Shared environmental effects are real, general and identifiable
A good family is good for many things
People coming from the same family have similar height, weight, intelligence, personality and income. This resemblance can be due to either due to genes that run in the family or the effects of the so-called shared environment. When we say “nurture” (as opposed to “nature”) it is essentially the shared environment we mean. The shared environment is everything siblings have in common in their environment, including many things sociologists are very interested in such as family income, the number of books at home, or the parents’ education and whatever that means for the type of environment they provide for their children. Parenting strategies, the schools the kids attend and some other factors can also be part of the shared environment, but only the if they are actually shared by siblings – it’s possible for siblings to be treated differently by parents or to go to different schools, in which case these are not part of their shared environment anymore.
It is often assumed that the similarity of family members is only due to the shared environment and never due to genes. This is very common among policymakers and in non-based sociometric studies. The OECD, for example, automatically assumes that all relationship between family socioeconomic status and PISA test scores is causal and never once mentions genes or intelligence. This error is so common that it has its own name: sociologists’ fallacy.
“Smoking is more common among the children of smokers” – the sociologists’ fallacy on a cigarette box.
On the other hand, among the based, the received wisdom is often that all or almost all family resemblance is due to genetics. The certainly very based late anthropologist Henry Harpending quipped that “It doesn’t matter who raised you, as long as they didn’t hit you on the head with a hammer.” The evidence for this were many family pedigree studies which showed much stronger genetic than shared environmental effects. But this received wisdom is not true either. Blogger Inquisitive Bird recently published a great piece on the reality of shared environmental effects. His points are the following:
- Large studies show that for traits like educational attainment, income, or crime, shared environmental effects are low and lower than genetic effects but not zero.
- Zero effects can be an illusion created by low statistical power. If a study is small, behavior geneticist find that assuming zero environmental effects doesn’t make their models fit worse. They then conclude that they are indistinguishable from zero which is true in a statistical sense but it doesn’t mean that they really are zero.
- Shared environmental effects also look smaller because of statistical convention. Normally, in genetics we talk about “variance accounted for”, but the true size of the effect is better expressed by its square root, the correlation coefficients. This makes genetic and shared environmental effects more equal.
I very strongly agree with this, and I had also wanted to write a piece with a similar conclusion. But there are two more things to add: shared environmental effects are not only real, but they are also general and they can be identified. A good family is good for many things, and we know what makes a family good.
Generalist genes, specialist environments
If you have multiple traits measured in the same twins, you can not only estimate how much the environment and genes affect them. You can also say if their correlation, if there is any, is genetic or environmental. In other words: you can say if these traits are similar because they are affected by similar genes, or because they are affected by similar environments. With molecular genetic data using tools like LDSC, you don’t even need to sample all traits from the same people.
Imagine a gene which changes both how an ion channel functions in a neuron and how well myelinated axons are. The former makes you slightly more neurotic and the latter makes you slightly faster. Such a gene with multiple effects is pleiotropic, and in the example neuroticism and speed are genetically correlated: their genetic background overlaps, the reason they correlate is in the genes. A genetic correlation is evident from the fact that the correlation of two traits is greater among more related people: say, it is 0 in unrelated people, 0.2 in DZ twins but 0.4 in MZ twins.
Now, imagine that parental abuse makes you both neurotic through trauma and faster due to the need to learn how do dodge fists and flying objects at home. In this case, neuroticism and speed have a shared environmental correlation: the reason they are similar is because they are both shaped by experiences during development in the rearing family. You can see a shared environmental correlation from the fact that the two traits in question are more similar among those growing up together, but do not scale further with within-family genetic similarity. All kinds of twins and even adoptees are traumatized and accelerated by abuse, but only those who live with abusive parents.
You can also imagine that the neuroticism-speed correlation has nothing to do with your family of origin at all. In this case, the correlation is nonshared environmental. You can see this from the fact that neither genetic similarity nor a history of living together changes how strongly they are correlated. I wrote about nonshared environmental correlations here before, where I said that a nonshared environmental between homosexuality and neuroticism would be in line in line with the latter causing the former (as opposed to both being caused by some genetic or shared environmental background effect).
Normally, when you look at trait pairs, you see strong genetic and weak nonshared environmental correlations. A famous paper by Kovas & Plomin interpreted this evidence for “generalist genes” and “specialist environments” Genes are usually pleiotropic and their effects ripple through the phenotype with the same genes affecting many traits. As it happens, pleiotropy is usually unidirectional considering human preferences and overlapping sets of genetic variants affect health, intelligence, income and antisocial behavior. There is not much in terms of tradeoffs between genes for various socially desirable outcomes, having intelligence-increasing genes normally also means better health, income-increasing genes reduce antisocial behavior and so forth. (The only exception is with fertility: modern society basically selects for dumb, poor, impulsive and sickly people.)
On the other hand, nonshared environmental correlations are weak or zero. This is a big deal because it shows that the strongest environmental effects on our traits are not big life-changing experiences which affect who your entire character, but rather small, random, nonsystematic, trait-specific effects which may or may not be just measurement error. If other than genes it was your experiences of trauma, personal growth or conscious learning that affected you the most, we would expect these to influence multiple characteristics (cognition, personality, mental and physical health etc). This would make it useful to look for intervention targets. We can’t change your genes or what kind of family you grew up in, but we could theoretically replicate the environmental experiences healthy and successful people have and give them to others as some kind of therapy or course to fix their problems.
The lack of general nonshared environmental effects made this look less feasible, resulting in a “gloomy prospect” of any non-genetic influence on being random, trait-specific and noise-like which is impossible to control and manipulate.
Shared environments are also generalists
Kovas & Plomin didn’t have much to say about shared environments. I found no studies searching for “generalist” or “specialist shared environments” either. I think shared environmental correlations have been under-studied because shared environmental effects are close to zero in most twin studies and due to low power they can be constrained to zero without much penalty to model fit,.
Low shared environmental correlations between traits would suggest that shared environmental effects can be safely ignored. Not only are they weak, they only affect single traits so the rearing family doesn’t even have a small general effect on the health or success of their children. On the other hand, high shared environmental correlations would suggest that the small influence families have on children is indeed general. While most of the reason success and health runs in families is due to genes, the small causal influence families have on their children affects them in a generally positive or negative way, much like how sociologists usually imagine it.
I plotted the results from three multivariate twin studies below. The first is about IQ subtest performance in the Vietnam Era Twin Study, the second about GCSE grades and the psychosocial characteristics of high-schoolers in the British TEDS, and the third only about GCSE grades. Two of the studies and the visualization idea is from here. All of these studied multiple metrics of social success, such as cognitive performance, school grades, and behavior. I gathered the correlations (phenotypic, genetic, shared and nonshared environmental) between all trait pairs and plotted their distributions.
There is also a very recent twin study, looking at intelligence, educational attainment and income which produced a similar chart:
As you can see, genetic correlations tend to be higher and nonshared environmental correlations tend to be lower than phenotypic correlations. In other words: genes make our traits more similar and experiences outside the family less similar. Genes put us on firm rails towards a certain degree of general social success, while environmental effects outside of the rearing family – guessing the IQ test right, winning the lottery, staying in school to benefit from a nice scholarship you just got – tend to have trait-specific effects.
Shared environmental effects are more similar to genetic than nonshared environmental effects. Shared environmental correlations, if anything, tend to be higher than not just phenotypic, but also than genetic correlations. Shared environmental effects, while small in magnitude, are also generalists: they affect many phenotypes. Social success not only depends somewhat on your rearing family (although much less so than on genes) but the realness of the effects of the rearing family is also shown by their generality. Good families can positively affect many indicators of social success.
What does a good family look like?
Shared environmental effects in a twin study are what we call an omnibus term. This means that they show the effect every single family characteristic shared by twins. Just like a GWAS can identify the specific genetic variants which cause a trait to be heritable, it is possible and interesting to identify the specific family characteristics that cause the effects of the shared environment.
The best paper I know that does this is Engelhardt et al 2019. This paper uses data from 1728 child and adolescent twins in an American sample that has data on lots of family characteristics. It focuses on shared environmental influences on cognition (full-scale IQ, verbal comprehension and perceptual reasoning) and academic outcomes (reading and math school scores).
The idea behind the paper is very smart: they first calculate the effect on the shared environment, and then see how much this drops if they first control cognitive or academic test scores for a family characteristic. Basically, they are asking: “How much would the effect of the shared environment drop if all kids had parents with the same SES, live in the same neighborhood, or identical teachers?” Shared environmental effects, like everything else in behavior genetics, are about differences: if there is no difference in a component of the shared environment, it cannot contribute to the omnibus effect of the shared environment. How much the effect of the shared environment is reduced after controlling for one of its elements can show how much that particular element matters. This is what the main results look like in this paper:
There are no big surprises here. Most of the effect of the shared environment comes from “Home composites”, which means a score calculated from things that a sociologist would be interested in: parental education, income, parental conflict, and changes in the twins’ lives such as moving or divorce. Next came either school-level differences or race – in a twin study, at least in the US, race is not a genetic but a shared environmental variable because both monozygotic and dizygotic twins have the same race. What kind of neighborhood the family lived in didn’t matter much. For IQ, the ~20% shared environmental effect was fully explained by these few factors the researchers measured. For other cognitive and academic outcomes, there was “missing shared environmentality” amounting to up to half the shared environmental variance in math scores. This is not that great but much better than how well we can enumerate the genetic variants which cause heritability. We understand shared environmental effects much better than genetic effects: they are only a bit more complicated than the “basic law of sociology” which Eric Turkheimer jokingly described as “poverty is bad for you”.
The shared environment has been on the backfoot for a long time among serious scholars of human differences because of the many blank slate fallacies sociologists committed and the politically motivated attacks on science motivated by these. My own impression for a long time was that traditional sociology was more or less completely discredited and genes explain all or very close to all differences between families. Genetic effects are indeed more important, but shared environmental effects are also real, general and relatively well understood, so they should also be on our horizon.