[beowulfey]

Let me try again. You receive three resumes for a job application, for three humans of three different races. They have the exact same credentials. You immediately assume one of the three is the brightest of them, because your understanding of Affirmative Action Policies says this particular race has the highest likelihood of having higher overall test scores.

One of the other two could have had the highest possible score of all time, but you have written them off by making an assumption about them, based on race. You have made a judgement based on statistical inference, when you should have treated them all equally.

You may not intend it in any ill-meaning way, but it is important to realize that minor assumptions like this are pervasive, and they have far-reaching effects.

[darawk]

> You immediately assume one of the three is the brightest of them, because your understanding of Affirmative Action Policies says this particular race has the highest likelihood of having higher overall test scores.

I think it's important to distinguish between probabilities and possibilities. It is possible that any of them has the highest score. However, it is most likely that the Asian does.

Let me articulate this phenomenon in a more neutral example. Suppose you start an elite academy for the game Go. All of the best Go players in the world come from places like South Korea, China, etc, who have a long history of playing the game. However, you would like to increase the appeal of the game internationally, so you institute an affirmative action policy that says 50% of your students must come from non-asian countries.

Let's say you have 100 slots to fill each year, and you operationalize your affirmative action policy as follows: You take all the asian applicants, rank them by ability, and take the top 50. You take all the non-asian applicants, rank them by ability and take the top 50.

It should be obvious that, in this example, the average absolute ability level of the two groups will be quite different. The incoming Asian group would crush the non-Asian group in competition. This isn't due to any innate racial capacity gap, but due to the historical and cultural relationship to the game of Go.

Now, you educate each group together for say, 4 years. That education process may homogenize ability a little bit - helping the lower performers improve more than the higher performers (though the opposite may also be true), but it's probably not sufficient to close the rather large incoming skill gap.

Now, if you were watching a match, and the only things you knew about the two competitors were that they both attended your elite academy, and one was from South Korea, and the other was from California, who would you bet on to win?

It's entirely possible that the Californian is better! It's just less likely, given no additional information. Critically, this isn't an argument against the affirmative action policy. The AA policy is doing just what it should do - helping to close the skill gap. But it does means that statistical reasoning about racism has to be sensitive to this confounding variable if it wants to make truly accurate inferences.

[beowulfey]

This is a good example, and I appreciate you trying to explain it further. But I think we are a bit like two ships passing in the night here. As I interpret it, you are trying to explain the effect of affirmative action on the likelihood that someone from a particular background is more likely to be skilled or not. I totally understand that this is an effect of AA, and that neither of us are arguing about the merits of AA.

However, the point that I am trying to make is that we, as a society, should be trying to ignore these obvious statistical likelihoods when we are choosing a candidate. Those statistical likelihoods have nothing to do with the candidate themselves. If we make these kinds of interpretations, we are no longer judging a candidate based on who they are, but rather who we think they might be. And who am I to make that judgement? I'm nobody special. That's all I'm trying to say, really.

EDIT Someone else in the thread brought up the idea of why there is AA for school, but not for the workplace as in my argument. It's kind of a different topic, but I think it's a good counterargument about the complexity of this. I don't really have a good answer, to be honest, but it will be on my mind for awhile now.

[darawk]

> However, the point that I am trying to make is that we, as a society, should be trying to ignore these obvious statistical likelihoods when we are choosing a candidate. Those statistical likelihoods have nothing to do with the candidate themselves. If we make these kinds of interpretations, we are no longer judging a candidate based on who they are, but rather who we think they might be. And who am I to make that judgement? I'm nobody special. That's all I'm trying to say, really.

Ah, ok I see. I didn't understand your point then. I think we at least kind of agree on that point. What I was trying to say is that, I don't think that it's accurate to characterize the resume study as proving racism or racial discrimination, given the bias induced by AA. At least, providing they are not going further than correcting for that bias.

I do agree with you that in an ideal world, people would try to avoid factoring that in. But, it is important to keep in mind I think that hiring decisions are often extremely consequential for the people that make them (in a way that university admissions are not), and as a consequence, asking the decision makers there to intentionally ignore pertinent information is almost always going to be a losing proposition.

I think, even if people are correcting a bit for this bias in the hiring pipeline, AA is still providing considerable value to historically disadvantaged candidates, by helping them get access to alumni networks, and presumably a higher quality education and hopefully that will be sufficient to close the remaining skill gaps over time.

[barry-cotter]

> However, the point that I am trying to make is that we, as a society, should be trying to ignore these obvious statistical likelihoods when we are choosing a candidate.

The truth is one. If you lie to other people and demand they lie to you it affects your entire model of the world. If there are facts about the world that you would prefer not to acknowledge they are linked to other facts. Lying consistently requires enormous effort.

[meowtimemania]

I think what GP said was logical. Imagine Harvard has 3 entrance criteria, and the criteria a student receives depends on the first letter of their first name:

* A name: must be in top 1% of test scores

* B name: must be in top 5% of test scores

* C name: must be in top 10% of test scores

The following 3 students are admitted:

* Allison (is in top 1%)

* Brian (is in top 4%)

* Caitlin (is in top 1%)

We can only safely assume that Allison is in the top 1% because her criteria certifies it. Even though Caitlin in actuality is in the top 1%, because her entrance criteria is more lax, we are not sure.

I think this is one downside of affirmative action, people are unsure if a person passes based on affirmative action or purely on merit. Now we consider the upsides and downsides of affirmative action, and decide whether it should be implemented.

[beowulfey]

Yes, I understand that affirmative action can have this consequence. I am not arguing for or against affirmative action. I am pointing out that it should not matter whether someone has an A, B, or C name when applications are being triaged. Because Allison, Brian and Caitlin all have the same credentials, they should be viewed as equally likely candidates.

Making assumptions about them based on probabilities is exactly the problem here, and it is one that we can easily avoid.

[mpweiher]

> same credentials ... making assumptions about them based on probabilities is exactly the problem here

Using the credentials is making assumptions about them based on probabilities.

[Judgmentality]

> you should have treated them all equally

This is obviously the golden standard we are trying to achieve, but how do we get there? It's theoretically impossible to treat everyone equally and apply affirmative action at the same time. I understand there is a difference between equality and equity, but I'm replying to the words you wrote.

Affirmative action may be the best solution we currently have to deal with systemic racism, but ultimately it's trying to fix prejudice with prejudice - and that is not a perfect solution. It also creates a lot of confusion because sometimes we say to treat people equally (as you say when trying to decide between hiring candidates), and other times we say we should help out the disenfranchised (such as when admitting students to schools). So where do we draw the line for when we want equality versus equity?

[beowulfey]

It is actually pretty simple (in this example anyway). If three candidates come to you with the same credentials, then do not assume one of them is the best candidate based on your interpretation of their background. You have to treat them all as equally likely candidates - interview all three. It is more work for you, but the effort is worth it, because it helps prevent the effect of possible biases.

[Judgmentality]

Sure, I understand in that example what to do. But as a society, where do we draw the line? Why is it okay to apply affirmative action for selecting students but not okay when accepting employees (continuing the example from this thread)? Since we're trying to fix a systemic issue, we need a consistent response across society for it to be most effective.

My point is this is a complicated problem with no perfect solution, and people will correctly point out flaws with it both theoretically and (more relevant for this discussion) how we implement it.

Anyway, I think we mostly agree. Cheers.

[beowulfey]

This is a really good point, and I appreciate you bringing it up. What I am saying directly conflicts with affirmative action itself, so in effect I am arguing against it. I don't really have a good answer to that. Thanks for pointing it out, I guess I'll ponder that for awhile.