Maybe Parents Matter After All

Just before New Year’s, we lost the great Judith Rich Harris, author of The Nurture Assumption, the 1998 book arguing that parenting doesn’t affect kids’ outcomes very much. Parents are similar to their kids, and siblings are similar to each other, she argued, predominantly because of genes, not because parents shape how their kids turn out. We can see this most strikingly in the twin and adoption studies of behavioral genetics, which show striking similarities between identical twins even when they’re reared apart, but modest similarities between adoptive siblings sharing the same household.

A new paper, while not mentioning her by name, argues strenuously that Harris got it wrong, if not when it comes to kids’ personalities in general, then at least when it comes to their “social capital”—measured by whether they passed a matriculation exam indicating that they’re college-ready. It demonstrates convincingly that, at least in the cohort of Israeli students it analyzes (born in 1974 to 1991), kids do become more like the parent they spend more time with. It also reconciles some different findings from other research, including those twin studies. For reasons we’ll get to in a bit, it would be a stretch to call this a wholesale refutation of The Nurture Assumption, but it’s a nice reassurance to parents that they do matter. Well, except for those of us who don’t want our kids to turn out like we did.

Familial resemblance is a hard thing to study, simply because most people get both their genes and a lot of their environment from their parents. Parents who invest a lot of time and effort in their children might be genetically different from those who don’t, and even within families, parents might spend the most time with kids who share their interests for genetic reasons. Morbidly, but ingenuously, the paper gets around such problems by looking at cases where a parent died and thus was abruptly prevented from interacting with the child at all.

Looking at children who did not lose a parent, the authors find the unsurprising result that children’s chances of passing the exam (which only 57% do overall) is correlated with their parents’ education levels and incomes. For instance, for each year of schooling the father has, the child’s probability of passing goes up 1.7 percentage points, and for each year of schooling the mother has, it goes up another 1.8 points.¹ By itself, of course, this says nothing about whether it is genes or the environment that transmits the parents’ abilities to the kids.

But then they look at cases where a parent died. When it’s the father who passed away, each year of the father’s education increases the chance of passage only 0.9 points, while the mother’s education grows in importance, to a 2.2-point increase in passage for each year. Crucially, the opposite happens when it’s the mother who dies: Each year of her schooling adds just 0.9 points, vs. 1.9 for the dad. This is pretty clear evidence that parental influence does matter; if genes were the only reason kids resembled their parents, dead parents would have the same impact as living ones, and we wouldn’t see influence patterns flip genders depending on which parent passed away.

Things get complicated from there, as the authors provide a bunch of complicated results littered with statistical “interactions,” not to mention assorted ways of restricting the data to rule out alternative explanations. (There are separate results for cases where Dad didn’t remarry, where Mom is less educated than Dad or vice versa, where the kid has two or fewer siblings . . .) But some of the main lessons are that the effect gets stronger the younger the child was when the parent died, that girls are more affected than boys, and that the effect is stronger in families where the parents have lower levels of education.

This last part is important when it comes to reconciling the results with those of twin studies. Twin studies tend to involve better-educated parents, so they might be missing effects at the lower end of the distribution. Studies that look at changes to compulsory-schooling laws, by contrast, tend to focus on this lower end and find stronger effects: Forcing someone to stay in school a little longer seems to have good effects not just on them, but also on their kids in the future, another sign that familial resemblance isn’t purely genetic.

Then, they look at families where the parents divorced. Since moms usually get the kids, the effect of moms’ traits should be relatively stronger here. Indeed, that is what they find.

They then return to families in which both parents survived, comparing households with different numbers of kids, which purportedly disentangles whether parents influence their kids by spending time with them, or just by earning a certain amount of money and providing an environment of that quality. Here’s their reasoning: If money is what matters, “the effects of parental education should be declining in family size because financial investments must be spread across more children,” but if time investment is what counts, “mother’s education becomes more important relative to father’s education"—because moms tend to spend more time with kids as families get bigger. (I found something similar in U.S. data: The more kids, the more likely a mother is to stay home entirely.)

This is more tenuous. Differences in family size don’t happen randomly, and while mothers may spend more time with kids in larger families, they also have to divide their attention up more, the same way that income is stretched thinner. I found the regression-result tea-leaf-reading far less compelling on this one, simply because the theory they set out to test is fuzzier.

More broadly, if the point here is simply that parenting matters, it’s well-taken. Yet this study works less well as a measure of how much parenting matters, especially for purposes of comparing the results to those of twin studies and the like.

Against the baseline of a 57% passage rate for the exam, parenting-based changes in the ballpark of 1 or 2 percentage points per parental year of education are considerable—especially if comparing, say, a couple of high-school grads with a couple of college grads, in which case there’s an eight-year education gap—but far from overwhelming. And these regressions include a variety of controls, including the socioeconomic status of the locality, which school the child attended, the number of siblings the child has, and how old his parents were when he was born (though these effect sizes are not reported). These things are, in part, the result of parental decision-making, which like everything about humans is partly genetic. So, the changing connections between parents’ education and income and their kids’ test-passing abilities are measured only after some genetic influence has been removed from the equation.

Twin and adoption studies, reductive as they are, conveniently end with a finding of “X percent genetic,” “Y percent shared environment,” and “Z percent who knows.” Few if any of this study’s results can be interpreted in an analogous fashion.² Indeed, since behavioral genetics doesn’t categorically deny parental influences on something like test-taking ability or college readiness—the effect of the “shared environment” is perhaps 16% for IQ in adulthood and 36% for educational attainment, according to one recent review—it’s not even clear exactly how great of a difference there is.

As researchers try new methods and genetic science advances, we’re learning a lot about why kids turn out the way they do. This paper serves as another piece of the puzzle, scrounged up improbably from deep within the couch cushions.

---

1. Yes, you read that right: They’re using linear probability models!

2. A possible exception is a result suggesting that “the effect of a mother’s education is essentially zero if she dies right after the child is born,” though, which suggests a very big difference indeed.