Rational Discrimination

Discrimination is a puzzle for economists. The very definition of discrimination is that it separates groups unjustly. In the labor market this might occur when hiring males rather than females, or whites rather than blacks, although there is no difference in productivity between those groups. I will present an example of a firm discriminating and why, in theory, this firm cannot survive in the long run. Next, I will try to resolve the puzzle by arguing that some discrimination is rational, so-called statistical discrimination. I will also argue that statistical discrimination might also explain discrimination in social settings. Lastly, I will tell you how I use discrimination to maximize my gains in the love life.

Let’s assume there is a digital marketing agency (A) that prefers to hire men over women. They chose to ignore female talent. They thus have to hire expensive males (more or less talented). Then comes digital marketing agency (B). They do not have the discriminatory hiring policy of (A). Agency (B) hires wherever they see talent, regardless of gender. They now have the opportunity to hire cheaper females with same productivity. By hiring cheap female talent, they have lower cost than agency (A) and can thus undercut the price service. Firm (B) will get all the revenue since firm (A) failed to hire cheap talent in the form of female labor. How can firm (A) that failed to employ otherwise identical but lower-paid female workers survive (Hamermesh, Beauty Pays)? It can’t! If there is discrimination in the market, firms can take advantage of it, hire cheap discriminated labor, thus increase the demand for discriminated labor, thus raise the price to non-discriminated labor price. Thus, discrimination should logically lead to non-discrimination. As long as profit maximizing firms can take advantage of cheap labor there will not be any discrimination. So where does discrimination come from?

To answer this question we have to loosen some assumptions. The example above rests on the notion of perfect information. It assumes that profit maximizing firms know the productivity of an individual. This is obviously not true. They get a CV, maybe a 30 minute interview, that’s it. From this they have to decide whether to hire someone. This is a huge investment with limited information. It might now be worth for the firm to incorporate group averages of attributes. The average productivity of gender, age, race (and many others) can indicate whether the individual is more likely to productive than an applicant with an identical CV, but different age, gender, and race. The classic example is hiring a just-married female worker versus an otherwise identical just-married male worker. Both say they are dedicated to work in the firm tirelessly for the next 5 years. On an individual basis this might be completely true. In some individual cases the female worker is going to work more tirelessly than the male worker. But, on average, it is statistically very likely that the female will get some kids within the next years, thus having to leave the firm for some time at least. This means that it is rational to discriminate against females in their late 20s and 30s when just married, in the labor market. As a profit maximizing firm, facing imperfect information, going with probabilities is the rational thing to do. This is the essence of statistical discrimination.

The individual female who just married in her late 20s might truthfully be the best worker, with no child-wish, but there is no way for an employer to know that for sure. This means she will (and from a firm’s prospective should be) discriminated against. The likelihood of her taking a year off, times the cost of that will be the difference of her pay versus a male pay. It is the lack of information, that leads to statistical discrimination.

Statistical discrimination has been well documented in the literature. It turns out that the more information you give an employer the less likely you are to get statistically discriminated. There are great randomized control trials with CVs supporting this notion. When sending a one page CV with Turkish names to German companies, you are more likely to get rejected than sending a 4 page CV (these results are compared to otherwise identical German name CVs).

I think the way forward against “discrimination” might lie less in blaming the “perpetrators” but rather doing two things: (1) increase information on individuals (create a better information environment), (2) work towards creating equal averages between groups (subsidize paternity leave in the case above). These two things can decrease statistical discrimination. If after that attempt discrimination still persists, we have to go back and look at irrational discrimination. In my view it is hard to work against irrational discrimination if we cannot even identify rational discrimination. It is counter productive to call people bigots, when in reality they act rationally. Thus, I would propose: lets fix rational discrimination first, see where there is still irrational discrimination, then we can do some bigot shaming.

Lastly, this model of statistical discrimination can be used, not only in the labor market, but also in social settings. Police brutality, homogeneous group and friendship formations, sexual partnership preferences, all these might be the result of statistical discrimination.

An example of statistical discrimination in a social setting is my dating life. I try to bring economics into every part of my life, including my love life. I see dating as equivalent to a hiring process and the potential job equivalent to a relationship. Thus, I face the same problem as the profit maximizing firm. I want to maximize my relationship outcome constraint by the time and effort I must put into dating. Going on dates and trying to pick up girls in clubs and bars is tiring. I have to chose wisely who to go on a date with. I have very limited information on who might fit me. So, what I look for are attributes that might exclude candidates, merely based on the average behavior of the people with those attribute. My classic example are girls with a high degree of body modifications. These include plastic surgery, tattoos, piercings, and hair colour changes. The literature is pretty clear that tattoos and piercings correlate with a higher degree of risk taking behavior and instability. The verdict isn’t in on how these modifications are related to mental health issues, but there is some evidence leaning towards a positive correlation. Since I am not necessarily a very risk-loving person myself, and, everything else being held equal, I prefer a partner with no mental health issues , I can exclude candidates. Although I think tattoos are very attractive, I decided to statistically discriminate against women with a lot of tattoos. On average it seems that they don’t fit to me and thus I can save time and effort trying to date them. If the love of my life is the lizard women, the one with the full body tattoo and the split tongue, then my discrimination backfired. But with limited information and limited dating resources, I think statistical discrimination is the optimal dating strategy for me.