January 25, 2017

The Heritability of IQ

  1. Introduction

Monozygotic twins, who share 100% of their DNA, have IQs that correlate, on average at .76, meaning their IQs are far more similar than average, even when they were adopted away at birth and raised in totally different homes. MZ twins that grew up in the same home have IQs that correlate at .86, meaning that their IQs are only slightly more similar than they would have been if they never knew each other.

Over the last 100 years, a huge numbers of studies have been conducted which looked at how the degree of IQ similarity between MZ twins, DZ twins, siblings, half siblings, and parent-offspring pairs changes depending on whether or not they grow up in the same home. By comparing the degree of IQ similarity found among these relatives when they live in the same home, and comparing that to how similar it is when they are reared apart, and then comparing this to how similar the IQs are of people who were adopted into the same home but are biologically unrelated, behavioral geneticists can estimate the amount of IQ variation which is due to genes and to the environment. Large reviews of more than 200 such studies have determined that the heritability of IQ lies somewhere between .5 and .7, meaning that 50-70% of IQ variance in the population is explained by genes (Bouchard and McGue, 2003Hunt 2011 ). By contrast, shared environment, which refers to those aspects of the environment which make people who live in the same home more similar than average, accounts for around 20% of IQ variance. Unshared environment, or environmental factors which do not make members of the same home more similar, accounts for even less.



Source: Hunt (2011)

  1. The Wilson Effect 

Reviews of this large literature can be a little misleading though, because they don’t separate studies by age and most studies are done on children. When you do segregate the studies by age, a clear pattern emerges: the heritability of IQ rises with age, hitting around 85% in late adulthood, and the influence of the shared environment falls to around 0% by young adulthood. In fact, in adulthood the IQs of people who are biologically unrelated but who were raised in the same home aren’t significantly more similar than the IQs of any random pair of people picked from the general population. In other words, differences in the homes people grew up in explains basically nothing about IQ variance in the adult population. This fact, which behavioral geneticists have labeled “The Wilson Effect”, has been shown in studies using a wide variety of methods over several decades utilizing data on thousands of twins and siblings (Bouchard 2013).


A = Additive genetics, C = Shared Environment, E = Unshared Environment

Source: Haworth et al (2010)


Source: McGue et al. (1993)


Source: Bouchard and McGue (2003)


Source: Plomin et al (1997)

Many people find this evidence to be highly counter-intuitive. Common sense tells us that facts about the home we grew up in, such as parenting style, the food our parents gave us, and our socio-economic status, impact how smart we are as adults. And yet, this just isn’t true.

The best explanation for this fact is that people with “high IQ genes” tend to put themselves into more cognitively enriching environments than those who are less genetically gifted. These environments do boost people’s IQs, but, as soon as they are able to choose their own environments, people genetically inclined to be less intelligent will chose environments that “correspond”, so to speak, with their genotypic IQ. Thus, while parents may be able to help their children while they are at home, and therefore subject to the environments that their parents force on them, this will not change the environments that children end up choosing for themselves once they leave home. And these environments will largely reflect their innate cognitive potential.


Several lines of evidence suggest this explanation is correct. First, the same genes explain intelligence during childhood and adulthood (Trzaskowski, Visscher, and Plomin, 2014). This disqualifies hypotheses which explain the Wilson Effect with reference to genes which don’t begin to impact intelligence until adulthood. Secondly, many environmental conditions which are known to impact intelligence, and which correlate with intelligence, such as socio-economic status and educational attainment, have been shown to be significantly heritable (Krapohl et al., 2014;  Branigan, McCallum, and Freese, 2013; Hyytinen et al., 2013Trzaskowski et al., 2014). In other words, a person’s genes predict their likelihood of being in environments which are themselves known to increase IQ. This is what we would predict if “high IQ genes” caused people to seek out cognitively stimulating environments. And thirdly, major interventions that attempt to give disadvantaged children better environments produce IQ gains while these interventions are ongoing, but these IQ advantages completely goes away by adulthood (Protzko 2015). This is easily explained by this model IQ development: these interventions put children in IQ boosting environments without motivating them to put themselves into IQ boosting environments once the intervention has ended and, consequently, their IQs fall right back down to where they were before once the intervention ends.

  1. The Basis of Objections 

Expert opinion on this topic sides with the hereditarians. For instance, a letter signed by 52 leading intelligence experts entitled “Mainstream Science on Intelligence“stated the following: “Individuals differ in intelligence due to differences in both their environments and genetic heritage. Heritability estimates range from 0.4 to 0.8 (on a scale from 0 to l), most thereby indicating that genetics plays a bigger role than does environment in creating IQ differences among individuals.” Similarly, when the American Psychological Association commissioned a task force to summarize the science on intelligence it concluded “Across the ordinary range of environments in modem Western societies, a sizable part of the variation in intelligence test scores is associated with genetic differences among individuals.” (Neisser et al., 1997)

However, the mainstream view does have its critics. The criticisms are mostly aimed at two assumptions. First, that the only reason why MZ twins are more similar in IQ than DZ twins is due to genes, and second, that the studies used by behavioral geneticists capture the full range of environments experienced by the general population. Both of these propositions are assumed by the mathematical models behavioral geneticists use to measure IQ’s heritability. To the extent that they are false, said models will tend to over-estimate how heritable intelligence is.

  1. Sign Up Bias 

A common objection along these lines is that really abusive parents aren’t going to want to sign up for psychological studies. As a result, behavioral genetics studies really only measure how much IQ variation is due to genes after excluding the “really bad” environments from their data set. If they included the really bad environments, shared environment would explain more IQ variation than the mainstream studies claim it does.

This argument sounds reasonable, but it has been totally refuted by studies which have either used the military or the national school system to measure the IQ of everyone (or, for the military, every male) in an entire country. These studies produce the heritability figures that are totally consistent with the general literature on the heritability of IQ (Sundet et al., 1988Benyamin et al., 2005).

This is not to say that really bad environments can’t damage IQ. Rather, this indicates that environments which are so bad that they have a negative impact on IQ which lasts into adulthood are so rare, in first world nations at least, that they explain very little of the total IQ variation across the population.

  1. Restriction of Range

Similarly, some people argue that adoption agencies typically favor middle to upper class married couples with no criminal records and with a basic knowledge of how to parent properly when selecting which applicants to give children to. Because of this, people adopted into separate homes mostly get placed into pretty nice homes, and therefore experience environments which are both more similar, and better, than average. To the extent that this increased similarity leads to increased IQ similarity, it will cause heritability to be over-estimated.

The only study to ever directly compare adoptive and non-adoptive families from the same sample found that, yes, adoptive homes are better and more similar than average in terms of income, parenting styles, and parental mental health, but statistically correcting for this didn’t change the IQ heritability figures one iota because these variables were also seen to not effect IQ in this adoptive sample (McGue et al., 2007). Moreover, as just mentioned, studies which do include a fully representative sample of entire nations produce the same heritability figures as other studies do.

  1. Non Total Separation 

A somewhat different argument posits that twins are often adopted considerably after birth, and get to know each other substantially prior to their enlistment in a study, and so, contrary to what adoption studies assume, have abnormally similar shared environments to some degree. It is true that some adoption studies have had less than perfect separation criteria. However, several studies have shown that the amount of time that twins adopted into separate homes spend together prior to a study does not impact their IQ similarity and so does not inflate heritability estimates (Bouchard 1983Bouchard et al., 1990).

This is what we should expect given what we know about IQ in general. As was previously noted, early interventions don’t matter after they stop and the shared environment only matters until kids leave home. This suggests that intelligence is like a muscle, and, once your parent’s aren’t controlling you, differences in the degree and ways in which people exercise that muscle are largely heritable. Normally, this is only seen once in a lifetime because we only leave home once. But this doesn’t need to be so. People that are adopted “leave home” more than once, and the fact that IQ similarity between twins is unaffected by how much time they spent together before they were adopted suggests that the cognitive “exercise regimen” they got from their first home doesn’t stick with them once they are adopted, and it stops impacting their IQ as soon as they stop practicing it.

  1. Twin Treatment 

A common objection raised against estimates of IQ heritability based on twins who were reared together suggests that MZ twins will be treated more alike than DZ twins and, as a result, be more similar in IQ. This implies that MZ and DZ twins do not have equally similar shared environments and so a major assumption of twin studies is violated.

This argument was first empirically addressed by Loehlin and Nichols (1976) who looked at data on over 850 pairs of twins. Loehlin and Nichols measured the degree to which parents treated these twins the same way, the degree to which they were dressed alike, whether they had been put into the same classes, whether they slept in the same room, etc. They then measured the correlation between how similarly the twins were treated by their parents to how similarly they were in IQ. They found that increased similarity of treatment predicted almost no increased similarity in IQ.

More evidence on this question comes from studies of misperceived zygosity. “Misperceived zygosity” refers to a situation in which an MZ twin pair is thought to be a DZ twin pair or vice versa. Then, often when these twins enter into a behavioral genetics study, they find out that they are actually the other kind of twin pair. Until this reveal, these sets of DZ twins were treated as MZ twins by everyone they knew, and the MZ twins were treated as DZ twins by everyone they knew. So, if this critique of twin studies is valid, we would predict that mis-classified MZ twins will be less alike in IQ than correctly classified MZ twins and mis-classified DZ twins will be more alike in IQ than correctly classified DZ twins. However, several studies have shown that this is not the case (Scarr and Saltzman, 1979Conley et al., 2013) .

Thus, both lines of evidence show that this critique of twin studies does not stand up to empirical scrutiny.

  1. Prenatal Effects

One common problem with the arguments we’ve considered thus far is that they presuppose that the shared environment matters. As we’ve already seen, in adulthood, the behavioral genetic evidence says that the shared environment’s impact is basically zero. Of course, zero times ten is still zero. Given this, arguments focused on the idea that MZ twins have a shared environment which is X times more similar than average, or that the full range of the shared environment isn’t captured in studies, shouldn’t matter much for estimates of the heritability of IQ in adulthood.

This last objection isn’t like that. It posits that MZ twins have more similar IQs than DZ twins because they have more similar pre-natal environments than DZ twins do. There is no evidence to suggest that the effects of the pre-natal environment go away with age. Moreover, twins who are separated at birth still share the same pre-natal environment, and biological non-relatives adopted into the same home matured in totally different pre-natal environments. Thus, this objection, if true, is the most damning argument we’ve encountered thus far.

Let’s look at this objection in more detail. MZ twins normally share one chorion and, therefore, one placenta. By contrast, DZ twins each have their own chorion and, therefore, their own placenta. Thus, MZ twins do have a more similar prenatal experience than DZ twins do, but does this make a difference for IQ?

Luckily, this is easy to answer. About 1 in 4 MZ twins do not share a single chorion and so have separate placentas. Multiple studies have confirmed that “monochorionic ” MZ twins are no more similar, in terms of IQ, than “dichorionic” MZ twins are (Beijsterveldt et al., 2015Jacobs et al., 2001Rose et al., 1981). Thus, the more similar pre-natal experience of MZ twins relative to DZ twins does not lead to increased IQ similarity.

  1. Genome Wide Complex Trait Analyses

Beginning just a few years ago, sequencing people’s genomes became cheap enough that researchers could afford to bypass all the concerns about twin and adoption studies by just measuring the genetic similarity of pairs of unrelated individuals and seeing how well that predicted IQ similarity. This technique is called genome wide complex trait analysis (GCTA). These newer studies largely confirm the findings of early twin and adoption based studies.

Before reviewing this literature, it is important to note that GCTA gives us a lower bound estimate of the heritability of a trait. That is, it tells us an estimate of the lowest possible value of heritability. This is for three reasons.

First, in GCTA not all of the genome is measured. GCTA will only detect that proportion of IQ variation which is explained by the parts of the genomes that researchers measure (or which are correlated with parts of the genomes they measure). Since the whole genome is not measured, IQ variation caused by unmeasured parts of the genome will not be detectable by GCTA and, therefore, attributed to the environment.

Secondly, IQ variation that is explained by “rare variants” may not get picked up by GCTA. A rare gene variant is one that exists at a very low frequency in the population. Because of their inherent rarity, a rare gene variant which impacts IQ may simply not exist in the sample of a given study.

Thirdly, unlike MZ twins, people who share a little more DNA than average are unlikely to share combinations of genes. This is important because sometimes the effect of a gene on a trait is dependent upon which other genes are in a person’s genome. To the extent that this is true of IQ, GCTA will under-estimate IQ’s true heritability.


With these caveats in mind, let’s look at the actual research that has been done in this area. We’ll start by looking at the results of GCTA on IQ with a study that measured the Wilson Effect. Trzaskowski, Visscher, and Plomin (2014) utilized participants in the Twins Early Development Study (TEDS) which included over 11,000 twin pairs born in England between 1993 and 1996. GCTA was performed on 2,875 unrelated individuals of which included 1,334 individuals who had IQ measured at both ages. Twin based heritability estimation had a sample of 6,702 which included 2,269 twin pairs who had IQ measured at both ages. GCTA heritability rose from .26 at age 7 to .45 at age 12. Twin based heritability rose from .36 at age 7 to .49 at age 12. Thus, GCTA estimates accounted for 74% of the twin estimate at age 7 and 94% of the twin estimate at age 12.

Trzaskowski, Visscher, and Plomin is the best study on this topic because it is the only study to measure heritability using both twin methods and GCTA in the same sample and to follow participants as they age. Here are, so far as I know, the only three other GCTA studies on intelligence:

Benyamin et al (2014) genotyped a sample of 18,000 children which were broken into several samples. GCTA based heritability ranged from .22 to .46.  Beyamin cites Haworth et al (2010), a meta-analysis of over 11,000 twins, as giving a heritability estimate of .41 for the same age group. Thus, twin based heritability fell within the range of CGTA based heritability.

Marioni et al (2014) genotyped 6815 individuals with a median age of 57. The traditional heritability estimate was .54 while the CGTA based heritability estimate was .29. Thus, the CGTA estimate accounted for 54% of the traditional heritability estimate.

Davies et al (2011) genotyped 3511 unrelated adults and found that the CGTA based heritability of crystallized intelligence was .44 and the heritability of fluid intelligence was .51. (Crystallized intelligence refers to people’s level of stored knowledge while fluid intelligence refers to their ability to perform more novel cognitive tasks). This suggests a heritability of over all intelligence was found to be in the high 40’s, which accounts for roughly 2/3 of twin based estimates of adult heritability.


As we can see, GCTA based heritability is normally lower than twin based heritability, but that is what we would expect since it is a lower bound estimate. More specifically, on average, studies that compare GCTA and twin or adoption based heritablity estimates for samples matched for age find that the GCTA heritability estimate accounts for 75% of the twin/adoption based estimate. Thus, the heritability figures generated by GCTA are similar enough to twin based ones such that they provide strong evidence for twin and adoption studies.

  1. Find the Genes

Hopefully, it is clear by now that claims about the heritability of IQ make predictions which can be tested without ever identifying a single gene that impacts IQ. In other words, we don’t need to “find the genes” to provide scientific evidence that IQ is largely heritable. That being said, we’ve found a lot of the genes.

From the outset, it is important to realize that IQ is influenced by thousands of genes that have tiny effects. Because of this, huge sample sizes are required in-order for a study to have the statistical power to register these small effect sizes as statistically significant.  Until recently, this was impossible to do due to technological limits. As a result, all the studies on IQ related genes that I will reference have been released within the last 3 years or so.

The basic strategy behind these studies is simple: find gene variants which correlate with higher or lower than average IQ in one sample and then replicate that association in a second sample. If the replication is successful, then we have evidence a a particular gene variant impacts IQ.

It may occur you that lots of genes might correlate with intelligence because they correlate with other traits that correlate with intelligence such as race or SES. Researchers typically avoid this problem by relying on racially homogeneous populations and by preforming a statistical technique called principal component analysis on gene covariance. If you don’t know what PCA is that’s fine, but it is beyond the scope of this post to explain it. So just know that some magic math is done and that for some reason a lot of “expert” people think that this helps alleviate this concern.

So, with that said, here is a summary of the gene variants that has been have been found to predict IQ in more than one sample:

The first set of studies worth mentioning are Reitvald et al (2013) and Ward (2014). In combined samples of over 200,000 people, they found that the same 3 gene variants were consistently associated with measures of intelligence and with grades. This not only tells us something about the genes that impact IQ but also tells us that, to some degree, the same genes that effect intelligence also impact educational achievement.

Reitvald et al (2014) involved three samples. One contained 107,736 individuals and measured educational attainment. Another contained 24,189 individuals and measured general intelligence. A third sample contained 8,652 old individuals for whom dementia and memory was measured. Reitvald found another three gene variants associated with educational attainment which were also found, in a separate sample, to be associated with intelligence.

Davies et al (2015) utilized three samples (Ns= 53,949, 5,976, and 5,487) which were genotyped and had their general intelligence measured. 13 gene variants were found to predict cognitive ability at a high level of statistical significance across the samples.

Ibrahim-Verbaas (2015) utilized 40 samples which ranged in size from 1,311 to 32,070 people. The cognitive traits looked at were executive functioning and processing speed. One gene variant  was found to be a significant predictor of intelligence in some samples but not others. However, Benjamin (2013) found this same gene variant predicted intelligence in a discovery sample of 12,441 individuals and a replication sample of 5,548 individuals. So this gene variant probably does predict intelligence to some degree.

As can be seen, we are quickly gathering a growing list of IQ related genes. However, each of these genes explains a tiny fraction of IQ variation. In the years to come, we will no doubt have a much larger list. We may even be able to use said list to predict someone’s IQ based on their genome.

  1. Conclusions and Implications

To summarize, the major findings on the heritability of IQ are that IQ variation between individuals is mostly explained by genes and, among adults, basically no variation is explained by the shared environment. These results have been confirmed by twin studies, adoption studies, and new studies in genetics.

It is important to understand what these results do and do not imply about group differences in intelligence. The fact that genes which impact IQ vary substantially from person to person make it plausible that genes play some role in racial IQ differences. However, the heritability of differences within two groups is not the same as the heritability between them. For instance, within two gardens height differences might be entirely due to genes, but between them they might be caused by the fact that one garden has good soil while the other has poor soil.

On the environmental side, the data reviewed in this article makes it unlikely that shared environmental factors, such as socio-economic status or parenting style, explain much about why some groups are smarter than average in adulthood. After all, if your childhood shared environment basically stops impacting your IQ once you are an adult, and therefore explains basically nothing about individual differences in intelligence, it is hard to see how it could explain group differences.

Of course, this does not mean that group differences are caused by genes. They could also be caused by unshared environmental factors, such as racism or sexism. That being said, on the whole this data obviously makes the hereditarian case on group intelligence differences stronger than it would be if IQ had a very low heritability or if it was more strongly impacted by the shared environment.


Facebook Comments
  • Doom Guy

    Will you guys ever make Youtube videos again? I think they’d be a little bit more palatable for much of your intended audience.

    • Yes, we plan on starting to make videos again pretty soon.

  • Emil Kirkegaard

    Okbay et al 2016 found some 165 SNPs associated with education in their combined sample. Most of these alleles will actually be for cognitive ability and some will be for other traits.


    Gwern has meta-analyzed the GCTA studies for IQ:


    Overall estimate about .32. There debate about some problems with GCTA. A study using another method across 19 traits found overall 42% higher heritabilities. So, this brings the value up to about .45. I think the new method only captures additive effects too, so this is still the additive/narrow heritability, not the broad heritability (about 80% in adults).