Harris’ Major Malfunction

If upbringing matters little, peers matter even less.

Judith Harris’ The Nurture Assumption was a huge influence on me, and the top inspiration for my Selfish Reasons to Have More Kids. Her book’s first main lesson is that family resemblance, defined in the broadest possible way to include physical, psychological, and social outcomes, is mostly driven by genetics rather than upbringing. Her book’s second main lesson, though, is that children’s peers explain a large share of remaining variation — defined, again, in the broadest possible way to include physical, psychological, and social outcomes.

Almost everyone takes Harris’ two lessons as a package deal. Her fans typically believe that genetics and peers matter a lot, while critics stick with the conventional view that parenting is what really counts. My view, in contrast, is that the very evidence that Harris uses to establish her first lesson undermines her second lesson.

Why? Let’s back up. Behavioral geneticists partition human variation into three categories. The first is heredity. The second is family environment. The third is a pure residual, usually just called “non-shared environment” or “non-family environment.” Mathematically, these shares must sum to 1. If heredity’s share of the variance is .7 and family environment’s share is .1, then non-shared environment has to equal .2.

If you’re trained in philosophy, you may be tempted to pose counter-examples. Suppose that (a) beauty is 100% hereditary, and (b) the sole cause of human happiness is being perceived by the people in your environment as beautiful. So is variation in happiness 100% hereditary, 100% environmental, or what? Behavioral geneticists answer in this scenario, by construction, is: 100% hereditary. Why? Because they’re analyzing observed variation, not variation in all conceivable worlds. If heredity fully explains trait A, which interacts with a complex environment to produce trait B, the math says that heredity fully explains trait B as well. Harris is fully aware of this point:

Direct genetic effects have consequences of their own, which I call indirect genetic effects — the effects of the effects of the genes. A child's timidity causes his mother to reassure him, his sister to make fun of him, and his peers to pick on him. A child's beauty causes her parents to dote on her and wins her a wide circle of admiring friends. These are indirect genetic effects. Identical twins lead similar lives because of indirect genetic effects. (The Nurture Assumption)

By construction, behavioral geneticists’ categories partition logical space. Variation explained by hereditary is “Everything caused either directly or indirectly by genes.” There are multiple ways to calculate this statistic for a trait, but the simplest is just the phenotypic correlation between identical twins raised apart. When researchers do these calculations, they usually come up with numbers between .4 and .8, depending on the trait.

Variation explained by shared environment, similarly, is “Everything caused either directly or indirectly by family environment.” There are multiple ways to calculate this statistic for a trait, but the simplest is just the phenotypic correlation between biologically unrelated adoptees raised together. When researchers do these calculations, they usually come up with numbers between 0 and .2, depending, again, on the trait.*

Taken together, these estimates are the foundation for Harris’ first main lesson. Nature crushes nurture, at least in the First World in the long-run.

Her evidence for her second main lesson, however, is much more diffuse. Behavioral geneticists show that non-shared environment usually ranges from .2 to .4, so we know there is much more to observed variation than either nature or nurture. If nurture is environment inside the home, non-shared environment would be stuff outside the home. Common sense suggests that children’s peer group is an important chunk of this extra-familial environment, and Harris finds a lot of supportive interdisciplinary research.

While this sounds ever-so-plausible, there’s one dire problem: “The effects of the effects of family environment” count as family environment — and family environment plainly has a massive effect on your peer group.

Take Sacerdote’s famous Korean adoption study. Some orphans got randomly assigned to rich families, while others were randomly assigned to families only modestly above the poverty line. The first group mostly grew up in rich neighborhoods with good schools, low crime, and rich peers. The latter group mostly grew up in much poorer neighborhoods with worse schools, higher crime, and poorer peers.

If peers had a large effect, the measured effect of shared family environment would be large. Even if the direct effect of a rich family were zero. Since estimates for shared family environment are in fact low, we must conclude that peers don’t have a large effect. QED.**

As usual, to repeat, we must remember that if we drastically broadened the sample, this result could change. Within the observed range, adoptive American families’ incomes had little effect on Korean orphans’ adult outcomes. But growing up in America rather than Korea obviously had a massive effect on the adoptees’ fluency in English versus Korean.

Harris’ most compelling examples of powerful peer effects fit the same mold. They’re not wrong, just outside the observed range she trying to explain.

I know a Jewish woman whose Orthodox grandparents immigrated to the United States from Poland and then took their children back to Poland when they saw them turning into godless Americans. The grandparents and all but one of their children perished in the Holocaust.

It is possible for Orthodox parents to rear children in the United States without having them turn into godless Americans. In Brooklyn, New York, there are Hasidic Jews who have preserved the religion, customs, and even styles of dress and adornment that came from Eastern Europe several generations ago. The way they do it is to educate the children themselves. The children go to religious schools called yeshivas; they do not mingle with children from other cultures either in school (where all the children are offspring of Hasidic Jews) or in the neighborhood (most of their neighbors are also Hasidic Jews). (The Nurture Assumption)

If Harris is wrong to think that peers have large effects, what does explain the big residual of behavioral genetics? Why are identical twins raised together still so different in so many ways? Harris’ follow-up book, No Two Alike, tried and failed to resolve this puzzle. The real story, I suspect, is that many small factors add up — and we should never discount the awesome metaphysical power of free will.

Whatever the answer may be, though, Harris didn’t just fail to show that peers are the crucial factor. Indeed, the very evidence she used to show that nurture is weak shows that peer effects are weaker still.

* Crucial caveat: All of this research uses First World data. Using global data, heredity would almost certainly matter much less because global environments vary so much more.

** As Harris knew, genes can also easily affect peers. A math genius is likely to make friends with fellow math geniuses. My point is not that all peer effects count as family environment, but that family environment definitely has a large effect on peer effects.