Elite College Fems Earn Less

David Leonhardt:

A decade ago, two economists — Stacy Dale and Alan Krueger — published a research paper arguing that elite colleges did not seem to give most graduates an earnings boost. … Ms. Dale and Mr. Krueger have just finished a new version of the study — with vastly more and better data … and the new version comes to the same conclusion. (more; HT Tyler)

Ezra Klein quotes David approvingly. But as I reported two years ago, many, like David and the original paper’s abstract, quite misleadingly ignored its (statistically significant) finding that:

Men who attend the most competitive colleges [according to Barron’s 1982 ratings] earn 23 percent more than men who attend very competitive colleges, other variables in the equation being equal. …

Ack!  I was almost conned by elite journal editors and media reporters into believing a comforting lie!  What saved me was becoming puzzled by actually reading the original paper.

The new study’s abstract is also seriously misleading, suggesting that the study finds no effect of college average SAT scores on graduate earnings:

When we adjust for unobserved student ability by controlling for the average SAT score of the colleges that students applied to, our estimates of the return to college selectivity fall substantially and are generally indistinguishable from zero. There were notable exceptions for certain subgroups, [namley] for black and Hispanic students and for students who come from less-educated families.

To find the truth, you have to study Table 4 carefully, and note footnote 13:

For both men and women, the coefficient was zero (and sometimes even negative) [in] the self-revelation model.13 …

[footnote:] 13 This lower return to college selectivity for women is consistent with other literature. Results from Hoekstra (2009), Black and Smith (2004) and Long (2008) all suggest that the effect of college selectivity on earnings is lower for women than for men.

Table 4 shows that attending a college with higher SAT scores clearly lowered the wages of women 17-26 years after starting college (in 1976) — a school with a 100-point higher average SAT score reduced earnings by about 6-7%!  The two estimates there are significant at ~0.01% level! (The other three, for other periods after starting college, are significant at the 5% level.)

One obvious explanation is that women at more elite colleges married richer classmate men, and so felt less need to earn money themselves. Why don’t the study’s authors want us to hear about that?

The new study conflicts with the earlier one in finding no significant effects of college higher tuition or Barron’s selectivity rating on later earnings. The authors attribute those differences mostly to the earlier study using reported earnings, while this new study used Social Security Administration data. So do elite college folks do earn more, but hide it better from the government, or do they just lie more about their income? The relevant section from the new paper on this conflict:

[Regarding] the Barron’s Index and the log of net tuition … estimates …are statistically insignificant at the 0.10 level. These results are partly a contrast to Dale and Krueger (2002), in that the earlier analysis of self-reported earnings data showed a statistically significant relationship. … When we estimated the same regression for the same sample, but used SSA’s administrative earnings data … over the full study period (1983 to 2007) the coefficient on net tuition was generally between 0 and .02 (and never greater than .033) in the self-revelation model based on earnings drawn from SSA administrative data as the outcome measure.

Added 8a: I started college in ’77, so this cohort is pretty much my cohort.

The ’98 D&K working paper nearly as large a sample, but didn’t see any sort of negative effect for women. Suggests something isn’t reliable here.

Thorfinn has a nice long post, where he points out that 1) this new data ignores capital gains income, 2) the point estimates for tuition and selectivity imply high rates of return, 3) those estimates would be be more significant if the data were pooled across future earning years, and 4):

The results do subset among full-time earners, so it’s unlikely that this [fem earning less] result is being generated by women withdrawing from the workplace altogether.

Added 9a 24Feb: The authors respond below:

The return to college selectivity was not significantly greater than zero for men or for women. In a handful of specifications, the return was less than zero for women, but in the vast majority of specifications that we examined it was not statistically different from zero, so any explanations about why the payoff may be negative for women is pure speculation, and probably unwarranted. The most likely reason for the occasional negative payoff for women is that it was due to random sampling error.

The only estimates presented in your new paper are in Table 4, which considers five time periods, and these are the five t-stats – the second row is using robust errors.

-3.08 -2.27 -3.28 -3.45 -1.67 -1.61 -1.79 -3.69 -4.31 -1.94

You really attribute a t-stat of 4.3 to sampling error?!  Even the lowest entry here, -1.61, is noteworthy.  What are the “vast majority” specifications where t-stats are much lower?

Added 1p 24Feb: The authors respond again saying that in models tried but not mentioned in the paper,

The estimated return to school characteristics was generally not statistically significant for women.

Thorfinn notes:

I was surprised by their comment, “The paper is not about gender differences from college selectivity, and we have little reason to suspect that there are such differences.” Well, all three drafts of this paper that are online emphasize the results for attending College on various subgroups — for instance, by race, parental education, and parental income. Surely gender is an equally interesting subgroup.