Eric Turkheimer is Hugh Scott Hamilton Professor of Psychology at the University of Virginia. In 2009, he won the James Shields Memorial Award for outstanding research in Twin Genetics. In 2024, he was awarded the Dobzhansky prize for lifetime achievement in behavioral genetics. He is a past president of the Behavior Genetics Association. His book, Understanding The Nature-Nurture Debate, was published by Cambridge University Press in November 2024.
Awais Aftab is a psychiatrist in Cleveland, OH, and clinical assistant professor of psychiatry at Case Western Reserve University. He is interested in conceptual and philosophical issues in psychiatry and writes online at Psychiatry at the Margins. His first book, Conversations in Critical Psychiatry (OUP, 2024), is an edited collection of interviews.
See:
Philosophical case conference in Philosophy, Psychiatry, & Psychology on Eric Turkheimer and Sarah Rodock Greer’s paper, “Spit for Science and the Limits of Applied Psychiatric Genetics.”
Quotes and excerpts from Turkheimer’s Understanding the Nature‒Nurture Debate
“[What have we learned from the project of behavioral genetics?] We learned, first, that behavioral differences among humans are correlated with genetic differences and are thus “heritable” in the statistical as opposed to the literal genetic sense of the word. Second, we learned that heritability coefficients don’t tell us much about what Galton wanted to know. They don’t tell us the extent to which genetic differences determine behavioral differences, how environmentally malleable those differences are, or how any one of us might have been different if we had been born with different genes. Non-zero heritability coefficients show that one way or another, genetic differences contribute to the variance of behavioral differences; the world would be less diverse if we were all clones.” (p 112)
“Modern genomics of human behavior has had a very surprising result: it has decreased estimates of heritability from where they stood 50 years ago in the twin study era. The common assumption back then was that the heritability of pretty much everything was around 0.5, if not higher, and no matter how skeptical one becomes about the causal meaning of the second decimal place of heritability coefficients, more than half the variance is a lot to reckon with. But with SNP heritabilities, and then PGS as real-world instantiations of SNP heritabilities, and then SNP heritabilities and PGS that have been appropriately corrected for between-family effects, those numbers are a lot closer to 0.1 than they are to 0.5.” (p 115)
“The empirical end to the Galtonian project is more whimper than bang. Genetic differences are a cause of all behavioral variability in a very nonspecific way, greater than zero but smaller than we might have thought at the height of twin study heritability 30 years ago. Neither the big promises of health care and education transformed by genomics nor the dire warnings of a dystopian eugenic nightmare have come to fruition.” (p 115)
“The 20 years of DNA-based genomics that have followed have mostly borne me out: everything is still heritable, and we still don’t have anything resembling a genetic explanation for differences in behavior.” (p 115)
“Individuals who are more similar in their overall genetic makeup, across all the genes influencing all the processes along the way, are more likely to follow similar paths, and thus more likely to wind up close to each other. Even though there is nothing resembling a divorce gene anywhere in the system, two people who are overall more similar in their genetic makeup are more likely to wind up similar in their marital status. It doesn’t matter that there are many thousands of genes, influencing many thousands of intermediate outcomes, in interaction with many thousands of environmental differences. When you add it all together, genotype is correlated with – “influences” if you insist – where you wind up.” (p 151)
“I propose to call this kind of heritability, arising as a statistical remnant in developmental systems that lack specific genes that cause specific outcomes, outcome heritability. Outcome heritability is universal, the explanation of the First Law of Behavior Genetics. The most important characteristic of outcome heritability is that it does not entail a representation of a complex outcome down in the genes somewhere. Heritability notwithstanding, you can look all you want for a genetic system that causes divorce, but you will never find it because it doesn’t exist.” (p 156)
“In contrast to outcome heritability, I propose the term essence heritability to refer to a system in which a heritable phenotype is represented in a specific way in a recognizable genetic system. Essence heritability is what we imagined all heritability to be before the history of complex human behavioral genetics forced us to recognize that not all heritability represents straightforward genetic causation and transmission. The obvious cases of essence heritability are single-gene disorders like Huntington’s disease. The point is not that the underlying biology of HD is simple – it isn’t – but rather that there is a correspondence between the phenotypic disease entity we observe and the genetic system that causes it. (p 157)
“That is the way the genetics of complex outcomes works out. Identical twins find similar paths through the myriad twists and turns toward eating habits, with thousands of remote genes nudging them in one direction or another along the way. In individual people without identical twins, however, those tiny effects become almost undetectable.” (p 158)
Aftab: You are often regarded as a skeptic within the behavioral genetics field. Unlike heritability deniers, you do not view twin heritabilities as illusory. Genetic clones and first-degree biological relatives are indeed more similar in behavioral outcomes due to their genetic similarity. However, you diverge from the mainstream view in that you are wary of tendencies towards genetic determinism and genetic essentialism, and you are far more pessimistic about the explanatory potential of genetics. Behavioral outcomes of interest to us are not, in your view, meaningfully encoded in genes. Although thousands of genes nudge us in one direction or another, and some associations can be identified via GWAS and summarized in polygenic scores, you contend that none of this adds up to meaningful genetic explanations for behavioral differences. The term “gloomy prospect” has been used to capture this explanatory pessimism, which contrasts sharply with the optimism of thought leaders like Robert Plomin, who argue that genes “make us who we are” and that genetic technologies will soon offer tractable methods of understanding this influence. Is this a fair characterization of your position?
Turkheimer: Yes, that’s fair. It was actually Robert Plomin who coined the phrase “gloomy prospect.” At the time, he brought it up in order to dismiss it—the “gloomy” refers to the ambitions of scientists, not the people they are studying. He was sure that if we tried hard enough, we would find systematic causal processes that explained why siblings were different, which is to say how people become individuals after accounting for their genes. But when he tried—as did I—there was nothing there. My paper with Mary Waldron documented the lack of success. Apparently, we can detect the nonshared environment as a big, poorly specified variance component—our experiences have the net effect of making us different—but we can’t break that down into the individual causal effects of individual experiences. In the years that followed, something even more surprising happened: the same thing turned out to be true for genes. We knew from twins and adoptees that genes had a general effect on people, but once the Human Genome Project allowed us to look for the individual effects of individual genes on human behavior, there was once again nothing there. The fundamental theme of the book is that however frustrating this state of affairs is for social scientists, it is also a bulwark against determinism. I don’t think people would want to live in a world where scientists could predict their behavioral outcomes from either their genes or their environment.
Turkheimer: Our experiences have the net effect of making us different—but we can’t break that down into the individual causal effects of individual experiences. In the years that followed, something even more surprising happened: the same thing turned out to be true for genes.
Aftab: Folks like Ken Kendler, Carl Craver, Mark Reimers, and others have responded (Reimers et al, 2019; Craver et al, 2022) to your critiques of GWAS, proposing that coherence in genetic influence can be expected to come in degrees. At one end, there are highly coherent conditions like Mendelian genetic disorders; at the other, there are highly incoherent traits like divorce. However, they argue that some conditions will exhibit intermediate coherence, where clusters of discernible genetic mechanisms will exist—even if no “genetic essence” exists. While these clusters haven’t been demonstrated yet for psychiatric disorders or behavioral traits of clinical interest (Dozmorov et al, 2020), the cautious optimism underlying this perspective acknowledges the potential for mechanistic insights in principle. Do you find this approach plausible, or are you unimpressed by this line of argument?
Turkheimer: Yes, it’s plausible, but the idea has always worked better in theory than in practice. As you acknowledge, there aren’t a lot of good examples of partial reductions in the behavioral sciences. In some sense, the idea of partial reduction has to be true. As you note, there are well-understood single gene disorders like Huntington’s disease and catalogs full of rare single gene syndromes with profound (and always detrimental) behavioral effects. Even more interesting are “neuropsychiatric” syndromes like dementia, Parkinson’s, and Tourette’s. No one has ever mistaken these things for divorce-style lifestyle choices. We have strong intuitions that they must have essential genetic and neurological causes. Plus, they seem to exist on a detectable continuum with psychotic disorders. Yet they remain remarkably poorly understood. Schizophrenia genetics is very excited right now about rare mutations detected with whole exome sequencing. I am sort of generically skeptical about this sort of thing. Could be true, but biological psychiatry has been enthusiastic about some new flavor of the month for as long as I have been in the field. I distinctly remember my (very bio-reductionist) mentor, the late Lee Willerman, slapping a very early CT scan down in front of me. There is your biological cause of schizophrenia, he said. He meant enlarged ventricles. You never know, but these hypotheses have always turned out to be methodologically driven, and generally don’t work out. Not very much real-world schizophrenia is explained by the rare mutations, and no new drugs have been developed. I’ll believe it when I see it.
Aftab: Your work often grapples with a core question: How can twin heritabilities be high—often around 50%—while quantifiable direct molecular genetic influence (in the form of SNP-based heritabilities and rare variants) is very low, often below 10%? One story that can be told is epistemic: aggregate genetic influence exists, but the developmental processes involved are so complex and chaotic that they defy explanatory tractability. Much of your book expands on this idea. However, a second story revolves around the issue of “missing heritability.” Does a significant gap really exist between molecular genetics (SNP heritabilities plus rare variants) and twin heritabilities, and if so, what accounts for it? If SNP heritability is less than 10%, we cannot validly say that environment explains 90% of the phenotype variation. Plausibly, we’re looking at gene-gene and gene-environment interactions as explanations for this gap.
When you describe genetics as “relatively unimportant” in understanding conditions like alcoholism, do you mean that in an epistemic sense (the genetic influence exists but cannot be meaningfully explained) or an ontological sense (the real genetic influence on risk is genuinely minimal)—or some combination of both? At times, your writings suggest both perspectives. Could you clarify?
Turkheimer: Lot to say about this. The first thing to remember is where human heritability estimates come from. It’s not as though we specify that this little SNP does this, and that little SNP does that, and then add up those causal effects to find that they can’t account for the twin heritability. In fact, human heritability, in twins or SNPs, is estimated without any reference to DNA-level causal effects, by inferring heritability from phenotypic similarity. Imagine a plot with genetic similarity on the x-axis (MZ=1, DZ=.5, half-siblings=.25 etc) and phenotypic pair correlations on the y-axis. The slope of the line of best fit is an estimate of heritability. In the book, I derive the Falconer twin formula 2*(rMZ-rDZ) that way. The important point is that is all we know about humans—more genetically similar people are more similar phenotypically. Any talk of additive effects or interactions in actual DNA is speculation. I think of the whole process as a statistical model of phenotypic similarity, with only a distant relationship to real biological genetics.
In this model, identical twins play a very important role, because they are more similar for all phenotypes than anyone else on the planet. One reason for this, the one you refer to, is that MZ twins just capture more genetic variance than anything else you can observe. If you think of humans just as a kind of developmental engine for realizing a genome, twins are the best way to accomplish it. What were the phenotypic possibilities given my genome? Growing another person from it would be the best way to answer that question, but most of us don’t have identical twins, so there is no way to know. First-degree relatives still work, but it drops off rapidly after that. SNP heritabilities are on the other end of the spectrum. We are trying to estimate heritability from differences among people with a couple percent relatedness on a quantitative genetic scale, and it just doesn’t work very well. I once said that predicting your phenotype from a polygenic score is like predicting it from a third cousin you have never met. There is a genetic signal down there somewhere, but it is very faint.
I am working on an even more challenging way to think about the problem. The line of best fit I mentioned between genotypic and phenotypic similarity isn’t in fact a straight line. If you picture a line drawn through those points, MZ twins are almost always a high outlier, more similar than would be predicted from a linear model of everyone else. Heritabilities estimated from twins are almost always higher than heritabilities estimated from siblings and half-siblings, for example. If you continue this process all the way down to the difference in SNP similarity among unrelated people, the line would become very flat by the time you got down to the genetic similarity between you and me. You could attribute this to “epistasis” (interactions among genetic loci), but that is a good example of attaching a speculative biological explanation to a process that is fundamentally about the course of phenotypic similarity. I prefer to describe it without the speculative genetic basis: there is a non-linear relationship between genetic and phenotypic similarity. Ultimately, genetic identity can enforce a level of developmental similarity that can’t be approached by anything less. (These ideas are related to the old concept of “emergenesis” that was developed in the 1990s by David Lykken.)
Aftab: You write, “The impression left by the ACE model that there are systematic effects of genes and families, then a bunch of randomness, is an illusion.” (p 159) Could you elaborate on this illusion? What do you see as the implications for understanding behavioral outcomes?
Turkheimer: This follows from what I have been saying. When we talk about additive effects and randomness, what we mean is, “In a statistical model of phenotypic similarity, we can attribute some portion of the variance to additive effects, with the rest of the variance in the error term.” But in that sentence, both “attribute” and “effects” are hypotheticals. I would say, “The net emergent effect of the complex developmental processes encompassing genes and experience is to cause more genetically (or environmentally) similar people to be more similar phenotypically.” Plus the non-linear part above. The “illusion,” which can be traced to Fisher, is that variance components correspond to causal processes. Whether you want to call them the gloomy prospect, causal hairballs (Carl Craver), epistasis, emergenesis, uncontrolled developmental processes, or just “random” is arbitrary. Given that most of the time we can’t conduct controlled experiments, they might as well be random. As you say, they are epistemically random.
Turkheimer: The net emergent effect of the complex developmental processes encompassing genes and experience is to cause more genetically (or environmentally) similar people to be more similar phenotypically.
Aftab: You recently co-authored a provocative paper for Philosophy, Psychiatry, & Psychology with Sarah Rodock Greer, titled “Spit for Science and the Limits of Applied Psychiatric Genetics.” The paper has been published in print with several commentaries and your response to them. [As you know, I wrote one of the commentaries, “Dialectical Tension Between Gloomy and Rosy Prospects of Behavioral Genetics.”]
What do these commentaries, in your view, reveal about the current state of thinking in behavioral genetics and the reception of your ideas?
Turkheimer: The responses are quite diverse and I can’t characterize all of them together. I think the biggest thing they made me think about is how different the scientific politics of psychiatric genetics are from the genetics of individual differences. My training was very much oriented toward the latter. The book only mentions psychopathology in passing. For individual differences, especially for intelligence, progressive politics and anti-hereditarianism align in a natural way. Historically, hereditarian ideas about intelligence have tried to malign the unfortunate, keep immigrants out of the country, undermine programs for social improvement, and explain away the difficulties of ethnic groups that have been discriminated against. This makes it natural for an enlightened progressive to be an anti-hereditarian about intelligence.
For mental illness, it isn’t so simple. The old eugenicists may have had ideas about eliminating the mentally ill from the population, and psychiatric genetics did have some roots in Nazi Germany, but none of that is especially active today. Psychiatric geneticists, even hereditarian ones, are just trying to help people as best they know how, and that can motivate them to cut a lot of scientific slack to the effort. Several of the replies, yours included :), keep trying to find something positive to say about the recent outcomes of applied psychiatric genetics. Surely the genetic effects can’t be absolute zero! There must be something down there somewhere! And yeah, the effects aren’t zero, and there must be something down there somewhere, but Spit for Science tried pretty hard, with 12,000 participants, and didn’t find anything. All our paper did was state that plainly.
Turkheimer: Pushing back against well-intended scientists who are only trying to alleviate psychiatric suffering can make me seem small, I know… Nevertheless, I still insist that unrealistic exaggeration of genetic effects has negative consequences that are not mitigated by the good intentions of the investigators.
Pushing back against well-intended scientists who are only trying to alleviate psychiatric suffering can make me seem small, I know. The Spit for Science paper has produced an angrier reaction (not in the published replies, but in my personal inbox) than anything I have ever written. The truth is, I was tough on Spit for Science, and I understand why my paper made people angry. Nevertheless, I still insist that unrealistic exaggeration of genetic effects has negative consequences that are not mitigated by the good intentions of the investigators.
The most important of these consequences is that over-enthusiastic “optimism” about genetic effects encourages the biologization of things that are not fundamentally biological. I think the distinction between biological and psychological is much more important than the one between genes and environment. I am not an environmentalist; I am an anti-reductionist. Alcohol use is a psychological and cultural phenomenon. I can’t rule out that there are some low-level biological processes that contribute to it, but other than ALDH2 (a variant that causes alcohol-flushing in Asian populations) we don’t know about any of them. Moreover, I insist that it is not necessarily the case that the inexorable progress of science will reveal the “biology of alcohol abuse,” any more than it will reveal the biology of divorce or bankruptcy. Maybe, who knows, but show me the money.
This problem is especially acute at NIMH. When Thomas Insel retired, after a decade of spending hundreds of millions of dollars on reductionist brain and genetics research, he wrote a book in which he laments that NIMH spends far too much money on brains and genes, and not nearly enough on behavior. Yet the response at NIMH was to appoint a new director who was at least as reductionist in outlook as Insel. In my opinion, the most important breakthrough in psychiatry in my lifetime has been cognitive behavior therapy, which is a thoroughly psychological model of etiology and treatment. CBT is everything that biogenetic psychiatry isn’t. It is a unified model of etiology and treatment, and the treatments it has produced have demonstrated solid empirical effectiveness. It should be the model for psychiatric science moving forward, not an afterthought.
Aftab: In light of the gloomy prospect, what would you advise geneticists who hope to contribute to the development of meaningful scientific explanations of behavioral phenotypes? Should they abandon the pursuit of genetic methods, and should funding agencies reconsider their priorities? Alternatively, is there still room for future research in behavioral genetics, perhaps with greater methodological sophistication and humility?
Turkheimer: See above about funding priorities. I think there is still plenty of room for meaningful genetic research in the meantime.
In the social sciences, I think genetics offers a route to the kind of non-experimental research we have always done, just conducted a little bit better. There are still a lot of psychologists who observe correlations between parents and their biological children and assume that the relationship is environmentally causal. The fact that parents who speak in more complex sentences have children who do better in the third grade does not prove that listening to complex language improves academic performance, and the reasons it does not are, broadly, “genetic.” Twins and adoptees offer ways to examine these kinds of questions a little more realistically. For example, in the oughts we had a line of research in which we studied the effects of divorce on children by comparing the children of identical twin mothers, one of whom was divorced and the other not. Presumably, polygenic scores offer a way to do something similar without the twins, but there are practical challenges because PGS have such tiny effects for most human behaviors.
In my most recent contribution to the Kendler-Parnas volumes on the philosophy of psychiatry, I explore the practical implications of admitting that there are no biological essences of many psychiatric conditions. If, for example, depression turns out to be more like divorce than it is like Huntington’s disease. For one thing, there would still be opportunity for palliative approaches. SSRIs may not tell us much about the etiology of mental illness, but they are a big improvement on older psychiatric medications and have helped a lot of people. For better or worse, we are pretty good at designing drugs these days, and I expect that to continue.
Another place my “outcome” model of complex behavior leads is to traditional psychotherapy. In my model of marital status, divorce is an end point of a long pseudo-random walk from conception to adulthood, influenced along the way by our genetic dispositions, environments, and goals. It is very hard to conduct science about such human processes, but it can be important for individual people to work through how they wound up where they did, to understand backwards the life they lived forwards. I have always had a soft spot for Freudianism (Paul Meehl said he had a soft spot in his heart, which his colleagues thought was located in his head). In the old days I held Freud in opposition to cognitive behavior therapy, but I no longer do. They are different ways of accomplishing different things. Same is true of psychiatric meds, come to think of it.
Turkheimer: Individual genotypes create potentials and place limits on what humans can become, but unless you have an identical twin, it is impossible to know what those limits and potentials are.
That seems like a funny place to leave a conversation that was supposed to be about genetics. My book argues that the genetics of behavior is real enough but can only be understood in the context of humanist assumptions about the nature of personhood. Here is another way to put it: individual genotypes create potentials and place limits on what humans can become, but unless you have an identical twin, it is impossible to know what those limits and potentials are. On average, we will all turn out more or less like our parents and siblings, but we didn’t need advanced genomics to tell us that.
Thanks so much for giving me the opportunity to answer such challenging questions.
Aftab: Thank you for your answers!
See also:
It has taken me a while, but I think I'm gradually coming around to Eric's side of the debate about all things genes and genetic. I think his is the less commonly accepted position because it's the most sophisticated and nuanced. And as we all know, anything sophisticated and nuanced is rarely popular. Also, his arguments about heritability and its derivatives seem to me to display certain virtues that are critical to keeping science and scientific enterprises grounded: virtues like caution and humility seem rather lacking on the side that is asserting an inevitable El Dorado for applied genetics.
I understand the allure of this promise because I've also been infected by it. However, this cautionary warning by Eric strikes me as potentially prescient: "I don’t think people would want to live in a world where scientists could predict their behavioral outcomes from either their genes or their environment." This is a highly important warning but one most likely to be dismissed by the vast majority of people desperate for utopia or freedom from a miserable life, becoming easy target for Mephistopheles-style covenant.
One last point: I felt immensely gratified to read a confirmation of one of my correct logical intuitions (the vast majority of others are probably incorrect) about heritability: that it has to be a primarily statistical process when one observes the highly variable (less than predictable except in MZ pairs) phenotypic outcomes. This fact is so fundamental to our understanding of heritability because even in the most heritable conditions (eg Huntington, some cancers, even height), the outcomes are still statistical rather than given. Thus, the only difference between the heritability of say religiousity, intelligence, and height is in their statistical probability. In the former (religiousity) the within-family variance will be very high (similar to what is found in the general population), while it'll be very low in the latter (height). Thus we could summarize that for attributes/traits with high heritability, within-family differences is low, and for those with low heritability, within-family differences is high. We can use this principle to study heritability in reverse, that is, to determine the degree of heritability by aggregating and analyzing traits based on whether they possess high, mid, to low within-family variance across many families.
PS: My knowledge of genetics and heritability is very very basic (only perhaps slightly better than a layman's) and I can't boast of being familiar with the literature at all. Hence, it's possible that my suggestion in the last paragraph is (1) off the mark (2) already stale.
Fascinating discussion. My only quibble is with ‘the most important breakthrough in psychiatry in my lifetime has been cognitive behavior therapy’.
I can think of various psychiatric treatments from 70s onward which seem more transformational than CBT - clozapine for treatment resistant schizophrenia and lithium for bipolar disorder spring to mind.
If anything, recent research seems to challenge the superiority of CBT over other modalities of psychotherapy!