influences systematically in univariate, multivariate, and longitudinal analyses that are based on the entire sample at all ages for all measures for both the low extremes (disabilities) and the entire sample (abilities). All of the analyses in this monograph are new and based on the same complete dataset and the same models and analytic strategies at 7, 9, and 10 years. We found that new interpretations emerged from comparisons across measures and ages that were not apparent in previous analyses focused on one measure or one age.

In this chapter, we provide a brief overview of what is known about the genetic and environmental origins of learning abilities and disabilities in the early school years. In addition to the basic issue of nature and nurture, we introduce the three themes of this monograph: the etiological relationship between the normal (abilities) and abnormal (disabilities), genetic and environmental contributions to longitudinal continuity and change, and multivariate analyses of genetic and environmental heterogeneity and homogeneity.

In this monograph, we use the phrase learning abilities and disabilities but not to indicate an a priori position on the issues of achievement versus ability and nature versus nurture. We view achievement and ability as a continuum from learning specific skills and content (e.g., learning to read) to using these skills and contents for comprehension and problem solving (reading to learn). Especially strong views are held on the use of appropriate labels for children's low performance, with the pros and cons debated for such labels as challenge, delay, difficulty, disorder, and impairment. We use the word disability with its semantic link to the word ability because research presented in this monograph suggests that common learning disabilities are the low end of the normal distribution of learning abilities (Chapter IV).

Finally, we recognize that there are many possible ways to address the questions that we raise in this monograph. Moreover, there are many other possible questions that can be asked about this dataset. For this reason, all of the data included in this monograph are freely accessible as a zipped SPSS file at the following web page: http://www.teds.ac.uk/information/SRCDdataset.htm. The academic and cognitive measures are included in standardized form (following adjustment for sex and age at time of assessment as in our genetic analyses) and unstandardized form. Only data described in this monograph are included in the dataset and these data are only available at the level at which they are analyzed in this monograph. For example, in this monograph we analyze data at the level of scales rather than individual items. It is our hope that Chapter II will provide adequate annotation for use of the dataset.


Adoption and twinning provide naturally occurring experimental situations that illuminate the relative influence of nature and nurture on specific traits and on the relationship between traits. Although each method has its own limitations (described in detail elsewhere, e.g., Plomin, DeFries, McClearn, & McGuffin, in press), these limitations are generally complementary. Consequently, the convergence of results from the two methods provides strong evidence for the validity of the findings. This section provides a brief introduction to the logic of the twin method; more information on statistics and estimates is presented in Chapter II and in the following chapters on results. The method is based on comparison between identical and nonidentical twins. Identical or monozygotic (MZ) twins derive from one zygote and are genetically identical. If genetic factors are important for a trait, these genetically identical pairs of individuals must be more similar than first-degree relatives, who are only 50% similar genetically on average. The best comparison group among first-degree relatives for MZ twins is dizygotic (DZ) twins, who develop from separately fertilized eggs. Half of DZ twins are same-sex pairs and half are opposite-sex pairs. Like MZ twins, DZ twins experience together most prenatal and many postnatal experiential variables such as prenatal nutrition and family social class.

If a trait is influenced genetically, identical twins must be more similar than fraternal twins. However, when greater similarity of MZ twins is found, it is also possible that some or all of the greater similarity is caused environmentally rather than genetically. The equal environment assumption of the twin method assumes that environmentally caused similarity is roughly the same for both types of twins. If the assumption were violated because identical twins experience more similar environments and consequently develop more similarly than nonidentical twins, this violation would inflate estimates of genetic influence. There is, in fact, evidence that MZ twins are treated more similarly than their DZ counterparts. For example, as children, MZ twins are more likely to have the same playmates, share the same room, and dress alike. As adults, MZ twins are more likely to keep in contact than are same-sex DZ twins (Evans & Martins, 2000). However, the equal environment assumption would only be violated if this greater similarity for MZ twins leads to a greater similarity for phenotypes of interest. The equal environments assumption has been tested in several ways and appears reasonable for most traits. For example, environmental similarity during childhood does not predict twin similarity in personality, attitudes, intelligence, nor a range of psychiatric disorders (Evans & Martins, 2000). Moreover, both greater similarity of parental treatment of MZ twins and greater physical similarity between MZ twins are uncorrelated with twin similarity for personality, vocational interests, and cognitive abilities.

Another potential violation of the equal environment assumption, in the opposite direction from that just discussed, would occur if identical twins experience greater environmental differences than fraternal twins, such as greater prenatal competition. To the extent that identical twins experience less similar environments, the twin method will underestimate heritability. Despite some potential limitations, the twin study remains the best method for assessing the relative contribution of genes and environment to traits in human populations (Evans & Martin, 2000). However, it is important to remember that statistics derived from twin data, which estimate genetic influences (heritability) and environmental influences, have very specific definitions within the twin method (see Chapter II), and can be misinterpreted. For example, heritability refers to effect size, the extent to which individual differences for the trait in the population can be accounted for by genetic differences among individuals. Effect size in this sense refers to individual differences for a trait in the entire population, not to the effect of genetic factors on a specific individual (Plomin et al., in press). In other words, heritability is the proportion of phenotypic variance that can be accounted for by genetic differences among individuals. Like all statistics, heritability estimates include error of estimation, which is a function of the effect size and the sample size. Therefore, as with other methods, replication across studies and across designs is crucial.

It should be emphasized that heritability refers to the contribution of genetic differences to observed differences among individuals within a specific population, for a particular trait, and at a particular time. Moreover, heritability describes what is in a particular population at a particular time rather than what could be. That is, if either genetic influences change (e.g., changes due to migration) or environmental influences change (e.g., changes in curricula or in educational opportunity), then the relative impact of genes and environment will also change. Even for a highly heritable trait such as height, changes in the environment could make a big difference. For example, if an epidemic struck or if children's diets were altered for the worse by famine, average height would decrease, but genetic influence might actually increase due to diminished environmental variance.

We also emphasize that the causes of individual differences within groups have no implications for the causes of average differences between groups. Specifically, heritability is defined, both conceptually and statistically, as the genetic contribution to differences among individuals within a group. Differences between groups may have quite different causes, which are difficult to evaluate rigorously; twin studies have little use here. Finally, it is important to remember that genetic influence on behavior involves probabilistic propensities rather than predetermined programming.

Much of the research reported in this monograph takes advantage of important extensions of the basic, univariate twin method. The most important is the development of multivariate methods (Martin & Eaves, 1977). The univariate approach just described estimates the genetic and environmental contribution to the variance in a specific trait. Analogously, multivariate methods estimate the genetic and environmental contribution to the covariance, or correlation, between two traits. Many aspects of behavior and development are known to be phenotypically correlated; however, such correlations might be the result of shared genetic influence or environmental influences on both. Distinguishing those influences provides valuable insight into the mechanisms underlying each. Like univariate analyses, multivariate analyses contrast correlations for MZ and DZ twins, where the magnitude of the discrepancy between them indexes a genetic effect, and the magnitude of the correlations regardless of zygosity indexes a shared environmental effect. But in multivariate analyses, the relevant correlation is the cross-trait twin correlation, that is, correlating measure A for twin 1 with measure B for twin 2. Multivariate analyses also provide an estimate of the degree to which the same genetic influences are at play for two traits. A special case of multivariate analysis of particular interest to developmental science is longitudinal analysis, where measure A and measure B (possibly the same measure, possibly a different one) are obtained at different time points. Longitudinal genetic analyses estimate genetic and environmental contributions to continuity and change. The assumptions underlying the twin method, and the qualifications concerning interpretation, apply to multivariate and longitudinal as well univariate analyses.


In this section, we focus on individual differences in learning abilities throughout the normal distribution. In the next section, we review research relevant to low performance because the genetic and environmental etiology of abilities and disabilities can differ; this issue is the focus of Chapter IV.

The first twin study with test data on academic performance in childhood included 278 pairs of twins that ranged in age from 6 to 12 years (Thompson, Detterman, & Plomin, 1991). The published report indicated modest heritability (about 20% of the variance in test performance was accounted for by genetic influences) and substantial (about 60%) shared environmental influence (i.e., environmental effects shared by the twins), but the measures had not been corrected for age. Age correction is necessary because members of a twin pair are exactly the same age; failure to correct for age inflates estimates of shared environment (see Chapter II). With age correction, the results of this study suggest moderate heritability (about 40%) and moderate shared environmental influence (about 40%) (L. A. Thompson, personal communication, June 21, 2006).

Three other twin studies of a broad range of academic abilities have been reported for older children. The classic study in this area included bright high school–age twins in the United States, using data obtained from the National Merit Scholarship Qualifying Test for 1,300 MZ and 864 DZ twin pairs (Loehlin & Nichols, 1976). For English and mathematics, MZ twin correlations were about .70 and DZ correlations were about .50, again suggesting moderate heritability (about 40%) and shared environmental influence (about 30%). The second study, which included 190 twin pairs assessed on a Dutch national test of educational achievement at 12 years, reported greater heritability (about 60%) and similar shared environmental influence (about 30%) (Bartels, Rietveld, van Baal, & Boomsma, 2002b). The third study yielded yet another pattern of results (Wainwright, Wright, Luciano, Geffen, & Martin, 2005b). The study of 390 pairs of twins from 15 to 18 years reported substantial heritability (about 60%) and modest shared environmental influence (about 10%). However, rather than assessing achievement in particular subjects such as English, the tests used in this latter study assessed general cognitive abilities such as “comprehension of facts from a broad range of stimuli” and “deduction and induction among relationships” (p. 603).

Given how diverse the studies are in samples, ages and measures, their results are surprisingly consistent in suggesting at least moderate heritability (about 50% on average) and shared environmental influence (about 25%). All these studies were based on tests administered to the twins. Another result relevant to findings in this monograph comes from an early Swedish study of a thousand pairs of 13-year-old twins based on report card grades (Husén, 1959). Results for reading, writing, and arithmetic were similar: The average heritability was 50% and the average estimate of shared environmental influence was 30%. The consistent evidence for shared environmental influence would seem unremarkable except for the striking fact that little evidence has been found for shared environmental influence in other domains of behavioral development such as personality or psychopathology (Plomin, Asbury, & Dunn, 2001).

Reading has received the most attention among academic abilities in genetic research. The major twin study of reading is a Colorado study that focused on reading disability but also included a control sample of twins (Light, DeFries, & Olson, 1998). For the control sample of 223 pairs of twins from 8 to 20 years of age, individual differences in reading ability yielded moderate heritability (about 40%) and modest shared environment (about 25%). A review of five smaller twin studies of various measures of reading ability in childhood also suggests an average heritability estimate of about 40% but the shared environment estimate was much higher, about 45% (Stromswold, 2001). Two more recent studies included many measures of early reading, although the sample sizes were modest (Byrne et al., 2005; Petrill, Deater-Deckard, Thompson, Schatschneider, & DeThorne, in press). The studies found diverse results across measures but generally suggested moderate genetic and shared environmental effects.

For mathematics, the only genetic research other than the three studies mentioned above comes from the Colorado study of reading, which also included tests of mathematics. High heritability (69%) and negligible shared environmental influence (6%) were reported for mathematics ability (Light, DeFries, & Olson, 1998); a latent variable analysis yielded even higher heritability and negligible shared environment (Alarcón, Knopik, & DeFries, 2000). For science, the only genetic study is the study of bright high school students mentioned above which reported heritability of 40% and shared environment of 30% for a measure of critical reading of scientific material (Loehlin & Nichols, 1976). Although science is not one of the traditional educational domains—reading, writing, and arithmetic—science has increasingly become a focus for education. For example, in the United Kingdom, science became a compulsory subject in elementary teaching in 1989 with the introduction of the National Curriculum.

In contrast to the meager previous research on academic learning abilities, a massive amount of research has been conducted on general cognitive ability (“g”), which refers to the observed positive manifold among different cognitive (verbal and nonverbal) tasks (Plomin & Spinath, 2004). This research has been reviewed many times, including an influential review in Science (Bouchard, Jr. & McGue, 1981). An updated review yielded an average MZ twin correlation of .86, which is near the test–retest reliability of the measures, in contrast to the DZ correlation of .60 (Plomin & Petrill, 1997). This pattern of twin correlations again suggests heritability of about 50% and shared environmental influence of about 30%. Meta-analyses including all of the family, adoption and twin data on “g” also yield heritability estimates of about 50% (Chipuer, Rovine, & Plomin, 1990; Devlin, Daniels, & Roeder, 1997; Loehlin, 1989). Similar results continue to be found in more recent twin studies (Benyamin, Wilson, Whalley, Visscher, & Deary, 2005; Rietveld, Dolan, van Baal, & Boomsma, 2003; Wainwright et al., 2005b). However, this overall conclusion averages out two important developmental changes, as discussed later.


Even fewer genetic studies have addressed the nature and nurture of learning disabilities. It cannot be assumed that low performance is influenced quantitatively and qualitatively by the same genetic and environmental factors responsible for the normal distribution of variation in learning abilities. The same issues about etiology are relevant to the origins of high ability but they are beyond the scope of the present paper and have only been addressed in relation to high “g” (e.g., Ronald, Spinath, & Plomin, 2002).

In fact, twin studies of learning disabilities suggest results roughly similar to those for learning abilities. For example, a review of twin studies of learning disabilities reported twin concordances (the likelihood that one twin will be affected if the other twin is affected) of 75% for MZ twins and 43% for DZ twins for language disability and 84% and 48%, respectively, for reading disability (Stromswold, 2001). For mathematics disability, the concordances are about 70% for MZ twins and 50% for DZ twins (Oliver et al., 2004). No twin studies of low performance in the sciences have been reported.

Treating low performance categorically, that is, analyzing twin data dichotomously as normal versus not normal, loses information about quantitative variation in the normal distribution. As explained more fully in Chapter II, we have emphasized an analysis called DF extremes analysis that combines qualitative information about probands' low performance with quantitative variation in their cotwins. Using DF extremes analysis, a review of twin studies that reported results for both learning disabilities and abilities found that the average weighted “group” heritability was .43 for language disabilities and .25 for language abilities; .52 and .63 for reading disabilities and abilities, respectively; and .61 and .63 for mathematics disabilities and abilities (Plomin & Kovas, 2005). In these analyses, group heritability refers to genetic influence on the average difference between the low-performing group and the rest of the population. However, most of these were small studies that make it hazardous to compare the magnitude of genetic influence for disabilities and abilities, a comparison that makes daunting demands in terms of sample size for adequate statistical power. Despite the large number of twin studies of individual differences in “g,” there are scarcely any twin studies on low “g” or mental retardation (Spinath, Harlaar, Ronald, & Plomin, 2004).

One of the goals of the present monograph is to investigate genetic and environmental influences on learning abilities in a large and representative sample, which is the focus of Chapter III. For the first time, we systematically compare estimates of genetic and environmental influences on individual differences across the full distribution of ability with estimates of those influences for low-performing children within the same sample assessed on the same measures at the same ages. The first question is whether the magnitude of genetic and environmental influences is similar for learning abilities and disabilities. However, as explained in Chapter II, even if the magnitude of genetic and environmental influence is the same for disability and ability, completely different genetic and environmental factors could be responsible for the genetic and environmental influence. A feature of DF extremes analysis is that, by combining qualitative information about proband status and quantitative variation in cotwins, it can clarify genetic and environmental links between the abnormal and the normal. This second question is the focus of Chapter IV.


To what extent do genetic and environmental influences on learning abilities and disabilities change during development? There are two questions here—a question about quantitative differences in the magnitude of genetic and environmental influences and a question about qualitative changes in genetic and environmental influences. The first question about quantitative differences can be addressed with cross-sectional data. The second question about qualitative changes from age to age requires longitudinal data.

Concerning quantitative age differences, genetic research on “g” has yielded two fascinating developmental trends. First, heritability increases linearly from about 20% in infancy, to about 40% in middle childhood, to about 50% in adolescence and young adulthood, and even higher in middle age (Boomsma, 1993; McGue, Bouchard, Jr., Iacono, & Lykken, 1993; Plomin, 1986). The cause of this developmental increase in heritability is not known but one possibility is that as children increasingly make their own way in the world they move from experiencing environments largely created by other people to actively creating correlations between their genetic propensities and their experiences (Plomin & DeFries, 1985). Second, shared environmental influence decreases sharply from about 30% in childhood to near 0% in adolescence, perhaps as adolescents increasingly live their lives outside their family. To the extent that academic achievement reflects “g,” similar developmental trends would be expected for learning abilities and disabilities.

Although there are few studies of learning abilities, and their measures and samples differ considerably, the results reviewed in the previous section on learning abilities suggest a trend in this same direction. The only study in the early school years (middle childhood) yielded estimates of 40% heritability and 40% shared environment (Thompson et al., 1991). In early adolescence, two studies yielded average estimates of about 55% heritability and 30% shared environment (Bartels et al., 2002b; Husén, 1959). In late adolescence, two studies yielded average estimates of about 50% heritability and 20% shared environment (Loehlin & Nichols, 1976; Wainwright, Wright, Geffen, Luciano, & Martin, 2005a; Wainwright et al., 2005b). Nonetheless, although such cross-sectional comparisons across studies with different samples and measures can provide rough estimates of developmental differences in genetic and environmental influences, what is needed for precise comparisons is a longitudinal study with the same samples and measures at each age. It should also be noted that comparing genetic and environmental estimates across ages requires large samples. For example, in a sample of 200 pairs of twins, a heritability estimate of 40% is surrounded by a 95% confidence interval of 5–70%, which means that it has no power to compare heritability estimates with another study. Twin studies are especially underpowered to detect and compare estimates of shared environmental influence (Hopper, 2000).

The second question about qualitative changes in genetic and environmental influence from age to age requires longitudinal data. Rather than asking how much do genetic and environmental factors affect performance, here we are asking a logically independent question: To what extent are the same genetic and environmental factors influential at different ages? Only one genetically informative study has examined reading longitudinally over more than a 1-year interval. In the Colorado Adoption Project, word recognition was examined at 7, 12, and 16 years in a sample of adoptive and nonadoptive sibling pairs (Wadsworth, Corley, Hewitt, & DeFries, 2001; Wadsworth, Corley, Plomin, Hewitt, & DeFries, 2006). Longitudinal genetic analysis (see Chapter II) indicated that genes were largely responsible for the substantial stability from age to age. Moreover, genetic correlations from age to age—an index of the extent to which it is the same genetic factors that are operative across age—were 1.0 indicating that the same genetic factors affect reading performance from childhood to adolescence. (See Chapter II for descriptions of these analyses.) These rare data for adoptive and nonadoptive siblings are especially important because, unlike twin analyses, adoptive sibling correlations provide a direct test of the importance of shared environmental influence. The results indicate that, although shared environmental influence accounted for only 10% of the total variance in word recognition, all of this shared environmental influence contributed to continuity from age to age. Nearly all of the change from age to age could be attributed to nonshared environment, that is, environmental effects that are distinct for the siblings, not shared.

Two other longitudinal studies of early reading are in progress but as yet have only reported longitudinal analyses from kindergarten to first grade (Byrne et al., 2006; Byrne et al., 2005) or from first to second grade (Petrill, Deater-Deckard, Thompson, Schatschneider, & DeThorne, in press). These twin studies also suggested substantial genetic stability. They yielded mixed results concerning shared environmental influence, as expected given the confidence intervals surrounding twin study estimates of shared environment mentioned above, but on balance the studies suggest that shared environmental influences contribute to stability and that nonshared environment is largely responsible for change. We look forward to future reports from these two studies because they include diverse measures of reading and language-related skills such as phonological awareness, rapid automatized naming, and spelling.

We are aware of no longitudinal studies of learning abilities other than reading and none for learning disabilities. Similar to the studies of reading, longitudinal studies of “g” indicate substantial genetic stability in childhood (Bartels, Rietveld, van Baal, & Boomsma, 2002a; Petrill et al., 2004), adulthood (Loehlin, Horn, & Willerman, 1989), and even late in life (Plomin, Pedersen, Lichtenstein, & McClearn, 1994). Also similar to reading, “g” shows less shared environmental influence, but to the extent that shared environment can be detected it appears that it is largely stable from age to age. Change from age to age is due to nonshared environment.

Chapter V presents TEDS results that address these two issues of quantitative age differences and qualitative age changes at 7, 9, and 10 years for learning abilities and, for the first time, for learning disabilities.


The third way in which the present monograph goes beyond the basic nature–nurture question is to investigate genetic and environmental links between learning abilities. For example, to what extent do genes that affect reading ability also affect mathematics? In contrast to univariate genetic analysis that focuses on genetic and environmental contributions to the variance of a single variable, multivariate genetic analysis investigates the covariance between variables and estimates the extent to which genetic and environmental factors that affect one variable also affect other variables. (Chapter II describes multivariate genetic analysis.)

The surprise from the few extant multivariate genetic analyses of learning abilities is that genetic correlations are high, which suggests that the same genes affect different abilities. In a recent review, genetic correlations varied from .67 to 1.0 for reading versus language (five studies), from .47 to .98 for reading versus mathematics (three studies), and from .59 to .98 for language versus mathematics (two studies) (Plomin & Kovas, 2005). The average genetic correlation is about .70, which can be interpreted to mean that when genes are found that are associated with one learning ability such as reading there is about a 70% chance that the genes will also be associated with other learning abilities such as mathematics. There is only one small multivariate genetic study of learning disabilities and it reported a genetic correlation of .53 between reading disability and mathematics disability (Knopik, Alarcón, & DeFries, 1997). If genetic correlations are so high between learning abilities, it makes sense to expect that components within each learning domain (e.g., read words vs. reading nonwords) are also highly correlated genetically, and that is the case. Genetic correlations range between .60 and .90 within each of the domains of language, reading, and mathematics (Plomin & Kovas, 2005). Multivariate genetic research on cognitive abilities such as verbal, spatial, and memory abilities also consistently find genetic correlations greater than .50 and often near 1.0 across diverse cognitive abilities, including basic information processing measures (Deary, Spinath, & Bates, 2006). This genetic overlap across cognitive abilities becomes stronger later in the life span (Petrill, 2002). Phenotypic correlations among diverse tests of cognitive abilities led Charles Spearman in 1904 to call this general factor “g” in order to avoid the many connotations of the word intelligence. To what extent do genes for “g” overlap with genes for specific learning abilities such as reading? A review of a dozen such studies concludes that genetic correlations between learning abilities (mostly reading) and “g” are substantial but somewhat lower than the genetic correlations among learning abilities (Plomin & Kovas, 2005), which is consistent with a paper published since this review (Wainwright et al., 2005a, 2005b). This result suggests that most (but not all) genes that affect learning abilities are even more general in that they also affect other sorts of cognitive abilities included in the “g” factor.

Multivariate genetic analysis also provides information on shared and nonshared environmental links between abilities. The first multivariate genetic analysis of learning abilities in childhood was subtitled Genetic Overlap but Environmental Differences because it found a genetic correlation of .98 between reading and mathematics but a nonshared environmental correlation of .28 (Thompson et al., 1991). Other multivariate genetic analyses tend to be consistent with the conclusion that nonshared environments are specialists (Kovas & Plomin, 2007).

These multivariate genetic results led to the development of a theory called “generalist genes,” which proposes that the same set of genes affects individual differences in diverse learning and cognitive abilities (Plomin & Kovas, 2005). If true, the generalist genes theory would have widespread implications for molecular genetics, cognitive neuroscience, and education (Kovas & Plomin, 2006). However, the theory is based on a fragile foundation of a few small and diverse studies, especially for learning abilities. In particular, larger studies are needed because multivariate genetic analysis is especially demanding in relation to statistical power (Rhee, Hewitt, Corley, & Willcutt, 2005). Chapter VI presents multivariate genetic analyses using the large TEDS sample that investigate genetic and environmental links within each domain of learning abilities (e.g., reading words vs. nonwords), between domains of learning abilities (e.g., reading vs. mathematics), and between learning abilities and “g.”


Chapter VII discusses our findings in relation to the three themes of this monograph: the etiological relationship between the normal (learning abilities) and the abnormal (learning disabilities), genetic and environmental contributions to stability and change from 7 to 10 years, and genetic and environmental heterogeneity and homogeneity within and between learning abilities as well as their relationship to general cognitive ability. These three themes go beyond the fundamental nature–nurture question, but in Chapter VII we also return to more general issues related to nature and nurture, including some limitations of our study, findings that surprised us and some puzzles that remain, and implications of this research for theories of education and child development.