https://nypost.com/2018/10/17/harvards-gatekeeper-reveals-sa...
Let's say we lived in a world where racial spoils were being used as a way to distribute contested goods and services. How could I differentiate between this world and the world of affirmative action? In both places the defenders claim to me all the candidates were otherwise identical in aggregate and race was only used as a tie breaker in order to better represent demographics. I put forward some objective measures that show candidates did not appear to be equal in aggregate and they claim these are not a complete measure and once you include a more holistic evaluation then the non-AA candidates were either equal or less good.
That doesn't change the original definition of AA. Neither does a demonstration that any specific institution does things differently.
If you want to change your claim to "the first sentence precisely matches what Harvard does for undergrad admissions", I'm not going to disagree with you. But that's a very different claim than the one you made.
I think a lot of 'AA' in practice is similar to what is done with Harvard admissions either by deliberate design or poor incentives. For example say a business notices an incorrect distribution of employees and then decides to introduce the steelman AA policy in order to try and fix this disparity. Businesses prefer to measure outcomes over process so you end up having either implicit or explicit racial proportions as goals instead of having the 'steelman AA' process being the goal. This ends up leading to explicit racial preferences and the strawman AA as the implementation.
Correct, and this might be meaningful if there wasn't racial bias encoded in the SAT.
Just as a guidinf question here, what does the SAT measure?
Maybe. But I think that world would be very different than ours today.
> What is the racial bias encoded in the SAT?
I can't find a source for this now, but I recall reading that if you give versions of reading tests to black students that use AAVE instead of common English, the racial gap in reading performance mostly disappears.
In other words, something like the SAT measures, at best a sort of confirmative intelligence and not an objective one (if that's even a thing. I'd suggest that intelligence is always relative to society). So institutions looking to influence society can't simply use a test that measures in part how well one conforms.
0: https://en.wikipedia.org/wiki/Race_and_intelligence#United_S...
1: https://en.wikipedia.org/wiki/Raven%27s_Progressive_Matrices
Similarly, western women have better reasoning about color than western men (I can't remember exactly where I found this, but it was probably an article about the World Color Survey controversy), and western cultured-people score better in pattern recognition tests compared to people from cultures that have fewer color words.
Such motifs are similar: people from cultures whose languages don't express numbers (1, 2, many) or express time[2] differently have more or less trouble reasoning in those contexts. Even formal logic can be culturally tied[3] (and that article also mentions certain forms of spacial reasoning being culturally tied too), which is weird (you can think of this as an extension to the Sapir-Whorf hypothesis[4]: language and culture influence how we think, and thus trying to judge intelligence across-culture will run into...issues).
tl;dr: Fairly judging "general intelligence" in any way is difficult. We continually find that we're really bad at it. Something like IQ is useful in certain contexts: an individual with a much higher IQ is probably going to perform better in US life than someone with a comparatively lower IQ, but when you try to make strong conclusions about intelligence, especially about group-levels of intelligence, from something like IQ scores, you're veering off into pseudoscience.
[1]: (pdf) https://s3.amazonaws.com/academia.edu.documents/45855297/Sex...
[2]: https://www.ncbi.nlm.nih.gov/pubmed/21564254
[3]: https://www.nytimes.com/2000/08/08/science/how-culture-molds...
And at least to me... Spatial reasoning, color comparison and differentiation, and formal logic would all be parts of general intelligence. Penalizing someone for being innumerate might not capture a person's theoretical intelligence if raised in a different culture, but that doesn't make it inaccurate as a test of how they are now.
I'll be fully transparent, this does make the question of "well how can we judge people without insisting on equality of outcome" rather tricky. I resolve this personally by just concluding that the precise judgement of people such that we'd need a good test like that is rarely actually important in day to day life. In other words, I'm okay with wider error bars for the reduction of implicit bias.
Isn't this just "I'm OK with error bars in the direction I favor", which happens to be equity? Unless you're saying you're OK with accepting the tests we have, with error bars that might err too far on the side of "other racial/cultural groups are less intelligent".
Not exactly: Precisely, I'd prefer error bars that are equally sized for everyone, instead of error bars that differ based on cultural factors, especially if we aren't capable of acknowledging the cultural disparity in distribution of those error bars (and IME, people generally aren't).
In other words, if you have a test that has an error of X when comparing 2 white kids, and an error of X when comparing two black kids, but an error of 3X when comparing a white kid to a black kid, you don't have a test with a precision of X, you have a test with an error of 3X (or worse: you have a test with a precision of X, but that consistently biases in favor of one group).
If on the other hand we have a test that has errors of 2X, but has that across cultural boundaries, that's a better test and we should use it instead. This is because the errors are randomly distributed across the population (so anyone has the same chance of being positively or negatively impacted by the test's imprecision) instead of culturally tied.
If we're willing to acknowledge the different distributions of error, then maybe we can make do, but consistent problems with things like affirmative action suggest that, at a macro level, we aren't.
Of course, as with many things human, this original goal/definition isn't always the one that plays out in practice (somewhat like a number of articles of the US constitution, or adherence to the moral screed of certain theistic religions).
Affirmative action was (and largely is) defined as: given multiple candidates for a position who are equal when considered in aggregate of all matters that exclude race, and therefore an arbitrary choice is required in order to select one of them, choose one whose presence in the relevant population will bring it closer to broader demographics.
It means that when you have a black male and a white woman who are equally qualified for the job, you will inevitably have to use some criteria other than their qualifications in order to pick one of them. AA implies considering the impact on racial demographics as one of the criteria to use given that qualifications cannot be used (because they are equivalent).