Consider a small law firm interviewing two job candidates, one male and one female. Both candidates are recent Harvard Law School graduates and are eminently qualified for the position. If the "best" candidate for the job is the one who will earn the most money for the firm, which seems a reasonable assumption, then I will argue that the rational choice is to hire the man. The interviewer has no specific information on the family plans of the candidates at hand (and is forbidden by law from asking about them), but can make a reasonable inference based on what everyone knows about America at the beginning of the twenty-first century: Women still bear the bulk of child-rearing responsibilities. Demographics suggest that both candidates are likely to start families in the near future. Yet only the female candidate will take paid maternity leave. More important, she may not return to work after having the child, which leaves the firm with the cost of finding, hiring, and training another lawyer.

Is any of this certain? No. The male candidate may have dreams of staying home with his five children; the female candidate may have decided years ago that she has no interest in having children. But these are not the most likely scenarios. The female candidate is punished because the firm has no information on her specific circumstances but good data on broad social trends. Is this fair? No. (And it's not legal either.) Yet the firm's logic makes sense. In other words, it is rational to discriminate in this case, which turns the whole idea of discrimination on its head. . . .A law firm that minimizes employee turnover by playing the statistical averages may offend our sensibilities and violate federal law--but it is not foolish.

When we approach this situation as an information problem, there are several crucial insights. First, firms are not the only villains. When professional women choose to have a child, take paid maternity leave, and then quit their companies, they impose a cost, arguably unfair, on their firms. More important, they impose a cost on other women. Firms that feel they have been "burned" by employees who take maternity leave and then quit are more likely to discriminate against young women in the hiring process (particularly those who are already pregnant) and less likely to offer generous maternity leave.

. . . .Statistical discrimination, or so-called "rational discrimination," takes place when an individual makes an inference that is defensible based on broad statistical patterns but (1) is likely to be wrong in the specific case at hand; and (2) has a discriminatory effect on some group.