The concept of heritability is widely misunderstood. The result of this misunderstanding is a fundamental misconception as to how we came to be who we individually are, and more generally, the genesis of individual differences.
When told that a propensity for, say, substance abuse is highly heritable, most presume that the causes of an individual’s addiction were in the genes transmitted from the parents. But it doesn’t mean that at all. Nor, if heritability for addiction is low, does it mean that addicts are entirely victims of their environments. It is certainly true that the genes an addict inherits from her parents and the environments she experiences during the course of her life causally contribute to the development of her addiction. It is also true, however, that heritability analyses provide no insight as to the etiology of her addiction, genetic or otherwise.
More generally, as currently defined, heritability has little to do with genetic inheritance, how our genomes contribute to who we individually are. It was not always thus. Until the early 20th century, heritable was synonymous with hereditary (or genetic inheritance). But the term got hijacked by Galton’s heirs and repurposed to mean something quite different. Unfortunately, the hijackers covered their tracks, leaving us in the conceptual quagmire we find ourselves in today. For an excellent account of how this happened, see this article by David Moore and David Shenk (10.1002/wcs.1400). In the following section I will draw on their account.
The Bait and Switch
The altered meaning of heritability needs to be understood against the backdrop of the genetic determinism that prevailed at the time. On this view, development, the process through which a fertilized egg (zygote) became you, is a matter of actualizing the genetic instructions already present at your conception. The role of the varied environments you experienced along the way is secondary from this perspective, merely the external conditions under which these genetic instructions were realized. Fine tuning, in other words. As for the complex interactions of genetic and environmental factors, and their contingencies, forget about it.
As an example, Moore and Shenk refer to the growth of a plant from a seed. From the perspective of a genetic determinist the seed already contains the plant’s essence in the form of genetic instructions. The environment (water, soil nutrients, sunshine, etc.) only enables the unfolding of this essence from these genetic instructions. Back then, the metaphor for these instructions was the genetic blueprint, today it is the genetic program.
This was the background mindset when the meaning of “heritability” was altered by those with statistical expertise who sought a more solid foundation for Galton’s nature-nurture dichotomy. They presumed that it was relatively easy to separate the genetic contribution to phenotypes (manifest traits) from the environmental component. Their goal was to measure the genetic component of a trait by a new statistical method. Once the genetic contribution to a trait was accounted for, the remainder, what was left over, could be considered the environmental contribution.
But for the purposes of these calculations the actual developmental causes of a trait state (such as height) were tacitly ignored. When, as often occurs, it is claimed that 80% of a trait (such as height) is due to genetic influences and 20% due to the environment, what are we to make of it with respect to individual development? Very little, it turns out.
Consider this thought experiment devised by Evelyn Fox-Keller. (https://doi.org/10.1515/9780822392811) In the first case, Jack and Jill are jointly filling a barrel with water using separate hoses. Either because of differences in water pressure at the source, or differences in hose girth, Jill contributes more to the final volume than Jack. And it is possible, in principle, to determine precisely their relative contributions, say 60% for Jill and 40% for Jack. This is how the influences of genes and environments on phenotypic traits are viewed in heritability analyses.
Now consider a different case. In this scenario both Jack and Jill again both contribute to filling the barrel. Jill still holds the hose over the barrel, but Jack turns the water on at the the spigot. How, in this case do you partition their relative contributions to the final water volume. Obviously, you can’t. Their contributions are incommensurable. There is no common metric by which to measure their relative contributions.
Similarly, it makes no sense to say of a person who is six feet tall that five feet of the height is due to that person’s genome and one foot to his/her environment. There is no common metric by which to measure the contributions of genes and environment to height or any other trait. Heritability (new sense) analyses circumvent this problem by changing the subject and the object investigated. The new measure of genetic influence, and hence the new meaning of heritability, is now how much of the variation in height can be attributed to genetic variation.
The focus on height variation as it relates to genetic variation is not a causal analysis, as Richard Lewontin has so long argued (10.1093/ije/dyl062). Unfortunately, it became common in heritability analyses to conflate the meaning of measures of genetic variation as it relates to trait variation (the new sense of heritability), with measures of genetic causation (the old sense of heritability). The analysis of variation is not the analysis of causation. A failure to understand this distinction looms large in perpetuating the nature-nurture debate. Those who emphasize the influence of genes in making us who we are, are especially prone to misinterpret and over-interpret the meaning of heritability estimates.
The New Meaning of Heritability
Let’s begin with a pretty good definition offered in Wikipedia, which I will lightly paraphrase: Heritability is a statistic that estimates the amount of variation of a trait in a population that can be attributed to genetic variation among individuals in that population. I have bolded the parts of this definition than need emphasis. I’ll consider them in order.
Statistic: A statistic isn’t just a number; it’s a number arrived at through a specific method of analysis. The method most widely used in calculating heritability is called the analysis of variance, or ANOVA, which has built into it the assumption of additivity, that genetic and environmental effects can be measured independently and summed, as in the first case of Jack and Jill. Here is the most basic formulation: Phenotype (measured trait) = Genotype + Environment. Short form: P = G + E. The additivity assumption is problematic even with respect to variance analyses, it is a categorical mistake with respect to causal analyses, of the sort highlighted in the second case of Jack and Jill filling the barrel..
Variation: as the name fully discloses, ANOVA is designed to identify sources of trait variation in the study population, not the causes of those traits. As such, it should be formulated thus Pv = Gv + Ev (where “v” denotes variation). That is, what is measured in an ANOVA is the amount of phenotypic variation that can be attributed to the genetic variation or environmental variation in the study population. It has nothing to say with respect to what causes a particular person to be six feet tall.
Trait: a trait is any phenotypic attribute that you care to measure. But some traits more obviously cut nature at the joints than others. Height, for example, is a non-problematic joint-cutting trait. Intelligence, on the other hand, is not obviously the result of joint cutting. Many would argue that it is an artificial construct that cuts across the bone. So too, many behavioral and “personality traits”.
Population: A population is a group of individuals. There are varied ways to delimit a population from other populations. For the purposes of ANOVA, the population is whatever individuals were measured for the trait in question. Heritability is a population measure of deviations from the mean trait state (e.g. height) that are associated with measures of genetic deviations from the genetic mean in that population. As such, heritability has no direct implications with respect to the genetic or nongenetic developmental causes of individual differences in height or any other trait. Heritability is an attribute of populations, not persons.
Genetic Differences: Genetic (genotypic) differences among individuals are not measured directly, but as variation from the population mean. In contrast to trait variation, the genetic variation can only be inferred indirectly. The genotype itself is generally veiled in heritability estimations.
In That Population: Whatever the magnitude of the heritability measurement from a given study, it cannot be generalized. ANOVA is a local analysis; it applies only to the specific population sample investigated.
When any of these bolded terms is insufficiently digested, a variance-based heritability estimate is prone to misinterpretation. In the next post I will consider such misinterpretations in the context of twin studies, currently the gold standard in heritability analyses.