Greg Mankiw posts the above graph (courtesy of Economix) and opines:
bq. …kids from higher income families get higher average SAT scores. Of course! But so what? This fact tells us nothing about the causal impact of income on test scores. (Economix does not advance a causal interpretation, but nor does it warn readers against it.)
bq. This graph is a good example of omitted variable bias, a statistical issue discussed in Chapter 2 of my favorite textbook. The key omitted variable here is parents’ IQ. Smart parents make more money and pass those good genes on to their offspring.
I’ll just leave aside problems with the concept of IQ and with the notion that genes communicate intelligence. (See Cosma Shalizi.)
But you might think Mankiw would at least nod his head in the direction of omitted variables that are correlated with income and with SAT scores, but have nothing to do with genes. Like, I dunno, the quality of primary and secondary education, which could — and let me go out on a limb here — be a bit more sucky in poor neighborhoods than rich neighborhoods. Or, maybe, and again, this is just crazy speculation, rich parents do more to prepare their children for standardized tests.
That’s what I would do if I were writing a blithe little blog post. You know, hedge my bets. If I wanted to work hard, I would type “adopted children sat scores family income” into Google, and then I would learn just how difficult it is to separate out the effects of genes and the environment, even when you look at adopted children:
bq. Chapter 4…tries to estimate the effect of specific family characteristics on young children’s test scores. This is not easy. Hundreds of different family characteristics correlate with children’s test performance. Disentangling their effects is a statistical nightmare. Almost any family characteristic can also serve as a proxy for a child’s genes. We know, for example, that a mother’s genes affect her test scores and that her test scores affect her educational attainment. Thus when we compare children whose mothers finished college to children whose mothers only finished high school, the two groups’ vocabulary scores can differ for genetic as well as environmental reasons. Even when a child is adopted, moreover, the way the adoptive parents treat the child may depend on the child’s genes. Parents read more to children who seem to enjoy it, for example, and whether children enjoy being read to may well depend partly on their genes.
bq. The best solution to such problems is to conduct experiments. In the 1970s, for example, the federal government conducted a series of “negative income tax” experiments that increased the cash income of randomly selected low-income families. These experiments did not last very long, the samples getting any given “treatment” were small, and the results were poorly reported, so it is hard to know exactly what happened. Short-term income increases did not have statistically reliable effects on low-income children’s test scores, but that does not mean their true effect was zero. As far as we know, these are the only randomized experiments that have altered families’ socioeconomic characteristics and then measured the effect on children’s test scores.
bq. In theory, we can also separate the effects of parents’ socioeconomic status from the effects of their genes by studying adopted children. But because adoption agencies try to screen out “unsuitable” parents, the range of environments in adoptive homes is usually restricted. The adoptive samples for which we have data are also small. Thus while parental SES does not predict adopted children’s IQ scores as well as it predicts natural children’s IQ scores, the data on adopted children are not likely to persuade skeptics.
And if I truly wanted to sacrifice my blood, sweat, and tears for the Glorious Enterprise of Blogging, I would Google “adopted children sat scores socioeconomic status” and — hey, look! — the first hit is a potentially relevant article in some backwater journal called the “American Economic Review.” It’s called “The Nature and Nurture of Economic Outcomes.” Let’s see…
bq. In this paper, I use data on adoptees to measure causal effects on children’s outcomes from being raised in a high-education or high-SES family. My key identifying assumption is the random assignment of adoptees to families. I find that being raised in a high-SES family (or in a high-income town) greatly increases the probability that a child will attend college and increases the selectivity of the college attended…In the NCDS data, the effect of the nurturing parent’s SES on the child’s college attendance is similar for adoptees and non-adoptees. In results reported here and in Sacerdote (2000), there is some evidence that the effect of family environment may be greater on educational attainment than for test scores. These findings support the notion that environment can be incredibly potent in determining children’s outcomes and that environment’s potency may vary with the outcome considered.
And then I might _really_ hedge my bets.
UPDATE: A friend sends along this piece on the intergenerational transmission of wealth. The authors conclude:
bq. The results are somewhat surprising: wealth, race and schooling are important to the inheritance of economic status, but IQ is not a major contributor and, as we have seen above, the genetic transmission of IQ is even less important.
UPDATE: More from Brad DeLong.