Wednesday, July 28, 2010

Statistical discrimination

In response to posts on credit scores in hiring by Yglesias and myself, Matt Zeitlin points out that discrimination can be perfectly rational (but nevertheless deserving of censure):
As we all know, labor market discrimination by race is both immoral and illegal, but it happens, and it doesn’t just happen because people are evil, but because there is an empirical element. Tim Harford had a great piece from a while back in Forbes where he talks about taste-based and statistical discrimination. (note: the piece is here.) The former is when employers just don’t like ethnic minorities and are straight-forward bigots and the latter is discrimination that happens because certain racial and ethnic identities are used as markers for other traits that matter to employers. Both, of course, are morally wrong and illegal, but the latter is hard to root out because, absent regulatory or legal pressure, employers will benefit from statistical discrimination, or at least they won’t suffer from it. From the perspective of the qualified black guy who can’t a job, there’s no difference between the two types of discrimination: he’s still being judged based on his group identity as opposed to his individual qualities.
For reasons stated in follow-up posts, I'm skeptical about the particular case of credit scores, but as a general matter this is spot on. I'm particularly glad that Matt brings up statistical discrimination, because it's a subtle but deeply important feature of labor markets. Not all racial discrimination is the result of uninformed prejudice; some of it, sadly, is rational for a profit-maximizing business, even though it remains destructive and morally distasteful.

The effects of statistical discrimination go beyond the obvious unfairness of judging an individual by his group identity. As economists have realized, there's the possibility that an entire racial group will be stuck in a bad equilibrium. If an employer believes that very few people of a particular race have invested in their job skills, he's likely to discount a good interview or test result from a member of that race as a fluke. Then, realizing that they're unlikely to get a good job anyway, people in that race will not invest in skills, confirming the employer's belief. It's a simple but lethal self-fulfilling prophecy.

Some observers, like Bryan Caplan, argue just the opposite—that statistical discrimination may be self-reversing because it raises the return to education. Caplan points out that the return to education is empirically higher for black students than white ones; he surmises that statistical discrimination makes "counter-stereotypical" behavior particularly valuable. To an extent, I think he's right. The "bad equilibrium" story doesn't seem to lower the returns to obtaining easily observable credentials like college degrees. Perhaps this is because college degrees are such good indicators of competence that they overwhelm negative racial perceptions, or because in a world that erects massive social and economic barriers to obtaining a good education, managing to get a degree as a black student is much more impressive than it is for most white students.

Yet I think that Caplan's analysis is fundamentally incomplete, because it's too quick to equate skills with formal educational credentials. Suppose you're a student for whom college is out of the question (too much of a financial burden) but completing high school is likely. Even conditional on obtaining a high school diploma, there is plenty of variation in the effort levels you can choose. To the extent that it depresses job prospects for your entire racial group, statistical discrimination may also decrease the returns to verbal and quantitative skills that aren't captured in a simple diploma, and hence your incentive to acquire them.

And, of course, "skills" aren't just taught in school. There's a very important set of vocational, interpersonal, and general "noncognitive" skills that are equally decisive in determining job performance. Since these skills are particularly hard to measure in an objective way, they're the most susceptible to racial stereotyping and statistical discrimination.

In fact, this is where I suspect most of the problem lies. Certain categories of people are thought to lack the intangible skills necessary for a work environment. Of course, there is no innate reason why this should be true, but the perception damages incentives (and changes culture) enough to make itself true. And because there is no diploma for "having the right stuff," no formal accreditation can unravel the bad equilibrium.

It's a tough problem, and it's worthy of policymakers' attention.

No comments: