Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

US Supreme Court to weigh end to race-based college admissions

bbc.com

Post reply on HN
Print view View original

281–286 of 286 posts

Re: US Supreme Court to weigh end to race-based college admissions

#281

Post by hot_gril »

godelski wrote: post #280
hot_gril wrote: post #278

Earlier quoted context omitted.

If the output seems to strongly favor some races over others, the expert debugs the inputs by temporarily swapping some out and rerunning. This is only a second line of defense beyond checking the inputs manually. Maybe the expert doesn't realize that some factor has a huge correlation with race without saying much about the candidate's merit. Beyond that, if there's a divergence between races, it is what it is. The…

> If the output seems to strongly favor some races over others, You'd typically do this with a cost function. > the expert debugs the inputs by temporarily swapping some out and rerunning. The best way to do this is to use that cost function to optimize your parameters. > Of course, could do this without race data just by proofreading. I thought I've adequately demonstrated why this is much easier said than done. > B…

Is your domain expertise in college admissions, data science, or a combination? You assuming this model should try and undo some effects of historical racism, but that isn't the engineers' job to decide. If that's the intended behavior, I understand that the difficulties involved are beyond my experience, and I would leave it to the data scientists. That's not what I think the goal of college admissions should be, and our government is making a decision on that now.

> The best way to do this is to use that cost function to optimize your parameters. > I thought I've adequately demonstrated why this is much easier said than done.

I understand how this works, but that's specifically what I don't want, for race (or other data that serves no purpose but to indicate the race, like name) to be part of the training data. Those inputs are manually selected at some level.

Re: US Supreme Court to weigh end to race-based college admissions

#282

Post by godelski »

hot_gril wrote: post #281
godelski wrote: post #280

Earlier quoted context omitted.

> If the output seems to strongly favor some races over others, You'd typically do this with a cost function. > the expert debugs the inputs by temporarily swapping some out and rerunning. The best way to do this is to use that cost function to optimize your parameters. > Of course, could do this without race data just by proofreading. I thought I've adequately demonstrated why this is much easier said than done. > B…

Is your domain expertise in college admissions, data science, or a combination? You assuming this model should try and undo some effects of historical racism, but that isn't the engineers' job to decide. If that's the intended behavior, I understand that the difficulties involved are beyond my experience, and I would leave it to the data scientists. That's not what I think the goal of college admissions should be, an…

My domain expertise is in data science. We're not talking about specifically school admissions but something bigger: handling race in data. That's decently general. Were I to actually create a model for predicting admissions I'd definitely be working with admissions to see those specific nuances. But you're attempting to pigeonhole the conversation.

> I understand how this works, but that's specifically what I don't want, for race to be part of the training data.

Tough. The problem, is that it is really difficult to move. That's what you aren't understanding. Making a model that doesn't explicitly incorporate race doesn't mean you've removed the race variable. It is just hidden now. There's plenty of examples of this happening in real life and they're fairly easy to Google. The historical data you train your model on doesn't suddenly lose it's correlation to race (there's major confounding variables) just because you didn't explicitly incorporate it. Dealing with hidden variables is hard and that's all you're doing.

Re: US Supreme Court to weigh end to race-based college admissions

#283

Post by frumper »

rahimnathwani wrote: post #265
whimsicalism wrote: post #250

Earlier quoted context omitted.

At least since 2007, that sort of explicit race based discrimination undertaken voluntarily is not legal in public school districts, see PICS v. Seattle.

What about this, which has been on SFUSD's web site for the past 12 months? Tutoring for African-American SFUSD Students Citywide Posted on September 23, 2021 The Citywide Tutorial Program is partnering with after school programs across San Francisco to offer homework help and tutoring for African-American students enrolled in SFUSD schools, kindergarten through 12th grade. SFUSD Access and Equity Department and SFUS…

This sounds like it violates prop 209 unless either the program predates 1996, or anyone can get tutoring despite it's specific program name.

Re: US Supreme Court to weigh end to race-based college admissions

#284

Post by snapplebobapple »

Aunche wrote: post #182
legitster wrote: post #34

> Institutions like Harvard say they consider an applicant's race as one of several factors - including economic status and religious belief - in order to build classes that accurately represent the racial and ethnic diversity of the country. I mean, this is all kind of window dressing for elite institutions to be elite, right? There are millions of people qualified to attend these university, Harvard gets to hand se…

Everything makes sense when you think of Harvard and similar universities as selfish private institutions that want to maximize their reputation. They want the children of world leaders, billionaires, and actual super-geniuses. They don't want to dilute their brand with thousands of extra students with mundane backgrounds that end up being moderately successful professionals who make mid-six-figures. Affirmative acti…

It's more than this. It's way better to think of Harvard et al as large hedge funds abusing the school tax loophole. They do want the children who will be world leaders, billionaires etc. but it is because those are the most likely to donate to the hedge fund. They don't want to grow admission because it might hurt hedge fund donations but also because educating people is secondary to their actual purpose and should only be engaged in to the point that their primary purpose is maximized (middle management doesn't make the kind of donations worth worrying about for a hedge fund, they want a higher minimum efficient scale).

Re: US Supreme Court to weigh end to race-based college admissions

#285

Post by AnonCoward42 »

filoleg wrote: post #262
AnonCoward42 wrote: post #258

Earlier quoted context omitted.

> Also, "socio-economic status" is a spectrum between poor and wealthy, including all of them. Just like "race" includes all races. We don't call it "non-white-and-asian-based affirmative action", so i dont see why you have an issue with me not calling it "poor-based affirmative action". It is a sliding scale, with all groups within it having different weights attached. So calling it "poor-based affirmative action" m…

Would that make my definition clearer and less arbitrary, if i said that by "socioeconomic status", in this specific scenario, I am referring to where you fall on the spectrum of "your family's income + savings"?

Yes, in that case you're actually calling for/proposing something specific. Socioeconomic is a placeholder term, where everyone can project something into it (and this is also regularly done).

Re: US Supreme Court to weigh end to race-based college admissions

#286

Post by filoleg »

AnonCoward42 wrote: post #285
filoleg wrote: post #262

Earlier quoted context omitted.

Would that make my definition clearer and less arbitrary, if i said that by "socioeconomic status", in this specific scenario, I am referring to where you fall on the spectrum of "your family's income + savings"?

Yes, in that case you're actually calling for/proposing something specific. Socioeconomic is a placeholder term, where everyone can project something into it (and this is also regularly done).

I appreciate your response. I genuinely was assuming it was well and clearly understood what I meant by "socio-economic status", but looks like I was wrong. In light of that, I totally understand your initial pushback to it. I hate vague and non-explicitly-defined terms, when it comes to policymaking, as well.
Post reply on HN