The Missing Heritability Problem is a Twin Study Misinterpretation Problem: Comments on Scott Alexander’s “Missing Heritability: Much More Than You Wanted to Know”

Missing heritability is a term that human genetic researchers invented around 15 years ago to acknowledge unexpected causal gene discovery failure, and to describe the large discrepancy between heritability estimates derived from twin studies versus those derived from DNA-based (molecular genetic) methods such as genome-wide association studies (GWAS). Importantly, as behavioral geneticist Eric Turkheimer emphasized in his recent book (see my review here), GWASes of psychiatric conditions and behavioral characteristics such as educational attainment (EA, often seen as an IQ “proxy measure”) identify (potentially spurious) gene-behavior “associations” (correlations), not causes.

While sometimes conceding that twin studies might “overestimate heritability,” most researchers and commentators addressing the “missing heritability problem” believe that the assumptions underlying twin research are sound, and that estimating heritability is a valid and important task. A misplaced faith, as we will soon see.

In a June 26, 2025, “Astral Codex Ten” post, psychiatrist Scott Alexander published “Missing Heritability: Much More Than You Wanted to Know.” In his post, Alexander grappled with the missing heritability problem and the “mysteries [that] remain.” In the process, he gave due consideration to recent writings by anti-hereditarian genetic researchers such as Turkheimer and Sasha Gusev.

My task is to address some aspects of Alexander’s analysis, with a focus on his support of behavioral twin research and its underlying assumptions. I begin with his brief historical description of human genetic research, and then move on to other areas.

The mid-20th century was the golden age of nurture. Psychoanalysis, behaviorism, and the spirit of the ‘60s convinced most experts that parents, peers, and propaganda were the most important causes of adult personality.

Starting in the 1970s, the pendulum swung the other way. Twin studies shocked the world by demonstrating that most behavioral traits - including socially relevant traits like IQ - were substantially genetic. Typical estimates for adult IQ found it was about 60% genetic, 40% unpredictable, and barely related at all to parenting or family environment.

By the early 2000s, genetic science reached a point where scientists could start pinpointing the particular genes behind any given trait.

This story aligns with the highly problematic history of genetic research presented by top behavioral geneticist Robert Plomin in his 2018 book, Blueprint (see my review here). In this book, Plomin airbrushed eugenics, German “racial hygiene,” IQ hereditarianism, and genetic determinism out of the history of human genetic research. He claimed that genetic research in psychology began in the early 1970s. Seeming to follow Plomin’s lead, Alexander described “the story so far” in a way that would lead unknowing readers to conclude that genetic theories and research began in the 1970s, and that gene searches began in the early 2000s. Not true on both counts, as I showed in this brief history.

Twin studies suggested that IQ was about 60% genetic, and EA about 40%.

In our review of the supposed genetic evidence presented in the 1994 book The Bell Curve, my colleague Ken Richardson and I provided a detailed explanation of why this claim is unfounded. A major reason is that twin studies, which form the foundation of such claims, are based on a key assumption that is not supported by the evidence (see below). Although Alexander did not mention so-called “twins reared apart” (TRA) studies in the main body of his post, the massive flaws in the five existing TRA studies lead to the conclusion that these studies fail to provide evidence in favor of genetic influences on IQ (see here, here, here, and here). The analysis I present below does not prove or suggest that heredity has no impact on within-group IQ scores or human intelligence, but only that twin studies provide no scientifically valid evidence that it does.

Scientists came up with two hypothesis [sic] for the gap [between twin study and DNA-based heritability estimates], which have been dueling ever since:

1. Maybe twin studies are wrong.

2. Maybe there are genes we haven’t found yet

Since 1998, in four books and numerous other publications, I have argued that Hypothesis #1 is the correct one. So have many others, going back to the 1930s. Allow me to briefly explain why.

The classical twin method (or “twin method”) compares the behavioral resemblance of reared-together MZ (identical) versus same-sex DZ (fraternal) twin pairs. MZs are assumed to share 100% of their segregating genes, whereas DZ pairs share on average 50%. Twin method results typically show that MZ pairs behave more similarly or correlate higher on psychological tests compared to same-sex DZ pairs at a statistically significant level. I designate this finding as “rMZ > rDZ” (where r represents the twin-group IQ score correlation). A behavioral twin study finding of rMZ> rDZ is rarely disputed. What is disputed is how this finding should be interpreted.

The MZ-DZ equal environment assumption (EEA) has always been the most controversial twin method assumption. According to twin researchers, MZ and same-sex DZ pairs grow up experiencing roughly equal environments, and that the only behaviorally relevant factor distinguishing these pairs is their differing degree of genetic relationship to each other.

Some critics, on the other hand, have argued that the EEA as it relates to behavioral and psychiatric twin studies is not valid or even “reasonable” because, when compared with same-sex DZ pairs, MZ pairs grow up experiencing:

• More similar treatment by parents, teachers, and others.

• More similar physical and social environments.

• More similar treatment by society due to sharing a very similar physical appearance.

• A much greater tendency to spend time together, to have common friends and peer influences, and to model their behavior on each other.

• Identity confusion and emotional attachment to each other to a far greater extent than DZs.

As even most twin researchers recognize, MZ twin pairs grow up experiencing much more similar environments than DZ pairs experience. Ironically, writing in support of twin studies in their field of criminology, one group acknowledged in 2014 that “critics of twin research have correctly pointed out that MZ twins tend to have more environments in common relative to DZ twins, including parental treatment…closeness with one another…belonging to the same peer networks…being enrolled in the same classes…and being dressed similarly.”

Because environmental influences confound genetic interpretations of behavioral rMZ> rDZ, non-genetic explanations should become the default explanations.

Nevertheless, Alexander believes that twin studies are valid and produce reasonably accurate heritability estimates, thereby falling into the same trap that “missing heritability” theorists have set for themselves since 2008. In a 2012 article, I concluded that the missing heritability problem is mainly a problem of misplaced faith in twin research. Heritability is still “missing,” which, in a better world, would have led to a serious and objective multidisciplinary assessment of twin studies of behavior and their underlying assumptions. Instead, billions more dollars were spent in a continuing effort to identify elusive “genes for behavior” that may not even exist, and would provide questionable benefits to society were they to be found.

Heritability is a crucial concept in behavioral and psychiatric genetics. According to field leaders, heritability is “the proportion of phenotypic variance that can be accounted for by genetic differences among individuals.” Heritability estimates range from 0% (0.0) to 100% (1.0). Critics, however, argue that heritability estimates in behavioral research depend on several questionable or unsupported theoretical assumptions. Heritability estimates were developed as an animal breeding statistic, in cases where genetic and environmental conditions can be controlled. They do not measure the strength or magnitude of genetic influences on psychiatric conditions or human behavioral characteristics such as IQ, nor do they measure the relative importance of genetic and environmental influences. Developmental psychologist David Moore and David Shenk concluded in their article “The Heritability Fallacy” that the “term ‘heritability,’ as it is used today in human behavioral genetics, is one of the most misleading in the history of science.” The behavioral sciences should have abandoned heritability estimates decades ago.

Twin studies claim that twins are a uniquely powerful genetic laboratory; both fraternal and identical twin pairs have equally concordant environments, but identical twins have more concordant genes. Therefore, the more similar identical twin pairs are relative to fraternal twin pairs, the more heritable a trait must be. But this conclusion falls apart if identical twin pairs actually have more similar environments than fraternal twin pairs do, maybe because parents (knowing their twins are identical) treat them more similarly than they would fraternal twins.

Twin studies don’t “find,” “claim,” or “suggest” anything—people do. Twin researchers interpret twin data genetically, whereas we have seen that others view the same data and argue that non-genetic interpretations are plausible. Continuing his (and many others’) tendency to attribute to research methods what are in fact human interpretations and theories, Alexander wrote that the GWAS method “accepted that most interesting traits were polygenic.” It is more accurate to say that researchers perform behavioral GWASes based on their belief that most interesting behavioral traits are polygenic.

As I have shown, the evidence critics put forward in support of MZ and DZ environmental dissimilarity extends far beyond the simple fact that parents treat MZs more similarly than DZs. MZs much more often experience identity confusion and attachment and see themselves as two halves of the same whole, and spend much more time together. Twin studies do indeed “fall apart” because, as we have seen, MZ and same-sex DZ twins grow up experiencing very unequal environments. Based mainly on twin method MZ-DZ comparisons, Alexander puts IQ heritability at roughly 60%. Yet other twin researchers have produced similar or higher heritability estimates for behaviors such as “dog ownership,” “adult loneliness,” and “vegetarianism.” Twin study red flags, I would argue.

Remember, twin studies find many behavioral traits like IQ are >60% heritable, so you would need to prove not only that parents treat identical twin pairs differently from fraternal, but that this was an overwhelming effect.

This sentence is an awkwardly worded endorsement of twin researchers’ 60-year-old “trait-relevant” defense of the EEA. While recognizing that MZs grow up experiencing much more similar environments than DZs experience, twin researchers often argue that the EEA remains valid until critics are able to identify “trait-relevant” aspects of the environment that cause MZ pairs to behave more similarly.

A basic principle of science, however, is that the burden of proof falls squarely on the people making a claim, not on their critics. Therefore, given that they recognize that MZ and DZ environments are different, behavioral twin researchers using the trait-relevant argument—and not their critics—are required to identify the specific and exclusive trait-relevant environmental factors involved in the behavioral characteristic or psychiatric disorder in question. After identifying such factors, they then must show (1) that MZ and DZ pairs did not experience these factors, or (2) that MZ and DZ pairs experienced these factors to roughly the same degree. Until they are able to do so, Alexander’s and others’ trait-relevant defense of the EEA fails completely.

Parents of identical twins would have to obsessively expose them to the exact same stimuli in the exact same order; parents of fraternal twins would have to send one to the Gifted Advanced Placement Acceleration program while locking the other in a box and force-feeding them lead pellets. Common sense tells us there are no such differences, and studies confirm this: when parents are wrong about their twins’ status (eg they have fraternal twins, but falsely think they’re identical, or vice versa) their trait similarity matches their real status, rather than the incorrect status that determined how their parents treat them; parental treatment explains less than 1% of why identical twin pairs are more concordant.

The first part is a straw man argument. Since parents of MZs don’t expose both twins to the exact same stimuli in the exact same order, and since parents of DZ twins usually don’t send one twin to a gifted program while force-feeding the other twin lead pellets, Alexander suggests that genetic interpretations of rMZ > rDZ must be right.

Science has been embroiled in a “replication crisis” for over a decade. The crisis began in academic psychology, where anonymous surveys suggested that researchers often engaged in “questionable research practices,” leading to desired or publishable conclusions. A key concept of the replication crisis is p-hacking, which is the practice of researchers, either consciously or unconsciously, manipulating definitions, data, and comparisons—either openly or behind the scenes—to transform non-findings into publishable “findings” that fall below the conventional .05 level of statistical significance. The replication crisis has revealed that leaders of major behavioral science fields have passed off and accepted obviously bad science as good science for decades. It has called into question decades of publications throughout science and has led to the retraction of thousands of research articles, including those by eminent field leaders (see the Retraction Watch website).

Twin researchers have published “EEA-test” studies since the 1960s, where they usually concluded in favor of the EEA’s validity. But can we really expect twin researchers to override their confirmation biases and conclude in favor of invalidating their and their colleagues’ life’s work? Building on calls by previous authors, some journals now require research “preregistration,” where investigators submit an introduction, along with their proposed definitions, methods, and analyses, before collecting their data. This important step can help reduce p-hacking in behavioral and other areas of research.

Alexander cited and briefly described EEA-test studies using the “mistaken zygosity” method. He linked a 2013 study by Dalton Conley and colleagues, whose supposedly EEA-skeptical lead author concluded in favor of the EEA’s validity. Like other non-preregistered behavioral research, we have no way of knowing, after reviewing the data, whether they conducted and described their study as initially planned, based on their original definitions, hypotheses, and intended comparisons. In his 2017 book The Genome Factor, Conley championed behavioral genetic research and defended the twin method and the EEA. Discussing his 2013 study, Conley described himself and his research team as “social scientists who questioned the veracity of the equal environments assumption and assumed that the behavioral geneticists were making a fundamental error.” Nevertheless, it is hard to accept that Conley’s evaluation of behavioral genetic twin studies flipped so radically based on one 2013 EEA-test study.

As I discussed in my 2023 book Schizophrenia and Genetics: The End of an Illusion, EEA-test researchers “tested” the validity of the EEA in ways other than the only way it can be tested, which is a simple determination of whether MZs and same-sex DZs grow up experiencing roughly equal environments—yes or no? If no, the EEA is false, and genetic interpretations of behavioral rMZ > rDZ are probably wrong. Most behavioral geneticists understand that finding that a behavioral characteristic “runs in the family” by itself cannot disentangle potential genetic and environmental influences. The same conclusion also applies to a twin method finding of rMZ > rDZ. The only relevant question in assessing the validity of the twin method and the EEA is whether MZ and DZ environments are different, not why they are different.

[Twin studies assume] there is little assortative mating: We discussed this one above in the earlier section on GWAS - smart/pretty/kind/whatever people tend to marry other smart/pretty/kind/whatever people. Why would this bias twin study results? Identical twins share 100% of their genes. Fraternal twins ought to share 50% of their genes - but they get half their genes from their mother, and half from their father. In the degenerate case where the mother and father have exactly the same genes (“would you have sex with your clone?”) even fraternal twins will be extremely similar (although not quite identical, since they’ll get different alleles from each clone). In the more plausible case where mothers and fathers are just a little more alike than chance (eg because smart people tend to marry other smart people), fraternal twins will share a genetic tendency towards a trait somewhat more than their 50% shared genes suggest. Since this makes fraternal twin pairs more (genetically) like identical twin pairs, and twin studies assess heritability as the difference in fraternal-identical-twin-pair concordance, this bias would make twin studies underestimate heritability. But this is the opposite of what you would need to “discredit” twin studies - if this bias is true, then everything is more genetic than twin studies think. And unlike the previous two biases, this one seems real and important, so much so that when you adjust for it, the heritability of educational attainment rises from ~40% to ~50%.

I’m only mentioning this one here because some anti-hereditarians argue that you can’t trust twin studies because of assortative mating, without mentioning that this can only bias them down.

Non-random (assortative) mating is usually defined as the tendency for people to choose mates who are more similar to each other in characteristics (“phenotypes”) than would be expected by chance. However, “anti-hereditarians” usually focus on the EEA and not the random mating (no-assortative mating) assumption.

Twin researchers and their supporters (Conley in The Genome Factor, as one example) sometimes argue that twin studies should be interpreted genetically because violations of the random mating assumption and violations of the EEA roughly cancel each other out in favor of genetics and “heritability.” Like Alexander, they claim that, whereas unequal MZ and DZ environments might lead to an overestimation of heritability, the existence of non-random mating patterns among the biological parents of twins leads to an underestimation of heritability.

Here’s a major problem with this argument. Let’s suppose that the “environmental null hypothesis,” which states that genes that directly cause human behavioral differences do not exist, is true. In this case, mating patterns would have no direct genetic influence on human behavioral differences, and rMZ > rDZ would be caused entirely by non-genetic factors. The contention that non-random mating patterns lead to an underestimation of genetic influences in twin studies assumes, in advance, that the environmental null hypothesis is false. A twin study, however, is an experiment designed to test whether the environmental null hypothesis is false. The findings of this experiment cannot be based on the assumption that it is false, especially since this assumption is based largely on earlier twin studies. Genetic interpretations of previous twin studies cannot be used to validate genetic interpretations of subsequent twin studies. For this and other reasons, supposed violations of the non-random mating assumption cannot be used to validate twin studies.

An alternative way of validating twin studies…is to check them against their close cousins, adoption studies and pedigree studies….I think the twin / pedigree / adoption estimates are mostly right. They are strong designs, their assumptions are well-validated, and they all converge on similar results.

In my 2022 critical analysis of Thomas Bouchard and colleagues’ “Minnesota Study of Twins Reared Apart” (MISTRA), I wrote that a psychological study or method must stand or fall on its own logic and soundness, and cannot be validated by supposed “converging evidence” from other studies and methods. In his 2023 reply, Bouchard strongly disagreed and argued at several points that the supposed MISTRA findings “triangulate” with or should be evaluated in the context of other types of behavioral genetic research, including animal research. Bouchard thereby implied that his IQ study’s findings don’t stand on their own and depend on supposed findings from other studies, as well as from other species. Quite an admission.

I cited psychologist Scott Lilienfeld and colleagues, who observed in 2003 that the “proponents of pseudoscientific claims. ...typically maintain that scientific claims can be evaluated only within the context of broader claims and therefore cannot be judged in isolation.” A classic example is Robert Plomin’s rarely cited 1998 adoption study of personality, in which he and his colleagues framed negative genetic findings as occurring in the context of twin studies supposedly finding 40% heritability.

The correlational IQ adoption studies Alexander cited contain their own set of major methodological problems and questionable assumptions. In other behavioral adoption study areas, there are instances of p-hacking and even fraud, which have been revealed in the replication crisis (see here, here, here, and here). Alexander failed to mention a different IQ adoption study design showing that the adopted-away biological children of poor or working-class parents achieve a roughly 14-point IQ score increase when raised in the homes of families in the upper ranges of the socioeconomic (SES) scale. This finding does not square with Alexander’s belief that IQ scores are “barely related at all to parenting or family environment.”

In any case, twin and adoption studies of IQ and other types of behavior can be evaluated only through a careful analysis of a study’s data, methods, researcher practices, and assumptions—just as claims by the authors of a questionable phrenology study cannot be validated by grouping their study with other methods that supposedly “converge” to predict mental characteristics.

I would add the evidence from some less formal “adoption studies.” During residency, I spent a few months working in a child psychiatric hospital for the worst of the worst - kids who committed murder or rape or something before age 18. Many of these children had similar stories: they were taken from their parents just after birth because the parents were criminals/drug addicts/in jail/abusing them. Then they were adopted out to some extremely nice Christian family whose church told them that God wanted them to help poor little children in need. Then they promptly proceeded to commit crime / get addicted to drugs / go to jail / abuse people, all while those families’ biological children were goody-goodies who never got so much as a school detention. When I met with the families, they would always be surprised that things had gone so badly, insisting that they’d raised them exactly like their own son/daughter and taught them good Christian morals. I had to resist the urge to shove a pile of twin studies in their face. This has left me convinced that behavioral traits are highly heritable to a level that it would be hard for any study to contradict.

Alexander implied that his personal anecdotal experiences constitute scientific evidence. I’m glad to hear that he didn’t “shove a pile of twin studies” in the families’ faces, since we have seen that those studies are scientifically dubious as well, because the EEA is false. I should also note that twin studies of criminality were pioneered by the German “racial hygiene” and criminal biology movements, and flourished in the Third Reich. I documented these origins in my 2004 book The Gene Illusion. A 1929 book describing a German twin study of criminal behavior carried the title Crime as Destiny (some English translations read “Crime and Destiny”). Four years before the 1933 Nazi seizure of power, twin researcher author Johannes Lange concluded, “We must try to make it impossible for human beings with positive criminal tendencies to be born.”

Alexander provided a screenshot of a second-hand account of a well-known 1984 Danish adoption study of criminality by Sarnoff Mednick and colleagues, published in Science Magazine. The author of the X (Twitter) screenshot was IQ hereditarian @cremieuxrecueil, who, as I write these lines, is under fire for supplying hacked information on a progressive candidate in the 2025 New York City mayoral race, which The New York Times decided to publish. @cremieuxrecueil continues to claim official anonymity, although his identity is widely known.

As I documented in The Gene Illusion and a 2001 article, the Mednick adoption study is massively flawed due to selective placement biases, findings that cannot be generalized to the non-adoptee population, how “criminality” is defined in such studies (almost everyone has broken a law at one time or another that could have led to prosecution and conviction if detected by the authorities), and other major problems. Mednick and colleagues’ adoption study is a replication crisis classic, yet the investigators still found no evidence in favor of genetic influences on violent crime. As they stated in a 1988 article, “We consider the research on genetic influence and find no evidence of hereditary transmission of violent criminal behavior.”

A refreshing break from the overall disturbing history of “criminal biology” and twin studies of criminal and anti-social behavior occurred around 50 years ago. In their 1976 twin study publication, O. S. Dalgard and veteran Norwegian twin researcher Einar Kringlen concluded that genetic interpretations of their “minor” findings are wrong because the EEA “cannot be accepted.” The “data do show higher concordance figures in MZ than in DZ with regard to criminal behavior,” they wrote. “However, the differences reported are of a minor character, and statistically non-significant. The natural conclusion would appear to be that, if there does exist a genetic predisposition to criminal behavior, the disposition is a weak one. Even such a modest conclusion, however, is based on the underlying assumption that the environmental conditions for MZ pairs do not differ from those of DZ pairs, an assumption which today cannot be accepted….These findings lead us to conclude that the significance of hereditary factors in registered crime is non-existent” (emphasis in original).

If Alexander gets another opportunity to properly inform adoptive parents about twin research on criminal behavior, I hope he sends them a reprint or PDF of Dalgard and Kringlen’s 1976 study with these and other key passages highlighted.

Although polygenic scores, GWAS, GREML, RDR, and Sib-Regression are also strong designs, they’re newer, have less agreement among themselves, and have more correlated error modes in their potential to miss rarer variants and interactions. Although it’s hard to figure out a story of exactly what’s going on with these rarer variants and interactions, there seems to be some evidence that they exist…and it seems easier to doubt this new and fuzzy area than the strong and simple conclusions from twin / pedigree / adoption work.

Alexander recognized that the candidate gene era (roughly 1990-2015) “mostly failed,” which is a generous assessment. It actually should be a major scandal, as billions of dollars were wasted, and thousands of scientific papers and countless media articles reported supposedly exciting new gene discoveries. Yet the entire candidate-gene enterprise turned out to be, as Plomin wrote in Blueprint, a “flop” and a “fiasco.”

For example, a highly publicized link between major depression and a candidate gene was proposed in a widely cited 2003 study by Avshalom Caspi and colleagues (according to Google Scholar, cited over 11,400 times). Caspi and colleagues concluded that people experiencing “stressful life events” are more likely to be diagnosed with depression if they carry 5-HTTLPR, a variant genetic sequence within the SLC6A4 gene that encodes a protein that transports serotonin within neuronal cells. For many people, the Caspi study provided a sensible explanation for the causes of depression, where life events and genetic predisposition combined to explain why some people become depressed, while others do not. However, despite the publication of at least 450 scientific research papers on this genetic variant, by 2019, it became clear that the 5-HTTLPR depression theory did not hold up.

The rise and fall of the 5-HTTLPR-depression link was described in psychiatric drug researcher Derek Lowe’s aptly-titled 2019 Science article, “There Is No ‘Depression Gene.’” The depression candidate gene literature, he wrote, turned out to be “all noise, all false positives, all junk.”

In a 2019 post, Alexander nicely described years of subsequently unsubstantiated 5-HTTLPR-depression claims in the scientific literature, and how the media popularized these claims by calling “5-HTTLPR and a few similar variants ‘orchid genes,’ because orchids are sensitive to stress but will bloom beautifully under the right conditions.” (Who could say a bad word about orchids?) In 2019, Alexander described the 5-HTTLPR trainwreck as follows:

First, what bothers me isn’t just that people said 5-HTTLPR mattered and it didn’t. It’s that we built whole imaginary edifices, whole castles in the air on top of this idea of 5-HTTLPR mattering. We ‘figured out’ how 5-HTTLPR exerted its effects, what parts of the brain it was active in, what sorts of things it interacted with, how its effects were enhanced or suppressed by the effects of other imaginary depression genes. This isn’t just an explorer coming back from the Orient and claiming there are unicorns there. It’s the explorer describing the life cycle of unicorns, what unicorns eat, all the different subspecies of unicorn, which cuts of unicorn meat are tastiest, and a blow-by-blow account of a wrestling match between unicorns and Bigfoot.

Most likely, future commentators will tell a similar story about behavioral polygenic scores, GWAS, GREML, RDR, and Sib-Regression. Alexander’s post merely continues (1) the 100-year fallacy of assuming that behavioral twin (and adoption) studies are based on sound assumptions and should be interpreted genetically; (2) the 55-year fallacy of assuming that twin studies are sound, so let’s spend billions of dollars trying to find the genes; and (3) the 15-year fallacy of believing that twin studies are sound while DNA-based methods failed, so “heritability must be missing.” It’s time to abandon behavioral and psychiatric research based on twin studies after a disastrous and harmful 100-year run. Once this is accomplished, and after confounding variables and researcher genetic confirmation biases are accounted for, the seemingly contradictory puzzle pieces that perplex Alexander and others will fit together quite nicely.