Download The Paradox of Intelligence and more Exams Genetics in PDF only on Docsity!
The Paradox of Intelligence: Heritability and Malleability Coexist
in Hidden Gene-Environment Interplay
Bruno Sauce and Louis D. Matzel Department of Psychology, Program in Behavioral and Systems Neuroscience Rutgers University
Abstract
The heritability of intelligence is extremely high, but it can also be malleable, a paradox that has been the source of continuous controversy. Here we attempt to clarify the issue, and offer a solution to this paradox that has been frequently overlooked: Intelligence is a trait with unusual properties that create a large reservoir of hidden Gene-Environment (GE) networks, allowing for
the contribution of high geneticand environmental influences on individual differences in IQ. GE
interplay (consisting of both interactions and correlations between genes and environment) is difficult to specify with current methods, and is underestimated in standard metrics of heritability (thus inflating estimates of “genetic” effects). We describe empirical evidence for GE interplay in intelligence, with malleability of intelligence existing on top of heritability. The evidence covers IQ gains consequent to adoption/immigration, changes in heritability across lifespan and socio- economic status, gains in IQ over time consequent to societal development (the Flynn effect), the slowdown of age-related cognitive declines, and IQ gains via early education. The GE solution has novel implications for enduring problems including our inability to identify intelligence-related genes (also known as IQ’s “missing heritability”), and the dissipation of initial benefits from early intervention programs (e.g., “Head Start”). The GE solution can be a powerful guide to future research, and may also aid in the implementation of policies to overcome barriers to the development of intelligence, particularly in impoverished and under-privileged populations.
Keywords Intelligence; Heritability; Gene-environment interplay; Adoption; Early intervention
“When you have eliminated the impossible, whatever remains, however improbable, must be the truth.” -Sherlock Holmes
1. The debate over the heritability of intelligence
Some of us have blue eyes, while others have green or brown eyes. Some are tall and others are short. Some people are outgoing while others are shy. Individual variations are far- reaching and can be found in both physical and psychological traits. “Heritability” is a
Contact Information: Louis D. Matzel, Department of Psychology, Rutgers University, Piscataway, NJ 08854, Phone: 732/735-0803; fax: 732/445-2263; matzel@rci.rutgers.edu.
HHS Public Access
Author manuscript
Psychol Bull. Author manuscript; available in PMC 2019 January 01.
Published in final edited form as:
Psychol Bull. 2018 January ; 144(1): 26–47. doi:10.1037/bul0000131.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
statistic that, as commonly interpreted, captures how much of the variation on a trait is due to genetic differences. Heritability can be estimated for any trait, and it ranges from 0. (meaning that the trait has no genetic component) to 1.0 (meaning that the trait is completely heritable). For example, the heritability of breast cancer is 0.27; the heritability of body mass index is 0.59; and the heritability of Type 1 diabetes is 0.88 (Hyttinen, Kaprio, Kinnunen, Koskenvuo, & Tuomilehto, 2003; Lichtenstein et al., 2000; Silventoinen, Magnusson, Tynelius, Kaprio, & Rasmussen, 2008). Using similar methods, the heritability of general intelligence is estimated to be as high as 0.8 (although, as we discuss below, this value will vary depending on where and when it is estimated). To put that in perspective, the heritability of other “highly heritable” psychological traits rarely approach the level of IQ (see Bouchard, 2004, for extensive examples). The most comparable is schizophrenia (heritability of 0.64; Lichtenstein et al., 2009), while alcoholism (0.50), neuroticism (0.48; Riemann et al., 1997), and major depression (0.40; Sullivan et al., 2000) are markedly lower.
The high heritability of intelligence has captured the attention of many researchers across diverse disciplines, and has spurred a century-long debate which still endures (e.g., Gottfredson, 1997; Jensen, 1969; Lewontin, 1970; Tabery, 2014). In retrospect, that controversy generated more heat than light, and confusion is still widespread. Even within the field of Psychology, many appear unclear about the implications of the heritability of IQ, and are unaware of the impact of gene-environment interplay on estimates of heritability. This confusion has profound functional implications. Not only is IQ a recognizably consistent measure, it also independently predicts real outcomes such as academic grades, income, social mobility, happiness, marital stability and satisfaction, general health, longevity, reduced risk of accidents, reduced risk of drug addiction, and reduced likelihood of committing violence and crimes (Gottfredson, 1998; Mackintosh, 2011). A clear understanding of the causes of variation in intelligence is critical for future research, and its potential applications to society are self-evident.
Recently, Robert Plomin, a prominent behavioral geneticist who has performed seminal
studies on the heritability of intelligence, said in an interview with the magazineThe
Spectator: “At one time, people thought family members were similar because of the
environment, but it turns out that the answer — in psychopathology or personality, and in cognition post-adolescence — the answer is that it’s all genetic! What runs in families is
genetic!” (Wakefield, 2013). When the interviewer interjected that the environmentmust
have an effect on what adults becomes, Plomin replied, “I did an adoption study on weight, IQ and cognitive abilities, and parents who don’t see their children after the first few hours of life are just as similar in terms of both weight and IQ to them after adolescence as are parents who reared their own kids. And adopted parents are zero similar!” (Wakefield, 2013 ). Plomin certainly did not mean “similar” in the highly deterministic way that the general public would have understood it. And, as we argue below, even what Plomin did mean by “similar” (as high independent genetic variance) is likely to be incomplete.
The influence of genes on IQ, are not as powerful or constrictive as might be assumed. As we describe here, intelligence seems to be quite malleable, and changes in the environment
can, byinteracting with genes, explain a great deal of differences in IQ across families,
lifespan, socioeconomic status, and generations.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
distributing family members with known genetic variations into distinct environments (or the same environment for all different families, which accomplishes the same goal). For humans, we look for cases where these splits happened naturally, such as cases of adoption, siblings raised apart, and families with twins.
For the present purposes, we can limit our discussion now to the “twin method”, a strategy that is frequently used to estimate heritability in humans. The strategy exploited by twin studies is to compare monozygotic, “identical” twins (henceforth called “MZ twins”) to dizygotic, “fraternal” twins (henceforth called “DZ twins”). MZ twins share approximately 100% of genetic material, while DZ twins share on average 50%, just like any other siblings. The assumption of twin studies is that the shared family environment experienced by MZ and DZ twins will be relatively similar, while only the shared genetic material will differ. Therefore, the discrepancy between how “similar” MZ and DZ twins are can be used to estimate how heritable a trait is. For IQ scores, if the values among MZ twins correlate more than the values among DZ twins, this discrepancy in correlation is inferred to be the consequence of genetic influences.
Here is a simplified formula to calculate heritability based on differences between MZ and DZ twins: Heritability = 2 (rMZ − rDZ) rMZ = the average correlation of a measured trait among MZ twins rDZ = the average correlation of a measured trait among DZ twins Since the difference in the resemblance of MZ and DZ twins is due to the difference in sharing either 100% or 50% of genes, the difference between rMZ and rDZ is multiplied by 2, yielding an index of the proportion of the trait that is heritable. (For more on the twin method and other models for estimating heritability in humans, see Tenesa & Haley, 2013).
There are a number of concerns one can raise about the estimates of heritability obtained from twin studies. MZ twins usually share a placenta, while DZ twins never do, and this may cause different patterns of gene expression (Gordon et al., 2012). And of course MZ twins may be treated differently than DZ twins, or might be encouraged to make more similar decisions in life (Kendler et al., 1993; Richardson & Norgate, 2005). But these often- discussed caveats (e.g., Eysenck, 1979) do not form the crux of our concerns here. The problems we have in mind are more substantive in nature, and they hinder people’s appreciation of the malleability of intelligence as well as its gene-environment interplay.
There is first a problem of interpretation: Heritability isnot the equivalent of genetic
determination. Heritability is an estimate of the causes of differences in a trait among people (a population statistic), whereas genetic determination is a matter of what causes a trait to be
expressed in an individual.Heritability is a measure of what is associated with variation in
the trait (its causes of variation), and not a measure of what causes the trait. Second, there is
a problem sometimes overlooked even by experts: estimates of heritability (owing to biases in the underlying calculations) grossly underestimate gene-environment interplay.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
In some instances, geneticallydetermined traits are associated with very low estimates of
heritability. For example, the number of fingers on a human hand is genetically determined: the genes related to this trait code for instructions that lead to five fingers in almost everyone in any normal environment. However, the heritability of number of fingers in humans is very low. That is because the vast majority of variation in finger number is purely environmental, with traumatic amputations and prenatal complications being the leading causes (while genetic coding for other than five fingers is rare in humans). In contrast, traits that are not genetically determined can sometimes be highly heritable. For example, the tendency to vote in an election has a heritability of 0.53 (Fowler et al., 2008), and the affiliation to Democrat (liberal) vs. Republican (conservative) parties in the US has a heritability of 0.46 (Settle et
al., 2009). If this strikes the reader as odd, remember: A heritablecharacteristic isnot one
that is passed down in the genes, and neither is political affiliation, at least not in any sense where this term has explanatory power. In fact, there might be dozens of correlations and interactions between genes and environmental factors such as wealth, education, and choices in life that contribute to one’s political affiliation. Unfortunately, typical measures of heritability grossly underestimate GE interplay, as we explore next.
2.2. First dibs to genetic effects: Why heritability can underestimate GE interplay The construct of “general intelligence” (or the g factor) captures the remarkable phenomenon that performance across diverse cognitive tasks are positively correlated. In other words, people who do well in one task tend to do well in many other tasks. This principle is a well-supported by decades of evidence, and is a central feature of many (if not the majority of) modern theories of intelligence (Mackintosh, 2011). At first glance, the existence of general intelligence might suggest that one “thing” explains much of the variability in intelligence. Such a casual assumption would severely limit our appreciation of the basis for intelligence. It could bias us into a straightforward genetic view where there is no space (nor need) for a multitude of correlations and interactions with the environment. Although general intelligence is a single, abstract construct, it needn’t be a single, concrete
property of the brain. Instead, the higher-order construct ofg may emerge from the interplay
between a multitude of systems, including geneticnetworks, influences from psychological
traits like working memory capacity, attentional capacity, motivation and personality, and reciprocal relations between cognitive and environmental processes (as held, for example, by on the “mutualism” model of intelligence originally proposed in van der Maas et al., 2006). Intelligence is a hugely complex trait, and we should expect a high number of moving parts behind differences across individuals.
At this point, it is important to define “GE interplay”. GE interplay denotes both correlations
as well as interactions between genes and environment. Gene-environmentcorrelation
means that individuals with particular genotypes for a trait are more likely to experience
particular relevant environments. Meanwhile, a gene-environmentinteraction means that
there is a difference in theeffect of a given relevant environment on individuals depending
on their genotypes for that trait. These effects will be explored in more detail below.
While historically ignored (or at least minimized), the study of GE interplay as a cause of variation in a trait is starting to gain momentum in our current scientific age of big data,
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
As we discussed above, in the simple equation VP = Vg + Ve, Fisher defined Ve as the leftover variance (residual variance) that was not explained by Vg. Similarly, indirect effects such as genetic interactions, gene-environment correlations, and gene-environment interactions are also relegated to the residual genetic variance left over after subtracting off the additive genetic variance (Templeton, 2006). Because additive genetic effects (Ga) are accounted for before rGE and G×E, any contribution from genes to the variation in a trait will be overestimated as the direct effect from genes (Templeton, 2006). And, of course, any contribution from the environment to the variation in a trait will be underestimated because its fair share of the variance was already appropriated by Ga. In other words, Fisher designed
the quantitative genetic parameters in his model in such a way that (as we like to say)first
dibs went to genetic effects. This is a problem intrinsic to partitioning variance under any
classic statistical tool (such as ANOVA or simple regression): Variance (outcomes) with multiple correlated sources (predictors) can be mispresented as belonging to the source first/ better specified in the analysis. In the case of intelligence, while researchers can precisely specify genetic relatedness, they do not understand, posit, and/or search for all of the rGE and G×E sources, and so default to the assumption that the variance comes from G. As pointed by Tenesa and Haley (2013), lack of modeling frequently leads to an overestimation of genetic effects. And this problem is not limited to twin studies. Any model (e.g., the ACED model for the estimation of heritability) where GE interplay is not specified, there is a high risk of GE becoming incorporated into additive genetic effects, which of course bias the estimates of heritability (Tenesa & Haley, 2013).
What exactly are additive, independent genetic effects (Ga)? Does it imply that a single isolated gene produces the trait? In any literal sense, this cannot be true. If you put a strand of DNA double helix in a tube and wait, no distinguishable trait will emerge from it. And what about the GE interplay of gene-environment correlations (rGE) and gene-environment interactions (G×E)? Do these parameters merely refer to the evident truth that an egg needs both genes and environment to create an adult chicken? The answer is “no”. As described
above, the effects that these parameters capture are related tocauses of variation, not
deterministic causation. Additivity or independence is conceptually analogous to what a statistician would describe as a “main effect”: it is how much a predictor/input variable (a gene) relates to a predicted/output variable (a trait) when averaging across the other predictor variables (environment and other genes). An rGE effect is a statistical correlation between two predictor variables. In turn, a G×E effect is conceptually close to a statistical interaction: it is when the effect of a variable (gene) with another variable (an environmental factor) has multiplicative consequences, with the result (variation in a trait) being more than the sum of its parts^2.
Now that we understand what additive effects and GE interplay mean conceptually, it is useful to explore rGE and G×E more concretely.
(^2) Technically, the interaction in G×E is similar, but not the same as a statistical interaction. A statistical interaction will only occur when there is variation in both G and E (Rutter & Silberg, 2002). Cases of potentially high G×E where an important environmental factor is the same to most people (e.g., exposure to pollen when studying the trait of hay fever) will result in no statistical interaction (Rutter & Silberg, 2002). In addition, as we already noted, the interactions of G×E can be masked by main, independent effects.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Gene-environment correlation (rGE)— Behavior geneticists commonly classify the gene-environment correlation into three different types: passive, reactive (or evocative), and active (Dick, 2011; Plomin et al., 1977). The philosopher Ned Block (Block, 1995) gives an interesting example to illustrate these types. Suppose there are genes that predispose humans to musical abilities. Now suppose that children with those genes tend to have parents who provide them with an environment conducive to developing those abilities, including music lessons, concerts, access to an extensive music collection, musical discussion over dinner, etc. Assume also that the children who have a genetic disadvantage also have an environment that hampers their musical abilities (as could reasonably be expected, since their parents were likely to be musically disadvantaged). In this scenario, there will be a correlation between genes and environment that will move children towards both extremes of the distribution. These types of gene-environment correlations are called “passive covariance” because they do not depend on what the child does. Parents create a home environment that is influenced by their own heritable characteristics, which correlates with the genetic material they pass to their biological children. Passive covariance can be controlled in heritability estimates by using, for example, cases of adoption, since adopted children with musical ability genes will not be more likely to be raised by music aficionado
parents. In contrast, “reactive covariance” is a matter of the environmentreacting to the
person’s traits, as when a school gives advanced tutoring classes to children who exhibit musical talent, or when friends devote a lot of time to help practicing for a performance. (In a real example, Tucker-Drob & Harden (2012) found a case of reactive covariance in children using a twin design. In it, genetic predispositions to higher cognitive abilities in 2- year-old children lead to more and higher quality cognitive stimulation by parents in the form of a dyadic task. This stimulation, then, led to children to a better reading ability at 4
years of age.) Finally, “active covariance” occurs when a personcreates the correlation
between genes and environment by self-selecting his/her experiences, as when a musically- able child practices musical themes in her imagination or pays attention to the songs created by other musicians.
As noted by Block (1995), reactive as well as active covariance cannot be estimated without specific hypotheses about how the environment affects a trait. Due to our vast ignorance regarding the development and expression of human intelligence, a significant portion of rGE is mostly beyond the reach of current (and ethically permissible) methods in genetics.
As we mentioned above,when estimating the heritability of IQ, those gene-environment
correlations that we don’t recognize or don’t know how to measure will be attributed to the
genetic component. (For a more sophisticated treatment on the genetics and mathematics of
this issue regarding hypothetical traits, see Templeton, 2006.)
Gene-environment interactions (G×E)— Distinct from cases of gene-environment correlations, gene-environment interactions mean that genotypes vary in their environmental
sensitivity. Continuing the example on musical abilities, a musically talented child who
frequently listens to a rap song might have her musical ability sharpened if she attends to the song’s complexity and the depth of the lyrics. In contrast, a child who is musically impaired might exhibit no sharpening of musical ability when listening to the exact same song because she ignores all cognitively and artistically engaging aspects of the song, focusing
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
is not far from what researchers with the gene-centric view of nature actually claim, such as
the ethologist Richard Dawkins in his bookThe Extended Phenotype (see Jablonka, 2004
for a critical analysis of the book). At a nominal level, the assertion may be “true”, but in such a trivial and vague way that it would have no explanatory value. The range of opportunities and stimuli that individuals might encounter are not merely static features of the environment always available to those genetically disposed to exploit them. As noted by the neurogeneticist Michael Meaney, heritability would be better described as representing the exclusively genetic influence on the variation of a trait if, and only if, there is zero Gene × Environment contribution (Meaney, 2010).
2.3. The iceberg of hidden interactions on evolutionarily critical traits As we have seen, the hidden nature of rGE in the variation of intelligence comes from our inability to specify factors that contribute to active and reactive correlations. The hidden nature of G×E reflects, like for rGE, a similar lack of knowledge on the modulation of intelligence. But in addition to that, as we see now, the search for G×E is made even harder due to a probable “hidden iceberg” in the past and current evolution of intelligence. Evidence from evolutionary genetics strongly suggests that traits related to survival and reproduction (like intelligence) have a large reservoir of hidden variance in the form of interactions (Merilä & Sheldon, 1999). This type of diversity comes from both gene-gene and gene-environment interactions, and is generally referred to as “hidden” variation because it has the potential to affect a trait, but is not expressed under typical/current conditions (Le Rouzic & Carlborg, 2008). These interactions capture genetic variation and accumulate mutations that stay latent for long periods of time, since natural selection is “blind” to anything other than additive, independent effects (Hermisson & Wagner, 2004). And since additive genetic variation is what fuels evolution by natural selection, important traits like intelligence might only continue to evolve at a fast pace because of constant new G×G and G×E hidden effects. A useful analogy to represent this is the iceberg. The tip of the iceberg (the independent genetic variance) is the only type of variance that can be seen by natural selection. And it is also the only effect that is not “residual” in most methods in quantitative genetics, like the ones seen for heritability here. Just as we see only the tip of the iceberg, the effects of interactions cannot be seen by natural selection and are much harder to detect during empirical measurements.
A recent meta-analysis on interventions to increase intelligence in humans (Protzko, 2015) suggested that “under increased demands from the environment (due to an intervention) we can raise IQ. Once those demands are removed, the system adapts to the new, reduced demands. Protzko’s conclusions suggest that G×E interactions influence IQ in a way similar to muscular mass, where some people are more genetically prone to build muscle, but that tendency also continuously interacts with circumstances: if you play in a professional league, your environment will buff you further (until you get fired or retire), and if are an astronaut in the ISS, your environment will lead to muscular atrophy (until you return to Earth). So, one could ask: What are the advantages of being weak? And why aren’t all humans “blessed” with hypertrophied muscles cells like gorillas? No one knows for sure, but it is not hard to imagine how our ecology and history might have sometimes benefited weak humans. Or, better, humans that could match their strength according to their needs in a
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
period of a few months (thus conserving resources). Similarly, one could ask: What are the advantages of having a low IQ? There might be many. And maybe intelligence’s plasticity is useful to match its environment as to save precious resources from our energy-hungry brain. 3
In fact, evolution is not only a part of our argument here on the hidden iceberg. The t present article is also inspired by recent changes in the field of evolutionary biology. Some psychologists, and more so behavior geneticists, claim that there is nothing new about the recent evidence on GE interplay as it relates to the interpretation of what heritability means and to the interpretation of the broader questions regarding the mixing of nature and nurture in the development of complex traits. We disagree, and are immediately reminded of an analogy in the field of evolutionary biology. Studies of G×G interactions were only recently possible, as were the assessment of epigenetic effects, niche construction, and interacting phenotypes (such as in the case for social evolution, seen in Moore, Brodie III, & Wolf, 1997 ). These findings and ideas make up today what modern researchers call The Extended Evolutionary Synthesis, which provides a new framework to think about and understand evolutionary phenomena that differs from the gene-centric conception that dominated evolutionary thinking for almost a century (Pigliucci, 2007). The extended evolutionary synthesis revisits some neglected and/or unknown factors at play in evolution, including developmental processes, the role of epigenetic inheritance and plasticity, as well as networks of interactions between genes and environment (Danchin et al., 2011; Laland et al., 2015 ). In sum, the goal of the new synthesis is, in Danchin et al., (2011) words: “to go beyond DNA in order to build a broader conception of evolution”.
Something analogous, we imagine, could be in the making for the fields studying the genetics of psychological (and other complex) traits, and GE interplay might be critical to it. Intelligence is a prime candidate, since this trait presents a unique case of established moderate-to-high heritability while at the same time exhibiting remarkable malleability. Intelligence has potential for gigantic GE interplay, although the way this interplay manifests itself in the variation in IQ has contributed to its earlier dismissal (as we discuss in the next section).
We can conclude so far that there is not only a common problem with the interpretation of
heritability’smeaning (genetic determination versus genetic cause of variation), but
heritability estimates themselves overlook indirect environmental/genetic effects. Furthermore, intelligence is a trait with a huge potential for gene-environment correlations
and interaction given its vast psychological and biological complexity. But how do weknow
that the environment or GE interplay exerts a central influence on the variation of
(^3) The reader might assume that there is convincing evidence that intelligence is highly stable (i.e., not plastic) across the life span (Deary, 2014). And indeed, that is the case. However, that stability does not mean that actual cognitive capacity does not change (an adult is clearly superior to a child). Rather, it means that an individual’s IQ relative to his/her age cohorts tend not to change with age. While this is an interesting fact, it says little about the malleability of intelligence or the presence of GE interplay. The relative stability of IQ allows for changes in environmental factors and of GE interplay in the same way that it needs to allow for changes in genetic factors (given that humans are not born already expressing all their genes). As we describe below, massive environmental manipulations can underlie similarly large changes in the IQ of some individuals, despite the fact the environments and IQs are relatively stable formost individuals.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
educated family can substantially increase IQ. In a recent adoption study done in Sweden, Kendler et al. (2015) assessed the IQs of 436 pairs of separated siblings where at least one member was reared by biological parents and the other by adoptive parents. Adoption by parents with higher level of education was associated with a significant increase of 4. points in the child’s IQ in adulthood. Interestingly, the authors also found that in families with at least 2.5 steps higher education status than biological parents (i.e., the difference between no high school and some postsecondary education), the adopted-away siblings had 7.6 IQ points higher on average than their home-reared adopted siblings. On the other extreme, sibling sets in which the biological parental educational status was at least 2 steps higher than that of the adoptive parents, the adopted-away siblings had an IQ on average 3.
pointslower than their home-reared siblings (Kendler et al., 2015). This result suggests
some role of GE interplay, as it seems that higher biological parental education (which is itself likely to correlate strongly with the biological parents’ IQ) was worse than higher adoptive parental education for stimulating intellectual development.
The studies above can be contrasted with a small study (n = 38) involving a full cross-
fostering design by Capron and Duyme (1989). Like in the above, the authors here found that the mean IQ of children reared in upper SES homes was significantly superior to those reared in low SES homes. The IQ gains from high SES environments was 12 points; smaller than the 19.5 points from Duyme et al. (1999), but still remarkably high! However, children born to upper SES families but adopted early into lower SES families had mean IQ of 107. whereas the mean IQ of children born into low SES families but adopted early into high SES families had a mean IQ of only 103.6 (Capron & Duyme, 1989). One would expect the means to travel extensively in the opposite direction if the environment was as potent as it appears in aggregated data because these contrasting environments (top vs. bottom 13% of French society) are several standard deviations apart. So, while these data are compatible with GE interplay, it suggests that additive genetic effects might sometimes be strong enough to counteract improvement in SES.
In a meta-analysis, van Ijzendoorn et al. (2008) considered 75 studies (totaling more than 3800 children in 19 different countries) to compare the intellectual development of children living in orphanages to that of children living with adoptive families. On average, children growing up in orphanages had an IQ that was 16.5 points lower than their peers who were adopted. Not surprisingly, orphanages in countries with a higher Human Development Index (a combined measure of life expectancy, literacy, education, standards of living, and quality of life) had smaller detrimental effects on children’s intelligence (reduction of 11.9 IQ points) than countries with a lower Index (reduction of 21 IQ points). Also, children in orphanages with the most favorable caregiver-child ratio (maximally three children per caregiver) did not significantly differ from their adopted peers. These observations suggest that the typical orphanage has environmental conditions that are detrimental to the development of intelligence. In other words, environmental conditions related to orphanages are causes of variation in IQ.
Environmental effects on intelligence from international adoption are potentially much more powerful than those observed within a country or region. Economically prosperous countries can have up to 9X the GDP per capita of economically undeveloped nations. (By
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
comparison, adoption within the US from a low SES family to a higher SES family is commonly associated with a 2–4X increase in family income.) The poorest 5% of the US population, for example, are richer than 60% of the world (Milanovic, 2013). In addition, there are many other environmental factors relevant to IQ that differ drastically across nations, such as educational opportunities, parental expectations and pressure, motivation, a culture of intellectualism versus anti-intellectualism, the availability of cognitively demanding jobs, etc.
Winick et al. (1975) examined 205 Korean orphans (all of the viable cases from a single adoption service from 1959–1967) who were adopted during early life by US parents, and divided cases into three categories (according to the conditions of the children before adoption): malnourished (n = 59), moderately nourished (n = 76), and a control of well- nourished children (n = 70). (Keep in mind that Korea as a whole was a poor and underdeveloped country at the time of the adoptions during the 1960s.) After at least six years with their American parents, the children were assessed for IQ. The mean IQ of the previously malnourished group was 102; the moderately nourished group, 106; and the well- nourished group, 112. Strikingly, the mean IQ of the children from the previously malnourished category was 10 to 40 points higher than the IQ of malnourished children living with their biological families in Korea or other poor populations (Galler, et al., 1983; Hertzig et al., 1972; Liu et al., 2003; S. A. Richardson, 1976). In a similar analysis, O’Connor et al. (2000) examined 111 children from Romania who were adopted (after the collapse of the Soviet Union) by families in the UK at the age of four. The authors found a considerable catch-up in children’s cognitive abilities from the time at the adoption to just two years later, at age six (although these adopted Romanian children were still slightly below the average IQ of adopted UK children.)
A meta-analysis of 62 studies from a multitude of countries (totaling 18,000 adopted children) found an average increase in IQ of 17.6 points within several years of adoption (van Ijzendoorn et al., 2005) – a remarkable cognitive gain over their biological, nonadopted siblings and their peers who stayed behind. The size of the IQ gains described in this meta- analysis is higher than typical for studies of adoption. That, we believe, is because the meta- analysis considered studies where children came from extremely low SES and were adopted by families in a developed country (so, as pointed out above, they experienced a more dramatic environmental change than would be typical for within-nation adoptions). Furthermore, no significant differences were observed between the ultimate IQs of the adopted children and their environmental siblings/peers. In other words, the new environment made the initially “dull” children climb up the IQ ladder as high as the typical child in their new environment.
While early adoption associated with immigration can have a dramatic positive impact on IQ, the effects of adult immigration (absent adoption) are less encouraging. Rindermann and Thompson (2016) compared immigrants to natives worldwide across a 10–15-year period. Rindermann and Thompson reported that generally, immigration was associated with a small drift in IQ toward that of the native inhabitants. Depending on the quality of the host country’s educational system and economic resources, increases in IQ of 1–4 points were typical, a level much lower than observed in typical adoption studies (Rindermann &
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
adopted individuals, even many years after the adoption, is still highly correlated with the biological parents (and, as we pointed above, the environmental influence of the adoptive family is drastically reduced). How can these patterns be reconciled with the massive IQ gains following adoption?
Indeed, substantial evidence shows that the correlations of adopted children with their biological parents are high, and the correlations with their adoptive family members are near zero by late adolescence (examples in Horn, 1983; Plomin et al., 1997). However, this pattern of correlations does not imply that the impact from the environment must be small. In fact, massive IQ changes consequent to adoption are entirely compatible with high IQ correlations with biological parents. Put more generally, the magnitude of a correlation is independent of changes in means. (While this is a mathematical truth, its implications can be easily overlooked when interpreting correlations.) A recent study explored this issue empirically by performing a longitudinal assessment of intelligence in thousands of twins, first at age 20, and later, at age 55 (Lyons et al., 2009). Using standard methods for separating the causes of variation between genetic components, shared environment, and unique environment, the authors were able to infer that the genetic component was responsible for most of the observed stability in IQ from age 20 to age 55 (71.3% of the longitudinal correlation was genetic). There was, however, a change of 10 IQ points in half the individuals, and a change of at least 20 points in one fifth of the individuals. The authors concluded that these massive changes were overwhelmingly (83.1% of the longitudinal correlation) due to aspects of the environment not shared by twins. Therefore, genetic factors were primarily responsible for stability, and environmental factors were primarily responsible for changes in the actual value of IQ. Does that mean, then, that genes are at least the main force for the stability of IQ? As discussed next, the answer is “it depends”. The very correlation between genetic effects and IQ (or, in other words, the heritability of IQ) also varies across lifespan and SES.
3.2. Heritability changes across lifespan and socio-economic status The heritability of body weight is quite high at five years of age (heritability = 0.95), but it
typicallydecreases across the lifespan, plateauing at approximately 0.60 by 50 years of age.
This decline in heritability reflects the fact that while genes set some initial parameters for body weight, lifestyle choices have more dominant later influence. While genetics can determine the physiological predisposition for body fat, such influence is less pronounced as lifestyle choices accumulate with age. The changing heritability of body weight illustrates
that this trait is not geneticallydetermined, but rather, is quite malleable despite the initially
strong genetic influence.
Like body weight, intelligence is also subject to a change in its heritability across the lifespan. However, the pattern of change for intelligence is quite different than that observed for body weight, and many other traits. Intelligence can be reliably quantified beginning at about age 4–5. In populations of this age, the heritability of IQ is estimated at approximately 0.22. By 16 years of age, the heritability of IQ is estimated to be 0.62. Even more striking, by age 50 (at which time the heritability of body weight has declined precipitously), the heritability of IQ is commonly estimated to be 0.80 (with estimates ranging as high as 0.90;
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Bouchard, 1997; Haworth et al., 2009). This increase in heritability is not simply an artifact of changes in our ability to measure IQ. While IQ can be difficult to measure accurately at age 4–5, its measurement at age 16 is as reliable as it is later in life. Instead, the increase in IQ’s heritability with age probably reflects an underlying role of GE interplay in creating IQ differences between individuals.
Gene-environment interplay may underlie the increase in IQ’s heritability in a way that may not be immediately intuitive. Importantly, the genome is largely established at birth, so a population is not gaining much genetic variation as it ages (except for age-dependent genes). This suggests that the increase in heritability across lifespan cannot be explained by pure changes in independent genetic effects. Alternatively, it is reasonable to expect that an individual’s intelligence influences that individual’s attraction toward a particular cognitive environment. Individuals with disparate intelligence are likely to find themselves pursuing very different cognitive challenges, while those with similar cognitive abilities are likely to
gravitate to more similar cognitive challenges. Thus, those with similar IQs becomemore
similar, while those with disparate IQs become less similar, with the net effect being an increase in the estimate of heritability. As this runaway process occurs (via the passive, active, and reactive rGE described above), genetic differences that underlie early differences in IQ can be amplified by the accumulation of cognitive challenges offered by different environments (Lykken, Bouchard, McGue, & Tellegen, 1993; McGue, Bacon, & Lykken, 1993 ). Children with slightly higher IQs end up mated with the environment (and choose an environment) that is appropriate for their cognitive abilities, which can in turn promote further gains in intelligence. Conversely, children with slightly lower IQs may gravitate toward less challenging environments, and may come to have relatively lower IQs as adults.
Part of the increase in heritability across lifespan could also be due to Ga×Ec interactions as well, and not only correlations. Interactions with the shared/familial environment are known to inflate estimates of Ga for a trait (Lathrope, Lalouel, & Jacquard, 1984). The Ec behind IQ variation is quite high during infancy (around 0.55) and early adolescence (around 0.30), but falls to single digits by ages around 17 to 23 (Flynn, 2016). So, during late teenage years and early adulthood, Ec could be channeled towards heritability by interacting with differences in genes to explain the sudden and quick raise in IQ’s heritability. And this process could be cumulative, as intelligence is expected to have a high number of G×E interactions later in life (as feedback loops and networks accumulate), and have fewer of them early in life (which as we have described is the opposite for what is expected for heritability in most traits). This theme is also discussed by Tucker-Drob and Briley (2014), and may have repercussions for the way we understand the genetic stability of intelligence. As Ga×Ec compound with age, the effects of shared environments become ever more tied to genotypic differences. The authors suggest that the accumulation of Ga×Ec “may also help to explain why the stability of the shared environment increases to such a high level. As recurrent objectively shared experiences increasingly differentiate individuals on the basis of their genes, it is possible that the only remaining shared environmental main effects are those that have resulted from particularly severe and lasting early environmental experiences that all humans respond similarly to” (Tucker-Drob & Briley, 2014).
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
for 39% of the variance in cognitive abilities (with 45% attributed to shared environment). Meanwhile, among the twins from the wealthiest families (94th^ percentile of US income, equivalent to $200,000 today), genetic effects accounted for 55% of the variance in cognitive abilities, and 35% was attributable to shared environmental influences. The adolescents sampled were all takers of the National Merit Scholarship Qualifying Test, and so we expect very few to be living in extreme poverty. Because of that, Harden et al. (2007) concluded that “genotype-by-environment interactions in cognitive development are not limited to severely disadvantaged environments, as has been previously suggested.”
The results above suggest that environmental differences between low, middle, and high SES families influence the expression of genetic potential for intelligence. Differences in genes are more accentuated in favorable environments, while on the other extreme, differences in familial environment are strongest for IQ’s variation among poor families. In fact, Bronfenbrenner and Ceci (1994) predicted these results and the importance of GE interplay in their bioecological model a decade before any empirical support (see discussion above). This pattern was confirmed in a more recent meta-analysis (Tucker-Drob, Briley, & Harden, 2013 ). Based on aggregated data from 11 studies that followed twin and adopted samples from birth to 18 years of age, Tucker-Drob et al., reported that in infancy, genes accounted for less than 25% of the variability in IQ, whereas the shared family environment accounted for approximately 60%. By late adolescence, this pattern had reversed, with genes accounting for 70% of the variability in IQ and the shared family environment accounting for near 0%. In the same paper, Tucker-Drob et al. traced a beautiful parallel between the meta-analysis results for heritability changes across life span with previous results (already detailed here) of heritability changes across socio economic status by Harden et al., 2007, and Tucker-Drob et al., 2011. In the parallel made by Tucker-Drob et al. 2013, the graph for heritability by lifespan and the graph for heritability by SES are remarkably similar! In both cases, the heritability of IQ is low precisely when GE is expected to be low (in young age due fewer relevant experiences, and in poor populations due to lack of opportunities), while heritability of IQ is high when GE is expected to be high (in old age due to accumulated experience, and in wealthy populations due to more opportunities). It’s hard to imagine how Ga alone could have affected the continuum of SES and lifespan in a similar manner to explain the same pattern of heritability changes. Independent genetic effects cannot be the only important way in which genes affect IQ’s heritability – G×E and/or rGE effects matter.
Common descriptions give the misimpression that IQ’s heritability is always high, and that the shared environment plays only a trivial role^5. Given the data on IQ and SES discussed above, the opposite is more likely true. In a global scale, 80% of humans are below US’s and
(^4) Note that quantitative conclusions based on the studies of Turkheimer et al. (2003) and Tucker-Drob et al. (2011) are somewhat complicated by a lack of detail regarding the range of incomes sampled, or the average incomes of the upper SES brackets. However, the reported results are qualitatively consistent with those reported by Hardin et al. (2007), and this latter study does not suffer from the same ambiguities. (^5) Some have argued that the relationship between environmental and phenotypical variance has been tested for, and those tests generally show little or no reason to suspect a G×E interaction. Were we to accept those earlier results at face value, they would be in clear conflict with the data reviewed above (as well as our broader conclusions). These arguments against G×E influences are based on what has been described as studies of “environmentality” (e.g., Plomin & DeFries, 1979; Thompson, Detterman, & Plomin, 1993; Thompson et al., 1993). To precisely perform such analyses, it would be necessary to quantify environmental variance. The difficulty arises in that while genetic variance (or its proxy, familial resemblance) can be precisely specified, estimates of environmental variance are at best qualitative. Environmental history is vague or unknown, it changes with time, the critical components of the
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Europe’s poverty lines (for an in-depth and insightful analysis on global inequality, see Milanovic, 2013.), and so the shared environment is likely to be an extremely powerful cause of variation in intelligence in most populations. Moreover, it is plausible that these high values of the shared environment component reflect (the same way high values of the genetic component probably do) a high G×E (and/or rGE) contribution to the estimate of heritability.
3.3. The Flynn Effect IQ scores are standardized, and so the average score of a population is necessarily 100. As tests are revised, the test maker designs each revision to maintain that average (based initially on large preliminary test samples). Consequently, if the population were to become “smarter”, the test maker would need to increase the test’s difficulty in order to maintain the average score of 100. This is exactly what occurred throughout the 20th^ century, where in France for instance (where IQ scores were obtained for the entire population of 18-year-old males), IQ increased at least 20 points just between 1950 and 1980. Thus, although the
average IQscore remained constant, the actual intelligence of test takers had increased. This
phenomenon, referred to as the “Flynn Effect”, was first formally described in detail in 1984 by James Flynn (Flynn, 1984).
Although the Flynn effect is a worldwide phenomenon, it occurs predominantly in countries transitioning into what we consider today to be a “developed” society, both in social aspects like education and health access, as well as in economic aspects like per capita GDP and industrialization. In the United States, IQ increased by approximately 14 points between the years 1932 and 1978 (Flynn, 1984), and similar gains of three IQ points per decade were observed during the last century in France, Great Britain, the Netherlands, Australia, Canada, Germany, and Japan (Lynn & Hampson, 1986). In recent decades, gains in IQ also began to emerge in developing countries such as Turkey, Sudan, and Dominica (Khaleefa et al., 2008; Meisenberg et al., 2005; Rindermann et al., 2013). Worldwide, a meta-analysis of 53 studies conducted in industrialized societies showed an increase of 17.6 IQ points occurred between 1951–2011, translating to an average increase of approximately 2.9 IQ points per decade (Trahan et al., 2014). These increases are far from trivial. According to the Wechsler IQ classification scheme, this increase translates to an equivalent shift (when comparing across the 20th^ century) from “average” to “superior” intelligence.
Many explanations for the Flynn Effect have been considered, such as genetic heterosis (the “hybrid vigor” effect from miscegenation) (Mingroni, 2007), improvements in nutrition/ health (Lynn, 2009), reduced pathogen stress (Eppig, Fincher, & Thornhill, 2010), reduced family size (Sundet, Borren, & Tambs, 2008), and test-takers’ familiarity with formal testing methods (Tuddenham, 1948). Although each of those factors may have some small
environment are a matter of speculation, and any measurement of environmental variance would be “nominal” at best. Ronald Fisher was aware of this problem, and his revolutionary quantitative techniques were only able to split a trait’s variation by giving precedence to what could be known: thegenetic effects (see Section 2.2 above). In this same tradition, rather than directly measure environmental variance, studies of environmentality compute the ratio of genetic and environmental covariance to the phenotypic correlation, yielding estimates of bivariate heritability and environmentality. In other words, this strategy assumes that the “left over” environmental variance is the inverse side of heritability. Thus, it is unavoidable that environmental effects will necessarily be low;all of the G×E interaction (which was not empirically determined) was already assigned to Vg!
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
