Abstract
In Jay Joseph’s review of my book, Understanding the Nature-Nurture Debate, the term quantitative genetics is usually placed in scare quotes, as in, … I have reviewed Eric Turkheimer’s book in the context of problematic “quantitative genetic” studies assessing the behavioral resemblance among relatives, such as family members, adoptees and their biological and adoptive relatives, and twins.
Quantitative genetics is a large and well-established field of study, existing mostly within the safe confines of the biology department, where it is applied without controversy to yeast and drosophila on the one hand, and human physiology and medicine on the other. Human behavior genetics is a tiny corner of the broad domain of quantitative (also statistical, population, and evolutionary) genetics. If the scare quotes are not intended to imply skepticism about the overall enterprise of modern genetics, then the need for them must arise somewhere between drosophila and human behavior.
This state of affairs is odd when you stop to think about it. All of us involved in the nature-nurture discussion, from Galton to Joseph to me to contemporary hereditarian racists and eugenicists, are unquestioning Darwinians as regards the broad outlines of human biology. Acceptance of Descent of Man involves believing that biology and genetics apply to human beings, and that means quantitative genetics just as much as it does Mendelian transmission of eye color. The scare quotes around quantitative genetics imply that humans somehow retain a little extra-biological wiggle room that other organisms lack. I don’t disagree, and this is a good place to point out that across most of Joseph’s review he doesn’t disagree with me, either. Our disputes are not so much about the empirical facts as they are about what to make of them. Our testy intellectual relationship (he is correct that my review of his book was unnecessarily unkind) is explained by our broad agreement about many things, which has a way of heightening the salience of our remaining differences.
Quantitative genetics was invented in one fell swoop by R. A. Fisher in Fisher (1918). Understanding its historical origins is essential because it shows how behavior genetics inherits its strengths and limitations from the same origin. Here is the first paragraph of Fisher’s famous paper: Several attempts have already been made to interpret the well-established results of biometry in accordance with the Mendelian scheme of inheritance. [1] It is here attempted to ascertain the biometrical properties of a population of a more general type than has hitherto been examined, inheritance in which follows this scheme. [2] It is hoped that in this way it will be possible to make a more exact analysis of the causes of human variability [Numbers added].
Fisher proposed two goals. The first one is well-known and led to the establishment of quantitative genetics as we know it today: synthesizing the great debate between the “Mendelians” who based their science on biological observations of the laws of genetic transmission of discrete traits and “Biometricians” who observed the variances and covariances of continuously distributed traits. The second goal, however, comes with a surprise: it is about discerning the causes of human variability. Why humans? Didn’t we just agree that quantitative genetics applies to all organisms?
Part of the answer is that Fisher was a committed eugenicist, for whom analysis of the causes of human variability had abominable practical implications. It runs deeper than that, however. In the next several paragraphs of the paper, Fisher invents analysis of variance, and thus most of the common statistical practice of the remainder of the century. Analysis of variance holds together Fisher’s two goals: [1] Mendelian relatedness and transmission, summed over many infinitesimal genes, produce normal distributions of known variance and covariance; [2] non-experimental observation of those variances and covariances, referring now to humans, can be used to infer the unobserved biological processes that caused them.
It is important to maintain the separation between these two conjectures. That Mendelian biology can produce biometric covariance structures is the unquestionable basis of un-scare-quoted modern quantitative genetics. The empirical foundation of that quantitative enterprise is that genetically related family members are more phenotypically similar than random individuals: the closer the genetic relationship, the more similar the phenotype. Fisher’s ratio of genetic to phenotypic variance is an intraclass correlation that quantifies clustering within genetically related families. The broad and narrow heritability coefficients derived by the animal breeder Jay Lush (1935) are exactly such ratios, and under conditions on the farm, narrow heritability is the mainspring of the breeder’s equation quantifying the lawlike response of phenotypes to artificial selection.
Now comes the disagreement between me and Jay Joseph. It turns out that the fundamental observation of correlation between genotypic and phenotypic similarity applies to free-ranging humans just as it does to plants and animals on a farm. Not only that, it applies to the behavior of humans just as much as it does to their physical characteristics. In fact it applies to all human differences, which are therefore, in the Fisher–Lush statistical sense, heritable. As Joseph notes, I (Turkheimer, 2000) named this observation, “The First Law of Behavior Genetics.”
That sounds hereditarian, the sort of genes-forward belief that would justify Joseph’s dissatisfied and disappointed review, but it is not. The analysis above is based only on Fisher’s first principle, not the second. Yes, more genetically related humans are more phenotypically similar, but no, one cannot use that fact to reach decisive conclusions about the causes of human similarity and difference. We have now located the human wiggle room we were seeking, but how can it be true? Why does something as biologically universal as quantitative genetics not apply to humans the same way it applies to corn and pigs?
The answer is that human beings are not raised on farms. We can’t breed humans, we can’t raise them under controlled conditions, and we can’t sacrifice and dissect them. The analysis of variance is an engine for detecting and quantifying correlations; the experimental method is required to establish causation, and most human behavior is outside the reach of the experimental method. The prohibition on human experimentation is, one hopes, permanent, which violates our usual expectation that the inexorable progress of science will eventually explain everything. We do not await a time when some newly imagined experimental science will allow us to draw determinist causal conclusions about human behavior. Indeed we have already experienced such a science, in the form of eugenics, and rejected it as a moral disaster.
As a consequence, we can observe correlational regularities in human behavior (The Analysis of Variance, in the first half of Richard Lewontin’s famous Lewontin, 1974 title), but we cannot reach determinist causal conclusions about them (The second half: The Analysis of Causes). The counter-intuitive and disturbing implication is that a large domain of human behavior is and will remain outside the bright explanatory light of natural science. Borrowing from Robert Plomin, I have dubbed the self-imposed limitations on the possibility of decisive explanations of human behavioral differences, “The Gloomy Prospect” (Turkheimer & Waldron, 2000). Notwithstanding this discouraging scientific outlook, the human desire to understand ourselves is not extinguished. The various correlations that can be observed among human behaviors—between genetically related family members, between individuals who share a culture, and between different behavioral tendencies within people or across developmental life-courses—may be causally underdetermined, but they are all we social scientists have, and they deserve to be taken seriously.
Taking correlations seriously means not selecting which relationships one attends to or overestimating one’s ability to understand their causal origins. The ultimate source of my disagreement with Jay Joseph (and along with him the larger absolutist opposition to the very idea of behavior genetics) is that I insist that the fundamental observation of correlation between genotypic and phenotypic similarity is real, with at least some real consequences, even as we remain unable to specify its genetic and environmental origins. I have been around family data for a long time by now, and I have never—not once—observed a dataset in which the correlation between genetic and phenotypic similarity did not occur. Monozygotic twins are more similar than dizygotic twins. Siblings are more similar than half-siblings. Cousins are more similar than unrelated people. People who share some genetic group status, however defined, are more similar than those who don’t. Unrelated people who are more similar on the basis of the SNPs are more similar than those who are less similar. It’s the First Law again, and although Joseph is correct that there aren’t any laws in the human behavioral sciences, this is as close as one could hope to come.
It is true enough that behavior geneticists of all stripes, from hard-core racists and eugenicists to well-meaning clinical scientists who only want to alleviate the suffering of unhappy people, have too often followed Fisher into a straightforward assumption that genetic causation is the basis of the First Law of Behavior Genetics. That way lie some of the great historical calamities of the modern West; more recently it led to (Arthur) Jensenism; to this day it is deployed in the service of malignant (pseudo) scientific racism; astonishingly, eugenics itself is making a comeback. Although Joseph doesn’t acknowledge it, all this is documented in my book; yet I still contend that reaction against the consequences of genetic determinism must not drive us into the arms of its biologically naive environmentalist twin. Darwin will not allow us to forget that humans, no different than any other organism in this regard, must reckon with our genes.
Moreover, it is also true that environmentally oriented developmentalists have misinterpreted familial correlations in the opposite direction, without any stronger causal justification than the hereditarians. Accounts of this environmentalist error are legion in the behavior genetic literature, ranging from the twin study theorists like Sandra Scarr (1996) and Judith Rich Harris (2011) to more recent authors like Paige Harden (2022). If Behavior X in parents is correlated with Behavior Y in their biological children, one cannot assume that X causes Y via environmental pathways. Try the same study in adoptive families and see what happens. The necessity of this interpretive caution neither requires believing that uncontrolled adoption is a watertight experiment nor declaring that either X or Y is “genetic.” It is just common-sensical, if necessarily weak, causal inference.
These considerations make me a skeptic about genetic explanations of human behavioral differences without turning me into an environmentalist. In fact, one could reverse “environmental” and “genetic” in the argument above and it would remain just as true. On the genetic side, we cannot intentionally breed human beings; on the environmental side, we cannot assign human beings to environments. On the genetic side, we observe effects at a general level (identical twins are very similar, more so than fraternal), but not at a specific level (there are no individual genes that cause behavioral differences). On the environmental side, we can observe general effects (poverty is bad for children), but not specific causal relations (is the poverty correlation caused by diet, schools, or neighborhoods?). Human families are the same tangle of uncontrolled genetic and environmental associations from one point of view as they are from the other.
Joseph’s criticism of my book is based on the contention that despite a forty-year career as a critic of mainstream behavior genetics, I was finally unable to pull the plug. He finds fault with my loyalty to my great mentor Irving Gottesman. He quotes me as describing candidate gene studies of human behavior as, “a complete and utter failure,” but then finds that I was insufficiently dismissive. After repeating almost verbatim my history of the failure of DNA-based explanation of human behavioral differences in the form of the unsuccessful predictions of the behavior geneticist Robert Plomin, he declares that in the end I haven’t been critical enough, a conclusion that will no doubt come as a surprise to Plomin (e.g., Turkheimer, 2019).
Unsurprisingly, I disagree that either misplaced loyalty to my mentors or timid deference to my opponents is the fundamental cause of the difference between me and Joseph. The real difference is that I am a working social scientist, and he is not. I don’t say this to disqualify him; I have certainly preached about domains where I don’t practice. I intend it as a challenge. Along with being a critical history of science as it has been applied to the nature-nurture problem, Understanding the Nature-Nurture Debate is an account of my own efforts to conduct that science just a little bit better. I have spent a career wondering why some children do better than others in school and why children from divorced families drink a little more alcohol than children from intact ones. It is easy to criticize oversimplified genetic or environmental explanations of that sort of thing, but what next? Should we just compare children in divorced and non-divorced families, declaring that any differences we observe represent the causal consequences of divorce, and leave it at that? If not, then why not? Isn’t the First Law of Behavior Genetics part of the problem?
What I, with my students and colleagues, did do, was to identify pairs of adult twin mothers, one of whom was divorced and the other not, and compared their children (d’Onofrio et al., 2005). It was in that context that we found that the children of the divorced mothers drank a little more than the children of the still-married twin mothers. Although we didn’t make much of it, the models we used to analyze the twin data had something like heritability coefficients lurking in the background, and those heritabilities were based in part on Joseph’s bête noir of the equal environments assumption. Worse yet, only the mothers were twins; the fathers (who probably differed in their drinking behavior) were uncontrolled. So we weren’t randomly assigning children to divorced and intact families, but that isn’t possible, and the children of twins design offers some imperfect control over processes that would otherwise just be unanalyzed confounds. Some of those processes and confounds that we isolated with the twin mothers can be broadly characterized as “genetic.”
Joseph may not like that characterization, but criticisms of other peoples’ solutions to intractable scientific problems ring hollow without some description of how one might do it better. Joseph limits himself to one “to be sure” paragraph at the end of his review: Saying the critics [of behavior genetic methods] were right all along does not necessarily mean that humans begin life as psychological “blank slates,” or that there are no inborn/genetic within-group individual differences in human intelligence and other behavioral areas, or that causal genes will never be discovered.
But if we agree that humans don’t begin life as psychological blank slates, then how should we go about studying how parental divorce affects children, or (closer to Joseph’s own interests) why some people develop schizophrenia and others do not? Joseph’s only answer is to revert to garden variety environmentalism: But it does mean that perinatal, family, social, cultural, religious, educational, geographical, and political environments (including institutionalized oppression/privilege, economic inequality, and neocolonialism) play a decisive role in shaping human behavior.
Whatever “decisive” may mean in that passage, I certainly don’t deny it in the book. Joseph’s catalog of socioenvironmental standards is as easy to rattle off as “genetic influence” and “polygenic prediction,” but don’t environmental structures require the same kind of rigorous non-experimental analysis as purported genetic effects? Can one be meaningfully investigated without the other? Simply stating a preference for genetic or environmental explanations of human behavioral differences gets one exactly nowhere. The proper response to the gloomy prospect is neither hereditarianism nor environmentalism: both attempt to simplify human self-determination and wind up cheapening it. Neither is the solution scientific nihilism, my own occasional vice. The proper responses are methodological pluralism and epistemic humility. The empirical and philosophical struggle with the complexities of genetic and environmental differences, combinations, and interactions leads to answers outside the hoary “nature vs. nurture” distinctions that Francis Galton established 150 years ago, and to which Jay Joseph still clings.
