The Genetics of Intelligence

There were an endless debate about whether IQ heritability was spuriously inflated, while the growing evience, notably from modern techniques (e.g., GCTA), shows a non-negligible heritability in the narrow sense. A large part of the twin method estimates has been confirmed.

To define it, the heritability is just the proportion of the phenotypic variance in a trait that is attributable to genetic variance, or the genetic variance divided by the total variance of the phenotype. There is broad, and there is narrow heritability. The first one involves additive and non-additive effects, as is the case in the twin methods, notably because of assortative mating causing an artificial rise in additive genetic variance due to children receiving correlated genetic influences from their parents, which means that broad heritability includes some interaction effects. The narrow heritability however is the ratio of additive genetic variance to phenotypic variance, which involves additive genetic effects that are caused by independent effects of alleles, that add up consequently in their effect on a trait, and thus, additivity means that genes and environments act separately. Or, to cite Jensen (1998, pp. 175-176) :

Narrow heritability includes only the additive genetic variance VA, that is, the part of the total genetic variance responsible for the resemblance between parent and offspring. The fixable component of the additive variance is the only part of it that “breeds true” (hence it was referred to by R. A. Fisher as the essential genotype). Therefore, it is only the fixable part of the additive genetic variance that affords the leverage for selective breeding, which can occur either by natural selection or by artificial selection by animal and plant breeders. The nonfixable part of the additive variance results from assortative mating (some degree of genetic correlation between parents on a specific trait). [7] (The coefficient of assortative mating [also called spousal correlation] for IQ in our present population is between +.40 and +.50.)

Broad heritability includes all sources of genetic variance. Besides the additive variance (which is the largest part of broad heritability), there is genotype X environment (GE) interaction: Different genotypes may react differently to the same environmental factor; an environmental condition that is favorable to the-phenotypic development of a certain genotype is less effective or even unfavorable for a different genotype. Also there is genotype-environment (GE) covariance, or the correlation between genetic and environmental factors that affect the development of the phenotype: Genotypes that are more favorable than average for the development of a trait are found with greater-than-chance frequency in environments that are also more favorable than average; likewise for genotypes and environments that are less favorable.

Then there is the nonadditive genetic variance. It results from two types of genetic interactions: (1) genetic dominance (e.g., a dominant and a recessive allele paired at the same locus on the chromosomes might have the same phenotypic effect as two dominant alleles at that locus), and (2) epistasis (a gene at one chromosomal locus affects the phenotypic expression of a gene at some other locus). Dominance and epistatis cause lower correlations between directline relatives (parents-offspring and full siblings) than would be the case if purely additive genetic effects were the only source of genetic variance.

Now, as an example of heritability in the narrow sense, Davies et al. (2011) estimated in a large sample (N=3,511) that crystallized (gc) and fluid (gf) intelligence are heritable (40%, and 51% respectively) and polygenic, establishing that the heritability of a complex trait like intelligence is due to a multitude of genes of very small, tiny effects. They report that the proportion of the explained variance for either gc or gf was correlated with the length of the chromosome, “We subsequently partitioned additive genetic variation to individual chromosomes using the software package GCTA, fitting all chromosomes simultaneously, and found that, on average, longer chromosomes explain more variation (Figure 2)”. Their estimates must be considered as being the lower-bound perhaps because their technique (GWAS) that captures the variance in a trait that is due to linkage disequilibrium (LD) between the genotyped SNPs and the unknown causal variants, does not detect all of the genetic variance due to causal genes of low frequency, as Visscher et al. (2010) stated : “If most causal variants for human height have such low frequency in the population that they are not in LD with the (common) SNPs on the commercial SNP arrays then the method we used would not detect much more additional variance than already accounted for by the published genome-wide significant loci”. This study, of course, weakens the Dickens-Flynn model (2001), which postulates that the strong heritability of IQ is explained by the gene-environment correlation (rGE), translating in a positive (or negative) feedback loop. In other words, their model implies non-additive effects.

Intelligence, Highly Heritable and Polygenic - Figure 2

The robustness of these estimates has been advanced by the authors :

What do our results imply about the heritability of intelligence? If our estimated relationships had been based on all causal variants instead of being derived from SNPs that may be in LD with such variants, then we would have had an unbiased estimate of the full narrow-sense heritability. Therefore, our estimates provide a lower bound for the narrow-sense heritability, due to imperfect LD between the genotyped SNPs and unknown causal variants. Our estimates are based upon realized relationships [i.e., the actual amount of genome sharing directly estimated from the SNP data] between very distant relatives and not on pedigree relationships [i.e., the expected relatedness inferred from the family pedigree] between close relatives. This breaks up a possible correlation (confounding) between genetic and environmental factors, since the variation in realized relationships given pedigree relations is large for distant relatives. Our estimates of the phenotypic variance explained by all SNPs are ~0.4–0.5, and we therefore conclude that the narrow-sense heritability for human intelligence is large and consistent with the inference from twin and family studies.

Davies et al. (2011) study has been replicated by Chabris et al. (2012). They applied the same procedure and estimate that about 630 000 SNPs are responsible for 47% of the variance in g, in the Swedish Twin Registry sample. These results corroborates the idea that g is a highly polygenic trait for which common genetic variants have, individually, modest effects. Chabris tried to replicate some earlier studies on the significance of the SNP-g association, with no success, but they concluded that the most likely reason for this was because of the insufficient sample size, and not because of any error in design or execution. That is the inherent problem of a polygenic trait, since very large sample size is required to reach the statistical power needed to detect small effect sizes. Nevertheless, according to newscientist, it seems recently that a single gene, HMGA2, has been found to substantially alter IQ.

Plomin et al. (2013), also applied a method called Genomewide Complex-Trait Analysis (Yang et al., 2011), GCTA, on a sample (N=3,154) of 12-years old twins. Unlike the twin-method estimates that capture both additive and non-additive genetic effects, the GCTA adds up the effect of each SNP. This method only captures additive effects. In short, the technique consists in estimating the genetic influence on a trait (e.g., intelligence) “by predicting phenotypic similarity for each pair of individuals in the sample from their total SNP similarity”. They estimate by that means all the genetic variance accounted for by all the SNPs that have been genotyped. This method requires a very large sample, so that the numerous genes associated with cognitive abilities can be detected by using the common SNPs on genotyping arrays. The ratio of GCTA/twin heritability is about 0.76 for general cognitive ability (and 0.48 for non-verbal ability). That means GCTA accounts for a great portion of twin heritability.

… GCTA provides a lower-limit estimate of heritability because it misses genetic influence due to causal variants that are not highly correlated with the common SNPs on genotyping arrays.

… As mentioned earlier, one possible explanation of the missing heritability is that rare genetic variants have not been considered in addition to the common SNPs that are detected by available DNA arrays.

These results suggest that research using current DNA arrays with their common SNPs could identify genes that account for about two thirds of the heritability of cognitive abilities simply by including larger samples. But why is the cup only two-thirds full? Accounting for the rest of the missing heritability is likely to require other DNA variants not well tagged by the common SNPs on current DNA arrays (Gibson, 2012). Although such data are not currently available, this situation will eventually be resolved by whole-genome sequencing data (Plomin, 2012). Until then, researchers need to consider the possibility that twin heritability estimates are inflated. One argument against this possibility is that twin-based heritability estimates for cognitive abilities are in line with estimates from adoption studies and family studies, even though the adoption and family designs have different assumptions than the twin design does (Plomin et al., 2013).

There is no doubt of course that IQ heritability in adulthood is higher than 40% or 50%, approaching 70%. On can argue that the rise in heritability is mostly due to active gene-environment (GxE) correlation, that is, individuals build their own environment on the basis on their genotype. An alternative hypothesis is the genetic amplification, which consists in that same genetic influences at an early age will become increasingly important in later age, as they amplify with development, and thereby, the variance explained by earlier genetic influences has become higher. It is said that the high genetic correlation will persist from childhood to adulthood, even with an increase in heritability. Note that amplification can actually occur even with a drop in heritability (e.g., van Soelen, 2012, p. 3873). Plomin (1986) explains the theory in these terms :

Genes that affect IQ make only a small contribution to phenotypic variance at first, but their effects are amplified throughout development. Suppose, for example, that genetic differences among infants are responsible for differences among them in the formation of dendritic spines during the first few years of life and that the complexity of dendritic spines is related to information processing capabilities. At first, these structural differences do not have a chance to cause function differences because so little information has been processed at this point. Gradually, the functional differences are amplified as more and more information is processed by children. If we were to measure differences in categorizing ability early in childhood, the genetic differences among children due to the complexity of dendritic spines would contribute a negligible amount of variance to observed variability among children in categorization ability. The differences snowball as development proceeds, so that a study of the children when they are older will show more genetic variance. Yet the genetic correlation between the two ages is near unity because the genetic portion of observed variability at both ages originates with the same set of genes whose effect become amplified during development.

In any case, these genetic studies are important, because GxE theories imply that IQ heritability is somewhat inflated. But even so, an active GxE model should be seen as being a genetic variation because in this case, the individual creates his own environment on the basis of his genotype; we talk about self-realization (Rowe, 2003, pp. 79-80; Sesardic, 2005, pp. 93-95).

Given the genetic amplification, a concept that deserves to be treated is the genetic correlation, which could be interpreted as being the probability that a gene influencing one trait will influence another. Jensen (1980, pp. 193-195, see also Jensen, 2006, p. 128, on pleiotropy) dedicated a paragraph to this :

5. Genetic Correlation. Variables x and y may be correlated because of common or correlated genetic determinants. There are three kinds of genetic correlation that are empirically distinguishable by the methods of quantitative genetics: correlated genes, pleiotropy, and genetic linkage.

Correlated genes, through selection and assortative mating – segregating genes that are involved in two (or more) different traits, may become correlated in the offspring of mated pairs of individuals both of whom carry the genes of one or the other of the traits. For example, there may be no correlation at all between height and number of fingerprint ridges. Each is determined by different genes. But, if, say, tall men mated only with women having a large number of fingerprint ridges, and short men only with women having few ridges, in the next generation there would be a positive genetic correlation between height and fingerprint ridges. Tall men and women would tend to have many ridges and short persons would have few. Breeding could just as well have created a negative correlation or could wipe out a genetic correlation that already exists in the population. A genetic correlation may also coincide with a functional correlation, but it need not. Selective breeding in experimental animal genetics can breed in or breed out correlations among certain traits. In the course of evolution, natural selection has undoubtedly bred in genetic correlations among certain characteristics. Populations with different past selection pressures and different factors affecting assortative mating, and consequently different evolutionary histories, might be expected to show somewhat different intercorrelations among various characteristics, behavioral as well as physical.

Pleiotropy is the phenomenon of a single gene having two or more distinctive phenotypic effects. For example, there is a single recessive gene that causes one form of severe mental retardation (phenylketonuria); this gene also causes light pigmentation of hair and skin, so that the afflicted children are usually more fair complexioned than the other members of the family. Thus, there is a pleiotropic correlation between IQ and complexion within these families.

Genetic linkage causes correlation between traits because the genes for the two traits are located on the same chromosome. (Humans have twenty-three pairs of chromosomes, each one carrying thousands of genes.) The closer together that the genes are located on the same chromosome, the more likely are the chances of their being linked and being passed on together from generation to generation. Simple genetic correlation due to selection can be distinguished from correlation due to linkage by the fact that two traits that are correlated in the population but are not correlated within families are not due to linkage. Linkage shows up as a correlation between traits within families. (In this respect it is like pleiotropy.)

One thing must be said, as well, is that a genetic correlation of 1.0 simply means that both traits have the same genetic determinants, so that any variation in the changes in intelligence between two points in time would be of purely environmental origin (Deary et al., 2012). Less-than-unity correlations mean that genetic influences explain both IQ stability and IQ changes during the development (Bishop et al., 2003, pp. 44-45). Because shared environment explains a modest portion of IQ stability, nonshared environment a substantial portion of IQ instability, while g correlates positively with heritability and negatively with nonshared environment, it is predictable (Beaver et al., 2013, p. 436) that intervention would fail to improve g, as the evidence shows.

All this leads us to the generalist genes hypothesis (Kovas & Plomin, 2006a, 2006b), advanced by Robert Plomin, which consists in that the same genes that influence, say, school achievement, will also account for IQ test scores, or a cognitive area (math) correlated with another cognitive area (verbal) due to the same genetic influences. In this way, the correlation between these two variables is due to genetic effects. The two key concepts to understand the theory are pleiotropy (a single gene affecting several traits) and polygenicity (each trait is affected by a multitude of genes), the latter reinforcing the former. The theory received a considerable support from numerous studies on very large sample size (Plomin & Kovas, 2005; Plomin et al., 2007; Haworth et al., 2009; Docherty et al., 2010; Davis et al. (2009b); Calvin et al., 2012; Chow et al., 2013; Trzaskowski et al., 2013b, who used GCTA method). With regard to disability, which also displays non-trivial genetic correlations between language, reading and mathematics disabilities, Plomin et al. (2007) write :

A common reaction to this conclusion about generalist genes is disbelief because it goes against the common observation that specific disabilities exist. That is, some children with reading problems have no problem with mathematics and vice versa. If genes are generalists, why do specific disabilities occur? There are three reasons. First, genes are also specialists – genetic correlations are not 1.0. Second, nonshared environments are largely specialists (Plomin & Kovas, in press-b). Third, there is less specificity than it might seem. Even though reading and mathematics correlate phenotypically 0.65 in TEDS, some children with reading problems have no problems with mathematics and vice versa. However, this so-called double dissociation is to be expected on statistical grounds alone and has no bearing on the extent to which different causal processes affect reading and mathematics. A related issue concerning the acceptance of these findings is that, although genetic correlations between learning abilities are greater than their phenotypic correlations, we cannot see genetic correlations in the population in the way that we can see phenotypic associations and dissociations.

Posthuma (2003) and Deary (2006) propose a review of studies showing genetic correlation between g and brain volume, g and cognitive processing (reaction time, inspection time), and in particular the IQ-IT covariance (Luciano et al., 2005) as evidence that a pleiotropic gene model, unlike causal models, has been proved to be the best fitting model, in a large sample (N=2012), hence validating the genetic g hypothesis: “In short, there is no causal relationship between IT and IQ; instead, both processes/abilities are partially dependent on the same underlying cause, which analysis has shown to be genetic”. With regards with genetic g, there were strong genetic correlations between different cognitive domains (g factor, math, language, reading), usually higher than 0.50, sometimes approaching 1.0. From Deary et al. (2006, pp. 692-694) we read :

From infancy to adulthood: twins. An analysis of first to sixth grade twins (148 MZ, 135 same sex DZ) from the Western Reserve Twin Project suggested that, ‘abilities may be differentially affected by genetic and environmental variation. However, these differential patterns may be simply reflecting the degree to which specific abilities measure general intelligence’. Using 17 ability measures from the Wechsler Intelligence Scale for Children (WISC) and another test battery, they found that all the tests were influenced by genetic sources common to all tests: in other words, they found a genetic g. They also found some genetic effects that were specific to domains of cognitive functioning such as verbal, spatial, perceptual speed, and memory functions. Correlations between phenotypic g loadings and genetic g loadings were 0.88 and 0.76 for the two mental test batteries.

This was investigated further in a Dutch Twin Study in which 194 pairs took Raven’s Progressive Matrices (a test of nonverbal reasoning with a high g-loading) at age 16.1 years and the WAIS at age 17.6 years. The heritability estimates for Full Scale IQ, Verbal IQ and Performance IQ were 0.82, 0.84, and 0.68, respectively. There were no significant effects of shared environment. There were substantial unique environmental contributions, specific to each subtest. The principal interest from these data is the contribution to each subtest from genetic factors. This followed the hierarchical model of mental abilities and, thus, genetic contributions were divided into contributions shared by all tests, those shared by tests covering the same cognitive domain, and contributions to individual tests (Table 3). A general genetic factor contributed a mean of 30% of the variance to all tests (range 8–53%). Note, for example, that 48% of the variance in Raven scores comes from a genetic factor shared with all of the WAIS tests. There are modest contributions from genetic factors at the level of the cognitive domain and the individual test. The heritability of the individual tests ranges from 27 to 76%, with a mean of 56%. The contribution of unique environment to subtests ranges from 24 to 73% with a mean of 44%. The authors concluded: ‘the factorial structure of the WAIS subtests is determined by individual differences in genetic structure (phenotypic g is strongly related to genetic g)’ (p. 207); ‘The covariation among the WAIS subtests and the covariation between the subtests and the Raven in our data are predominantly influenced by a second-order genetic factor and thus strongly support the notion of a biological basis of g’ (p. 209).

Analyses of a Dutch Twin Study have also addressed the changing genetic contribution with age. Twins (N=209 pairs) were assessed by the RAKIT test battery at ages 5, 7, and 10 years, and on the WISC-R at age 12 years. For Full-scale IQ (general intelligence), the contributions (percent variance) were as follows at ages 5, 7, 10, and 12 years: genetics, 26, 39, 54, 64; shared environment, 50, 30, 25, 21 (for the latter three values, the 95% confidence interval includes zero); and unique environment, 24, 31, 21, 15. This decrease in the shared environmental contribution and increase in genetic influence with age from childhood to adolescence was congruent with previous studies. The best-fitting model showed an additive genetic influence which was a common factor, but with age-specific factor loadings; thus, ‘continuity in cognitive abilities is mainly due to additive genetic factors’ (p. 245). Shared environment contributed to continuity and change in cognition, and unique environment contributed to change in development. …

The increase in importance of genetic effects from infancy to childhood has also been demonstrated in longitudinal analyses of twin data from different research groups. For example, in data from 2824 twins analysed using a genetic longitudinal latent g model, heritability increased from 0.17 for a composite score across ages 2, 3 and 4 years to 0.47 at age 7 years. The same genes appeared to affect IQ across age. The term (genetic) ‘amplification’ has been used to describe this pattern of effects.

From infancy to adulthood: adoption studies. The Colorado adoption project included adopted children and their adoptive parents, and also their biological mothers and some biological fathers, as well as control parents and their children. Parents undertook a 3-h test battery, with cognitive, personality, and other assessments. Children were tested at ages 1, 2, 3, and 4 years in the home. At 7 and 12 years, they were seen in a lab. At 9, 10, and 11 years, they undertook a telephone interview. At age 4 years, the h² for specific cognitive abilities were: verbal=0.12; spatial=0.31; perceptual speed=0.21; and visual memory=0.06. These were not significantly different. The heritability of general mental ability increased over time, with the 1, 2, 3, 4, and 7 year h² estimates of 0.09, 0.14, 0.10, 0.20, and 0.36, respectively. A further report applied a Schmid–Leiman-type hierarchical model to the analysis of the genetic and environmental contributions to verbal, spatial, perceptual speed, and memory domains in the year 7 assessment data. A genetic g factor influenced all four domains, with additional domain-specific genetic influences on verbal, spatial, and memory domains. There were no significant shared environment effects; the nonshared environment effects were principally domain-specific, with a shared effect between spatial and memory domains. By age 12 years, with 175 adoptive families and 209 control families, the h² for ability domains derived from a mixture of WISC and Educational Testing Service tests was as follows: verbal=0.26, spatial=0.35, perceptual speed=0.38, memory=0.53. Genetic correlations between the ability domains ranged from 0.27 to 0.58. A simple model, which assumed that the genetic correlations among the four areas were identical, fitted well. Thus about half of the phenotypic association between the cognitive domains was caused by genetic factors and the authors concluded that, ‘specific cognitive abilities appear to be influenced by a pervasive genetic factor whose contribution to each ability does not differ substantially’ (p. 262). The effects of familial environment transmission were nonsignificant.

A more recent analysis of the Colorado Adoption Project asked, ‘what is the pattern of genetic and environmental influence on the stability of cognitive skills from early childhood through late adolescence’. There were 245 adoptive and matched control families. Children by that stage had taken cognitive tests at age 16 years (the WAIS). Phenotypic stability coefficients were moderate to high from age 2 years onwards. For example, the correlation between ages 7 and 16 years was 0.68, and between 12 and 16 years 0.80. At age 16 years, the mean correlation between adoptive siblings’ intelligence test scores was 0.11, and between control siblings was 0.30. Genetic sources were responsible for stability of general cognitive ability from age 1 years to age 16 years. For nonshared environment, only age-specific effects were required, suggesting that they contribute mainly to age-to-age instability or test-error. The mean of the genetic correlations between all ages from 2 to 16 years was 0.78 (range 0.57–1.0).

The Texas Adoption Project involves about 300 families in Texas who adopted children through a church-related scheme for unwed mothers. Children went to adopted homes within a few days from birth and were adopted permanently. Birth and adoptive parents tended to be middle class. Children took Stanford-Binet or age-appropriate Wechsler tests at around age 7 years, at which time adults, excluding birth fathers, took the Adult Wechsler and/or the Revised Beta test. Children were tested on the adult tests at a 10-year follow-up. The correlations of the Beta test between adopting fathers and mothers and their adopted children (with whom they had spent 17 years on average in the same home) were 0.08 and -0.02, respectively. Correlations between fathers and mothers and their biological children were 0.20 and 0.21, respectively. The correlation between the birth mothers and their adopted-away children was 0.33. Three of the six subscales of the Revised Beta exam had specific genetic contributions beyond a general genetic factor. A later analysis of the Texas Adoption Project examined both parent–offspring and sibling correlations. Sibling correlations were higher for biologically related siblings than for adopted siblings, whose scores correlated near to zero. The estimated additive genetic effect on general intelligence was 0.78, for true scores in the population. The authors concluded that, ‘The major contributor to familial resemblance is the genes. Shared family environment has an appreciable effect on IQ when children are small, but this becomes minor by the time they are late adolescents.’

Deary et al. (2012) also reported a rather strong genetic correlation (0.62) between IQ measured at age 11 and IQ measured at age 65-79 in a large scottish sample (N=1,940). In another longitudinal sample, Brant et al. (2009; pp. 401-402) report positive genetic correlations between IQs measured at age 1, 2, 3, 4, 7, 12, 16, that tend to increase with age, “signifying that the same genetic effects are evident from infancy to late adolescence, but that these influences increase in importance across development”. Lyons et al. (2009) show a phenotypic correlation for AFQT, a highly g-loaded test, measured at age 20 (N=7,232) and 55 (N=1,237), at about 0.74, for which the genetic correlation was 1.0. The fact that Lyons (2009) found no new genetic influences over the adult development is interesting. Brant et al. (2013) found that the genetic influence during the cognitive development comes both from previous, existing genetic influences and new influences; the authors also report a lower heritability for high-IQ people, and a higher heritability for lower-IQ people (N=11,000). Brant, nevertheless, seems to reject a commonly invoked theory for explaining the rise in heritability, namely, the active GxE correlational model :

The most prominent theory of developmental increases in the heritability of IQ posits that across development, individuals gain more scope to shape their own environments on the basis of their genetic propensities (active gene-environment correlation), which causes an increase in genetic influence over time (Haworth et al., 2010; Plomin, DeFries, & Loehlin, 1977). Our results challenge this explanation, as they show a later increase in heritability for individuals of higher IQ. To explain our results in the context of active gene-environment correlations, one would need to posit, counterintuitively, that higher-IQ individuals seek out environments concordant with their genetic propensities later in development than do lower-IQ individuals.

The plausibility of their hypothesis is supported by the fact that IQ differences certainly reflect differences in mental age (Jensen, 1980, pp. 559-562, 1998, pp. 370-371). The authors are still open to the possibility of the amplification model which is, at present, perhaps the most likely and tenable hypothesis. van Soelen et al. (2011) confirmed the amplification model in a pediatric longitudinal sample, consistent with Hoekstra et al. (2007, p. 112) analysis, in which no clear pattern of GxE interaction has been found and in which they show a correlation between verbal and nonverbal abilities that is entirely explained by genetic effects, with stability in nonverbal ability attributed completely to genetic influences between age 5 and 18 (pp. 105-109). Subsequently, Leeuwen et al. (2008, pp. 78, 84-85, 86-87) also report weak effect of GxE interaction (due to nonshared environmental influences), in line with Jensen (1973, pp. 173-174), and no evidence for cultural transmission between generations, as was also the case in most previous studies. In their best fitting model, assuming phenotypic assortment (i.e., mating based on observable characteristics such as intelligence), as opposed to the model assuming social homogamy (i.e., mating based on environmental resemblance only because spouses of same IQ level meet each other given that people with the same IQ level live in the same social environment), genetic variation contributes 58% to the variation in IQ among children and adults, with the remaining 42% accounted for by unique, nonshared environmental variation. Also, Leeuwen et al. (2008) found no evidence of cultural transmission from parents to children, consistent with previous findings (Plomin et al., 1997). Reminding this is essential because under the GxE correlational model, as is the case in Dickens & Flynn, it is assumed that parents transmit no only their genes but also their environment.

More recently, Briley & Tucker-Drob (2013) meta-analysis demonstrates that new genetic influences (innovation; e.g., novel biological changes such as hormonal changes associated with puberty, or environmental changes such as transition from the home to grade school) are likely to explain the increase in heritability at the very early stage of development, but beginning at age 8, genetic amplification becomes predominant, and the effect of new genetic influences drops to zero; the widening of the shaded area (i.e., standard error) with regard to amplification suggests that this parameter is not estimated with much precision, but the authors maintain that their alternative models show the very same phenomenon but more precisely (Figures S3 & S4). In any case, this does not square well with the Dickens-Flynn hypothesis. And thus the fact that Briley & Tucker-Drob (2013) cited Dickens & Flynn (2001) as an illustration of the evidence in favor of the amplification model is a curious non-sequitur.

Explaining the Increasing Heritability of Cognitive Ability Across Development - A Meta-Analysis of Longitudinal Twin and Adoption Studies - Figure S3

Explaining the Increasing Heritability of Cognitive Ability Across Development - A Meta-Analysis of Longitudinal Twin and Adoption Studies - Figure S4

Although human brain, a very dynamic organ, undergoes considerable developmental changes from childhood to adolescence, van Soelen et al. (2011, p. 125) point out that those changes in cortical thickness is under genetic influences, partly overlapping genetic influences on IQ. van Soelen et al. (2012) demonstrate subsequently the existence of a strong genetic influence in cortical thinning during the development, so that “The same genetic factor operates in language areas (e.g., left inferior frontal (superior of) Broca’s and left parietal (superior of) Wernicke’s area). In these areas amplification of the same genetic factors across ages 9 and 12 was observed. Between these two regions, the genetic factors acting on cortical thinning were completely overlapping and this might suggest that the genetic influences acting on cortical development in these areas are implicated in language. We did not find significant influences of common environmental factors on cortical thickness. This is consistent with other neuroimaging studies in both adults (Peper et al., 2007), and children (Lenroot et al., 2009; Peper et al., 2009; Yoon et al., 2010) of both volumetric and cortical thickness measures.” (p. 3878).

In parallel, Smit et al. (2010) found that at the earliest stages in life, common environmental factors can play a certain role in head circumference, but disappear very early in life (4 months). And then, from childhood to adolescence, over a period of about 10 years, individual differences in head circumference were genetically highly stable, as evidenced by the higher genetic correlations for girls (0.78) and for boys (0.85). Recall that brain size correlates with IQ at 0.40 and with g-loadings at 0.63 (Rushton & Ankney, 2009; Jensen, 1998, pp. 146-147), meaning that brain size increases by 1 SD with any 0.40 SD increase in IQ (Rushton, 1997, p. 282). Also noteworthy is that Schmitt et al. (2007b, p. 688) indicate that multivariate genetic analyses established a strong genetic relationship between brain areas; the same authors (Schmitt et al., 2007a) indeed discovered that the great majority of variation in cerebrum, cerebellum, thalamus and basal ganglia was determined by a unique genetic factor, as shown by the high genetic correlations, “As expected from the previous finding that a single genetic factor dominated inter-structure covariance, the genetic correlations between structures dropped substantially when adjusting for total brain volume (Table 7).” (p. 76), “most of the genetic variance is determined by genes that are shared between the major gross neural subdivisions.” (p. 78). Overall, a strong support for the ‘genetic g’ hypothesis, and thus a biological g, seems to be warranted (Plomin, 2003), which weakens the claims made that g is a mere statistical entity.

Incidentally, Davis et al. (2009a), included in the Briley & Tucker-Drob (2013) meta-analysis, show that the genetic and shared environmental influences affecting g from early to middle childhood are quite similar, as illustrated by the genetic and shared environmental correlations, of about 0.57 and 0.65 respectively (N=8,791). Deary (2006) reported a similar result from a large longitudinal study. If the Dickens-Flynn model requires new (genetic) influences to account for the increase in heritability, then the rather strong evidence for amplification weakens their model.

Dramatic Increase in Heritability of Cognitive Development From Early to Middle Childhood  An 8-Year Longitudinal Study of 8,700 Pairs of Twins - Figure 1

Trzaskowski et al. (2013a) used an advanced technique, GCTA as a tool allowing to estimate the genetic variance due to linkage disequilibrium between unknown causal variants and genotyped SNPs, which aims to correlate the genomic similarity across hundreds of thousands of single nucleotide polymorphisms (SNPs) with phenotypic similarity in a large group of unrelated individuals (N=2875), children aged 7 through 12. What they discover is a genetic correlation of 0.73 for g. For the sample of the 6702 pairs of twins, it was 0.75. In conclusion, they note : “As noted above, our GCTA estimates of genetic influence account for 74–94% of our twin-study heritability estimates, which implies that most of the missing heritability can be found with additive effects of common SNPs.” (p. 4). The same genes were responsible for genetic influences on g at age 7 and 12, although heritability increased. The established IQ heritability from GCTA studies is interesting in itself when compared to the possible absence of genetic influence for childhood behavior problems from DNA analysis (Trzaskowski, 2013c, pp. 1053-1055).

Numerous problems with GxE correlational theories, commonly used to dismiss IQ heritability, are yet to be underlined. Here is how Neven Sesardic (2005, pp. 113-114) refutes the GxE theory :

This kind of scenario involves four causal factors: G (genetic difference) → C (characteristic that is not directly trait-relevant) → E (environmental influence) → P (trait difference). Whether something like this really occurs can be tested in different ways.

(i) The scenario implies that C will be correlated with P: genetically related individuals are claimed to be more alike with respect to both C and P. One way, then, to sort out the causal connections is to check whether the correlation between C and P will still exist when G is controlled for. If the correlation disappears, this would strongly suggest that G is actually the common cause of C and P, and that G is not causing P via C. The four-link scenario would be undermined.

Essentially this is what Sandra Scarr did when she tested a very popular environmentalist hypothesis (H) in which the externally perceived identity of appearance of MZ twins plays the role of C. According to H, MZ twins are more similar with respect to some phenotypic trait(s) P merely because their genetic relatedness (G) makes them look similar (C), which in turn makes others treat them similarly (E), and this similar treatment by others (E) is the crucial causal influence on P. Scarr (1968) and Scarr and Carter-Saltzman (1982) compared MZ twins (who are 100 percent G-similar and strongly C-similar) with those DZ twins who, being almost indistinguishable from one another, are incorrectly classified as MZ twins (who are strongly C-similar despite being only 50 percent G-similar). Obviously, H predicts that these two kinds of twins should exhibit an increased degree of twin-to-twin similarity with respect to P, because they share the same degree of C (the similar external appearance), the hypothesized “main cause.” But although many phenotypic traits were explored, the prediction was disconfirmed: despite their increased perceived similarity, DZ twins incorrectly classified as MZ were far less similar phenotypically than real MZ twins.

(ii) An alternative testing method would be to check whether another prediction of the above four-link scenario holds, namely whether P-differences within twin pairs are systematically related to their E-differences. For example, if it is twins’ being dressed alike that makes them phenotypically similar, it would follow that those twins who are not dressed alike should show less trait similarity than those who are. John Loehlin and Robert Nichols (Loehlin & Nichols 1976) conducted such a study with a large sample, and the result was essentially negative. They collected information about twins’ different treatments (as reported by their mothers), but the average correlation between a composite measure of these various possible E-factors and a number of psychological traits was very close to zero (0.056).

(iii) Yet another approach is based on a simple idea (Loehlin 1992: 109; Neale & Cardon 1992: 223): take a number of pairs of MZ twins reared apart, and then just check whether there is a correlation between a given environmental measure of one twin and the other twin’s phenotype. If there is no correlation, this would indicate that the MZA phenotypic similarity is not mediated through this particular environmental measure. (In active and reactive G–E correlation the causal relation is G → E → P, and since the twins have the same genotype, G could not influence P via E if there were no cross-correlation between E and P.) Plomin (Plomin et al. 2001: 311–312) used a similar research design in the context of the Colorado Adoption Project but with parent–child pairs instead of twins. Only meager evidence was found for reactive or active G–E covariance: just a few and fairly low correlations were found between the biological mother’s personality traits and the adopted children’s environmental measures.

(iv) If it is the environments shared by MZ twins that cause their phenotypic similarity, one would expect that their similarity would decrease with earlier age of separation, lower degree of contact, and less time spent together. But the massive study of Swedish twins found no effects of this kind (Lykken 1995: 78).

(v) Finally, the most effective way to empirically investigate the issue of G–E covariance is the multivariate genetic analysis. Here, the basic strategy is to look simultaneously at the impact of genetic differences on two variables (environment and phenotype) in the attempt to estimate whether the two effects overlap (and if yes, to what extent). Explorations with this method have not unearthed strong G–E correlations of the kind envisaged by Block and others, and certainly nothing of the order that could put into serious doubt the accepted high heritability estimates for IQ and many other psychological traits.

It is evident that g shows numerous biological correlates (Jung & Haier, 2007; Peper et al., 2007; Schmitt et al., 2007b; Deary et al., 2010; Goh et al., 2011, p. 310; Menary et al., 2013, p. 603), neural mechanisms such as glucose metabolism, biochemical activity, cortical development, and contradicts the ‘anti g-theorists’ who tend to build g from the outside, reducing it to a simple mathematical abstraction. Explanations of these biological correlates have been proposed. For instance, the fact that glucose consumption correlates negatively with IQ would strongly support the idea that more intelligent individuals use their neurons more efficiently. A correlation between callosal thickness and IQ is an indication that additional or better-myelinated callosal pathways can facilitate a more efficient inter-hemispheric information transfer, which would benefit the integration and processing of information. Likewise, individuals with more gray matter (neurons, synapses, and dendrites) and more white matter (myelinated axons) in these areas usually display higher IQ scores. For a review of the studies, see Luders et al. (2008) and Neubauer & Fink (2009). Besides, evidence points out that some common genetic variants are associated with infant head circumference (Taal, 2012).

The existence of additive genetic effects between brain volumes and IQ is also suspected. van Leeuwen et al. (2009), by fitting a series of nested models in which the means and variances between MZ and DZ twins were equated, have tested several assumptions (e.g., equality of means and variances between MZ and DZ twins). They continued constraining parameters until the most parsimonious model with still acceptable fit was established, with the choice for the best fitting model based on likelihood-ratio tests. They found that “dropping the A component in the saturated AE model led to a significant deterioration of fit”, meaning that additive genetic factors make a significant contribution to the variance and covariance in the three brain measures (TBV, GMV, WMV) and the four intelligence measures (Raven, VC, PO, PS). In their discussion section, they write :

Under the causal hypothesis both genetic and environmental correlations should be significant, whereas a significant genetic correlation in the absence of an environmental correlation falsifies the hypothesized causal effect of intelligence. However, when traits are highly heritable (in the range of 90–100%), as is the case in brain volumes, causality (brain volume causes intelligence) cannot be distinguished from pleiotropy (the same set of genes affects brain volume as well as intelligence).

… However, our study shows that only the genetic correlations are significant. In fact 85% to 100% of the covariation between brain volume and intelligence are caused by shared genetic factors. This leaves two options: 1) the relation between brain volume and intelligence is caused by a set of genes which influences variation in brain volume and this variation in turn leads to variation in intelligence 2) pleiotropy: there is a set of genes which influence brain volume as well as intelligence.

But whereas Leeuwen failed to detect a relationship between processing speed (PS) and brain volume, Betjemann et al. (2010) succeeded, in a study showing that the correlation between IQ and brain volume is mostly due to genetic influences. The reason is simply because Leeuwen used only 2 measures of PS while Betjemann used 4 measures. Therefore, Betjemann’s composite score of PS is much more reliable.

In an analysis at the g level, Grandy et al. (2013) provide evidence that individual alpha peak frequency, IAF, which had been discussed by Posthuma et al. (2003), correlates with g, and that g mediates the relationship between IAF and some specific ability factors. As they write, “After establishing a substantial correlation between IAF and intelligence at the level of second-order g, we examined whether there was any evidence within our data for specific associations between IAF and first-order factors of perceptual speed, memory, or reasoning. Importantly, estimating residual associations between the IAF latent factor and each of the three first-order latent factors in the presence of the correlation between IAF and g did not indicate any significant correlation between IAF and cognitive abilities beyond the correlation between IAF and g.” (pp. 15-16). Just like g, the authors reported that IAF is not affected by extensive cognitive training. Their association, anyway, supports the theory that differences in intelligence are related to differences in general oscillatory properties in the brain. Although earlier studies reported conflicting result on this matter, Grandy speculates that a possible reason is that the correlation has been computed at the level of individual tests, thus containing a larger portion of error and specific variance. Instead, Grandy uses a g unbiased by measurement errors.

Penke et al. (2010) established that a general factor was not only present among specific cognitive abilities, there is also a general factor of white matter integrity, supporting the views that impaired cortical connection (lowered white matter integrity) is a global process affecting many tracts simultaneously. Eight matter tracts have been factor analyzed, yielding a so-called “common integrity factor” which explains about 45% of the individual differences across all eight tracts. Individual tracts showed no associations beyond what the common integrity factor explained. Besides, a simple RT, 4-choice RT, and inspection time have been factor analyzed, yielding a general information processing-speed factor. This processing speed factor is found to be (modestly) correlated with the common integrity factor (while general intelligence, obtained by factor analyzing mostly the Gf or PIQ subtests of the Wechsler, was not correlated with the integrity factor). When controlling for the integrity factor, the correlations between the speed factor and the eight individual tracts were not significant (excepted one). According to the authors, the absence of correlation between the g-factor and the integrity factor could be explained by the fact that “speed tends to be affected earlier in life by age-related decline … whereas higher abilities are more likely to be maintained by compensatory processes” in the face of age-related cognitive decline.

Subsequently, Penke et al. (2012, Figure 2) conducted SEM analysis to test the possible pathways from three different indicators of white matter tract integrity to general intelligence (g). They discovered that this path was fully mediated by information-processing speed, which is relevant to Jensen’s hypothesis (2006, pp. 207-208, 212, 216-217) of mental speed. They made this assumption since efficient information processing between distal brain regions is thought to rely on the integrity of their interconnecting white matter tracts. Well-connected white matter favors efficient information processing. Jung & Haier (2007) also emphasized on brain functional connectivity to explicate the neural basis of intelligence.

Still related with the causal entity of g in mental abilities, a multivariate SEM study by Shikishima et al. (2009, see p. 258, Fig. 1 & 3 for details of their applied method) indicates that a common pathway model AE among the other competing models (Table 8), in which a latent factor ‘g’ was posited as a higher-construct (i.e., genetic and environmental factors influencing logical, verbal and spatial abilities through a latent factor), has been proven to be the best fit model to their twin data, confirming g as existing at the etiological level, and thus they were able to represent g as a causal entity.

Is g an entity - A Japanese twin study using syllogisms and intelligence tests

The common pathway model implies that genetic and environmental effects contribute to a psychological entity, g, per se, rather than directly to each ability, and that each ability is subdominant to g. Alternatively, if g is not an entity but merely an artifact, the model that does not include the latent factor should exhibit a better goodness-of-fit. …

Given the background mentioned thus far, we propose the following two hypotheses. First, the ability that has historically been referred to as the symbol of human intelligence, namely, syllogistic deductive reasoning ability, will be strongly g-loaded. Second, general intelligence g will be identified not only at a phenotypic level but also at its genetic and environmental factor level. …

The better goodness-of-fit for the common pathway model (in which a higher-order construct is hypothesized as an entity) compared with the independent pathway model (in which commonalities are not postulated as an entity) lends plausibility for the existence of g in terms of its genetic and environmental origin.

This is not consistent with van der Maas et al. (2006) mutualism model because they posit that the positive correlations (positive manifold) of cognitive tests, represented by the g factor, are the result of mutually, reciprocal (beneficial) causal interactions amongst distinct elements, during childhood development, that finally give rise to g. They were building g from the outside, making it a hollow concept; g was not a causal entity. According to the model, at the initial phase of the development, there is no positive manifold, cognitive processes being not correlated, meaning that “intelligence, which standard psychometric intelligence tests purport to measure, takes some time to develop in children. Consequently, we expect that it takes some time for the positive manifold, and thus the psychometric g factor, to emerge.” (p. 851).

Then Shikishima et al. (2009) remind the reader that the empirical evidence and reality of a biological and causal g does not mean that g itself involves a unique process, as was wrongly believed by some researchers (Hampshire et al. 2012). According to Plomin (2003), “It should be noted that genetic g does not necessarily imply that there is a single fundamental brain process that permeates all other brain processing, such as a ‘speedy brain’, neural plasticity, or the quality and quantity of neurons. It has been proposed that g exists in the brain in the sense that diverse brain processes are genetically correlated. For example, gray and white matter densities in diverse brain regions are highly heritable, substantially intercorrelated across brain regions, and correlated genetically with g”.  Jensen (1998, pp. 130-132, 259-261) specifically stated the following :

The g factor, which is needed theoretically to account for the positive correlations between all tests, is necessarily unitary only within the domain of factor analysis. But the brain mechanisms or processes responsible for the fact that individual differences in a variety of abilities are positively correlated, giving rise to g, need not be unitary. … Some modules may be reflected in the primary factors; but there are other modules that do not show up as factors, such as the ability to acquire language, quick recognition memory for human faces, and three-dimensional space perception, because individual differences among normal persons are too slight for these virtually universal abilities to emerge as factors, or sources of variance. This makes them no less real or important. Modules are distinct, innate brain structures that have developed in the course of human evolution. They are especially characterized by the various ways that information or knowledge is represented by the neural activity of the brain. The main modules thus are linguistic (verbal/auditory/lexical/semantic), visuospatial, object recognition, numerical-mathematical, musical, and kinesthetic. …

In contrast, there are persons whose tested general level of ability is within the normal range, yet who, because of a localized brain lesion, show a severe deficiency in some particular ability, such as face recognition, receptive or expressive language dysfunctions (aphasia), or inability to form long-term memories of events. Again, modularity is evidenced by the fact that these functional deficiencies are quite isolated from the person’s total repertoire of abilities. Even in persons with a normally intact brain, a module’s efficiency can be narrowly enhanced through extensive experience and practice in the particular domain served by the module. …

But at some level of analysis of the processes correlated with g it will certainly be found that more than a single process is responsible for g, whether these processes are at the level of the processes measured by elementary cognitive tasks, or at the level of neurophysiological processes, or even at the molecular level of neural activity. If successful performance on every complex mental test involves, let us say, two distinct, uncorrelated processes, A and B (which are distinguishable and measurable at some less complex level than that of the said tests) in addition to any other processes that are specific to each test or common only to certain groups of tests, then in a factor analysis all tests containing A and B will be loaded on a general factor. At this level of analysis, this general factor will forever appear unitary, although it is actually the result of two separate processes, A and B. … However, the fact that g has all the characteristics of a polygenic trait (with a substantial component of nongenetic variance) and is correlated with a number of complexly determined aspects of brain anatomy and physiology, as indicated in Chapter 6, makes it highly probable that g, though unitary at a psychometric level of analysis, is not unitary at a biological level.

To summarize, the additive genetic influences on g is quite high, the GxE correlational model not clear, and the very fact is that g seems really to be a biological entity, and a causal entity as well, meaning that strong inferences can be drawn from it.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s