Some interesting points about affirmative action

“Affirmative action” refers to a set of policies to ensure that certain disadvantaged groups get equal access to certain kinds of opportunities. It is typically used in the context of lowering standards, in the sense of giving opportunity to people from disadvantaged groups whose credentials in other respects may be lower than those of people from other groups.

The system of reservations for Scheduled Castes (SC) and Scheduled Tribes (ST) in government jobs and college admissions practiced in India is an example of affirmative action at its most explicit. A fixed fraction of the available seats is reserved for the group that is considered to be at a disadvantage. Competition within this fraction of the seats, as well as in the “general category”, is through the same procedure. Thus, if a college selects students through an entrance examination, the same examination is administered to all candidates. The admission cut-off for general category students and for SC/ST students, however, is determined separately. Thus, an SC/ST student may get selected even though he/she scored lower on the admission test than a general category student who did not get selected.

In principle, it could happen that the cut-off in the SC/ST category is higher, or more stringent, than the cut-off in the general category. In such a situation, a separate cut-off is not implemented, and a common cut-off point is determined. I am not aware of any instance in which this has happened, indicating that the disadvantaged/backward status is far from redundant.

Recently, I had a chance to read Tim Harford’s book The Logic of Life, where Harford devotes a chapter to affirmative action and related issues. In the chapter, Harford talks of the work of Roland Fryer, an economist now at Harvard University (Fryer is also mentioned in passing in Freakonomics, the famous book on economics by Steven D. Levitt and Stephen J. Dubner). Curious to learn more, I visited Fryer’s faculty page. A quick link to his papers led me to discover a wealth of insights on affirmative action. This paper by Fryer and Glenn Loury on the myths surrounding affirmative action was particularly enlightening. I’ll take the liberty of highlighting some of the points I found particularly interesting and giving my own take on them.

Over-resentment?

A reservation or quota for people in a certain disadvantaged group can mean that certain other capable people in the non-disadvantaged group are deprived of opportunities. This is particularly true, for instance, for cases where there are strict entry limits and that are highly competitive.

However, if entry into a place is highly competitive and the number of positions is limited, then allocating a small fraction of that to a certain disadvantaged group deprives at most that small fraction of people of opportunities. For instance, a university that has an intake of 5000 students, and sets aside a quota for 1000 students, deprives at most 1000 people of potential positions in the university. Thus, the size of the set of people who resent the quotas should be limited to 1000. In practice, the degree of resentment far exceeds the number of people deprived.

Why? Fryer offers the “parking” analogy. Imagine a parking lot where one slot is reserved for parking for cars with people needing wheelchairs. Suppose all other slots are full. Every non-handicapped driver who arrives at the parking lot finds all slots except the wheelchair slot empty, and curses the “reservation” system that prevents him/her from taking the empty slot. In fact, if the slot had not been reserved, only one driver would have been able to park there. In other words, each driver ends up over-resenting because he/she assumes that he/she “just missed it”, rather than being more realistic.

Similarly, students applying for admission and employees seeking jobs may tend to believe that they “just missed getting in” and hence may believe that they are in the narrow window of people who have been adversely affected by the quota.

Dumb quotas versus blind affirmative action

Suppose a university has an affirmative action target. The university needs to get at least 20% of its intake from a specified disadvantaged group. There are two options the university can use: first, an explicit quota that simply ranks all disadvantaged people and all other people according to the same criteria but chooses different cut-off levels to meet the quota. This is the admission test scenario I described earlier.

Second, the university can try to tweak its admission criteria in order to ensure that a larger fraction of disadvantaged people get through. In fact, before determining its precise admission criteria, it can examine and data-mine the applicant pool and then determine a formula for weighting different factors to achieve the quota as best as possible.

The former is what Fryer calls “color-sighted” affirmative action, while the latter is what Fryer calls “color-blind” affirmative action, because it does not explicitly acknowledge any quota (Fryer discusses color-blind affirmative action in a separate paper with Glenn Loury). Fryer’s key insight, which is again one of the things that should be obvious after a little thought, is that color-blind affirmative action is always worse than its color-sighted counterpart.

The reason is that in the color-sighted case, choices within each group are made optimally. Thus, among the disadvantaged group, the correct criteria are used, and ditto for the non-disadvantaged group. However, tweaking the criteria too much can result in making bad choices in both groups. When the disadvantaged group is not too disadvantaged compared to the other group, then the problem is not so severe, since only a little tweaking is necessary.

It gets worse. If criteria are tweaked too much away from the criteria that determine what a good student or employee should know and be, then this alters the incentive systems, both for the disadvantaged and non-disadvantaged. For instance, a criterion that relative ranking within one’s school should be given undue weight so that people coming from poor schools have a fighting chance, may lead students across the board to get obsessed with beating their school fellows on tests, which is arguably not a good thing. Similarly, a criterion that favors certain extra-curricular activities for no good reason other than that people in the disadvantaged group are more likely to do those activities, may lead a lot of people to waste their own time doing such activities to bolster their admission chances.

I think there is one situation in which tweaking criteria in the light of affirmative action might be good. This is where the spirit of the affirmative action policy really requires an application of new criteria that change the meaning of the value or relative worth of a student. For instance, affirmative action for a disadvantaged group that has been historically poor may result in a realization that factoring in the family’s poverty against the student’s performance may result in better quality decisions even for general applicants.

Good people do not discriminate against others, or discrimination is not something that we do, so we should not be subject to affirmative action

A classroom experiment by Fryer, along with Goeree and Holt, discovered something eerie. In the experiment, there were two kinds of workers — “green” and “purple”, and a bunch of employers. The employers were given information about whether a worker was green or purple, and the worker’s score on a test. The test score was in turn determined by how much “education” the worker got (workers had to “pay” for education) along with some random factor. The more education a worker chose to purchase, the higher the likelihood of a good test score. The employer’s goal was to try to hire the “best” workers in the sense of those who had the most education — but the only thing the employer saw was the color and score.

It so happened that on the first round, the purple workers ot somewhat lower test scores. This made employers more reticent to hire purple workers. Employers started hiring more green workers. Even purple workers with high test scores started getting by-passed because employers found the color a stronger indicator of ability than the test score, which was partly random. After a few rounds, purple workers stopped bothering to purchase an education. By the end, purple workers were shouting at the employers that they weren’t being hired, and employers were shouting at the purple workers that they weren’t getting educated enough.

This discrimination arose spontaneously, from an equal beginning, with just a bit of randomness tipping people off.

Discrimination usually hurts employers — but it is in their best interests to do so

Gary Becker, a professor at the University of Chicago and also a Nobel Laureate in Economics, pioneered the economic analysis of discrimination, back in the 1950s, when such subjects were considered outside the domain of economics. Becker made a simple observation, backed by statistical analysis and arguments: discrimination against a certain group of workers usually improves the bargaining power of the workers competing with them, but hurts potential employers, because they have a smaller labor pool to draw from. For instance, laws that forbid blacks from mining in South Africa were good for white miners but bad for mining companies that were forced to pay higher rates because competition in the labor market was reduced. This led Gary Becker to make the bold prediction that the more free and competitive the market, the more the pressure to discriminate less.

The study of discrimination has advanced a lot since Becker’s original work on the subject. A new understanding of discrimination has surfaced, whereby in the short run, employers benefit from discriminating. Tim Harford calls this “rational discrimination”. Others have called it “statistical discrimination”.

The idea behind rational discrimination is that employers, faced with limited information about employees, and limited resources to collect more information, are likely to use demographic and other statistical factors that correlate well with effective employees. This does not merely apply to employers. It also applies, for instance, to insurance providers. Health insurance is cheaper for non-smokers, auto insurance is cheaper for women, even though certain men may be very careful drivers and certain non-smokers may be very callous about their health in other ways.

This creates a vicious cycle, because once people in the disadvantaged group feel that their application will not be given fair consideration, they become less inclined to work to acquire the credentials needed. This is precisely what happened in the classroom experiment. The classroom experiment described above in fact demonstrates how a small initial accident got perpetuated into something collectively destructive, even while each player was acting fully rationally.

The way to get out? One thing that can be done is to have true information about employee capabilities more easy for employers to access and verify. In fact, related ideas have been proposed in many other related areas. For instance, some recent work has suggested that if employers are allowed to have access to the criminal record of potential employees, they may be more inclined to hire people from the disadvantaged groups who do not have a criminal background. The precise criminal record makes “racial profiling” more redundant.