2000 words

Introduction

For the better part of 100 years, twin studies have been used to prove that there is a genetic influence to all traits. The claim is that “all human traits are heritable”, which is the “first law of behavioral genetics” as stated by Turkheimer. Using twin, family, and adoption studies, it is claimed that some traits are “more genetic than others”, especially IQ. But a meta-analysis 10 years ago found that heritability was 49 percent for all traits (Polderman et al, 2014; see here for critique). However due to the falsity of the EEA—which holds that environments are similar for both MZ and DZ twins—the so-called heritability is shared environment (h2=c2). (See Joseph, 2014 for arguments against twin interpretations and Joseph 2022, 2023 for arguments against Thomas Bouchard’s twin studies. See Bouchard 2023 for response to Joseph 2022, and Joseph 2023 is Joseph’s response.)

But imagine we could bypass such devestating issues for twin studies that invalidates them for their stated aims. What would we find? Here, I will create a thought experiment in which 100 MZs and 100 DZs were placed in a metabolic ward immediately after birth. In this scenario, every single aspect of the environment is meticulously controlled—as is the case in animal breeding studies—which would then eliminate shared environmental influences and therefore would give us a “real look” into the (obviously context-dependent) heritability of traits.

However unethical this study is, though, even if it were possible there are a slew of conceptual and empirical issues that would still invalidate the estimates. The issues include the limitations of assuming additive genetic effects, the presence of GxE, the non-representative nature of twins, and the missing heritability problem. The argument I will mount here will show that even with perfectly controllable environments, we still wouldn’t be able to “estimate heritability” and furthermore due to the interaction problem that this study still wouldn’t overcome, that would further make any such results meaningless.

The unethical twin metabolic ward study

A metabolic ward is a controlled environment which is usually found in a hospital, research facility or clinic in which the participant’s can be closely monitored and their metabolic processes studied under tightly regulated conditions. Researchers can control diet, diet quality, sleep, physical activity, and environmental influences to investigate nutrient absorption, energy expenditure, hormone regulation and metabolic disorders. Controlled settings like this allow researchers to get precise measurements, ensure that a participant is following a diet correctly as to study their metabolic processes on that diet, and understand the physiological mechanisms better. (See Hodges and Bean, 1960.)

Now imagine all of that but imagine that 100 MZ twins and 100 DZ twins were—after birth—immediately whisked away to a metabolic ward and from that moment on, they were in a tightly regulated environment as to attempt to minimize or completely eliminate shared environmental influences. What would happen?

We could learn about the genetic influence of genetics on metabolic processes, like the influence of genes on nutrient absorption, energy expenditure, hormone regulation and metabolic disorders. So by comparing MZ and DZ twins on the similarity and differences in these traits, we could then ascertain the genetic influence that underlies the variability between them. This would also allow researchers to completely control sleep, eating, physical activity levels and other environmental exposures which would then allow researchers to tease a part the relative influenced of genes and environment. This would also allow us to see how GxE affects a trait. We could learn about how genetic predispositions interact with dietary factors or how lifestyle choices modulate metabolic health outcomes and disease risk. It could also give us insights into individual variability so we can give individualized metabolic approaches.

However, even if IRB can be passed, there are of course potential psychological and social implications for those studied. The issue is, we have minds and we therefore react differently to things independently of our genetic makeup. So although we could potentially learn some interesting things about human metabolic processes, when it comes to psychological ones, due to the unrepresentative environment, these will be less valid than metabolic processes.

We technically do “measure” heritability of traits now, but they’re heavily confounded, and even though—intuitively—it may seem like such an experiment would be valid and show “true heritability” (whatever that means), we know that h2 is context-dependent and not useful for individuals. Such a study would be about perfectly controlling every aspect of the environment as best we could to decrease whatever environmental effect would bias the h2 upwards. But I think a slew of conceptual issues would still even then invalidate such interpretations.

Cattle and humans are quite obviously different in their propensities, and even then, if we could pass IRB ethics and lock 100 MZs and 100 DZs in a metabolic ward and perfectly control every aspect of the environment, the issue there would be our experiences in society and our interactions with others are extremely meaningful to how we develop as humans. So if that kind of experiment were possible, then the h2s would be way lower than twin, family and adoption studies give us (we see this in animal breeding studies—way lower h2 in those traits than for IQ, see Schonemann, 1997).

There are also further conceptual issues like the falsity of the addivitiy assumption, GxE, GxG (Zuk et al, 2011), psychological distress, and the fact that we become human and gain our minds through our interactions with people in broader society would also further confound such interpretations. Furthermore, since the EEA is false, h2 equals c2, meaning the heritability is shared environment. That’s the best interpretation we have of twin studies. And the missing heritability issue that plagues GWAS and other molecular approaches to trait heritability further throws a wrench in this, since it’s completely possible that GxE (and other interactions) contribute to trait variance, and this—along with epigenetic and non-genetic factors—aren’t accounted for by h2 estimates (since they assume additive independent genetic effects). So even if such an experiment were possible and the conceptual issues I raised weren’t an issue, that alone would invalidate the study.

One big issue is the fact that twins aren’t representative of the population at large. For example we know that twins have a higher chance of having low birth weights, are more likely to be born premature than non-twins (Isakkson, Ruchkin, and Ljungstrom, 2023). Low birth weight is also predictive of health outcomes in adulthood (Hassan et al, 2021). Twins are also more likely to experience complications during pregnancy and childbirth like preterm birth, intrauterine birth restriction and childbirth trauma. The intrauterine birth restrictions could also influence their metabolic development and responses within the ward, which This then would influence their outcomes in adulthood compared to non-twins. We also know that there are a slew of environmental effects that increase the chance for DZ twinning between groups (MZ twinning is relatively the same between them). Moreover, due to the complications that twins face before, during, and after childbirth, this could then affect their metabolic health and responses within the metabolic ward. Therefore, while this thought experiment proposes ways in which researchers can use a metabolic ward to study the relative influence of genes and environment on traits (assuming that there are no other conceptual issues with such an endeavor), researchers would still need to take into account the ways in which twins differ from the general population which would then influence the results that would be observed in the metabolic ward study.

Joseph (2009) proposed an interesting thought experiment:

Finally, we could conduct a thought experiment on political behavior and social attitude correlations among reared-apart MZ twin pairs who, while genetically identical, grow up in truly uncorrelated environments in different eras. Suppose one male MZA twin is placed at birth in an aristocratic Japanese family in 1802. The other male MZA twin is placed at birth in a poor peasant family living in the highlands of El Salvador in 1965. Unlike previous TRA studies, inwhich the investigators calculate correlations among partially reared-apart twins sharing many cultural influences, in our thought experiment we eliminate cultural influences such as family (which most MZAs share to some extent), mutual association and influence (which most MZAs also share to some extent), nation, region, ethnic, religion, economic class, and birth cohort (of course, they would still be the same sex).75 I conclude this thought experiment by posing the following question: Would we expect a study of genetically identical pairs of this type to find sizable correlations for political behavior and social attitudes?

The correct conclusion to draw here is that although the twins share genetic makeup (to whatever degree they do), but they are raised in different environments and different time periods, the divergence in their experiences shows that heritability can’t fully capture the role of environment in capturing in shaping not only these traits but all traits studied by twin researchers.

Conclusion

In the end, if such a study were logistically possible and ethical, it wouldn’t show h2 anywhere near twin, family and adoption studies. We’ve basically reached an upper bound on molecular h2 estimates. And of course twins aren’t representative of the general population at large (since they have lower birth weights/injuries and a higher mortality rate). And even then, twins in these studies aren’t even representative of all twins. This is, yet again, a conceptual failure of hereditarianism—the assumption that twin studies will tell us anything about “the genetic architecture” of traits merely due to genetic similarity and being so-called “reared apart.”

Even if such a study were possible, there would be a lack of generalizability due to the fact that the observations occurred during rightly controlled conditions. Such a study is fraught with conceptual, empirical and ethical issues. Biases in sample selection, the health differences between twins and non-twins, prenatal and peri-natal factors, psychological and social dynamics and the lack of generalizability highlight why twins aren’t a representative of the population at large. Moreover, the complexity of metabolic traits and the assumption of genetic independence further complicate interpretations in the ward setting—especially since DNA interacts with all other developmental resources (Noble, 2012).

But even in a possible world where such limitations can be addressed and made meaningless, there are still inherent limitations of heritability estimates as a measure of the genetic determination of traits (in the popular culture, see Moore et al, 2016 for a critique if the heritability concept). Heritability estimates are context-dependent and subject to numerous confounding factors.

But disregarding this thought experiment and looking to the real world, we know that there is no way to estimate interactions in population studies. The so-called solution is to assume that there are NO gene-environment and gene-gene interactions, and this therefore biases such estimates upwards (Ho, 2013). This shows that the assumption of additive effects and no GxG or GxE interactions would still pose a problem—an insurmountable one. Thus, even estimates from this study would still be subject to upward bias due to the nature of the assumptions of heritability. This shows that even in a perfectly controlled environment that measuring h2 is impossible due to the disregarded biological facts inherent in the heritability formula. Joseph and Richardson (2024) also show that Herrnstein and Murray (1994) didn’t give any valid evidence that genes influence IQ scores within groups.

So the conclusion here is very simple—since heritability studies can’t tell us anything about the relative contributions of genes and environments due to the inherent nature of the heritability formula which disregards interactions, and due to the whole slew of other conceptual and empirical issues raised here, even in a perfectly controlled environment we still wouldn’t be able to get “true h2” estimates of any kind of trait in the metabolic ward. So the goal of behavioral genetics is an impossible one.