Affirmative action must be comprehensive because otherwise it disproportionately comes at the expense of those with hidden disadvantages

Let us start with a simple model. A university is considering admissions. It has about twice as many applications as spots. There are three populations of equal size.

1. Those who face no or insignificant unfair barriers to success.

2. Those who face recognized and unfair barriers to success, for whom affirmative action or other supportive provisions can be implemented.

3. Those who face unrecognized and unfair barriers to success, for whom affirmative action or other supportive provisions are not implemented.

Each person's “score” for admission to university consists in the sum of three random numbers evenly sampled between 0 and 1, except that the disadvantaged populations have 0.5 subtracted from their score. AA reverses this for the second population, but not the third.

Unsurprisingly, implementing AA for group 2 at a level that cancels their disadvantage will increase the total proportion of disadvantaged people attending the university, which is good. Also, unsurprisingly, implementing AA will reduce the proportion of people in category 1 who get in. Perhaps somewhat surprisingly, though for clear reasons, it will also reduce the proportion of Category 3 that get in.

Here’s the kicker though, a larger proportion of the unrecognized disadvantaged “lose their spots” in order to enable AA than those without disadvantages. That is to say, when more spots go to Category 2 because of AA, those spots disproportionately come from Category 3 rather than Category 1.

I ran the simulation many times. On average 28 members of the hidden disadvantage population got an offer before AA. After the implementation of AA, on average 18 got an offer. On the other hand, on average of 51 of those facing no unfair disadvantage received an offer, and after the implementation of AA, 42 received an offer. Proportionally the hidden disadvantaged were much more likely to lose out on an offer. This is because their disadvantages meant that those who had gotten over the threshold often did so marginally.

We can generalize this point. If:

1. There exist disadvantages and

2. There exists a mechanism to counterweigh these disadvantages

3. But not all disadvantages can be detected, or some are detected, but weighed less than they should be.

Then

4. Disproportionately, the people who lose their “Spot” to enable AA will be those with undetected disadvantages, or underweighted disadvantages.

Now AA still increases, overall the number of people with disadvantages who get in. Hence it is overall a good thing- at least in my view. However this is cold comfort for those with hidden disadvantages, who are first disadvantaged by their circumstances, and then not only unaided by AA but particularly likely to be knocked off by it. The solution is to commit seriously to the measurement of disadvantage and work urgently to ensure all forms of unfair disadvantage are counteracted. Of course, this can only be an ideal, but approximating this ideal is important for the reasons I’ve outlined.

Socioeconomic background, catastrophic life events, disability, and mental illness are, I think, often under-discussed in these conversations, at least partly because it is hard to measure them as forms of disadvantage and to verify their effects on a person. These difficulties are real, but that’s no excuse.

It is also, always, worth making the point that the fairest possible job market is a tight job market. When there are more good jobs, the proportion of disadvantaged people who can get a good job goes up. That’s obvious. What may be less obvious is that the proportion of good jobs going to disadvantaged people goes up. Once again, this is because disadvantaged people who are potentially in a position to qualify for X are particularly likely to be just on the borderline of qualifying. Purely as an illustration, I ran a simulation of increasing the number of “good” spots from 70 to 105 in a population in which half were disadvantaged. The proportion of “Good” spots going to the disadvantaged rose from 36% to 42%- about half way to equality. Of course, everything depends on the specification of the parameters, but this illustrates the point nonetheless.

I am quite poor, spend many hours a week on this blog, and make it available for free. Your paid subscription would be greatly appreciated.