Behavioral Genetics and Human Agency: How Selectively Deterministic Theories of Free Will Drive Unwarranted Opposition to Behavioral Genetic Research and Undermine Our Moral and Legal Conventions, Part I

Abstract

This article argues that a pervasive but confused theory of free will is driving unwarranted resistance to behavioral genetic research and undermining the concept of personal responsibility enshrined in our moral and legal conventions. We call this the theory of ‘free-will-by-subtraction’. A particularly explicit version of this theory has been propounded by the psychologist Eric Turkheimer, who has proposed that human agency can be scientifically quantified as the behavioral variation that remains unexplained after known genetic and environmental causes have been accounted for. This theory motivates resistance to research that suggests genetic differences substantially account for differences in human behavior because that is seen to reduce the scope of human freedom. In academic philosophy, free-will-by-subtraction theory corresponds to a position called ‘libertarian incompatibilism’, which holds that human beings are not responsible for behavior that has extrinsic causes yet maintains that free will nonetheless exists because some fraction of human behavior is self-caused. However, this position is rejected by most professional philosophers. We argue that libertarian incompatibilism is inconsistent with a secular materialist outlook in which all human behavior is understood to have extrinsic causes whether those causes are known to science or not — an outlook Turkheimer shares. We show that Turkheimer sustains this contradiction by adopting an untenable position we call ‘epistemic libertarianism’, which holds that extrinsic causes of our behavior only infringe on our freedom if we know about them. By contrast, the overwhelming majority of secular materialist philosophers support a position called ‘compatibilism’, which maintains that free will is compatible with the comprehensive extrinsic causation of human behavior. We show that compatibilism neutralizes the threat that genetic explanation poses to human agency and rescues a generous conception of personal responsibility that aligns with our moral intuitions.

Executive Summary

In section 1, we propose that a pervasive but confused theory of free will is producing systemic bias against research that shows substantial genetic influence on socially valued behavioral traits. We call this the theory of ‘free-will-by-subtraction’ because it holds that genetic and environmental causes of behavior partly infringe on human agency. We showcase recent examples where behavior genetic research on cognitive and socioeconomic traits has been obstructed or singled out for special oversight by data custodians and journal editors. We also show how prominent behavioral geneticists have been unfairly maligned by other scholars as ‘genetic determinists’, ‘genetic reductionists’, or even ‘eugenicists’ for highlighting the role that genetic differences play in socially valued behavioral traits. We introduce a particularly explicit formalization of this free-will-by-subtraction theory developed by the psychologist Eric Turkheimer, in which he proposes that twin study estimates of the nonshared environment can be used to quantify the degree of agency that people have over different traits after genetic and shared environmental influences on their behavior have been accounted for.

In section 2 we explore weaknesses and inconsistencies in Turkheimer’s treatment of heritability as it relates to this theory of free will. For example, Turkheimer (2011a) argues that the heritability of a given behavior is not a valid measure of the extent to which our agency is infringed by genetic causation because heritability is not a valid measure of how genetic a trait is. He argues this on the grounds that the heritability of a trait is population-specific and because all traits require both genes and environments in order to develop (and are therefore equally and unquantifiably genetic). Contra Turkheimer, we show that heritability corresponds to what is meant in ordinary language when people ask how genetic a trait is, which is always a question about how much genetic differences account for trait differences in an implicit reference population. On this basis, heritability retains many of the profound implications for behavioral science and social policy that Turkheimer denies, and the specific magnitude of heritability estimates matters for assessing those implications. While we will ultimately agree with Turkheimer that heritability does not limit human agency (as we discuss in sections 8 and 9), this is not because heritability fails to capture genetic causes of individual differences.

In section 3, we show that Turkheimer’s (2011a) theory that nonshared environmental variance is a yardstick of free will contradicts his argument that heritability does not infringe on human agency, as the nonshared environment is simply the variation remaining after heritability and shared environmental influence have both been subtracted. This is one of several instances we document where Turkheimer contradicts his own thesis that heritability does not signal genetic causation of human behavior. As this interpretation of the nonshared environment necessarily implies a narrower scope for human agency than the theory that heritability alone infringes it, Turkheimer (2011a) thereby presents a vision of human agency that is more deterministic than those he accuses of ‘genetic determinism’. This implication is drawn out in Kathryn Paige Harden’s recapitulation of Turkheimer’s theory in The Genetic Lottery where she explicitly states, ‘there is remarkably little territory left for “personal responsibility”’ (Harden, 2021b, p. 204) after genetic and shared environmental influences on behavior have both been accounted for. In addition, we highlight that nonshared environmental influences can be just as population-specific as heritability estimates, undermining Turkheimer’s critique of heritability on this basis.

In section 4, we show that Turkheimer (2011a) covertly advances the incendiary argument that people are more responsible for their sexual orientation than their criminal behavior on the basis that the nonshared environment can be used to order traits by their ‘controllability and moral relevance’ (p. 827). While Turkheimer makes clear that he is personally opposed to the pathologization of homosexuality, his idea that we have more control over our sexual preferences than whether we commit crimes risks buttressing homophobic viewpoints that see homosexuality as a morally charged lifestyle choice. In this way, Turkheimer’s theory of free-will-by-subtraction presents downstream political risks quite as serious as those that are said to follow from the ‘hereditarian’ viewpoint that heritability quantifies genetic causes of individual differences in behavior. This example highlights the problem with evaluating philosophical arguments (or lines of empirical research) based on their perceived downstream risks rather than on their merits — namely, that doing so might have unintended consequences worse than those that we were trying to avoid. Contra Turkheimer, we argue that the nonshared environment is not experienced as a special domain of choice or control, illustrating how events manifestly outside of their control produce behavioral differences between identical twins for which they cannot be held responsible.

In section 5 we situate Turkheimer and Harden within the academic free will debate, introducing readers to the three major schools of thinking on this topic: hard incompatibilism, the view that agency is incompatible with the total extrinsic causation of human behavior; libertarian incompatibilism, the view that humans have agency because some fraction of our behavior is uncaused; and compatibilism, the view that agency is compatible with the total extrinsic causation of human behavior. While Turkheimer describes himself as a compatibilist, we show that he is, in fact, a libertarian incompatibilist who views human beings as having agency except where their behavior is shown to have genetic and environmental causes. Moreover, while Turkheimer nominally rejects metaphysical explanations for free will, we show that his theory that human agency can reside in the unexplained variation in human behavior effectively reproduces the supernatural logic that some fraction of human behavior is miraculously self-caused. Like Turkheimer, Harden accepts the incompatibilist logic that extrinsic causes of behavior infringe on human agency, but she is prepared to recognize more extensive genetic and environmental causation of human behavior. Nevertheless, she balks at the prospect of hard incompatibilism and embraces a de facto libertarian perspective in which free will and personal responsibility are held to reside in some sliver of the nonshared environmental variation that remains inexplicably uncaused. We argue these selectively determinist, libertarian accounts of free will are fundamentally irreconcilable with a secular materialist outlook in which all behavior is understood to have extrinsic causes (whether those causes are simple, complex, deterministic or indeterministic).

In section 6 we discuss an important variation on Turkheimer’s theory of free-will-by-subtraction which we call the ‘gloomy prospect’ theory of free will. Turkheimer (2000) proposed that the behavioral variation left unexplained by specific measures across any variance component could be interpreted as a yardstick of human agency. This theory implies a more generous scope for human agency than Turkheimer’s (2011a) nonshared environmental theory of free will, mainly because it replaces the threat posed by large heritability estimates with the more modest threat posed by the direct effect of current molecular genetic predictors of behavior. However, both theories rest on the same supernatural logic that some fraction of human behavior is self-caused — a logic that is inconsistent with Turkheimer’s professed secular materialism and with his admission that unexplained sources of human behavioral variation have extrinsic causes that might someday yield to science. We argue that Turkheimer sustains this contradiction by substituting metaphysical libertarianism, the idea that some human behavior does not have extrinsic causes, with epistemic libertarianism, the incoherent idea that extrinsic causes of behavior only infringe on our freedom when we know about them. We show that free-will-by-subtraction theories inherently motivate ideological resistance towards the scientific explanation for human behavior.

In section 7 we explore specific cases where Turkheimer’s ‘gloomy prospect’ theory of free will has led him to make unduly pessimistic predictions about what behavioral geneticists will discover and to downplay the importance of advances that have already been made. One of Turkheimer’s signature predictions was that associations between genetic variants and complex behaviors would ‘fail to replicate and accumulate’ (Turkheimer, 2000, p. 163). We discuss results from recent genomewide association studies (GWAS) that suggest this prediction was inaccurate. We proceed to evaluate Turkheimer’s recent assertion that ‘GWAS of complex human behavior [have] turned out to be a disappointment’ (Turkheimer, 2022a) in light of earlier predictions he had made about what GWAS would and would not deliver. We argue that the alignment of Turkheimer’s views on a wide range of scientific questions in behavioral genetics is more easily explained by his philosophical anxieties that genetic causation of human behavior infringes on human agency than by the weight of empirical evidence. We discuss how a one-sided, philosophically motivated skepticism against scientific results that suggest strong genetic influence on human behavior risks distorting the scientific record, diverting resources away from productive lines of scientific inquiry, and postponing important public conversations about the societal implications of those findings.

Having established that, for secular materialists, libertarian incompatibilism necessarily collapses into hard incompatibilism once it is recognized that all human behavior has extrinsic causes, in section 8 we show how our intuitive conceptions of human agency and personal responsibility can be salvaged under a ‘compatibilist’ framework. Compatibilism — the view that free will is compatible with comprehensive extrinsic causation of human behavior — is the position supported by the majority of academic philosophers but remains underrepresented in the behavioral sciences where (libertarian or hard) incompatibilist views appear dominant. Drawing heavily on the work of the philosopher Daniel Dennett, we argue that incompatibilists erase the concept of personal responsibility by abusing the ordinary meaning of the words ‘control’ and ‘luck’. Whereas ‘control’ ordinarily refers to the deliberate manipulation of another entity’s behavior by an intentional agent, incompatibilists routinely extend the term to agentless causes of behavior, such that we are said to be ‘controlled’ by the past in much the same way as a marionette is controlled by a puppet master, leaving no space for self-control. Similarly, whereas ‘luck’ ordinarily refers to fortuitous and unforeseeable events outside of our control, incompatibilists routinely invoke the term as a universal explanation for all phenomena, erasing the distinction between luck and skill. We seek to show that the concept of personal responsibility is grounded in the recognition of a wide class of persons whose behavior can be positively influenced by the expectation of praise, blame, reward, and punishment, and who are thereby considered morally competent. Under this view, it is not all extrinsic causes of our behavior that diminish our responsibility but only a tiny subset of those causes which render us indifferent to normal incentives.

In section 9 we explore how society should appropriately assign criminal responsibility under conditions in which crime has identifiable genetic and environmental causes. We highlight the irrelevance of the nonshared environmental variance — a population statistic — to the question of whether a particular individual is criminally responsible. We argue, contra Turkheimer and Harden, that genetic and environmental causes of behavior should only be considered exculpatory under narrow conditions where they have severely impaired someone’s responsiveness to normal incentives. We revisit the familiar argument that a person cannot be held responsible for criminal behavior that was caused by a brain tumor and propose that the legitimacy of the brain tumor defense rests on the tumor impairing the subject’s responsiveness to normal incentives rather than on the tumor’s biological causation of behavior, as such. We proceed to discuss how the notions of legal and criminal responsibility are imperiled by the incompatibilist logic that extrinsic causes of criminal behavior are exculpatory.

In section 10, we discuss how selectively determinist perspectives on free will can motivate illiberal and coercive social policies. When normal adults are not considered fully responsible for their actions, that belief can be used to justify controlling their access to information and severely reducing their liberties in the name of public welfare or public safety. We showcase examples where the neuroscientist David Eagleman and the biosocial criminologist Adrian Raine have invoked incompatibilist arguments to respectively support unequal treatment under the law and indefinite, pre-emptive detention of people who have never committed a crime. We also show how selective determinism is leveraged by Harden to support a coercively paternalist political program. We describe how the celebrated, 19th century, Russian novelist Fyodor Dostoyevsky successfully predicted that a political utopianism built upon a statistically circumscribed conception of human agency would lead to the brutal and repressive totalitarian regimes that emerged in the 20th century. We conclude that it is selective determinism that poses the greater risk of indirect social harms, not the research on genetic causes of human behavior that selective determinists routinely seek to suppress.

Genetic explanations for human behavioral differences pose little concern to either compatibilists or hard incompatibilists because both perspectives are reconciled with the comprehensive extrinsic causation of human behavior. By highlighting the fatal inconsistencies in selectively determinist, secular libertarian theories of free will we hope to remove unnecessary philosophical and psychological barriers to behavior genetic research. More ambitiously, we hope to inoculate readers against a pervasive idea that threatens the social order by undermining the concept of personal responsibility and justifying authoritarian social policies.

1. Introduction

There is no more potent source of anxiety about free will than the image of the physical sciences engulfing our every deed, good and bad, in the acid broth of causal explanation, nibbling away at the soul until there is nothing left to praise or blame, to honor, respect, or love.

– Daniel Dennett, Freedom Evolves (2004)

[T]he difficulties that people have had concerning the meaning of voluntary action and responsibility do not at all spring from any necessary consequence of the belief that human action is causally determined but are the result of an intellectual muddle, of drawing conclusions which do not follow from the premises…the whole issue is a phantom problem.

– Friedrich Hayek, The Constitution of Liberty (1960)

The economist John Maynard Keynes once wrote that ‘The ideas of…philosophers, both when they are right and when they are wrong, are more powerful than is commonly understood’ adding that ‘practical men, who believe themselves to be quite exempt from any intellectual influences, are usually the slaves of some defunct [philosopher]’ (Keynes, 1936). Scientists are pragmatists who do not generally preoccupy themselves with academic philosophy, but second-hand philosophical ideas can nevertheless have profound influence upon the direction and interpretation of scientific research. This holds especially true for the controversial science of human behavioral genetics where philosophical assumptions concerning the nature of free will can render certain scientific results morally and politically unpalatable.

When genetic explanations for human behavioral differences are perceived to infringe upon human agency, to undermine support for the least fortunate, or to legitimize dangerous political ideologies, the temptation to downplay or trivialize the role of genetic influences can become almost irresistible. But these implications hinge on contentious philosophical assumptions over which reasonable people disagree and which need not follow from the facts. Extra-scientific concerns about the nature of free will can therefore pose a significant barrier to the advancement of behavioral genetics as a science.

The perceived unacceptability of results that show substantial genetic influence on human behavioral differences can potentially bias the direction and reception of behavioral genetic research across the scientific ecosystem. Funders may be less willing to finance such research, data custodians and principal investigators might decline applications to access certain samples, reviewers might be more critical of such research, and other scientists more likely to dismiss and disparage it. Conversely, we can expect research that suggests a limited, ambivalent, or highly contingent role for genetic influences on human behavior to be more favorably received. We can expect these dangers to be greatest concerning socially significant behavioral traits.

These risks are not just hypothetical. Several important genomic datasets have already denied access to researchers seeking to investigate cognitive traits and socioeconomic outcomes on the basis that such research might be ‘stigmatizing’ to certain individuals or groups (Lee, 2022; Ritchie, 2022b). These restrictions have become more acute following a racially motivated mass shooting in Buffalo, New York in May 2022 after it was discovered that the shooter’s manifesto had included references to some behavioral genetic studies (Smart, 2022). Additionally, in the wake of that incident, several concerned scientists began calling for scientific journals to carefully police how behavior genetic results are communicated (Carlson et al., 2022; Wedow et al., 2022). They also called for Institutional Review Boards to be granted new powers to reject studies based on their potential downstream social impact rather than simply assessing whether ethical procedures were followed when conducting the relevant research.

Partly in response to these campaigns, in August 2022, Nature Human Behaviour (NHB) published an editorial proclaiming that ‘The same ethical considerations should underlie science about humans as apply to research with human participants’ (NHB, 2022, p. 1029 [emphasis added]). In accordance with that principle, they announced that editors across the prestigious Nature publishing group would henceforth reserve the right to reject, amend, or retract papers whose findings have the potential to ‘stigmatize individuals or human groups’ (NHB, 2022, p. 1029).¹

The editorial elicited strident criticism from several prominent scientists and scholars who saw it as a charter for the partisan political censorship of science. The biologist Jerry Coyne wrote that the editorial is ‘a threat that unless articles conform to a specific ideological stance, they can be rejected even if the data themselves are worth publishing’ (Coyne, 2022). The psychologist Stuart Ritchie wrote, ‘vague wording throughout is open to abuse [and] could be used to shut down research anyone finds offensive for any reason’ (Ritchie, 2022a). The cognitive scientist and author Steven Pinker proclaimed that ‘Nature Human Behavior is no longer a peer-reviewed scientific journal but an enforcer of a political creed’ (Pinker, 2022). Finally, the Brookings Institute scholar Jonathan Rauch wrote, ‘However professional and well-intentioned NHB’s editors may be, they are not qualified to decide on society’s behalf whether research is socially harmful or desirable. In fact, they have no idea how a piece of research will ramify’ (Rauch, 2022).²

While these well-intentioned initiatives are overtly framed around reducing the potential harms behavioral genetic research might indirectly cause — for example, by providing ideological ammunition to violent hate groups — the perceived source of that harm ultimately remains the prospect that behavior genetic research will show that there is a substantial genetic basis for variation in socially valued traits.³ Results supporting that conclusion, so the reasoning goes, must therefore be subject to special scrutiny, deliberately camouflaged, or actively suppressed — even if true.⁴ As the biologist Luana Maroja wrote in response to the NHB editorial, ‘They are not referring to the importance of protecting individuals participating in research. They are saying that the study of human variation is itself suspect’ (Maroja, 2022). Rather than condemning racism, classism, sexism, and other forms of prejudice as illegitimate responses to behavioral differences irrespective of their genetic or environmental origin, several important scientific institutions have elected to turn on genetic explanation instead.

Beyond these formal obstacles, behavior genetic researchers are obliged to routinely contend with distempered accusations from other scholars that they are engaged in morally suspect pseudoscience. A review of Robert Plomin’s 2018 book Blueprint in Nature magazine accused him of ‘vintage genetic determinism’ (p. 461), ‘crude hereditarianism’ (p. 461), and of writing ‘a roadmap to regressive social policy’ (p .463) (Comfort, 2018). A review of Kathryn Paige Harden’s 2021 book The Genetic Lottery in the New York Review of Books described behavioral genetics as ‘genomic astrology’ and charged Harden with ‘resurrect[ing] the misconceived science underlying…scientific racism’ and ‘reproduc[ing]…important steps in…eugenic…logic’ (Riskin & Feldman, 2022). In addition, the review accused Harden of ‘bogus scientism’, ‘spurious reductionism’, and ‘biological essentialism’.

But living in glass houses has not prevented behavioral geneticists from throwing stones at each other. Harden has herself accused Robert Plomin of ‘genetic determinism’ (Harden, 2018) and recently wrote a comment piece in Nature Reviews Genetics that seeks to rehabilitate ‘genetic essentialism’, ‘genetic determinism’, and ‘genetic reductionism’ as legitimate pejoratives in scientific debate after rigidly redefining those terms in such a way that they do not apply to her own published views (Harden, 2023).⁵ Nevertheless, these efforts have not prevented Harden from being accused of ‘nudging readers towards genetic determinism’ by other geneticists (Coop & Przeworski, 2022), or of being bracketed with Robert Plomin as a ‘center-hereditarian’ by her PhD supervisor and longtime collaborator Eric Turkheimer, who also described her central genetic lottery metaphor as ‘eugenic’ (Turkheimer, 2022b).

For his part, Turkheimer has been levelling similar criticisms against fellow behavioral geneticists since the 1990s, repudiating researchers within the ranks whom he classifies as ‘genetic essentialists’ (Turkheimer, 2011a) or ‘[g]enetic and biological reductionists’ (Turkheimer, 1998, p. 789). In a 1997 comment piece called ‘The Search for a Psychometric Left’, Turkheimer issued a rallying cry for likeminded scientists to stand with him in ‘opposition to determinism, reductionism and racism’ against what he called ‘the psychometric right’ (Turkheimer, 1997, p. 783).

Again, the scholars who invoke this language clearly do so out of a profound concern about the social consequences that might follow if behavior genetic findings are interpreted in ways they disagree with. Nevertheless, in a community that professes to be anxiously concerned about the stigmatizing potential of certain empirical findings, it is seldom acknowledged that it is deliberately stigmatizing to label fellow scientists as genetic ‘essentialists’, ‘determinists’, or ‘reductionists’, let alone ‘scientific racists’ or ‘eugenicists’.

In this article we seek to show that widespread fears that genetic causes of human behavior imply ‘genetic determinism’ stem from an untenable theory of free will that is rejected by the majority of academic philosophers who formally study this question (Bourget & Chalmers, 2023). We will call this the theory of free-will-by-subtraction because it posits that free will is whatever we have left after the measured causes of our behavior have been accounted for.

This theory of free-will-by-subtraction is pervasive across the behavioral sciences where individuals are widely presented as victims of circumstance to the extent that their behavior can be statistically predicted by variables that are manifestly outside of their control, such as the socioeconomic characteristics of the household they were raised in (Hollander, 1973; Morse, 1993). It is also pervasive in lay moral and political debates, where individuals are often regarded as less responsible for their socioeconomic outcomes, their health outcomes, or the crimes they have committed based on their childhood circumstances.

The conspicuous feature of this free-will-by-subtraction position is that, beyond the ‘social determinants’ of various life outcomes, individuals are assumed to remain responsible agents with power over their own life choices. These statistical intrusions on human agency are treated as limited exceptions to the general rule of human autonomy. What is typically left unexplained is how human agency is supposed to arise under a secular materialist framework in which all human behavior is assumed to ultimately have extrinsic causes, or why scientific knowledge of extrinsic causes of human behavior should infringe upon our freedom more than causes that have not been identified.

Within this free-will-by-subtraction framework, empirical findings that suggest substantial genetic influence over important life outcomes arouse consternation on two main fronts. First, they threaten to contract the residual domain of human agency over these outcomes by adding ‘genetic determinants’ of behavior to the ‘social determinants’ already recognized. Second, they threaten to supplant social explanations for important life outcomes with genetic explanations, that is, by suggesting that ‘inborn’ differences in ability and proclivity partially explain differences in social outcomes rather than environmental advantages and disadvantages. The latter concern is often expressed as the danger that genetic explanations will ‘naturalize’ or ‘justify’ existing social inequalities and is often accompanied by the worry that behavioral genetic findings will undermine the case for redistributive taxation, or for government programs that support the socially deprived (e.g., Henn et al., 2021). This philosophical outlook can therefore motivate powerful moral and political resistance to scientific findings that suggest strong genetic influence on socially significant human outcomes and can motivate elaborate efforts to reinterpret those findings to ward off these perceived threats.

This usually implicit theory of free-will-by-subtraction has been developed into an explicit theory by Eric Turkheimer. In the pages that follow, we will critically examine the key writings in which he develops that theory, focusing mainly on ‘Heritability and Biological Explanation’ (Turkheimer, 1998), ‘The Three Laws of Behavior Genetics and What they Mean’ (Turkheimer, 2000), and especially ‘Genetics and Human Agency’ (Turkheimer, 2011a). Against this view, we will argue that free-will-by-subtraction represents an untenable position which, under scrutiny, collapses into full-blooded free will skepticism. For now, we note that, when all human behavior is seen to be determined by extrinsic causes, the question of whether those causes are genetic or environmental becomes morally irrelevant and specific concerns about genetic determinism lose their force.

In academic philosophy, free-will-by-subtraction theory and full-blooded free will skepticism would both be described as ‘incompatibilist’ positions on free will because they hold that human agency is ‘incompatible’ with conditions in which our behavior has extrinsic causes. Against both views, we will defend a ‘compatibilist’ theory of free will in which genetic and environmental causes of human behavior are both seen as ‘compatible’ with personal responsibility outside of a narrow range of special (and intuitive) circumstances.⁶ Compatibilism is currently the majority perspective on free will among academic philosophers (Bourget & Chalmers, 2023).

The clearest formulation of Turkheimer’s free-will-by-subtraction theory is given in ‘Genetics and Human Agency’ (Turkheimer, 2011a). In that article, Turkheimer argues that we should dethrone heritability as the central focus of behavioral genetic research and view the nonshared environment (as measured in twin studies) as the more socially important variance component.

For readers unfamiliar with these terms, ‘heritability’ is an estimate of the extent to which individual differences in a particular trait in a given population can be attributed to genetic differences between those individuals. It is estimated in conventional twin studies by assuming any additional resemblance identical twins show over nonidentical twins can be attributed to them sharing 100% of their additive genetic influences in common instead of the 50% that nonidentical twins are assumed to share on average. Any residual resemblance between identical twins after accounting for the heritability is attributed to environmental influences they hold in common: the so-called ‘shared environment’. Finally, any differences between identical twins are attributed to the ‘nonshared environment’, which is assumed to capture everything else that makes people different for a given trait other than their genetic differences and the kind of environmental influences they share with their twin siblings.

Turkheimer’s article makes the philosophically original claim that ‘The nonshared environment, in a phrase, is free will’ (Turkheimer, 2011a, p. 826) and that behavior genetic estimates of this variance component are capable of ‘quantifying human agency… as the variability that remains after genetic and environmental familial constraints have been taken into account’ (Turkheimer, 2011a, p. 827).⁷ In this way, Turkheimer develops the folk theory of free-will-by-subtraction that we have briefly described above into a formal theory, welded to behavioral genetics, wherein heritability and shared environmental influences are held up as omnibus measures of the total genetic and environmental influences that infringe on human freedom with the nonshared environment making up the remainder. Turkheimer goes on to say that ‘The nonshared environment captures what is at stake when we are concerned about whether people are able to control their own weight or choose their own sexual orientation or whether they should be held responsible for their criminal behavior’ (Turkheimer, 2011a, p. 827), and proposes that by ranking traits according to twin study estimates of the nonshared environment, we can ‘rescue our common sense intuition that people are more responsible for their life of crime than for their short stature’ (Turkheimer, 2011a, p. 826).

Turkheimer’s theory of free-will-by-subtraction has recently been brought to wider attention after being rehearsed by Harden in the Annual Review of Psychology (Harden, 2021a) and in her book The Genetic Lottery (Harden, 2021b). Both publications contain a detailed and sympathetic exposition of Turkheimer’s theory, inviting readers to view ‘twin studies through the lens of the non-shared environment’, describing the nonshared environmental variance component as ‘the free-will coefficient’, and suggesting the size of the nonshared environment can be used to assess the degree of moral responsibility we have for various traits after we account for the combined effects of genetic and shared environmental influences. In The Genetic Lottery, Harden goes on to develop a political interpretation of this theory of free-will-by-subtraction which — rather than seeing genetic causes of inequality as excuses for political inaction — agitates for an enlarged egalitarian program that seeks to ameliorate genetic, as well as environmental, inequalities of birth.

If correct, Turkheimer’s theory of free will would have momentous implications for science and for society. In this article we will argue that his theory is not correct.

***

The interdisciplinary subject matter of this article demands that we discuss empirical findings and specialist concepts from behavioral genetics alongside specialist concepts from academic philosophy. Some of the material that we discuss will therefore be unfamiliar to readers from either discipline. However, we have attempted to write this piece in language that curious readers from either discipline (or neither) can understand. We also forewarn the reader that central issues explored in this article, such as the meaning of heritability and its policy implications, or the nature of free will and personal responsibility, are among the most contested issues in both behavioral genetics and philosophy. Reasonable people from both fields disagree on these questions and often do so passionately. We cannot hope to persuade all readers to agree with our conclusions or to settle these longstanding controversies once and for all. Our modest hope is that readers interested in these controversies will feel their views are further refined by critically engaging with the argument that follows.

2. Does Heritability Tell Us How Genetic a Trait Is?

2.1. Turkheimer Argues That Heritability Is Not a Measure of Genetic Influence

Turkheimer begins ‘Genetics and Human Agency’ by addressing what he argues are some fundamental misconceptions about the implications of behavioral genetics for free will. He writes:

The obvious way to think about [free will] is to presume that the higher the heritability of a trait, the more ‘determined’ it is, the less we can control it, and the less we should be blamed for it. This presumption underlies the thinking of both genetic essentialists and naïve environmentalists, but it is incorrect. (Turkheimer, 2011a, p. 825)

Turkheimer’s first argument as to why it is incorrect to use heritability as an index of human agency is because ‘traits do not have heritabilities…heritability depends on the population in which it is measured’ (Turkheimer, 2011a, p. 825). As a case in point, Turkheimer admonishes Dar-Nimrod and Heine (2011), the authors of the target article he was responding to, for saying the heritability of adult intelligence is ‘typically estimated to range from .50 to .85’ (p. 805), writing: ‘The heritability of intelligence isn’t anything, and even placing it in a range is misleading’ (Turkheimer, 2011a, pp. 825–826). Turkheimer invites us to instead think of the heritability of intelligence as an unbounded statistic that can fall anywhere between zero and one depending on the population being measured. He uses a thought experiment to drive this point home: ‘Among cloned animals with widely varying diets, body size is perfectly environmental with heritability of 0; in genetically variable animals raised in identical environments heritability is 1.0’ (Turkheimer, 2011a, p. 826).

But Turkheimer has rather run away with this point. While it is true that the heritability coefficient for a particular trait is always population-specific, it is misleading to suggest it can range anywhere from 0 to 1 in any human population we would be interested in estimating that parameter. The tendentiousness of the thought experiment Turkheimer enlists reveals this: there are few conceivable circumstances under which a researcher would ever be interested in estimating the heritability coefficient for a population entirely comprised of monozygotic twins or ‘cloned animals’. In less tendentious contexts, it is so well established that the heritability of human behavior is consistently higher than zero that Turkheimer himself enshrined this as ‘the first law of behavior genetics’ (Turkheimer & Gottesman, 1991, p. 411). Moreover, in practical terms, the variation in heritability estimates we observe across a range of contemporary and historical contexts gives us an indication — albeit imperfect — of the heritability we can reasonably expect to encounter in a broadly similar range of environmental contexts. This point was emphasized by the late David Rowe when he insisted that ‘The views that (1) heritabilities are necessarily nongeneralizable, and (2) that population specific estimates have no policy relevance are both mistaken’ (Rowe, 1994, p. 24).

Turkheimer proceeds to leverage this point about the variability and population-specificity of heritability estimates to make some far-reaching claims about how these estimates should and should not be interpreted. ‘This is no mere statistical fine point,’ he says, ‘it means that the entire project of assessing how essentially genetic traits are in terms of measured heritability coefficients is a fool’s errand’ (Turkheimer, 2011a, p. 826). He elaborates:

[It] is not only misguided to use heritability as a means of quantifying the genetic and moral structure of a trait, but the whole project of differentiating traits into those that are more or less essentially genetic is hopeless from the start. In this, all traits are the same: they all require interaction of genes and environment to develop, and that’s that. Growing to a particular height is no more genetic than robbing banks or learning to play the oboe, so moral analysis of these traits will have to be found elsewhere. (Turkheimer, 2011a, p. 826)

Here Turkheimer has enlisted a second, slightly different argument. Where initially it was the population-specificity of heritability that was marshalled as the reason that heritability does not ‘quantif[y] the genetic and moral structure of a trait’, now it is because ‘all traits…require interaction of genes and environment to develop’ (Turkheimer, 2011a, p. 826). But it is an exemplary case of a straw man argument because it is a position that no-one disagrees with. As the philosopher of biology Neven Sesardić writes: ‘No one ever doubted this obvious and rather unilluminating truth…Everybody has always known that, as David Lykken puts it, ‘without environmental input, your genome would have created nothing more than a damp spot on the carpet’ (Sesardić, 2005, p. 14).

In a highly cited earlier piece, ‘The Three Laws of Behavior Genetics and What They Mean’ (Turkheimer, 2000), Turkheimer uses a quote from Gottlieb (1992), ‘that genes are an inextricable component of any developmental system, and thus genes are involved in all traits’ to refute the ‘implausible environmentalist contention that important aspects of behavior will be without genetic influence’ (Turkheimer, 2000, p. 161). But there are no environmentalists who deny genes are implicated in behavioral traits in the way that Gottlieb describes.⁸ To the extent that ‘naïve environmentalists’ are supposed to be defined by this egregious mistake these bogeymen are figments of Turkheimer’s imagination. What makes environmentalists ‘naïve’ isn’t that they fail to acknowledge that ‘genes are an inextricable component of any developmental system’ (Turkheimer, 2000, p. 161), it is that they fail to acknowledge that genetic differences between individuals substantially account for differences in their behavior (i.e., that complex human behaviors are substantially heritable).

Turkheimer draws a false equivalence between ‘naïve environmentalists’ and ‘genetic essentialists’ (a.k.a. ‘hereditarians’) in these passages. He chastises the environmentalists for denying the heritability of human behavior and chastises hereditarians for interpreting heritability as a measure of how genetic a trait is. But in fact, both camps interpret heritability as a measure of how genetic a trait is and — as we will presently argue — both camps are correct to do so. While Turkheimer is right to complain that environmentalists reject ‘undeniable if easily misinterpreted facts like heritability’ (Turkheimer, 1997, p. 783), it is Turkheimer who misinterprets what heritability means, not the environmentalists and not the hereditarians.

2.2. Heritability Does, in Fact, Tell Us How Genetic a Given Trait Is…in a Given Population

The upshot of Turkheimer’s argument then, is that no trait is more or less genetic than any other because ‘no trait can develop in any organism without both genetic and environmental inputs’ (Turkheimer, 2011a, p. 826) and because the heritability for a given trait is a population-specific, moving target. But this seems to fundamentally misconstrue what people ordinarily mean when they ask the question ‘How genetic is trait X?’ Under almost every conceivable circumstance in which anyone might be inclined to ask this question, what they mean is, ‘To what extent are the differences in trait X in this particular population due to genetic rather than environmental differences between the members of that population?’ This is precisely the question that heritability estimates answer.⁹ No-one would ask the question ‘How genetic is trait X?’ with respect to a characteristic that differed between things which are not biological organisms because genes cannot possibly be implicated in those differences. Neither would they ask the question in relation to differences in a population of genetically identical clones — as per Turkheimer’s thought experiment — because the differences between them cannot possibly be due to genetic differences. Nor would they ask the question in relation to a trait that did not vary at all in the population that is explicitly or implicitly under consideration.

This last point is worth teasing out in more detail in relation to species-universal traits which, to the extent that they vary at all, rarely do so for genetic reasons. In a separate paper, Turkheimer invokes the toy example of the heritability of having two arms:

Having two arms notoriously has a heritability of zero, for example, because the genetic mechanisms that cause us to have two arms don’t vary among individuals. Although developing two arms is intuitively and sensibly a biological process, variation in arm-number is primarily due to environmental events like accidents…It is therefore not a good idea to cite heritability coefficients as a measure of how ‘genetic’ or ‘environmental’ something is. (Turkheimer, 2011b, p. 45)

The heritability of having two arms is just one of many toy examples that behavioral geneticists are fond of using as educational devices to drive home the idea that heritability estimates are population-specific. Other popular examples include the negligible heritability of, for example, having ten fingers, having opposable thumbs, or walking upright on two legs. But scientists who use these toy examples rarely stop to think how unlikely it is that anyone would ever actually pose the question of ‘how genetic’ (or how heritable!) these traits are. They also rarely stop to think what underlies our inference that having ten fingers, or walking bipedally, and so forth, is ‘intuitively and sensibly’ biological or genetic (Turkheimer, 2011a, p. 45). This intuition rests not on the banal observation that ‘no trait can develop in any organism without both genetic and environmental inputs’ (Turkheimer, 2011a, p. 826) but rather on the observation that these characteristics vary between genetically distinct species (see Figure 1). Neven Sesardić addresses this confusion in more detail:

First, no…behavior geneticist ever took zero heritability to mean that genes are unimportant for development. What is taken to follow from zero heritability is only that genes are unimportant for explaining the existing phenotypic differences in the population in question. Second, even if a trait is shared by all organisms in a given population it can still be heritable – if we take a broader perspective and compare that population with other populations. The critics of heritability are often confused, and switch from one perspective to another without noticing it…in explaining species universals, what we account for is why a particular species has a given trait, while other species don’t. It is a question about (inter-species) variance, not invariance… The low heritability of opposable thumbs among humans reflects the fact that, indeed, in this population (humans) there is no genetic variation in that trait. However, the reason we regard the trait as genetic is that we actually think about the variation in this trait in a wider population (say, consisting of humans and monkeys like tamarins and marmosets). In that population, heritability of having opposable thumbs is high, and the trait is genetic…. Without such a contrasting case, … it is difficult to make sense of the statement that [a given trait] is genetically caused. (Sesardić, 2005, pp. 28–30)¹⁰

Figure 1. The interspecies heritability of limb number. While variation in the number of limbs within a species is overwhelmingly due to environmental influences (e.g., injuries causing limb loss), the variation between species is primarily due to genetic differences. It is the second type of variation that undergirds our intuition that developing four limbs (in humans) is genetic. Within-species heritability estimates fail to capture the contribution of genetic differences to interspecies differences in limb number because they have the wrong reference population. When the reference population is expanded to include different species, the heritability estimates will become large, reflecting the major contribution of genetic differences to the observed differences. This image was generated using DALL.E 3 (OpenAI, 2024).

Turkheimer risks leading his readers seriously astray, then, when he suggests the negligible heritability of species universals teaches us that heritability estimates are not an indicator of how genetic a trait is. Heritability estimates are a perfectly sensible answer — perhaps the only sensible answer — to the question of how genetic a trait is once we recognize that the question is always implicitly asked in relation to a specific population in which the trait varies. This becomes clear if we pose the question how genetic the number of fingers is in a population consisting of people who have five fingers or more on each hand. The heritability in this case would be 100%, accurately reflecting that polydactyly is an entirely genetic condition. In summary, the question ‘How genetic is trait X?’ is a question that is as population-specific as the question ‘How heritable is trait X?’, and the answers to both questions are one and the same.

2.3. Turkheimer’s Confusion About What Heritability Means

Turkheimer’s confusion on this point sometimes leads him to directly contradict himself about what heritability means. Despite repeatedly insisting that heritability cannot measure how genetic a trait is, he frequently invokes the first law of behavioral genetics as proof that everything is genetic. Indeed, he makes both (contradictory) points in the same sentence of ‘Heritability and Biological Explanation’, when he writes ‘Heritability …cannot be the basis for establishing whether behavior is genetic… because to do so leads only to the banal tautology that all behavior is ultimately based in the genotype’ (Turkheimer, 1998, p. 782 [emphasis added]). Why does heritability suggest everything is genetic? Because ‘everything is heritable and biological’ (Turkheimer, 1998, p. 789 [emphasis added]). Of course, this directly contradicts his own point that having two arms (or ten fingers or walking upright) is clearly ‘genetic’ despite having a heritability of zero. Nevertheless, Turkheimer insists:

If one accepts …that genes are a necessary component of all development, and if one reinforces the confidence in this conclusion by observing the ubiquity of heritability for variation in behavioral phenotypes, the existence of small covariations between alleles and behavior becomes a theoretical certainty, not a risky empirical prediction. (Turkheimer, 1998, p. 788 [emphasis added])

However, the ‘ubiquity of heritability’ for behavioral traits was not a ‘theoretical certainty’ as Turkheimer claims here. It was reasonable to expect that, for many traits, the behavioral differences between people might be entirely due to environmental differences between them. The fact that ‘genes are a necessary component of all development’ in no way guaranteed that human behavioral differences would covary with genetic differences. If it had, the discovery that ‘everything is heritable’ (Turkheimer, 1998, p. 789) would not, in Turkheimer’s own words, be an ‘astounding fact’ (Turkheimer, 1998, p. 785) that ‘has taken all sides of the nature nurture debate by surprise’ (Turkheimer, 2000, p. 160). Heritability cannot be both a ‘banal tautology’ and an ‘astounding fact’.¹¹

In truth, the discovery that virtually all human behavior is partly heritable is neither ‘a banal tautology’, nor an astounding fact. Given that the evolution of modern humans has been driven by natural selection acting upon behavioral differences in our hominin (and earlier) ancestors, it would be extraordinary if genetic differences did not produce behavioral differences in modern humans. That would suggest there is no genetic basis to behavioral variation in modern humans upon which evolutionary pressures could potentially act. Nevertheless, it remains an empirical question whether (and how much) genetic differences account for differences in behavioral traits in contemporary human populations. The ‘astounding fact’ is not that genetic differences partly account for human behavioral differences. It is that genetic differences tend to account for a large fraction of human behavioral variation.¹²

Across several of his most influential writings, then, Turkheimer confuses two fundamentally different meanings of whether a trait is ‘genetic’: one that is a ‘banal tautology’ and one that is an empirically testable question. The tautological sense in which any given trait is ‘genetic’ is the sense in which any trait in any individual organism requires both ‘genes and environments to develop’ (Turkheimer, 2011a, p. 826). The empirically uncertain question about whether a trait is ‘genetic’ is a question that concerns whether genetic differences account for phenotypic differences in a population. It is only the second question that heritability estimates answer. The ubiquitous heritability of human behavior does not simply corroborate the trivial observation that ‘genes are a necessary component of all development’ (Turkheimer, 1998, p. 788). In fact, it fails to corroborate that fact at all.

On the basis of this confusion, Turkheimer frequently mischaracterizes heritability as a dichotomous indicator of whether genes are implicated in the existence of a trait, rather than a continuous indicator of the extent to which genetic differences account for variation in a trait. He proceeds to mischaracterize ‘naïve environmentalists’ who resist the idea that human behavioral traits are heritable as rejecting the obvious fact that genes are necessarily implicated in the development of all human traits when, in fact, they are rejecting the proposition that genetic differences explain human behavioral differences. More importantly, he groundlessly dismisses the ‘hereditarian’ view that heritability is a measure of ‘how genetic’ a trait is, because he fails to recognize that the question is always implicitly posed in relation to a target population in which the trait varies.

2.4. Is Adult Height Really No More Genetic Than Robbing Banks?

Building on what we have said above, we can now re-examine Turkheimer’s claim that ‘the whole project of differentiating traits into those that are more or less essentially genetic is hopeless from the start’ (Turkheimer, 2011a, p. 826). It will help if we first dispense with this troublesome word ‘essentially’. Turkheimer is right to insist that there is no absolute scale of ‘geneticness’ we can use to determine whether a trait is more or less genetic than another trait across every population and circumstance. But, as we have sought to establish, the question ‘How genetic is trait X’ is also a population-specific question. If we compare the heritability estimates for two different traits estimated in large representative samples drawn from the same population, we will be able to assess whether genetic differences account for more of the phenotypic differences in one trait versus another trait in that population. Turkheimer is therefore mistaken when he says, ‘Growing to a particular height is no more genetic than robbing banks’ (Turkheimer, 2011a, p. 826). If someone were to ask the question in good faith, they would almost certainly be asking something like ‘Do genetic differences between people account for more of the observed differences in their height than the observed differences in their tendency to rob banks…in the contemporary United States?’ — and the correct answer to that question would be the intuitive one, namely, ‘Yes, they do!’¹³

While reference to a specific population is often unspoken in the questions ‘how genetic is trait X?’ or ‘Is trait X more genetic than trait Y?’, it is nevertheless implied, and that implicit population is almost always ‘…in societies like the one you and I currently inhabit’. Such unspoken context is a ubiquitous aspect of our language. When we ask a waiter in a London restaurant how much a glass of wine costs, we are not asking how much it costs in Timbuktu, or how much it cost in mediaeval England, or how much it will cost on a Mars colony 200 years from now. We would find it extremely grating if the waiter responded to our question with the answer, ‘It depends on the context’.¹⁴ Yet many critiques of ‘hereditarianism’ take precisely this form and are often mistaken for profound insights. But it is neither profound nor insightful to misread the obvious thrust of questions posed in ordinary language to belabor an irrelevant technical point. Indeed, it is actively harmful when it leads the questioner to regret posing a perfectly legitimate question or when it persuades her that remaining ignorant of the heritability of a trait that she was interested in is a form of enlightenment. Similarly, when laypeople look to experts to answer questions about how genetic a particular trait is, they do not expect to be told sententiously that ‘all traits…require…genes and environments to develop’ (Turkheimer, 2011a, p. 826). They knew that perfectly well already.

2.5. Should We Care About the Magnitude of Heritability Estimates?

Turkheimer seeks to persuade his readers that they should be indifferent to the magnitude of heritability estimates, writing, ‘my own view is that the heritability of adult criminality is very unlikely to be zero, and other than that its magnitude is indeterminate and of little theoretical interest’ (Turkheimer, 1998, p. 787).¹⁵ However, in an important concession, Turkheimer goes on to note that the nonzero heritability of criminality ‘…means that, on the day people are born, they do not all have an equal probability of becoming criminals. It means that if a deacon adopts the son of a gangster, the child will have a greater chance of following a life of crime than his adoptive sibling sprung from the deacon’s loins’ (Turkheimer, 1998, p. 788).

Here, Turkheimer has switched back from describing heritability as a tautological indicator that genes are implicated in all human traits, to describing heritability as an indicator that genetic differences between people produce differences in their traits. He concedes that the biological children of gangsters will tend to have a higher probability of becoming criminals than the biological children of deacons because of their higher genetic propensity for criminal behavior. What Turkheimer conspicuously fails to tease out, however, is that the higher the heritability of criminality in his locality the more the deacon should be concerned that his adopted son might go on to pursue a life of crime.

Turkheimer writes that heritability means ‘[i]t is no longer possible to interpret correlations among biologically related family members as prima facie evidence of sociocultural causal mechanisms’ (Turkheimer, 2000, p. 162) and insists that ‘if you are studying the effects of childrearing practices on [e.g.] adult criminality, you had better include adoptees in your research agenda’ (Turkheimer, 1998, p. 788). But again, the larger the heritability estimates that we find for a given trait, the greater the attenuation of the parent-offspring correlation we can expect in adoptive families. Indeed, heritability is sometimes estimated by doubling the difference between the parent-offspring correlation in traditional versus adoptive families.

Far from being of ‘little theoretical interest’, then, the magnitude of heritability estimates for criminality is something that has obvious relevance to prospective parents considering adopting children from troubled families. More broadly, the magnitude of these heritability estimates should also be a matter of profound interest to behavioral scientists and policymakers seeking to understand the intergenerational transmission of criminal behavior. That the heritability of criminality is ‘indeterminate’ in the sense it might be different in Rio de Janeiro than in Stockholm does not diminish its relevance to parents or policymakers from either jurisdiction so long as they draw on statistics relevant to their own population.

In Sweden, both twin- and sibling-difference designs find that genetic differences between people account for more than 50% of the variation in violent criminal offending, while environmental differences between families account for less than 15% of the variation (Frisell et al., 2012).¹⁶ Against a backdrop in which scientists and policymakers routinely interpret ‘causally ambiguous correlations’ between relatives as measures of the social transmission of (dis)advantage, the magnitude of these estimates should be of major theoretical interest, as they essentially reverse the empirical conclusions upon which many public policy decisions have been based.

To summarize, Turkheimer argues that all human traits require both genes and environments to develop and that all that we can infer from heritability is that all traits are equally genetic. However, in this he is mistaken: nonzero heritability neither confirms nor disconfirms the fact that genes and environments are inseparably intertwined in the development of each individual organism. He argues further that heritability does not quantify how genetic a trait is because heritability estimates are population-specific and there is no absolute scale by which the genetic influence on a trait can be measured or by which different traits can be compared. But when people ask how genetic a trait is or whether one trait is more genetic than another, they are always implicitly referring to a target population in which the trait varies. The question of how genetic a trait is is therefore always a relative question of how much genetic differences contribute to trait differences in a particular population — a question that heritability answers.

As Visscher et al. (2008) note, ‘Heritability is a simple dimensionless measure of the importance of genetic factors in explaining the differences between individuals, and it allows an immediate comparison of the same trait across populations and of different traits within a population’ (p. 259, emphasis added). Contra Turkheimer, it is not ‘misguided’ to ‘use heritability as a means of quantifying the genetic…structure of a trait’ (Turkheimer, 2011a, p. 826) nor is it ‘hopeless’ to ‘differentiat[e] traits into those that are more or less…genetic’ (Turkheimer, 2011a, p. 826). Nevertheless, we will ultimately agree with Turkheimer that heritability cannot be used to measure human agency and agree that ‘a moral analysis of [human] traits will have to be found elsewhere’ (Turkheimer, 2011a, p. 826). Unfortunately, we will show that the nonshared environment cannot be used to measure human agency either.

3. The Nonshared Environment Faces the Same Pitfalls Turkheimer Ascribes to Heritability

3.1. Turkheimer Decides That Heritability Does Measure Controllability After All

Having dispensed with heritability as a measure of genetic constraints on human agency (on rather dubious grounds), Turkheimer looks for an alternative way to ground our sense that we have more control over some traits than others: ‘Isn’t there some way,’ he asks, ‘to rescue our common sense intuition that people are more responsible for their life of crime than for their short stature?’ (Turkheimer, 2011a, p. 826). He nominates the nonshared environmental variance component as an alternative measure of agency, which, ‘From the point of view of the individual represents the degree to which [e.g.,] intelligence is under that individual’s control, as opposed to the degree to which it is limited by either genes or upbringing’ (Turkheimer, 2011a, p. 827 [emphasis added]).

In sections 4, 5, and 6, we will return to interrogate whether the nonshared environment can fulfil the metaphysical function that Turkheimer ascribes to it. For now, we note this passage seems to directly contradict Turkheimer’s previous claims about heritability. It suggests heritability is a measure of the degree to which an individual’s control over their intelligence is limited by their genes after all (while the shared environment indicates how much their control is further limited by their upbringing). This remark stands in direct contradiction to Turkheimer’s claims in the previous pages that ‘it is incorrect’ to ‘presume that the higher the heritability of a trait, the more ‘determined’ it is, the less we can control it, and the less we should be blamed for it’ (Turkheimer, 2011a, p. 825) or that ‘the heritability of [e.g.] height says nothing about… our control over our own height or our blameworthiness for our height’ (Turkheimer, 2011a, p. 826).

Ultimately, the nonshared environmental variation, which Turkheimer touts as a way of ‘quantifying human agency’ (Turkheimer, 2011a, p. 827) is nothing more than the residual variation after we account for both the heritable variation and the shared environmental variation for a trait. In proposing the nonshared environment as a substitute measure of human agency to heritability, Turkheimer is merely suggesting that heritability is an incomplete measure of the noncontrollability of a trait because there is an additional variance component that further limits the individual’s control: namely, the shared environment.

After his remonstrations about heritability not being a measure of controllability because ‘traits do not have heritabilities’ (Turkheimer, 2011a, p. 825) or because heritability is not a measure of ‘how…genetic traits are’ (Turkheimer, 2011a, p. 826), Turkheimer’s only departure from the cardinal sin of ‘genetic determinism’ here is to suggest that the ‘genetic essentialists’ have not been deterministic enough — that in practice, the use of heritability as an index of human agency leads us to overestimate the degree of agency we have for a particular trait because it fails to capture the additional constraints on our freedom imposed by our upbringing.¹⁷

3.2. What’s Left for Free Will After Everything Else Has Been Subtracted?

In fact, Turkheimer’s vision of free will is even more severely constricted than this because, by his own admission, the nonshared environmental variance component in twin studies captures two other things which are not free will, namely ‘measurement error’, and also ‘exogenous within-family environmental differences, [such as] differential parenting practices applied within sibling sets’ (Turkheimer, 2011a, p. 826). Importantly, for Turkheimer these ‘exogenous within-family environmental differences’ refer exclusively to the effects of measured environmental differences between twins of the kind reviewed in Turkheimer and Waldron (2000), a crucial point to which we shall repeatedly return in forthcoming sections.¹⁸ Referring back to that review, Turkheimer notes that ‘on average less than 3% of differences in outcome were explained’ (Turkheimer, 2011a, p. 826) by these measured differences in the nonshared environment. Now, 3% of the variation might not be much, but this figure varies across different traits and is sometimes substantially larger. After correcting for measurement error and measured nonshared environments, it is beginning to look like death by a thousand cuts, with a dwindling corner of nonshared environmental variation left remaining for human agency.

However, the prospects for free will get even worse in Harden’s recapitulation of Turkheimer’s theory of free will in her book The Genetic Lottery, where she cuts yet another slice off the nonshared environmental variation that can be attributed to free will. Harden states that the nonshared environment is only ‘an indication of the extent to which people potentially have agency over their outcomes’ (Harden, 2021b, p. 202), adding:

More precisely, [the nonshared environment] might be thought of as an upper bound of the extent to which people have agency. What the neuroscientist Kevin Mitchell calls ‘developmental variation,’ i.e., inherent randomness in processes of phenotypic development, will also pull twins away from one another, without either one of them exerting anything we would typically recognize as agency. (Harden, 2021b, p. 288)

To the extent that Kevin Mitchell and others are correct that stochastic developmental variation explains most of the nonshared environmental variation that we observe (Mitchell, 2020; Tikhodeyev & Shcherbakova, 2019), Harden’s amendment to Turkheimer’s theory would leave precious little variation remaining that can be attributed to human agency. Thus, we find that Turkheimer and Harden, two of the most vocal critics of ‘genetic determinism’ in contemporary behavioral genetics, turn out to be far more deterministic than the so-called ‘genetic essentialists’ that Turkheimer repudiates for suggesting that heritability alone represents a constraint on human freedom. After we subtract the variation due to heritability, the shared environment, measurement error, (measured) environmental differences between twins, and stochastic developmental variation, little or no behavioral variation is left for personal agency to explain under this free-will-by-subtraction theory.¹⁹

3.3. Nonshared Environmental Variation Is Population-Specific, Too

If we recall, the initial argument that Turkheimer enlists to dismiss heritability as a measure of personal agency is that ‘traits do not have heritabilities [because] heritability depends on the population in which it is measured’ (Turkheimer, 2011a, p. 825). He nevertheless goes on to unironically present, in tabular form, ‘a quick collection of nonshared environmental proportions for…. sexual orientation, criminality, mental illness, and body mass index…. height, schizophrenia, and personality’ from different studies drawn from different populations to argue that these ‘are plausibly ordered in terms of their controllability and moral relevance: weight more than height, personality more than intelligence, and depression more than schizophrenia.’ (Turkheimer, 2011a, p. 827). How can Turkheimer say it is misleading to even place the heritability of intelligence in a range, but not consider it misleading to supply estimates of nonshared environmental influence for these traits, let alone organize them in a list that ranks traits in order of nonshared environmental influence and (allegedly) of human agency?

In Table 1 we reproduce the table of nonshared environment estimates (1-rMZ) collated from Turkheimer (2011a), providing additional details on the countries the sample was drawn from in each study. Given the differences in national composition for these studies, it is unclear why nonshared environmental influences would be any more (or any less) comparable than heritability estimates for the same set of studies. There are important differences in gender composition and birth cohort across these studies as well.

Table 1. Nonshared environmental variation estimates reproduced from Table 1 in Turkheimer (2011a) supported with additional information about the nationality of the sample.

Note: *Rhee and Waldman (2002) was, in fact, a study of antisocial behavior. Criminality was only one operationalization of antisocial behavior among many. When analyzed separately 1-rMZ for criminality was only 0.25.

Turkheimer provides one qualification to this list of nonshared environmental estimates: ‘As proportions of variance, they are prone to the some of the same distortions as heritabilities, related to changes in the variability of the nonshared environment, but not to the same extent.’ (Turkheimer, 2011a, p. 827). To support this argument that the nonshared environment is less variable than heritability he writes, ‘as the heritability of intelligence goes up and the shared environmental portion goes down as a function of increasing age…the proportion attributable to the nonshared environment is relatively constant.’ (Turkheimer, 2011a, p. 827). But more recent research on this age moderating effect tells a rather different story to the paper Turkheimer originally enlisted (which was Haworth et al., 2010).

In the most recent longitudinal meta-analysis of developmental changes in the genetic and environmental influences on intelligence, Briley and Tucker-Drob (2013) reported that nonshared environmental influences increased significantly as the interval between intelligence tests lengthened (see Figure 2a). Moreover, Morris et al (2024) performed a primary, longitudinal study of this age moderation effect incorporating data from over 10,000 pairs of British twins tested on 9 occasions between the ages of 2 and age 16. In that study, they found a significant increase in the nonshared environmental influence on intelligence that was even more substantial than the increase in heritability (see Figure 2b) rising by 28 percentage points over 14 years while heritability increased by 24 points.

Figure 2. Nonshared environmental influences on intelligence rise substantially across development. Figure 2a is reproduced from a corrected version of Figure S1 from a meta-analysis of longitudinal twin and adoption studies of intelligence by Briley and Tucker-Drob (2013) that the authors subsequently published on the Open Science Framework (Briley & Tucker-Drob, 2021). The colored bands represent ± 1SD. Figure 2b is reproduced from Figure 3B from Morris et al. (2024), which was a longitudinal study of over 10,000 twin pairs from age 2 to age 16. A = the proportion of additive genetic influences on intelligence, C = the proportion of shared environmental influences, and E = the proportion of nonshared environmental influences. 95% confidence intervals are here shown as error bars. An inset panel in Figure 2b represents the ACE point estimates as a stacked area chart. Note that testing intervals in Figure 2b are not evenly spaced. Both charts show that the increase in the proportion of nonshared environmental variance across development is comparable with the increase in heritability across development.

But we can look further afield than age moderation of variance components for intelligence to disconfirm the proposal that the nonshared environmental variance remains relatively constant across study populations and contexts. If we look at historical trends in the nonshared environmental influence on educational attainment, we see a significant increase in nonshared environmental influence between the first and second half of the 20th century in a meta-analysis of international twin studies by Branigan et al. (2013). A significant increase in nonshared environmental influence across the 20th century was also found in a recent mega-analysis of international twin studies of educational attainment by Silventoinen et al. (2020) in the full international sample. Likewise, if we look at historical trends in the nonshared environmental influence for adult height, a mega-analysis of international twin studies by Jelenkovic et al. (2016) found that nonshared environmental influences decreased significantly across the 20th century both for men and for women.

For many traits, then, population-specificity is as much an issue for estimates of nonshared environmental influence as it is for heritability. Turkheimer’s argument that heritability cannot be used as an index of human agency because ‘the heritability of [X] isn’t anything’ (Turkheimer, 2011a, pp. 825–826) applies equally to the nonshared environment. The selective criticism of heritability on this account is illegitimate. What is more, the nonshared environment can only be interpreted as ‘the variability that remains after genetic and environmental familial constraints have been taken into account’ (Turkheimer, 2011b, p. 827) if heritability is interpreted as a genetic constraint on behavior, contradicting Turkheimer’s position in the same piece that heritability does not measure genetic causes of behavioral differences (see also Turkheimer, 1998, 2000).

So far, we have not questioned Turkheimer’s assumption that genetic causation of behavioral differences represents a constraint on human agency in order to highlight inconsistencies in his interpretation of heritability and to demonstrate that heritability measures the degree to which genetic differences cause observed behavioral differences in a given population.

That done, we can now state that we agree with Turkheimer when he says that heritability does not constrain human agency, albeit for very different reasons. Whereas Turkheimer argues that heritability does not constrain human freedom because it does not capture genetic causes of behavioral differences, we will argue that heritability does capture genetic causes of behavioral differences, but these do not limit our agency.²⁰ The details of that discussion take place in sections 8 and 9. In the next three sections (4, 5, and 6) we will discuss why the nonshared environment cannot be used as an index of human agency either.