A Hereditarian Edifice on a Foundation of Sand
Review of Charles Murray’s Human Diversity
[Note: This review is a slightly expanded and edited version of one that was originally posted on Amazon.com on March 1, 2020.]
Social constructionism holds that race and gender are shaped more by social forces and less by biological facts than people commonly realize. Charles Murray believes that this view has become an unscientific “orthodoxy,” and in his book Human Diversity, he launches a blistering attack against social constructionists, whom he portrays as an academic “elite” who bully and browbeat their critics into submission. In reality, social constructionism is a deep intellectual tradition with an extensive literature. Murray’s readers, however, will never know this, as Murray’s entire survey of this literature consists of a single sentence from Berger and Luckmann’s (1966) Social Construction of Reality (p. 3) and, later in the book, a few quotes from theorists, activists, and social critics. This tactic is unfortunate because by conflating the theory with the activists who invoke it, Murray is undermining legitimate critique of both. Questions about the value of social constructionism and the appropriate relationship between theory and activism are important ones, and Murray’s periodic descent into stereotyping and name-calling does nothing to clarify these important issues.
The main thrust of Murray’s book, however, is to show that there is another way—an alternative to social constructionism—for understanding human diversity. This other way is the biological approach, which Murray considers hardheaded and rigorous, as opposed to the fuzzy thinking of the constructionists. It turns out that the assumptions underlying this biological approach—particularly as applied by Murray—also rest on a great deal of fuzziness, though frequently concealed. In fairness to Murray, the genomic and psychometric literatures he is writing about are difficult and technical ones. Murray recognizes this and provides several lengthy sections explaining the technical issues in some detail so that non-specialists can understand them. It is obvious that Murray put a great deal of work into making these sections clear and accessible, and his success in doing so makes them the strongest parts of the book. The rest of the book, however, is considerably more problematic. The issues are complex, and Murray does not always present the whole picture. Many readers will assume that Murray’s representations are accurate. Few will directly access the sources he cites, and fewer still will know how to evaluate the methods and findings of these sources, or the conclusions that Murray draws from them. In what follows, therefore, I will detail some of the significant problems with Murray’s account. NOTE: Some of the following citations refer to sources cited by Murray (indicated with an asterisk); others refer to scientific sources that will not be found in Murray’s book. Full reference information for all these sources can be found on the “References” page of this website.
To begin with, Murray consistently overstates the evidence for genetic influence and understates the evidence for environmental influence on human diversity. He devotes large sections of the book to the former, often mentioning the latter only in passing, or in endnotes, or not at all. For example, Murray makes no mention of the Flynn Effect, one of the clearest indications of environmental influence on cognition (Mackintosh, 2011); he cites several sources on the validity of twin studies (pp. 215-217) but ignores Joseph’s (2015) extensive critique of that literature; he stresses the limits of early childhood interventions but says little about the social forces that undermine them (Protzko, 2015); and in discussing stereotype threat he emphasizes publication bias, yet says nothing about such bias in the publication of brain imaging studies, where it appears to be rampant (see Jennings & Van Horn, 2012). Along similar lines, the genetic methods and technologies that Murray admires often have serious reliability issues (for example, see de Gruijter et al., 2011 and Szpak et al., 2019), yet little or no attention is given to rigorous research designs finding environmental effects (e.g., Koch, D’Mello & Sackett, 2015). These examples are not exhaustive and several more will be given below. But Murray’s general stance is worth noting here—as is the fact that he frequently tags biological and genetic studies with adjectives like “seminal,” and “highly regarded” (pp. 102, 438), while ignoring or dismissing research widely recognized as supporting social construction (e.g., *Lewontin, 1972).
It should be kept in mind that nearly all the evidence that Murray cites to argue for genetic determination is correlational. Even the brain imaging studies he reviews typically show neural correlates whose causal relationships to developmental and environmental influences are complex, multi-directional, interactive, and largely unknown. Most college students understand that correlation does not prove causation; but they rarely grasp just how ubiquitous and persistent correlation/causation fallacies actually are. Even professional researchers commit these fallacies when they survey vast fields of interrelated variables and make conscious or unconscious assumptions about causation which they then import into interpretations of the data based on circular reasoning. Hereditarians like Murray are notorious for falling into these traps—and some, like Arthur Jensen, for diving into them. Such fallacies are particularly misleading when used to portray group differences as genetic, not only because they frequently scapegoat ethnic minorities and confuse the public, but also because they concern processes of such complexity (like the neural interactions described above) that inferences about genetic causation are essentially unfalsifiable—i.e., nothing can be decisively proven or disproven, so anything goes. At times, Murray tries to avoid these fallacies, but he nevertheless succumbs to them and their associated circular reasoning on a regular basis, as exemplified by (but not limited to): genetic interpretations of statistical heritability estimates (e.g., pp. 228-29); ignoring environmental population stratification (e.g., p. 189); and referring to the causal role and “biological reality” of the “g” factor (pp. 231, 427) (regarding the latter, see Horn & McArdle, 2007 and Sternberg, 2019; regarding the former two examples, see below). Murray does have some awareness of the correlation/causation problem, and in one section he attempts to minimize it by asserting that prediction alone, without knowing causation, is good enough for a science of genomics (pp. 285-92). There are two problems with this tactic: First, predictions based on gene studies are weak, and, despite Murray’s optimism, are not showing much progress in becoming stronger (Turkheimer, 2019). And second, Murray himself is not happy with mere prediction—he repeatedly envisions an advanced science of genomics comparable to physics. But Murray’s references to physics focus on the 19th century; he avoids 20th century physics with its discoveries of inherent limits like the Heisenberg Principle and chaotic processes. Biology may be encountering similar limits in the 21st century.
Turning now to Murray’s specific claims, he begins by examining whether psychological differences between males and females might be biological. In Chapters 2-4, he surveys much evidence for male/female psychological differences, and in each chapter he gives additional evidence that the differences persist worldwide, which he interprets as support for biological causation. But the meanings of this evidence are not always clear. Since these data are correlational, male/female differences may simply reflect socialization. Even the international studies, which are supposed to show pan-cultural gender differences, are problematic in that they focus on nations rather than cultures. This means that far from reflecting underlying biological causation, gender role differences—especially the larger differences in developed nations—may reflect, in part or in whole, choices of people in various cultures within these nations (e.g., working class and/or rural areas within the U.S.) to socialize their children along traditional lines. To supplement such findings, therefore, Chapter 5 covers biological and brain scan findings, which Murray feels to be the strongest evidence for male/female differences. As noted above, the brain scan data are correlational and are probably also inflated by publication bias. They have the same problems as the data in the first four chapters. However, other findings reported in this chapter do seem to present evidence for actual biological causation. These include studies of individuals with hormonal abnormalities and studies of responses to hormone injections. While the former are “natural experiments” with confounding variables, the latter are true randomized controlled trials, suggesting that hormones have a causal role. These findings do not contradict the role of social construction in gender roles and behavior, but they do indicate complex interactions between biology and socialization, or to put it another way, between sex and gender.
The next four chapters focus on race. Murray’s position here will strike some readers as peculiar. He begins by noting that the race concept arose in the context of colonialism and slavery (p. 129) and adds that the term “race” no longer has any legitimate scientific usage (p. 135). But a few pages later (p. 148), he asserts that human beings fall into genetic clusters that conveniently correspond to the racial categories currently used in the U.S. So we will be talking about races after all, only now we are calling them “ancestral populations.” This back-and-forth about race embodies Murray’s struggle with the internal contradictions of his own position. He wants to drop the word “race,” but not the category. Yet it is the category that is socially constructed and that invites the scientifically unproductive conflation of social and biological causation. In the past, racial categories served to preserve the subordinate status of specific groups whose members bore phenotypic markers like skin color—whereupon the responses of those members to their subordination (that is, their temperament, intellect, speech, etc.) were judged inferior and regarded as proof of the rightness of that subordination. Murray now hopes to redefine membership in those same groups, not by phenotypic but by genotypic markers—whereupon psychological measurements, bogusly interpreted as genetic (see below), will establish the same old circular justification. Meet the new conflation; same as the old one.
This conflation becomes more clear as Murray explains how his ancestral population clusters were discovered. Researchers led by Noah Rosenberg (*Rosenberg, Pritchard, & Weber, 2002) using a computer program called Structure found that the genomes of people whose ancestors came from each geographical continent are slightly more similar to each other than they are to those whose ancestors came from different continents. This is not surprising since genetic variants (alleles) change frequencies over time as populations disperse through migration, and these populations can therefore be expected to show greater similarity to neighboring populations and less similarity to more distant ones. But to Murray and others looking for a scientific basis for race, these clusters of genetic similarity hold great significance since they roughly align with the five major geographical continents (Africa, Europe, Asia, the Americas, and Oceania) and the racial associations of these continents for most Americans.
But there are problems with this interpretation. First, the Structure program presupposes that the data will form clusters and that the number of such clusters will be whatever the researcher tells it to find. There are different techniques for deciding if a particular input number identifies objectively real clusters—but these techniques are complex, often vary with sampling, and do not always agree with each other. Therefore, most researches, including Rosenberg, publish multiple outputs, with different numbers of clusters, so that people reading the results can make their own judgments. Murray is not too troubled by this, as he states that this format “usually makes it easy” to see which results are “substantive” and which are not (p. 149). The potential for circular reasoning here is obvious, especially for someone like Murray who is motivated to see evidence for races. Another problem is that several studies have found that the Structure program tends to underestimate the number of ancestral population clusters (see Wang, 2017). In fact, Murray cites a later study (*Li, 2008) that found two additional clusters (which he interprets as roughly supporting the five-cluster theory), but he says nothing about several more recent and extensive studies that found still larger numbers of ancestral clusters, including Tishkoff, et al. (2009) (14 clusters), Shriner et al. (2014) (19 clusters), and Baker et al. (2017) (21 clusters). Overall, therefore, the human ancestral structure appears to be much more complex and multi-leveled than Murray portrays it, and not particularly supportive of the continental race theory.
Another problem for Murray is that race categories only weakly differentiate people genetically. This was first discovered by *Lewontin (1972), who found that only about 6% of genetic variance in humans is associated with traditionally defined racial categories (Murray’s wording on p. 130 misleadingly implies that the number was close to 15%). This research has been replicated several different ways, with findings averaging around 4-10% (e.g., Feldman & Lewontin, 2008). Moreover, the interpretation of even this small portion of variance as support for racial clusters is questionable, given what was said above about migration and changes in allele frequency. In fact, geographical distance alone accounts for a far greater proportion of human genetic variance—over 75%. And when such distance is taken into account, the contribution of cluster variance drops to only about 2% (Handley, 2007)—and much of this 2% appears to be due to geographical barriers like oceans and deserts impeding migration and gene flow (*Rosenberg et al., 2005, p. 668). For such reasons, Rosenberg, like most other researchers, rejects a racial interpretation of continental clusters (*2005, pp.668-69).
Murray knows this, and he has already, in effect, preemptively admitted it at the beginning of his “race” section—where he acknowledges that cluster differences are “minor,” and even adds that there are “many ways in which race is a social construct” (p. 157). But Murray is determined to keep this construct alive, and to do so he must show that race is intrinsically biological. So Chapter 8 considers evidence that selective pressures recently operated on the human population. I have already noted the unreliability of this data (see de Gruijter et al., 2011 and Szpak et al., 2019), and even the researchers that Murray cites have acknowledged this (e.g. *Akey, 2009 and *Haasl & Payseur, 2016). I will only add that: (1) all of this evidence involves correlations among molecular elements; it has not been connected to any observable phenotypes, and because much of it involves noncoding DNA it may have no connection with them whatsoever; (2) while some phenotypes have been plausibly identified as recently selected (e.g., lactase persistence), these traits follow a clinal rather than a clustered pattern—that is, they change gradually, vary independently of each other, and do not distribute according to Murray’s continental races; and (3) these traits all involve physical adaptations rather than the cognitive adaptations (like intelligence) that are so central Murray’s larger argument—Murray’s sole piece of evidence for causally induced cognitive adaptation is Belyaev’s 1959 experiment breeding foxes (*Trut, 1999), a study which has been questioned on a number of grounds, including significant sampling bias (Lord et al., 2020).
In Chapter 9, Murray presents additional data in tables and scatterplots. Little is new here, since these data, in effect, simply re-present the cluster argument from Chapter 7. The one new development is that Murray relates this data to genes that correlate with psychological traits like positive affect and cognitive abilities. Since this analysis involves correlations, it has all the limitations already mentioned, plus an additional problem that I have not yet elaborated: environmental population stratification. Population stratification occurs when differences among populations cause researchers to draw erroneous conclusions. Murray does talk about this at times, but he almost always focuses on genetic rather than environmental confounds. Environmental population stratification arises from the fact that the environments of various genetically distinct populations are nonrandomly distributed—that is, people from different ancestral populations grow up in environments that almost always differ systematically from each other. This, in turn, means that correlations between biological genotypes and observed phenotypes are confounded by environmental factors, sometimes dramatically so, especially in the case of psychological traits, which are particularly susceptible to cultural influences. These environmental confounds cannot be factored out unless you know ahead of time exactly what each one is and how to test for it, which is essentially impossible (some writers, including Murray, hold that these influences can be statistically controlled, but this, once again, involves additional circular assumptions). All of this poses a serious problem that will haunt Murray’s analyses for the rest of his book—for example, in the “continental differences” that he surveys at the end of this chapter, and in his later advocacy of polygenic scores (pp. 280-283).
Murray’s final substantive section (Chapters 10-13) allegedly deals with social class, but it would be more accurate to say that his argument here is that intelligence (measured as IQ) is the main source of success, that it is primarily genetically determined, and that social factors like poverty and race play only a minor role. This is the same argument Murray made in The Bell Curve (*Herrnstein & Murray, 1994). It was subjected to a number of devastating critiques (e.g. Kamin, 1995; Heckman, 1995), so I will not reiterate them here. I will merely note that environmental population stratification and correlation/causation fallacies lie at the heart of much of Murray’s reasoning in these chapters, as they did in The Bell Curve. For example, IQ and measures of success are both phenotypes that correlate positively with each other and negatively with certain racial group memberships (genotypes). So the relationship between IQ and success can be highlighted and the ones between both of these phenotypes and race can be disregarded, and along with them, the unjust environmental conditions that contribute to the discriminatory way in which IQ and success are racially and economically distributed in the U.S.
Chapters 10 and 11 detail the statistical technique used to support many of the above claims—the so-called ACE model. Like many hereditarians, Murray relies heavily on this model, which parses measurements from twins and other related individuals into components of genetic and environmental variance that can then be added together to explain behavior. Most notable here is that Murray says nothing about the serious errors that routinely result from ignoring the limitations of this model. In an extensive analysis of all the most rigorous studies using the ACE model in the previous 30 years, *Turkheimer & Waldron (2000) found that designs based on that model had consistently failed to detect large portions of environmental variance. They concluded (pp. 91-93; see also *Turkheimer, 2000) that the additive assumptions of the ACE model rendered it unable to detect the complex and nonlinear ways that environmental forces actually interact with genetic factors and with each other to produce human behavior. To Murray’s credit, he does acknowledge these and other similar findings. But—characteristically—he then misrepresents them as implying that environmental factors can be disregarded as mere “noise” that can be ignored (p. 260). As with so many of Murray’s summaries, the reader will not know what these critiques actually say unless he or she goes to the original sources.
Overall, then, Murray’s argument, both in this section and throughout the book, can be described as a hereditarian polemic. It rests on skewed data, over-interpretations of favored sources, under-interpretations of critical ones, and subtle (or not-so-subtle) misrepresentations of research findings. The result is an impressive edifice of hereditarian ideology built on a conceptual foundation of sand.
In the last two chapters (14 and 15), Murray gives some final thoughts about where the science of genetics is taking us and how we should think about our fellow human beings. By now, the reader can probably guess what these thoughts will be. However, one section (pp, 275-279) is unexpectedly informative because it constitutes, in effect, a burst of honesty by Murray about the true complexity of genetic phenomena and the problems with making the kinds of inferences from genetic data that Murray has been making all along. This section makes worthwhile reading in its own right, and it would have constituted an excellent introduction to Murray’s book had it come at the beginning instead of the end. In that introductory position it might have encouraged the reader to adopt a cautious attitude about the state of our current knowledge about genetics and an appropriate skepticism about Murray’s inferences and speculations throughout the rest of the book.