Re:‘Culture wars’ are fought by tiny minority – UK study
by scythe on Monday October 26, 2020 @10:05AM
(#24896265)
>Or are you suggesting that unintentionally disadvantaging a group is not only not discriminating against that group, but that also it is is also excusable and needs no correction? i.e. that the consistent, albeit unintentional, disadvantaging shall continue?
The problem arises when the only opportunity is to replace measurable discrimination with non-measurable discrimination. For example, SAT results are easy to measure, and because of this we take correlations with X, Y, and Z, and find that SAT slightly discriminates against people in subpopulation Z.
So we replace the SAT with a holistic interview-and-quiz format that is only used at our institution. The data is kept internal (for student privacy) and there aren't enough datapoints to derive meaningful correlations with X, Y, or Z.
Is there less discrimination? We don't know! What you accomplish is to replace a standard within which you can detect discrimination with one where you can't.
So yes, using metrics that have small but measurable inequities may be preferable to using metrics where inequities cannot be measured. In the metrizable case the "disadvantage" is at least bounded.
Now, in the alternative case where you actually have options, i.e. you have one measure with some inequities and another measure that you know has fewer inequities, then of course it would be racist to choose the first measure over the second. But this is not an analogous situation to university entrance exams at all.
The problem arises when the only opportunity is to replace measurable discrimination with non-measurable discrimination. For example, SAT results are easy to measure, and because of this we take correlations with X, Y, and Z, and find that SAT slightly discriminates against people in subpopulation Z.
So we replace the SAT with a holistic interview-and-quiz format that is only used at our institution. The data is kept internal (for student privacy) and there aren't enough datapoints to derive meaningful correlations with X, Y, or Z.
Is there less discrimination? We don't know! What you accomplish is to replace a standard within which you can detect discrimination with one where you can't.
So yes, using metrics that have small but measurable inequities may be preferable to using metrics where inequities cannot be measured. In the metrizable case the "disadvantage" is at least bounded.
Now, in the alternative case where you actually have options, i.e. you have one measure with some inequities and another measure that you know has fewer inequities, then of course it would be racist to choose the first measure over the second. But this is not an analogous situation to university entrance exams at all.