How much variation in human behavior is due to variation in our genes? Answer: quite a bit.

May 24, 2021 • 9:45 am

The only people who claim that behavioral variation among people has nothing to do with their genes are ideologues: “pure blank slaters.” Based on studies of other species, we know that virtually all studies of traits that vary among individuals—be those traits morphological, physiological, or behavioral—show that some or even a lot of the variation in a population is based on variation in the genes of different individuals. (I know of only three studies in animals, out of thousands done, that failed to find a genetic basis for variation among analyzed traits.  Two of the three studies were mine, and all were on directional asymmetry: trying to see if there’s a genetic difference for, say, having more bristles on the right than on the left side of a fly.)

While humans have an extra source of inter-individual variation—culture—there are ways to get around cultural inheritance to see how much of human variation is based on genetic variation. There are several ways to determine and measure the contribution of genetic variation to variation among individuals.

First, though, let’s learn the technical term at issue: heritability. Heritability is a measure that ranges from zero to one (or 0% to 100%), and tells you how much of the observed variation among individuals in a population is based on genetic variation among those individuals.  The higher the heritability, the more genetic determination there is of variation in the trait. So, for example, if the heritability of human height for females is 0.7 (or 70%), that means that if you measure the variation in a population for height of women (the variation is conventionally estimated by calculating the variance—denoted by σ² or s²—then of the total variance for height, 70% of that estimate is due to variation of individual’s genes. In this case, there’s a high degree of genetic control of variation in height. (This is close to the actual figure for female height.) Confusingly, the symbol for heritability is h².

Now it’s a bit more complicated than this, for heritability incorporates only what we call the additive genetic variance rather than other kinds of genetic variance (usually minor). And course, there are other obvious sources of variation in height—most prominently nutrition and health.  The more that environment contributes to differences among individuals in a trait, the lower the genetic heritability. So, for example, the heritability of hair color among adults has been reduced by the introduction of an environmental source of variation: hair dyes. Less of the variation in hair color that we see, compared to, say, 200 years ago, is due to genetic variation.

It’s important to grasp several caveats about heritability. First, it is a figure that applies to one interbreeding population at one time—the time of measurement. You cannot apply heritability in one group to a different group that may have different genes and, importantly, different amounts and sources of environmental variation that affect a trait. A population undergoing famine, for example, may show a lower heritability of height because individuals’ heights may be altered by grossly different amounts of food they get.

Second, heritability says very little about how much a trait can be changed, for genetics isn’t destiny. Yes, the heritability of height may be 70%, but that doesn’t mean that we can’t make people bigger or smaller by feeding them a lot of good food, injecting them with growth hormones, or starving them. When people object to measuring heritability of IQ, for instance, they often mistakenly think that because IQ has a sizable heritability, which it does, it therefore can’t be changed. But that’s bogus; there are many possible interventions that can affect IQ.

Third, as implied above, measuring heritability within a group tells us very little, if anything, about the genetic basis of difference among groups. That’s because there may be environmental differences between groups that affect the character in profound ways and make extrapolations from one group to another useless. You cannot conclude, for example, that measures of behavior, mentation, and so on, in one population or ethnic group, even if the heritabilities are large within that one population, must therefore mean that big differences between groups must rest on genetic differences between those groups. The heritability of height in the Japanese population right after WWII, for example, was probably considerably smaller than 70% because of wide variation of nutrition in the Japanese. Thus you can’t conclude that their smaller height than Americans at the time was based on genetic differences. In fact, the average height of Japanese people went up three to four inches in about thirty years! So much for the idea that a substantial heritability for a trait in one group means that that trait can’t be changed!

Fourth, there’s a common misconception that heritability tells you “how much of your height (or weight, or IQ) is genetic”.  A heritability of 80% for height doesn’t mean that, if you’re five feet tall, four feet comes from genes and one foot from the environment. That conclusion is biologically meaningless since the height of an individual involves an interaction between genes and environment. The only sensible way to construe heritability is to say that it tells us how much of the VARIATION we see from individual to individual within a population is based on differences in their genes (or rather, forms of their genes: different “alleles”).

How do we determine heritability? There are several ways. In species like plants and animals in which we can perform artificial selection, we can estimate heritability by seeing how much a population responds to artificial selection for the trait. The bigger the response, the higher the heritability. There is an equation that tells you this: the response to artificial selection is roughly equal to the strength of selection (how much difference there is in the average trait in the group you choose for breeding and that of the population in general) multiplied by the heritability. If the heritability of head-to-tail length in pigs is 50%, and you choose for breeding a group of pigs whose average size is two feet longer than that of the population, you’d expect the next generation of swine to be one foot longer than the original population (0.5 X 2 feet). So if you know how strong you’re selecting, which you do, and what the response is in the next generation, you can back-calculate to estimate the heritability of the trait you selected.

Artificial selection isn’t practiced in humans, of course, so we usually determine heritability by looking at the correlation between relatives, including parent-offspring correlation and the correlation between twins. Parent-offspring correlation is dicey if there’s an environmental component to the trait that can also be inherited. I seem to remember that the two traits with the highest heritability in humans are religion and wealth, and that’s due to the passing on of these traits among generations via culture, not via genes!  In animals that have no transmissible culture, like fruit flies, one can, however, do these kinds of studies.

Humans researchers often use twin studies, comparing the similarity between identical twins (which have the same genes) with the similarity between fraternal twins, which share half their genes. We all know that identical twins are more similar for virtually every behavioral and morphological trait than are fraternal twins (just look at them!), implying that genes play a big role in these traits. You can in fact estimate the heritability of traits by looking at the difference in the correlations of identical vs. fraternal twins (you need a decent-sized sample of twins to do this).

There’s one caveat here, too, however. Identical twins often share more environmental commonalities than do fraternal twins. They may be treated more alike, dressed alike, brought up alike, and educated more alike than are fraternal twins. Thus an increased similarity of identical twins need not reflect the identity of their genes, but a greater similarity of their environments. At least for physical appearance, though, that doesn’t seem to be the case: you can’t “socialize” identical twins to look more alike than do fraternal twins!

One way around the possible environmental similarity is to compare fraternal twins raised together with identical twins separated at birth and raised apart. If the latter still show appreciably greater similarities despite their different environments, that’s a sure sign that the traits measured have substantial heritabilities. However, as you can imagine, there aren’t big samples of identical twins separated at birth.

Finally, we can measure heritabilities using DNA, by “genome-wide association studies”. This is more complicated, but involves finding those regions of the DNA associated with variation in a trait (like height), and then adding up the small effects of all known regions to see how much these known bits of the genome can contribute to variation among individuals. Heritabilities measured in this way are invariably smaller than those measured by correlations or selection, as the latter two methods take into account every region of the genome contributing to variation. Variation in most behavioral traits is due to many genes of very small effect, and it’s nearly impossible to find them all by association mapping.

This is all a very long prologue to a very short figure I’m going to show you—a figure that comes from this new paper in Nature Human Behavior. It summarizes heritability data for a number of behavioral traits, comparing heritabilities from twin studies to those from association studies. Click on the screenshot to see the paper, and I’ve put the full reference at the bottom:

And here’s Figure 2 showing the heritabilities measured both ways. Blue lines and dots give data from twin or family studies, orange lines and dots from association mapping.  You can see that association studies produce, as expected, lower estimates of heritability, but I’d expect the true values to be closer to the family-study data). 26 behavioral traits were measured, ranging from educational attainment, IQ as children and adults, amount of drinking and smoking, neuroticism, to mental disorders like schizophrenia.

Click on the figure to enlarge it.

The values in the colored triangle to the left are the “genetic correlations” between traits, which tells us the degree to which pairs of traits are affected by variation in the same genes. We need not concern ourselves with that.

For the moment, look at the lengths of the blue bars to the right, which are probably pretty close to accurate estimates of heritabilities. And for most traits they are pretty big, with over 25% of the variation in a trait due to variation in the genes within that population. For some traits, like adult IQ, number of sexual partners, alcohol dependence, autism spectrum placement, and schizophrenia, heritabilities are over 50%

What all this does is refute the “blank slate” view that the differences between people in their behavior is completely due to culture and socialization. It also shows that a substantial portion, however, is due to other sources of variation: developmental, post-birth environmental differences, and so on.  It also shows that you could select on any of these traits and get a response, increasing, for instance, the age of first intercourse (Christians take note!) or reducing alcohol dependence. NOTE: I AM NOT SUGGESTING THAT WE SELECT ON THESE TRAITS!

So have a look—and click on the figure!


Abdellaoui, A. and K. J. H. Verweij. 2021. Dissecting polygenic signals from genome-wide association studies on human behaviour. Nature Human Behaviour.

54 thoughts on “How much variation in human behavior is due to variation in our genes? Answer: quite a bit.

  1. I really enjoy content like this, maybe because I have an identical twin. It is interesting – looking at that chart, my brother and I are similar in many of the categories (education, income, sexual orientation [we’re both gay]), but quite different in others, for instance extroversion and alcohol/cannabis use.

    Also, he developed Type 1 diabetes in early adulthood. I did not.

    Really fascinating stuff.

  2. Thank you so much for posting this. When I was teaching Genetics and (especially) Evolution, I found heritability to be an extremely difficult concept for students to comprehend. The biggest problem was your point four – the idea that it’s a measure of an individual’s trait that is “due to genes.” If I were still teaching, I would assign this blog post as required reading; fortunately, however, I am 3.5 years into retirement and not missing the classroom whatsoever.

  3. Look at the heritiability of psychiatric disorders vs. the other sets of traits. I guess not too surprising, but still somewhat disappointing in terms of our ability to prevent them via changing environmental factors.

    you can’t “socialize” identical twins to look more alike than do fraternal twins!

    Ah, but let’s think about your earlier comment on hair color heritability going down because of hair dye. I’m going to make up an example, but this is actually pretty close to a real case of a family I know personally: let us say you have a set of twins with a non-twin sibling close in age. One of the twins gets their hair cut and dyed to look like the non-twin sib, always. The other twin gets their hair cut and dyed completely differently. So that with the right clothes, people often think that the one twin and their non-twin sibling are the twins, and the other twin is the non-twin sib.

    There’s an example of socializing twin appearance. And while the non-twin sib isn’t a fraternal twin, genetically a ‘fraternal twin’ isn’t any different than a full brother or sister. So yeah, it’s possible using things like hair dye to make fraternal twins more alike than identical twins, if you simultaneously use those same “environmental” factors to make identical twin appearance less alike. (And for an even more extreme change, think facial tattoo or facial scarring use.)

      1. My apologies because the delay in posting my comments is preventing me from editing, but I also intended to add: thanks for the interesting science post.

    1. One reason why heritabilities for autism and for schizophrenia are often very similar (and very high) is that many of the same genes contribute to both disorders, with a lot of sub-clinical genetic variation among neurotypical people along the continuum from autism to schizophrenia. Hamilton said that “People divide roughly, it seems to me, into two kinds, or rather a continuum is stretched between two extremes. There are people people and things people.” Those with autism are extreme things people (social and language deficits, avoidance of eye contact, fixation on objects and repetitive activities), and those with schizophrenia are extreme people people (paranoia, delusions, voices).

      1. The phenomenon you mention is described by “predictive processing” and covered brilliantly by Scott Alexander in the blog post below from Slate Star Codex.

        Top down versus bottom up processing and how this delicate balance, when out of wack (not well mediated by serotonin and other compounds), can land someone somewhere on the range between autism (bottom up overweight) and paranoia (top down overweight).

  4. Human behavior has a lot to do with what genes in one human do in response to genes in another human – is that beyond the scope of this topic, or, what is the name for that discipline?

    I am tired today so I am not writing or thinking clearly.

    1. what genes in one human do in response to genes in another human – is that beyond the scope of this topic, or, what is the name for that discipline?


      1. I suppose, but it seems genetics is an accurate science where everything stems from – bottom up. My poor understanding of sociology is that it started top-down, independent of genetics, so the researchers would not be trained in the methods necessarily.

        The idea I have sounds like “population genetics”, but the accurate sub field would be somewhere in there.

        1. Once you get past the first few thousand genes and into gene promoters, supressors, etc, and particularly into repeat counts, the interface between genetic instructions and biochemical outcomes becomes a lot less well-defined than the digital nature of ACG and T/U would imply.

  5. It’s somewhat disappointing that the two measures (GWAS and twin studies) diverge so much. I would be interested in hearing more about why the GWAS data are considered to be less “accurate” than the twin data — blank-slaters might claim that to be confirmation bias.

    Regardless, this is an endlessly fascinating topic that gets far too little attention in the popular media.

    1. It’s not disappointing but understandable given that behavioral traits are probably affected by many alleles of very small effect. To even detect those associations you’d need HUGE sample sizes of people and huge amounts of genome data. Thus we’re still missing many regions of the genome that affect behavior because we don’t have the statistical power to detect them. That, I think, will come in time.

      1. The “difference in differences (between technique results)” is an interesting subject in it’s own right. It can point out where traits are controlled by a few known genes (small difference in results of the two techniques), a wide number of genes (larger difference between results of the two techniques), and cases where we don’t really know very well which genes contribute (inconsistent results between the two techniques).

    2. “Genome wide” studies are not full-genome studies. They use only a *sampling* of genes strewn across the genome. Further — I think this is correct, but someone tell me if it’s not — they account for only one type of genetic variation, namely SNiPs (single nucleotide polymorphism) and are blind to other sorts of variation.

      Both of these limitations are practical ones, arising from what current technology allows you to do.

  6. “Artificial selection isn’t practiced in humans”. Maybe it is practiced, but only among the most educated and mainly to increase intelligence and good looks.

    1. intelligence and good looks.

      where the metrics of “good” are defined by the standards of the parents, who generally believe themselves to be close to paragons of perfection.
      Odd, the way that children, regardless of their personal inclinations and “innate” abilities are generally moulded into near-duplicates of their parents. Unless they escape.

      1. I read somewhere that 25% of PHDs marry other PHDs, and 71% of women who are college graduates marry other college graduates. I suspect that the difference in IQ between, say, the top 1% and average people is increasing, and it was smaller in the past.

        1. If the heritability of “intelligence” were high, across the hundreds (or thousands) or relevant alleles, then you would expect that. But it’s unlikely to be a simple correlation on which you can trivially calculate an expected value to compare with an observed value. It gets even more complex comparing your “experiments” with the general population because there is a high probability of your “CG” and/ or “PHD” groups contributing fewer offspring to the next generation than Joe SixPack “x” Sigourney SixSprogs. That’s going to hugely affect the contribution to the next population measure.
          Galton and Stopes might be un-people, but that doesn’t make the effects they were concerned about non-existent.

  7. Behavior similarities and differences are much harder to determine than other traits such as loss of hair. My grandfather and dad both went through this as I did. How far back I must go to establish blame, I don’t know. I have two sisters and both of them were terrible at picking husbands. One has been married twice and the other three times with all three going down the drain. There must be a gene for this type of performance.

      1. I guess it is. All my uncles from father’s side kept their hair, but went white early, while all uncles from my mother’s side got this Male baldness. Guess where I am.

      2. Odd. In my immediate bloodline the evidence is that the pattern of hair loss is inherited from Mum, not Dad. Mum still has a good head of hair (if white) in her 80s, as do I and my sisters, while Dad was well down the road to baldness before any of us finished potty training.

        What’s that word for the singular of “data”. It’s not “anecdote” …

  8. great explanation of heritabiltiy. I have found this to be one of the most difficult concepts to teach, because most folks think heritability = inheritance, which are actually very very different concepts.

    I think it should say in your 9th paragraph “(0.5 X 2 feet)”, not “(0.5 X 1 foot)”

  9. Thank you for this very clear explanation, I am opening up power-point to revise my human behavior lecture!

  10. Variation in most behavioral traits is due to many genes of very small effect, and it’s nearly impossible to find them all by association mapping.

    Do you think it’s possible that “some day” association mapping or another, more sophisticated type of mapping will be able to pinpoint the genes that cause variation in behavioral traits? Or is it more likely (like understanding consciousness) we may never fully understand the myriad genes which affect behavioral variation?

    Thanks for this fascinating post.

    1. Or is it more likely (like understanding consciousness) we may never fully understand … ?

      [OT] I had a similarly “mysterian” view of consciousness, and someone here recommended that I read Dan Dennett on the subject. I later read his 1993 Consciousness Explained, which set me firmly on a path that changed my entire view. YMMV.

      (BTW, I strongly prefer that book to his 2017 From Bacteria to Bach and Back : The Evolution of Minds. Again, YMMV.)

      1. Thanks for the recommendations. I’ve read quite a bit of Dennett, but I haven’t read Consciousness Explained. On the list.

  11. I have very little to contribute; but I would just like to thank PCC(E) for the trouble he has taken to compile such an informative (and wonderfully written) post.

    It’s interesting how particular traits manifest themselves. My mother was red-haired, but neither I nor my two sisters were. Two of my four children have red hair, and my eldest daughter has a couple of red-haired children as well – but then she’s married to a red-haired man. The genetics of all this are fascinating.

    And don’t get me started on blue eyes…

    1. Isn’t red hair a recessive trait? So it can be silent in both parent and still show up with probability 0.25 in F1 offspring.
      Not sure about eye colour.

  12. Is there any evidence for less variability in personality traits in hunter-gatherer societies? Their relatively small numbers alone must limit diversity, but I’m guessing that it may be further limited by their far fewer ways of “earning a living”.

    PS: This presupposes that an environment in which there are vastly different ways of supporting oneself—such as civilization—creates different coexisting selective pressures leading to genetic variation, and also that natural selection can happen faster than typically assumed.

  13. It’s important to realize several caveats about heritability. First, it is a figure that applies to one interbreeding population at one time … A population undergoing famine, for example, may show a lower heritability of height because individuals’ heights may be altered by grossly different amounts of food they get.

    Second, heritability says very little about how much a trait can be changed, for genetics isn’t destiny.

    Thank you so much for these remarks! Now instead of trying to explain them myself, I can just link to your post. I’ve used the famine -> height example, ironically enough, but I’ve never been able to explain the overall point quite like this.

  14. Just an extra comment to show Jerry how many of us read and appreciate these sort of posts even if we don’t have anything meaningful to add.

    Putting on my statistical graphics hat, its a nice picture but a bit confusing to use the same colour contrast for the two study types and for +ve/-ve correlations between traits.

  15. Great article. A couple of comments:

    1. “There’s one caveat here, too, however. Identical twins often share more environmental commonalities than do fraternal twins. They may be treated more alike, dressed alike, brought up alike, and educated more alike than are fraternal twins. Thus an increased similarity of identical twins need not reflect the identity of their genes, but a greater similarity of their environments.”

    This is in fact the claim of the blank slaters, who are numerous, to explain the better correlation of behavioral traits in identical vs fraternal twins. I’ve read a few papers from sociologists who periodically take a swipe at behavioural genetics and this along with the “remember the Nazis” argument are their stock in trade. Plus the “missing heritability” – see my point 2.

    There are a large number of papers within behavioral genetics that attempted to estimate the effect from identical twins vs fraternal twins by finding out how many dressed exactly the same, had the same friends, etc, etc. The result of 60 or so papers (I’m going from memory here) was 0% – 10% of the h^2 heritability estimate.

    But a much clearer validation was done by Dalton Conley, Professor of Sociology at Princeton, who apparently also thought the behavioral genetics people were naive.

    He used an interesting fact – many twins believed to be identical are in fact just fraternal – they just happen to look the same. Also, very surprising to me, some twins believed to be fraternal because they look a little different, are in fact identical – something to do with how they were lying in the womb.

    So, if it’s all about how you are treated – the blank slate explanation – then reassigning twins to the correct categories should make no difference to your heritability estimate.

    His conclusion in “The Genome Factor”, p. 27:

    “We, the social scientists who questioned the veracity of the equal environments assumption, and assumed that the behavioral geneticists were making a fundamental error, ended up confirming their ‘naive’ ACE models.”

    2. The missing heritability from GWAS and other genetics analysis is often cited by the blank slate crowd (or BS crowd for short) as “evidence” that behavioral genetics and the “equal environment assumption” is plainly wrong.

    And one of the comments earlier asked why heritability estimates from family studies should be used in preference to heritability estimates from genetic studies.

    If you look at height, which has a h^2 estimate of about 0.9, the GWAS estimate maybe 5 years ago was something like 0.2 or 0.3. I can’t remember the amount or the year. The key point is, as Jerry explained in the reply to that comment, that you need very large samples.

    No one would suggest that height is not highly heritable. I found it quite telling that the papers and articles that dismissed behavioral genetics and used “missing heritability” as a key argument never mentioned the missing heritability of height. Of course, that would knock a blow in their own counter argument.

    “Accurate Genomic Prediction Of Human Height” by Louis Lello and co-authors, including Stephen Hsu in 2017, basically demonstrated that you do need very large sample sizes. Some heavy maths from a previous paper (ref below) were used to do this, with a sample size of almost 1M. And they demonstrated very good prediction of height from genetic data only.

    The point has been demonstrated. “Missing Heritability” is not because heritability doesn’t exist, but you need very large datasets to “fill in the blanks”.

    Ref: Applying compressed sensing to genome-wide association studies, Shashaank Vattikuti et al, 2014. Stephen Hsu was a co-author.

    From that paper:

    Background: The aim of a genome-wide association study (GWAS) is to isolate DNA markers for variants affecting phenotypes of interest. Linear regression is employed for this purpose, and in recent years a signal-processing paradigm known as compressed sensing (CS) has coalesced around a particular class of regression techniques. CS is not a method in its own right, but rather a body of theory regarding signal recovery when the number of predictor variables (i.e., genotyped markers) exceeds the sample size.

  16. “Two of the three studies were mine”
    This seems like a suspicious coincidence. Could it be that there’s a “Murray Gell-Mann amnesia effect” going on and you happen to know about these because you did them, and don’t know about many other instances?

  17. “In fact, the average height of Japanese people went up three to four inches in about thirty years! So much for the idea that a substantial heritability for a trait in one group means that that trait can’t be changed!”

    There is a layer of complexity here which many people miss: the secular changes in height across time are not strictly comparable to those within populations at a given time. In psychometric jargon: there is a lack of measurement invariance. Specifically, dietary-related height changes are concentrated in limb length, the least heritable component of height.

    The upshot is that the inference you are drawing, from this example, is invalid, since the trait differences on which the heritability estimates were based is different, when understood from a latent factor perspective, from the trait differences between time.

    That is not to say that your general point is incorrect. Just that this example is not a good illustration of it. Similarly, clowns mounting stilts would also be a poor example, since this would be another apples to oranges comparison. Does that make sense?

    1. Well, I appreciate your analysis, but I still think the general point is correct: heritability of a trait within a population, even if substantial, does not show the trait cannot be changed by environmental intervention. That holds in this case because the heritability that is measured, and the trait that is measured, are for TOTAL HEIGHT. The fact that dietary changes are concentrated in the limb seems irrelevant if they are also reflected in total height, which I assume is the issue at hand.

      Note again heritability of total height is high. Total height increased substantially within a short period in Japanese people. Ergo, a high heritability doesn’t mean that the trait (total height) cannot change by environmental intervention.

  18. I have a question. Say there was transgenerational epigenetic inheritance in humans. I know that transgenerational epigenetic inheritance in humans is supposed to be very rare but what if this was not the case. Wouldn’t the twins studies give the same results even though inheritance was transgenerational epigenetic inheritance rather than genetic?

Leave a Reply