Request to econ and math people: solve the Delhi University admission problem | Hacker News Reader

Request to econ and math people: solve the Delhi University admission problem

jyotirmoy.net

59 pointsyomritoyj47 comments

If you're interested in matchmaking and markets, check out Al Roth's work. He's been involved in designing matching systems for schools, jobs, organ transplants and other areas.

Homepage: http://web.stanford.edu/~alroth/

Blog: http://marketdesigner.blogspot.com

And here's a review of his recent book: http://www.wsj.com/articles/matchmaker-make-me-a-market-1434.... I'm reading it at the moment and it's fascinating.

In 1999 I totally flunked my UCAS admissions process in the UK precisely due to one of the scenarios described herein (aiming slightly too high, getting zero offers, and then being stuck in the clearing totally random and haphazard clearing process with various universities competing for an unexpectedly high-grade candidate). This badly needs addressing.

It may be comforting to know that I did exactly the same thing a few years before that and ended up at a shitty university and quit after 6 months. Best thing I ever did.

This is a solved problem in the US, but we traded it for a different problem. We don't do guaranteed admissions using a score cutoff, except in some cases at the state level where a specific high school achievement level guarantees admission to the state university system... or a certain achievement level in a state's community college system guarantees transfer admission into the state university system.

The upside is that it puts the power to choose in the hands of the student (for better or worse -- many students & their families have no idea what they're doing. See the post from MichaelCrawford about his application experience.) and mostly eliminates the quota/population problem for schools. Most universities offer early admission here, which helps them fill a set number of spots with what they see as top students. One condition of applying early admission at a lot of schools is that you commit to attending their if they let you in. Of course, schools still offer admission to many more students than they know will attend, and I have no idea whether they do this algorithmically or using tribal knowledge gleaned from previous experiences, but I have very rarely heard stories of a school having serious difficulties because they let in too many kids. There have been a few, but the worst that happened was they had to put up some excess in off-campus apartments or hotels until attrition shrank the class to the point where campus housing became available again.

There are other problems with the US method of college admissions, but I prefer our more holistic process to something as rigid as described in this post.

In India, a much smaller proportion of students change their horses mid-stream... so, if there is excess, it will be a problem throughout the 4 years or so.

Just one year, you would have a smaller class the next year.

If the process were bad enough that it wasn't producing good results (targeted enrollment numbers, more donations, larger endowments), universities would have an incentive to deviate from the cutoff system.

As you describe, the system works the way that it does now. It's not a perfect system, and people who shoot to high or too low get screwed. That's life - perfect info, perfect matching, etc. don't exist, especially for a relationship as complex as the student-college one. There's no clear-cut signal for which students will be "the best" for colleges and vice versa, which colleges will benefit the students the most.

We can only try to reduce the number of students who get screwed over by adding more people to admissions departments, college counselors, reducing application fees, increasing awareness etc.

Adding more info to the student/college set is expensive. On the college side: Top college endowments are flourishing and there is no incentive to innovate at the most prestigious schools. On the student side: parents who can already pay massive amounts to game the system with private school, private tutors, admissions experts etc. It is the job of government to reduce the difference between wealthy and poor students. So an increase in spending on public school education is necessary, especially with regards to college prep.

The system is built around the information that schools collect from students: grades and extracurriculars. Unless other types of data are collected and used as signals during the admissions process (having some kind of score for EQ and social ability, for example), then nothing will change.

This is a process that happens millions of times per year, hundreds of millions of times if we see hiring as a similar problem. There is no silver bullet.

Should social ability determine university admission?

Let me ask you a question: Should SAT scores determine university admission?

There is no one clear signal for future "success", but it would be hard to argue that social ability does not contribute to success. Half the reason people go to college is to pick up social skills and fraternize/network with other young, talented individuals. If it were possible to quantify, it definitely wouldn't hurt to consider social skills data alongside academic ability.

Half the reason people go to college is to pick up social skills and fraternize/network with other young, talented individuals.

So then perhaps it should be targeted for low EQs so that they can improve?

If they can improve.

How can you tell whether a student with bad grades is smart and underachieving (possibly coachable/teachable) or just not that smart/not able to grasp concepts? Colleges will not spend the resources and take the chance. There are plenty of applicants with good grades: admitting these applicants is safe.

It would be the same argument for EQ.

The problem with the Gale-Shapley algorithm for stable matching is that it's inherently unfair. The askers/proposers slowly get downgraded to their lowest possible match, while the other side slowly moves up to their highest possible match. Not sure how different the outcome would be from the current system.

Which actually brings me to another point: The GS algorithm is roughly that members of one party(in this case the schools) work down their list of preferences, asking members of the other party(the students) to partner with them. The members of the other party always say yes to their first match, and will always leave their current match when someone higher on their list asks them to be partners. Sounds pretty similar to the system currently in place in India.

A final note on the GS algorithm: If an anecdote told by a professor of mine is correct, this has been tried before. It was how hospitals chose their residents after med school, but it was set up so the hospitals always got the best match. I believe the students sued them.

edit:formating

In 1995, the system was updated to make it applicant-friendly rather than hospital-friendly

https://en.wikipedia.org/wiki/National_Resident_Matching_Pro...

Thanks for the link. I wasn't sure how true the story was.

Drop the cutoff system completely or partially. The top scoring students aren't necessarily all the students you want.Quite often the gems are hidden throughout the spectrum. Selective breeding for people good at grades and tests isn't going to get you the people with the most potential for the future.

Agreed, but I think your suggested course of action is beyond the author's pay grade.

And, where's the fun in it? :)

How do you reliably and objectively identify the gems?

Note that an American-style "holistic" system is unnecessary since India explicitly allows racism into the process (so no need for convoluted schemes to hide it).

https://en.wikipedia.org/wiki/Reservation_in_India#Governmen...

As much as I hate the Reservation system in Indian colleges, I wouldn't call that 'racism'. They are two different things. They are trying to _mitigate_ the effects of a possible racism by doing this. Just that they don't understand they have to give that reservation advantage to them really early in life - not while admitting students for Doctor and Engineer degrees.

The fact that one might allude to cancelling one bit of racism with another doesn't make it not racism. (I'm defining "racism" as "discrimination based on race", not "any cause for a statistical disparity" as some other people do.)

I must say that I do prefer the Indian system to the US system, however. It's at least explicit and can be debated on merits. The purveyors of the US system are secretive so as to prevent opponents from gathering facts to argue against it.

I think you (and others) have espoused this viewpoint in other threads.

I'll just say that having an outsider (perhaps a politician) re-appropriate words that have a clear, technical definition (say surveillance or security) must be frustrating for experts who understand those words.

Redefining a well-understood technical word is a good rhetorical technique but side steps whatever logic underlies the word's original defintion.

Give everyone between 16 and 20 basic income, access to MOOCs (supported by advisers, TA's, etc.), dorm-like housing and let anybody try to do anything.

Select those who've made something of themselves for 2 years of university undergraduate finishing.

You've cooked up a scheme that's vastly more expensive than the current one (let me remind you that India is poor) and you still haven't even explicitly defined "made something of themselves".

Let me also remind you that unlike the US, India is a country with real problems - with the money you spend on this crazy vague subjective university admission scheme, you could instead do something useful like building more toilets.

Just because there are "real problems" doesn't mean other problems are insignificant and worthless. Because India is a poor country, cost of living is cheaper so less money would be needed for basic income, which somewhat reduces the effect of less tax income.

The cost of constructing toilets is also cheaper. The opportunity cost (measured in # of toilets built, or rapes prevented by police, or children who don't starve) remains the same.

>unlike the US, India is a country with real problems

Every country has "real problems" unless your definition of a real problem is the winner of a who-suffers-the-most pissing contest. (In which case perhaps the Congo is the only country with real problems.)

This argument is just like the one used against Mars program. In competitions, asymmetrical developments are often better than symmetrical ones.

We can build more toilets (aren't we already?) or we can educate next generation to aspire to do something great.

We are already educating the next generation. The question is whether we should adopt some incredibly expensive scheme to implement a vague and subjective university enrollment method when we already have a very effective one, or whether we should spend the money on toilets for the bottom half instead.

Every ~2500rs spent on colechristensen's basic income is one less toilet. That's the core calculation.

And yes, I do think the money spent on the Mars orbiter was just an exercise in national pride and would have been better spent on toilets, protecting women's dignity, literacy/arithmetic/english and similar low hanging fruit.

>money spent on the Mars orbiter was just an exercise in national pride

When start-ups compete against big corporations, do they try to follow corporate crap and other slow things? The advantage of being second is that you can learn from those ahead of you.

Although I agree that at least basic sanitation is more important than this, but this is not the case in general. In general, we need to invest in next generation technologies and advancements. Only then can we create income sources which dwarf all coming from the advancements of rest of the world.

I studied astronomybat caltech.

The valedictorian of the other high school didnt get an interview.

The valedictorian of my high school didnt apply.

I totally blew off high school to do my own thing.

However I was crushed that I was not accepted to MIT.

I didnt apply to stanford. I expect I would have gotten in but didnt think we could afford it. In reality stanford is so expensive so people like me can get great financial aid.

Caltech was quite cheap for a private school no doubt because it is small but with a hufe endowment.

astronomybat == astronomy at?

This seems to be very similar to the dating problem. Instead of asking students to give preference on college/course combination, maybe they should be asked about what matters most to them. Are they looking for a college close to their home, are they looking for a course that has many career options or are they looking for something they are passionate about, etc., Colleges could go through a similar set of questions and then you can build a machine learning algorithm that uses this information to match a college and student. It may not be straightforward but a few iterations of questions (or features in ML terms) can build a pretty robust model.

In Gujarat,India , we do have mock round before allocating actual admission to an undergraduate student. So student can have rough idea about where s/he can have proper idea about where he may get admission. After mock round there is seat allocation round; this is final round where all students get seat allocation. Now if a student is unhappy with the admission then s/he can opt for the next round and now the next round will be only for the vacant seat which are available from whoever opted for this round. This process is usually done for another two rounds after main seat allocation round. This all things are done through online account.

I'm not familiar with Gale-Shapley -- but maybe exploiting the new on-line nature of the application process would be good.

Have students rank, say 5 subjects and 5 colleges -- do a cut-off, and allow those cut to change or keep their choices (along with listing open positions after the current round).

Run the cut-off twice a day, for 20 days. If things don't balance out after 40 rounds, I suppose an entirely different system is needed.

I'm sure this could be modelled first, to check the viability.

But then you are asking students to collectively act out an algorithm. Why not leave it to a machine?

We could assume that each student has as close to perfect knowledge about his/her preferences as is attainable. The student will have to work to communicate that to the system one way or another -- I'm suggesting that using several rounds might mean less work (and/or less unpleasant work) for the students, than if they have to fill out an impossible number of choices/preferences upfront -- just so the system have the data it needs to complete in one round. Which would be optimizing the wrong part of the problem (or even the wrong problem).

No matter how interesting the problem might be from an algorithmic point of view.

Because the machine can't deduce each student's preferences.

I suspect the problem is essentially unsolvable, without requiring each student to fully write out the order of his/her school preferences. Otherwise there will inevitably (if the solution to the Stable Marriage problem is optimal in this regard) be undecidable cases that would have to be dealt with in a second round.

I believe that financial incentives can be added to the mix.

Colleges that are looking for a certain profile of students can provide a discount x% of their fees.

This is shown to the student who can similar rank the colleges for this.

On the other side, a student should not be allowed to pay his way in, so might be problematic?

Ask students to submit their preferences as a piece of code returning a score.

The problem is now outsourced to third parties who will build tools to generate scoring functions intuitively, for example based on constraints, or based on matching you with the preferences of similar students etc.

trade matching algos might work here.

and just one question: who the hell are the "econ" people?

Economists.

To know more about them, read Leijonhufvud's "Life Among the Econ" http://www.econ.ucla.edu/alleras/teaching/life_among_the_eco...

btw~, which platform do you use for your blog.

Hakyll, which is a static site generator written in Haskell http://jaspervdj.be/hakyll/

"Economists people" -- is that still English?

how are you so sure about plural too, huh?

Econ -> Econs , no?

Are you just here for a fight?

"Econ people" is short for "Economics people". Another word for "Economics people" would be "Economists", but it might also include people who have an interest in economics but don't consider themselves economists.

And as for the plural, An "econ person" would be an economist. The OP specified "Econ people".

Calm yourself down a bit.

I'm a bit confused about the point you are trying to make.

Based on your comment history, you seem comfortable with taking liberties with the English language to get your point across quickly (e.g. omitting implicit pronouns and articles in sentences such as 'Somehow never was true with Iceland.', or 'They should quit Euro (Merkel will never allow because others would follow suit) and debase currency. Do exactly as Iceland did.')

So I'm not sure why you seem to take particular issue with 'econ', which is a very common casual way to refer, e.g., to the economics major at uni. ('econ 101' etc)

Reply on news.ycombinator.com