The Answer of Signaling

Wow, I’m impressed by the quantity and quality of responses to my “A Question of Signaling.” I should do more posts like this.

The question, to recap, was:

If the observed education premium is 100% due to signaling, controlling for intelligence will NOT reduce estimates of the effect of education on earnings.

It’s True/False, but you get only 1 point for saying “True,” and 0-3 points for the quality of your response.

To get those 3 points, you needed to write a less verbose version of:

TRUE. If the education premium is 100% signaling, then if a non-smart person gets a degree, they will still get the full labor market bonus for possessing the degree. Similarly, if a smart person fails to get a degree, they will get none of the labor market bonus for possessing a degree. In a simple model where education is the sole way to signal intelligence, the 100% signaling assumption implies that intelligence has zero independent reward; you only get credit for your IQ if you “launder” it through the education system.

In the comments, some objected that employers will eventually figure out your IQ anyway, at least to some degree. Reasonable in the real world, but that contradicts the “education premium is 100% due to signaling” assumption. After all, much of the education premium comes from data for experienced workers.

Part of the point of the question is to see if students grasp the difference between signaling and ability bias. Signaling is a theory of how educational attainment genuinely causes higher earnings. Ability bias, in contrast, says that at least some of the apparent effect of education on earnings is not genuinely causal: While the well-educated are smarter, the labor market would have, at least to some extent, independently rewarded their smarts even if they lacked their credentials. In a “the education premium is 100% intelligence-based ability bias” story, smart yet uncredentialed workers would earn just as much as smart credentialed workers, implying no causal effect of education on earnings at all.

P.S. If education and intelligence were perfectly correlated, collinearity would prevent you from even estimating a standard education premium regression like:

ln(Labor Earnings) = a + b*Educ + c*IQ

In that case, controlling for intelligence still wouldn’t reduce estimates of the education premium. Rather, the software would say “estimation is impossible.”